Machine Learning (ML) and Artificial Intelligence (AI) have progressed rapidly in recent years.

Techniques of ML and AI have played important role in medical field like medical image processing,
computer-aided diagnosis, image interpretation, image fusion, image registration, image segmentation, ˙
image-guided therapy, image retrieval and analysis Techniques of ML extract information from the
images and represents information effectively and efficiently. The ML and AI facilitate and assist doctors
that they can diagnose and predict accurate and faster the risk of diseases and prevent them in time.
These techniques enhance the abilities of doctors and researchers to understand that how to analyze
the generic variations which will lead to disease. These techniques composed of conventional algorithms
without learning like Support Vector Machine (SVM), Neural Network (NN), KNN etc. and deep learning
algorithms such as Convolutional Neural Network (CNN), Recur-rent neural Network (RNN), Long Short
term Memory (LSTM), Extreme Learning Model (ELM), Generative Adversarial Networks (GANs) etc. For-
mer algorithms are limited in processing the natural images in their raw form, time consuming, based on
expert knowledge and requires a lot time for tuning the features. The later algorithms are fed with raw
data, automatic features learner and fast. These algorithms try to learn multiple levels of abstraction,
representation and information automatically from large set of images that exhibit the desired behavior
of data. Although automated detec-tion of diseases based on conventional methods in medical imaging
has been shown significant accuracies around for decades, but new advances in machine learning
techniques have ignited a boom in the deep learning. Deep learning based algorithms showed promising
performance as well speed in different domains like speech recognition, text recognition, lips reading,
computer-aided diagnosis, face recognition, drug discovery.

Deep Learning Over Machine Learning

Accurate diagnoses of disease depends upon image acquisition and image interpretation. Image
acquisition devices has improved substantially over the recent few years i.e. currently we are getting
radiological images ((X-Ray, CT and MRI scans etc.) with much higher resolution. However, we just
started to get benefits for automated image interpretation. One of the best machine learning
application is computer vision, though traditional machine learning algorithms for image interpretation
rely heavily on expert crafted features i.e. lungs tumor detection requires structure features to be
extracted. Due to the extensive variation from patient to patient data, traditional learning methods are
not reliable. Machine learning has evolved over the last few years by its ability to shift through complex
and big data. Now deep learning has got great interest in each and every field and especially in medical
image analysis and it is expected that it will hold $300 million medical imaging market by 2021. Thus, by
2021, it alone will get more more investment for medical imaging than the entire analysis industry spent
in 2016. It is the most effective and supervised machine learning approach. This approach use models of
deep neural network which is variation of Neural Network but with large approximation to human brain
using advance mechanism as compare to simple neural network. The term deep learning implies the use
of a deep neural network model. The basic computational unit in a neural network is the neuron, a
concept inspired by the study of the human brain, which takes multiple signals as inputs, combines them
linearly using weights, and then passes the combined signals through nonlinear operations to generate
output signals.
2.1 Neural Network and Deep Learning Architecture

Artificial neural networks structurally and conceptually inspired by human biological nervous system .
Preceptron is one of the earliest neural network that was based on human brain system. It consist of
input layer that is di-rectly connect to output layer and was good to classify linearly separable patterns.
To solve more complex pattern, neural network was introduced that has a layered architecture i.e.,
input layer, output layer and one or more hidden layers. Neural network consist of interconnected
neurons that takes input and perform some processing on the input data, and finally forward the
current layer output to the coming layer. The general architecture of neural network is shown in figure
2. Each neuron in the network sums up the input data and apply the activation function to the summed
data and finally pro-vides the output that might be propagated to the next layer. Thus adding more
hidden layer allows to deal with complex as hidden layer capture nonlinear relationship. These neural
networks are knows as Deep Neural network. Deep learning provides new cost effective to train DNN
were slow in learning the weights. Extra layers in DNN enable composition of features from lower layers
to the upper layer by giving the potential of modeling complex data. Deep learning is the growing trend
to develop automated applications and has been termed in 10 breakthrough technologies of 2013.
Today, several deep learning based computer vision applications are performing even better than
human i.e. identifying indicators for cancer in blood and tumors in MRI scans. It is improvement of
artificial neural network that consist of more hidden layer that permits higher level of abstraction and
improved image analysis. It becomes extensively applied method due to its recent unparal-lelled result
for several applications i.e. object detection, speech recognition, face recognition and medical imaging.

A deep neural network hierarchically stacks multiple layers of neurons, forming a hierarchical feature
representation. The number of layers now extends to over 1,000! With such a gigantic modeling
capacity, a deep network can essentially memorize all possible mappings after successful training with a
sufficiently large knowledge database and make intelligent predictions e.g. interpolations and/or
extrapolations for unseen cases. Thus, deep learning is generating a major impact in computer vision
and medical imaging. In fact, similar impact is happening in domains like text, voice, etc. Various types of
deep learning algorithms are in use in research like convolutional neural networks (CNN), deep neural
network (DNN), deep belief network (DBN), deep autoencodre (dA), deep Boltzmann machine (DBM),
deep conventional extreme machine learning (DC-ELM) recurrent neural network (RNN) and its variant
like BLSTM and MDLATM etc (as illustrated with their pros and cons in table 1. The CNN model is getting
a lot interest in digital imaging processing and vision. There different types of architectures of CNN such
as, Alexnet 1 (as shown in Figure 10, Lenet 2 , faster R-CNN 3 , googleNEt 4 , ResNEt 5 , VGGNet 6 , ZFnet
etc.

1 https://github.com/BVLC/caffe/tree/master/models/bvlcalexnet 2
http://deeplearning.net/tutorial/lenet.html 3 https://github.com/ShaoqingRen/fasterr cnn 4
https://github.com/BVLC/caffe/tree/master/models/bvlcg ooglenet 5 https://github.com/gcr/torch-
residual-networks 6 http://www.robots.ox.ac.uk/ vgg/research/verydeep/
 neural network architecture

Table 1: Comparison of different architecture of Deep learning models

3 Deep Learning: Not-so-near Future in Medical Imaging

Deep learning technology applied to medical imaging may become the most disruptive technology
radiology has seen since the advent of digital imag-ing. Most researchers believe that within next 15
years, deep learning based applications will take over human and not only most of the diagnosis will be
performed by intelligent machines but will also help to predict disease, prescribe medicine and guide in
treatment. Which field in medical has revolutionised Deep learning first? ophthalmology,pathology,
cancer detection, radiology or prediction and personalized medicine. Ophthalmology will be the first
field to be revolutionized in health care, however, pathology, cancer diagnosis have received more
attention and currently we have application with decent accuracy. Google DeepMind Health is working
with National Health Service, UK signed five year agreement to process the medical data of up to 1m
patients across the trusts five hospitals. Even its early days of this project, Deepmind already has high
hopes for the proposal. Researchers and vendors in medical sector are moving this field forward have a
bold recommendation i.e. IBM Watson recently boosted itself through billion-dollar entry into the
imaging arena by the acquisition Merge (imag-ing and Google DeepMind Health is another big
investment. Even though, huge investment and interest, deep learning future in medical imaging is not
that near as compare to other imaging applications due to the complexities involved in this field. The
notion of applying deep learning based algorithms to medical imaging data is a fascinating and growing
research area however, there are several barriers that slow down its progress. These challenges are
unavailability of dataset, privacy and legal issues, dedicated medical experts, non standard data machine
learning algorithms etc.

convolutional neaural network

 deep neural network

3.1 Dataset

Deep learning requires massive amount of training dataset as classification accuracy of deep learning
classifier is largely dependent on the quality and size of the dataset, however, unavailability of dataset is
one the biggest barrier in the success of deep learning in medical imaging. On the other hand,
development of large medical imaging data is quite challenging as annotation requires extensive time
from medical experts especially it requires multiple expert opinion to overcome the human error.
Furthermore, annotation may not be possible due to unavailability of qualified expert or availability of
sufficient cases are also issue in case of rare disease. Another issue major issue is unbalancing of data
that is very common in health sector i.e.rare diseases, by virtue of being rare, are underrepresented in
the data sets. If not accounted for properly, the class imbalance that ensues.

3.4 Black Box and Deep Learning

Medical imaging broke paradigms when it first began more than 100 years ago and deep learning
algorithms gave new birth to medical imaging appli-cation and open new possibilities. It solves the
problems previously thought to be unsolvable by a machine learning algorithms, however, deep learning
is not free form the problems. One of the biggest issues is so called black-box problem, although math
used to construct a neural network is straightfor-ward but how the output was arrived is exceedingly
complicated i.e. machine learning algorithms get bunch of data as input, identify patterns and build
predictive model but understanding how the model worked is issue. The deep learning model is offer
uninterpretable and most of the researchers are using it without know the working process that why it
provides better result.
4 Deep Learning in Medical Imaging

Many image diagnosis task requires initial search to identify abnormalities, quantify measurement and
changes over time. Automated image analysis tool based on machine learning algorithms are the key
enablers to improve the quality of image diagnosis and interpretation by facilitating through efficient
identification of finding. Deep learning is one extensively applied techniques that provides state of the
aft accuracy. It opened new doors in medical image analysis that have not been before. Applications of
deep learning in healthcare covers a broad range of problems ranging from cancer screening and disease
monitoring to personalized treatment suggestions. Various sources of data today - radiological imaging
(X-Ray, CT and MRI scans), pathology imaging and recently, genomic sequences have brought an
immense amount of data at the physicians disposal. However, we are still short of tools to convert all
this data to useful information. In the below discussion, we highlighted state of the art applications of
deep learning in medical image analysis. Though, the list is by no means complete however it provides
an indication of the long-ranging deep learning impact in the medical imaging industry today.

4.1 Diabetic Retinopathy

Diabetes Mellitus (DM) is a metabolic disorder in which pancreases cannot produce proper insulin
(Type-1) or the body tissues do not response to the insulin properly (Type-2) which results in high blood
sugar. The Diabetic Retinopathy (DR) is an eye disease due to diabetes which results in eye blindness
with the passage of time of diabetes in a person Manual process of detection of DR is difficult and time
consuming process at presence due to unavailability of equipment and expertise. As this disease shows
hardly any symptoms in early stage and a clinician needs to examine the colored fundus image of retina
which lead to delay the treatment, miscommunication and loss of follow up. Automated detection of DR
based on deep learning models has proven their optimized and better accuracy. In this section, we are
presenting the research work using deep learning approaches.

Gulshan et al. applied Deep Convolutional Neural Network (DCNN) on Eye Picture Archive
Communication System (EyePACS-1) dataset and Messidor-2 dataset for classification and detection of
moderate and worse referable [21]. The EyePACS-1 consists of approximately 10,000 retinal images and
the Messidor-2 data set consists of 1,700 images that collected from 874 patients. The authors claimed
97.5%sensitivity and 93.4% specificity on EyePACS-1; and 96.1% sensitivity and 93.9% specificity on
Messidor-1, respectively. Kathirvel [9] trained DCNN with dropout layer techniques and tested on
publically available datasets like kaggle fundus, DRIVE and STARE for classification of fundus. The
reported accuracy is up to 94-96%. Pratt et al. [6] employed NVIDIA CUDA DCNN library on Kaggle
dataset consisting of above 80,000 digital fundus images. They also validated the network on 5,000
images. The images resized into 512x512 pixels and then sharpened. Finally, the features vector fed to
Cu-DCNN. They classified the images into 5 classes using features like exudates, haemorrhages and
micro-aneurysms and achieve upto 95% specificity, 30% sensitivity and 75% accuracy.

Haloi [7] implemented five layers CNN with drop out mechanism for detec-tion of early stage DR on
Retinopathy Online Challenge (ROC) and Massidor datasets and claimed t Sensitivity, Specificity,
accuracy and area under the curve (AUC) up to 97%, 96%, 96% and 0.988 on Maddissor dataset and AUC
up to 0.98 on ROC dataset. Alban [8] de-noised the angiograph images of EyePACS and then applied
CNNs for detection of DR. They diagnosed five classes severities and provide 79% AUC and 45%
accuracy. Lim et al. [14] extracted features from identified regions using method proposed in [26] and
then the features vector passed to DCNN for classification. They realized the model on DIARETDB1 and
SiDRP datasets. All above works summarized in Table tableDiabetic.

Table 2: Summary of Deep Learning (DL) for Diabetic Retinopathy (DR).

4.2 Histological and Microscopical Elements Detection

Histological analysis is the study of cell, group of cells and tissues. When different changes come at
cellular and tissue level then microscopic changes, characteristics and features can be detected through
microscopic image technology and stains (colorful chemicals) [24] [30] [29]. It involves number of steps
like fixation, sectioning, staining and optical microscopic imaging. Different skin disease especially
squamus cell carcinoma, and melanoma Other dis-eases like gastric carcinoma, gastric ephitilial
metaplasia, breast carcinoma, malaria, Intestinal parasites and TB etc. Genus plasmodiums parasite is
the main reason of Malaria. Microscopical imaging is the standard method for de-tection of parasites in
stained blood smear sample. Mycobacteria in sputum is the main cause of Tuberculosis (TB). Smear
microscopy and fluorescent auramine-rhodamine stain or Ziehl-Neelsen (ZN) stain are golden
standanrds for detection TB.

Recently, the HistoPhenotypes dataset published [31] where DCNN classifier applied for diagnosis nuclei
of cells of colon cancer using stained histological images. Bayramoglu and Heikkil [1] conducted two
studies for detection of thoraco-abdominal lymph node and interstitial lung disease using transfer
learning (fine tuning) approach with CNN model. Due to limited histological data in [31], features vector
extracted using facial images [13] and natural images of ImageNet [11] using source CNN and then
transferred to object CNN model for classification. The CNN classifier employed for grading gas-tric
cancer by analyzing signet ring cells in tissues and epithelial layers of tissue. They also count the mitotic
figures for breast cancer [16].

In [23], authors employ shape features like moment and morphological to pre-dict malaria, tuberculosis
and hookworm from blood, sputum and stool sam-ples. Automatic microscopic image analysis
performed using DCNN model as a classifier and reported AUC 100% for Malaria and 99% for
tuberculosis and hookworm. DCNN also applied for diagnosis of malaria in [23] and intestinal parasites
in [20]. Fully CNN Deep learning has been used in [36] for auto-matic cell counting. Qiu et al. also
employed DCNN for detection of leukemia in metaphase [22]. Quin et al. conducted experiment using
DCNN for detec-tion of malaria in thick blood smear,, intestinal parasite like helminthes in stool and
mycobacteria in sputum [23]. Malaria detection is a crucial and important research area. In 2015,
438,000 people were died due to malaria according to World Health Organization. Dong et al. developed
four systems for detection infected and non-infected cells by malaria using CNN mod-els and SVM.
Three architectures of CNN named as GoogLeNet , LeNet-5 and AlexNet used for automatic features
extraction and classification and reported 98.13%, 96.18% and 95.79% respectively. SVM based system
got lowest accuracy upto 91.66%. Summary of all above written work is given in Table 3.

Table 3: Summary of Deep Learning (DL) for Histological and Microscopical Elements Detection.

4.3 Gastrointestinal (GI) Diseases Detection

Gastrointestinal (GI) consists of all organs involve in digestion of food and absorption of nutrients and
excretion of the waste products. It starts from mouth to anus. The organs are esophagus, stomach, large
intestine (colon or large bowel) and small intestine (small bowel). The GI may also divide into upper GI
tract and lower GI tract. The upper GI tract consists of esophagus, stomach and duodenum (part of small
bowel) and lower GI tract consists of most of small intestine (jejunam and jilium) and large intestine. The
food digestion and absorption is affected due to different ailments and diseases like inflammation,
bleeding, infections and cancer in the GI tract [23]. Ul-cers cause bleeding in upper GI tract. Polyps,
cancer or diverticulitis cause bleeding from colon. Small intestine has diseases like Celiac, Crohn,
malignant and benign tumor, intestinal obstruction, duodenal ulcer, Irritable bowel syndrome and
bleeding due to abnormal blood vessels named as arteriovenous malformations (angiodysplasias or
angioectasias).

Image processing and machine learning play vital role in diagnosing and analyzing these diseases and
help the doctors in making fast decision for treatment efficiently and accurately. Due to advancement in
computer aided diagnosis (CAD) systems, various kind of imaging tests is in practice for digestive
systems disease detection and classification. These imaging test are Wireless Capsule endoscopy,
Endoscopy and enteroscopy, colonoscopy or sigmoidoscopy, Radioopaque dyes and X-ray studies, deep
small bowel en-teroscopy, intra operative enteroscopy, Computed tomography and magnetic resonance
imaging (MRI).

Jia et al. employed DCNN for detection of bleeding in GI 10,000 Wireless Capsule Endoscopy (WCE)
images [7]. The WCE is a non-invasive image video method for examination small bowel disease. They
claimed F measure approximately to 99Pei etal. mainly focused on evaluation of contraction frequency
of bowel by investigation diameter patterns and length of bowel by measuring temporal information
[19]. The authors implemented Fully Convolutional Networks (FCN) and stacked FCN with LSTM using
small and massive datasets. FCN-LSTM trained on small dataset consisted of 5 cineMRI sequences
without labeling and FCN system realized on massive dataset consisted of fifty raw cine MRI sequence
with labeling Wimmer et al. learned features from ImageNet dataset and then the learned feature
vector fed to CNN SoftMax for classification and detection of celiac disease using duodenums
endoscopic images [33].

A popular approach of automatic feature extraction from endoscopy images adopted using CNN [38].
Then the features vector to the SVM for classification and detection of gastrointestinal lesions. The
proposed system realized on 180 imagesfor lesions detection and 80% accuracy reported. Similarly hy-
brid approach used by [5]. Fast features extraction using CNN architecture in [29] applied and then the
extracted features passed to SVM for detection of inflammatory gastrointestinal disease in WCE videos.
The experiments con-ducted on 337 annotated inflammatory images and 599 non-inflammatory images
of the GI tract of KID [30]. Training set containing 200 normal and 200 abnormal while test set
containing 27 normal and 27 abnormal and ob-tained overall accuracy upto 90%.

The work involved in detection of polyp in colonoscopy videos using representation of image in three
ways [32]. Number of CNN models trained on isolated feature like texture, shape, color and temporal
information in multiple scales which enhance the accurate localization of polyp. and then combined the
results for final decision. They claimed that their polyp dataset is the largest annotated dataset and they
decrease the latency of detection of polyps as compare with the stat of the art techniques. Ribeiro et al.
[32] also conducted three experiment using different CNN. They applied normalization (see details in
[33]). The size of dataset increased using data augmentation by making different variations of images.
They presented another pixels and CNN based work [34] for prognosis of polyp tumor staging using
colonic mucosa as a target attribute. The work available on GI are summarized in Table 4.

4.4 Cardiac Imaging

Deep learning has provided extremely promising result for cardiac imaging especially for Calcium score
quantification. Number of diverse applications has been developed, CT and MRI are the most used
imaging modality whereas common task for image segmentation are left ventricle. Manual identification
of CAC in cardiac CT requires substan- tial expert interaction, which makes it time-consuming and in-
feasible for large-scale or epidemiological studies. To overcome these limitations, (semi)-automatic
calcium scoring methods have been proposed for CSCT. Recent work on Cardiac images is focusing on CT
angeographic images based CAC computation using deep conventional neural network as shown in
figure 9

4.5 Tumor Detection

When cells of any part of the body have abnormal growth and make a mass then it is called Tumor or
Neoplasm. There are two types of tumor. One is noncancerous (Benign tumor) and other is cancerous
(Malignant tumor). Benign tumor is not much dangerous and it is remained stick to one part of
 Fig. 8: Microscopic blood smear
images

Fig. 9: calcium score

score classification from CT [34]

the body and do not spread to other part of the body. While malignant tu-mor is very harmful and it
spread to other part of the body. When it spreads to other part then it is difficult to treat and prognoses
also become very poor.

Wang et al [8] used 482 mammographic images having ages from 32 years to 70 years. Images of 246
women affected by tumor. First the images de-noised using median filter and then segment the breast
tumor using region growth, morphological operations and modified wavelet transformation. Then
morphological and textural features passed to extreme learning machine and SVM for classification and
detection of breast tumor. The total error rate was 84 using ELM and 96 using SVM. Authors in [35] have
limited data of malignant mass and benign solitary cysts. CNN model needs large amount of images for
better finding of cysts and mass. Therefore CNN fed by different variations of images and reported Area
under curve upto 87%.

Arevalo et al.[36] conducted experiment on a benchmark dataset having 736 mediolateral oblique and
craniocaudal mammographic views from 344 cancerous patients. They segmented the images manually
into 310 malignant and 426 benign lesions. First, the images enhanced then fed to CNN for identification
of benign and malignant lesions. They reported 82.6% AUC. Huynh et al. [37] used CNN for features
learning on breast ultrasound images having 2393 regions of interests form 1125 patients. They perform
two experiments. In first experiment, SVM model classified the extracted features into malignant,
benign and cystic and got satisfactory result. In second experiment, SVM classified on hand crafted
features. They obtained 88% AUC on CNN features and 85% AUC on hand crafted features. Antropova et
al. [38] investigated CNN for transfer learning of features form ImageNet dataset (nonmadical) and SVM
applied on 4,096 extracted features for classification of breast lesions as malignant and benign using 551
MRI images consisting of 194 benign and 357 malignant. AUC reported upto 85%.

SVM used for classification and CNN used for features extraction in [39]. They obtained AUC 86% on
dataset containing 219 lesions in 607 breast images. In [40], frozen approach of transfer learning used.
DCNN trained on mammographic images with drop out and jittering approaches with 99% AUC and then
validate on DTB images with 90% AUC after transfer learn-ing. The datasets containing 2282 digitized
film and digital mammograms [41] and 324 DBT volumes [42]. Training set containing 2282 with 2461 le-
sions and 230 DBT views with 228 masses. The remaining images used as independent test. Shin et
al.[43] conducted fine tuning approach of transfer learning on ImageNet using CNN. Then CNN model
applied as a classifier for detection of lesion in thoraco-abdominal lymph node and interstitial lung
disease. The authors got sensitivity upto 83% and 85% and AUC upto 94% and 95%, respectively. A brief
summary of published work is given in Table 5.

4.6 Alzheimer’s and Parkinsons Diseases Detection

Parkinsons disease (PD) is a neurological disorder associated with a progressive decline in motor
precision and sensor motor integration stemming presumably from a disorder of the basal ganglia [46].
Parkinsons disease is associated with the breaking up or the dying of dopaminergic neurons.
Neurological testing like MMSE [47], and UPDRS [48]) and brain scans are routinely used to determine
the diagnosis of AD [49]. The scale and shift invariant based features like shape of data, mean and
standard deviation using CNN model (LeNet-5) , carries out classification on functional MRI 4D in work of
Sarraf and Tofigh [50]. The proposed system trained on 270900 images and validated and tested on
90300 images in fMRI. The authors obtained 96.86% accuracy for detection of affected brains by
Alzheimer disease.

Suk [51] employed Deep Boltzmann Machine (DBM) for features extrac-tion and detection of
abnormalities from 3D patch of MRI and PET images. The results were validated on ADNI dataset [52] for
alzheimers disease us-ing dataset of PET, MRI and combination of PET and MRI and obtained accuracies
upto 92.38%, 92.20% and 95.35%. Hosseini-Asl [53] also explored 3D-CNN for diagnosis AD and
extracted generic features using CADDementia MRI dataset. The authors then fine tuned three fully
connected layers CNN for classification of AD using ADNI dataset. Sarraf et al. [54] diag-nosed Alzheimer
disease in adults (above 75 years old) using fMRI and MRI images. The authors conducted studies based
on research and clinical appli-cations. CNN model employed for detection of healthy or Alzheimers brain
and report 99.9 for functional MRI data and 98.84% for MRI data, respec-tively. They then performed
classification on subject level and finally decision making based algorithm applied. The final accuracy
improved upto 97.77% and 100% for fMRI and MRI subjects. In [55], sparse auto encoder (a neural
network) used for extraction of features and then 3D-CNN applied as a clas-sifier on ADNI dataset
consisting of neuron images. The dataset is divided into training set (1,731 samples), validation set (306
samples) and test set (228 sample) and achieved performance upto 95.39% for AD and 95.39%.

Liu et al. [? ] also used sparse auto encoder fro extraction generic features and then applied sofmax of
CNN for classification of affected brain by Alzheimer; or, prodromal stage or mild stage of Alzheimer.
They achieved accuracy upto 87.76% on binary images of MRI and PET for early stage detection of
Alzheimer disease. All methods are summarized in Table 6.

Table 4: Summary of Deep Learning (DL) for GI.

Table 5: Summary of Deep Learning (DL) for Tumor Detection.

Table 6: Summary of Deep Learning (DL) for Alzheimer Disease Detection.

5 Open Research Issues and Future Directions

Three trends that drive the deep learning revolution are availability of big data, recent deep learning
algorithm modeled on the human brain and pro-cessing power. While deep learning potential benefits
are extremely significant and so are the initial efforts and costs. Big companies like Google DeepMind,
IBS Watson, research labs along with leading hospitals and vendors are com-ing together and working
toward the optimal solution of big medical imaging. Siemen, Philips, Hitachi and GE Healthcare etc. have
already made significant investments. Similarly research lab such as Google, IBM are also investing
towards the delivery of efficient imaging applications i.e. IBM Watson is working with more than 15
healthcare provides to learn how deep learning could work in real world.Similary google DeepMind
health is collaborating with NHS, UK to apply deep learning on different healthcare applications ( for
example : anonymised eye scans analysis could help to find the sing of diseases that could leads to
blindness) on dataset of 1.6 million patient. GE Healthcare partnership with Bostons Children Hospital is
working to create smart imaging technology for detecting pediatric brain disorders. Further-more, GE
Healthcare and UC San Francisco has also announced a 3-year partnership to develop a set of algorithms
to differentiate between normal result and one that requires further attention by expert.

5.1 Requires Extensive Inter-organization Collaboration

Despite great effort done by big stakeholder and their predictions about the growth of deep learning
and medical imaging, there will be a debate on re-placing human with machine however, deep learning
has potential benefits towards disease diagnosis and treatment. However, there are several issues that
need to be solved to make it possible earlier. Collaboration between hospital providers, vendors and
machine learning scientists is extensively re-quired to windup this extremely beneficial solution for
improving the quality of health. This collaboration will resolve the issue of data unavailability to the
machine learning researcher. Another major issue is, we need more so-phisticated techniques to deal
extensive amount of healthcare data, especially in future, when more of the healthcare industry will be
based on body senor network.

5.2 Need to Capitalize Big Image Data

Deep learning applications rely on extremely large dataset, however, availabil-ity is of annotated data is
not easily possible as compared to other imaging area. It is very simple to annotate the real world data
i.e. annotation of men and woman in crowd, annotating of object in real world image. However,
annotation of medical data is expensive, tedious and time consuming as it requires extensive time for
expert (especially due the sensitivity of domain, annotation required opinions of different expert on
same data), furthermore annotation may not be always possible in case of rare cases. Thus sharing the
data resource with in different healthcare service providers will help to overcome this issue somehow.

5.3 Advancement in Deep Learning Methods

Majority of deep learning methods focus on supervised deep learning however annotations of medical
data especially image data is not always possible i.e. in case when rare disease or unavailability of
qualified expert. To overcome, the issue of big data unavailability, the supervised deep learning field is
required to shift from supervised to unsupervised or semi-supervised. Thus, how efficient will be
unsupervised and semi-supervised approaches in medical and how we can move from supervised to
transform learning without effecting the accuracy by keeping in the healthcare systems are very
sensitive. Despite current best efforts, deep learning theories have not yet provided complete solutions
and many questions are still unanswered, we see unlimited in the opportunity to improve.

5.4 Black-Box and Its Acceptance by Health Professional

Health professional wary as many question are still unanswered and deep learning theories has not
provided complete solution. Unlike health professional, machine learning researchers argues
interoperability is less of an issue than a reality. Human does not care about all parameters and perform
complicated decision, it is just mater of human trust. Acceptance of deep learning in health sector need
proof form the other fields, medical expert, are hoping to see its success on other critical area of real
world life i.e. autonomous car, robots etc. Even though great success of deep learning based method,
decent theory of deep learning algorithms is still missing. Embarrassment due to the absence this is well
recognized by the machine learning community. Black- box could be another one of the main challenge,
legal implications of black box functionality could be barrier as healthcare expert would not rely on it.
Who could be responsible if the result went wrong. Due to the sensitivity of this area, hospital may not
comfortable with black-box i.e. how it could be traced that particular result is from the ophthalmologist.
Unlocking of black box is big research issue, to deal with it, deep learning scientist are working toward
unlocking this proverbial black box.
Fig. 11: Deep Learning: A Black Box

6 Conclusion

During the recent few years, deep learning has gained a central position toward the automation of our
daily life and delivered considerable improvements as compared to traditional machine learning
algorithms. Based on the tremendous performance, most researchers believe that within next 15 years,
deep learning based applications will take over human and most of the daily activities with be
performed by autonomous machine. However, penetration of deep learning in healthcare especially in
medical image is quite slow as compare to the other real world problems. In this chapter, we highlighted
the barriers, that are reducing the growth in health sector. In last section, we highlighted state of the art
applications of deep learning in medical image analysis. Though, the list is by no means complete
however it provides an indication of the long-ranging deep learning impact in the medical imaging
industry today. Finally, we have highlighted the open research issues. Many big research organization
are working on deep learning based solu-tion that encourage to use deep learning to apply deep
learning on medical images. Looking to the brighter side of machine learning, we are hoping the sooner
human will be replaced in most of the medical application especially diagnosis. However, we should not
consider it as only solution as there are sev-eral challenges that reduces its growth. One of the big
barrier is unavailability of annotated dataset. Thus, this question is still answerable, that whether we will
be able to get enough training data without effecting the performance of deep learning algorithms.
Recent development on other application showed that bigger the data, better the result, however, how
big data could be used in healthcare. So far deep learning based application provided positive feedback,
however, but due to the sensitivity of healthcare data and challenges, we should look more
sophisticated deep learning methods that can deal complex healthcare data efficiently. Lastly we
conclude that there are unlimited opportunities to improve healthcare system.

References

[1] Neslihan Bayramoglu and Janne Heikkil¨a. Transfer learning for cell nu-clei classification in
histopathology images. In Computer Vision– ECCV 2016 Workshops, pages 532–539. Springer, 2016. [2]
Francesco Ciompi, Kaman Chung, Sarah J van Riel, Arnaud Arindra Adiyoso Setio, Paul K Gerke, Colin
Jacobs, Ernst Th Scholten, Cornelia Schaefer-Prokop, Mathilde MW Wille, Alfonso Marchiano, et al.
Towards automatic pulmonary nodule management in lung cancer screening with deep learning. arXiv
preprint arXiv:1610.09157, 2016.

Deep Learning for Medical Imaging 27

[3] Zhipeng Cui, Jie Yang, and Yu Qiao. Brain mri segmentation with patch-based cnn approach. In
Control Conference (CCC), 2016 35th Chinese, pages 7026–7031. IEEE, 2016. [4] Rahul Duggal, Anubha
Gupta, Ritu Gupta, Manya Wadhwa, and Chirag Ahuja. Overlapping cell nuclei segmentation in
microscopic images using deep belief networks. In Proceedings of the Tenth Indian Conference on
Computer Vision, Graphics and Image Processing, page 82. ACM, 2016. [5] Spiros V Georgakopoulos,
Dimitris K Iakovidis, Michael Vasilakakis, Vassilis P Plagianakos, and Anastasios Koulaouzidis.
Weaklysupervised convolutional learning for detection of inflammatory gastrointestinal le-sions. In
Imaging Systems and Techniques (IST), 2016 IEEE Interna-tional Conference on, pages 510–514. IEEE,
2016. [6] Varun Gulshan, Lily Peng, Marc Coram, Martin C Stumpe, Derek Wu, Arunachalam
Narayanaswamy, Subhashini Venugopalan, Kasumi Widner, Tom Madams, Jorge Cuadros, et al.
Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal
fundus photographs. JAMA, 316(22):2402–2410, 2016. [7] Xiao Jia and Max Q-H Meng. A deep
convolutional neural network for bleeding detection in wireless capsule endoscopy images. In
Engineering in Medicine and Biology Society (EMBC), 2016 IEEE 38th Annual International Conference of
the, pages 639–642. IEEE, 2016. [8] Konstantinos Kamnitsas, Christian Ledig, Virginia FJ Newcombe,
Joanna P Simpson, Andrew D Kane, David K Menon, Daniel Rueckert, and Ben Glocker. Efficient multi-
scale 3d cnn with fully connected crf for accurate brain lesion segmentation. Medical Image Analysis,
36:61–78, 2017. [9] C. T. R. Kathirvel. Classifying Diabetic Retinopathy using Deep Learning Architecture.
International Journal of Engineering Research Technology, 5(6), 2016. [10] Jens Kleesiek, Gregor Urban,
Alexander Hubert, Daniel Schwarz, Klaus Maier-Hein, Martin Bendszus, and Armin Biller. Deep mri brain
extrac-tion: a 3d convolutional neural network for skull stripping. NeuroImage, 129:460–469, 2016. [11]
Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classi-fication with deep convolutional
neural networks. In Advances in neural information processing systems, pages 1097–1105, 2012. [12]
Nikolas Lessmann, Ivana Isgum, Arnaud AA Setio, Bob D de Vos, Francesco Ciompi, Pim A de Jong,
Matthjis Oudkerk, P Th M Willem, Max A Viergever, and Bram van Ginneken. Deep convolutional neural
networks for automatic coronary calcium scoring in a screening study with low-dose chest ct. In SPIE
Medical Imaging, pages 978511– 978511. International Society for Optics and Photonics, 2016. [13] Gil
Levi and Tal Hassner. Age and gender classification using convolutional neural networks. In Proceedings
of the IEEE Conference on Com-puter Vision and Pattern Recognition Workshops, pages 34–42, 2015.

28 Muhammad Imran Razzak, Saeeda Naz and Ahmad Zaib

[14] Gilbert Lim, Mong Li Lee, Wynne Hsu, and Tien Yin Wong. Transformed representations for
convolutional neural networks in diabetic retinopathy screening. Modern Artif Intell Health Anal, 55:21–
25, 2014. [15] Pawel Liskowski and Krzysztof Krawiec. Segmenting retinal blood ves-sels with¡? pub
newline?¿ deep neural networks. IEEE transactions on medical imaging, 35(11):2369–2380, 2016. [16]
Christopher Malon, Matthew Miller, Harold Christopher Burger, Eric Cosatto, and Hans Peter Graf.
Identifying histological elements with convolutional neural networks. In Proceedings of the 5th
international conference on Soft computing as transdisciplinary science and technol-ogy, pages 450–
456. ACM, 2008. [17] Lua Ngo and Jae-Ho Han. Advanced deep learning for blood vessel segmentation
in retinal fundus images. In Brain-Computer Interface (BCI), 2017 5th International Winter Conference
on, pages 91–92. IEEE, 2017. [18] Rahul Paul, Samuel H Hawkins, Lawrence O Hall, Dmitry B Goldgof,
and Robert J Gillies. Combining deep neural network and traditional image features to improve survival
prediction accuracy for lung cancer patients from diagnostic ct. In Systems, Man, and Cybernetics (SMC),
2016 IEEE International Conference on, pages 002570–002575. IEEE, 2016. [19] Mengqi Pei, Xing Wu,
Yike Guo, and Hamido Fujita. Small bowel motil-ity assessment based on fully convolutional networks
and long short-term memory. Knowledge-Based Systems, 121:163–172, 2017. [20] AZ Peixinho, SB
Martins, JE Vargas, AX Falcao, JF Gomes, and CTN Suzuki. Diagnosis of human intestinal parasites by
deep learning. In Computational Vision and Medical Image Processing V: Proceedings of the 5th
Eccomas Thematic Conference on Computational Vision and Medical Image Processing (VipIMAGE 2015,
Tenerife, Spain, page 107, 2015.

[21] Harry Pratt, Frans Coenen, Deborah M Broadbent, Simon P Harding, and Yalin Zheng. Convolutional
neural networks for diabetic retinopathy. Procedia Computer Science, 90:200–205, 2016. [22] Yuchen
Qiu, Xianglan Lu, Shiju Yan, Maxine Tan, Samuel Cheng, Shibo Li, Hong Liu, and Bin Zheng. Applying deep
learning technology to automatically identify metaphase chromosomes using scanning microscopic
images: an initial investigation. In SPIE BiOS, pages 97090K–97090K. International Society for Optics and
Photonics, 2016. [23] John A Quinn, Rose Nakasi, Pius KB Mugagga, Patrick Byanyima, William Lubega,
and Alfred Andama. Deep convolutional neural networks for microscopy-based point of care diagnostics.
arXiv preprint arXiv:1608.02989, 2016. [24] Muhammad Imran Razzak and Bandar Alhaqbani. Automatic
detection of malarial parasite using microscopic blood images. Journal of Medical Imaging and Health
Informatics, 5(3):591–598, 2015.

Deep Learning for Medical Imaging 29

[25] Masaharu Sakamoto and Hiroki Nakano. Cascaded neural networks with selective classifiers and its
evaluation using lung x-ray ct images. arXiv preprint arXiv:1611.07136, 2016. [26] Gilbert Lim Yong San,
Mong Li Lee, and Wynne Hsu. Constrainedmser detection of retinal pathology. In Pattern Recognition
(ICPR), 2012 21st International Conference on, pages 2059–2062. IEEE, 2012. [27] Saman Sarraf, John
Anderson, Ghassem Tofighi, et al. Deepad: Alzheimer s disease classification via deep convolutional
neural networks using mri and fmri. bioRxiv, page 070441, 2016. [28] Santi Segu´ı, Michal Drozdzal,
Guillem Pascual, Petia Radeva, Carolina Malagelada, Fernando Azpiroz, and Jordi Vitri`a. Deep learning
features for wireless capsule endoscopy analysis. In Iberoamerican Congress on Pattern Recognition,
pages 326–333. Springer, 2016. [29] Syed H Shirazi, Arif Iqbal Umar, Saeeda Naz, and Muhammad I
Razzak. Efficient leukocyte segmentation and recognition in peripheral blood image. Technology and
Health Care, 24(3):335–347, 2016. [30] Syed Hamad Shirazi, Arif Iqbal Umar, Nuhman Ul Haq, Saeeda
Naz, and Muhammad Imran Razzak. Accurate microscopic red blood cell image enhancement and
segmentation. In International Conference on Bioinformatics and Biomedical Engineering, pages 183–
192. Springer International Publishing, 2015. [31] Korsuk Sirinukunwattana, Shan E Ahmed Raza, Yee-
Wah Tsang, David RJ Snead, Ian A Cree, and Nasir M Rajpoot. Locality sensi-tive deep learning for
detection and classification of nuclei in routine colon cancer histology images. IEEE transactions on
medical imaging, 35(5):1196–1206, 2016. [32] Nima Tajbakhsh, Suryakanth R Gurudu, and Jianming
Liang. Automatic polyp detection in colonoscopy videos using an ensemble of con-volutional neural
networks. In Biomedical Imaging (ISBI), 2015 IEEE 12th International Symposium on, pages 79–83. IEEE,
2015. [33] G Wimmer, S Hegenbart, A Vecsei, and A Uhl. Convolutional neural network architectures for
the automated diagnosis of celiac disease. In International Workshop on Computer-Assisted and
Robotic Endoscopy, pages 104–113. Springer, 2016. [34] Jelmer M Wolterink, Tim Leiner, Bob D de Vos,
Robbert W van Hamersvelt, Max A Viergever, and Ivana Iˇsgum. Automatic coronary artery calcium
scoring in cardiac ct angiography using paired convolutional neural networks. Medical image analysis,
34:123–136, 2016. [35] Jelmer M Wolterink, Tim Leiner, Max A Viergever, and Ivana Iˇsgum. Automatic
coronary calcium scoring in cardiac ct angiography using convolutional neural networks. In International
Conference on Medical Image Computing and Computer-Assisted Intervention, pages 589–596.
Springer, 2015. [36] Weidi Xie, J Alison Noble, and Andrew Zisserman. Microscopy cell counting and
detection with fully convolutional regression networks.

30 Muhammad Imran Razzak, Saeeda Naz and Ahmad Zaib

Computer Methods in Biomechanics and Biomedical Engineering: Imag-ing & Visualization, pages 1–10,
2016. [37] Yixuan Yuan and Max Q-H Meng. Deep learning for polyp recognition in wireless capsule
endoscopy images. Medical Physics, 2017. [38] Rongsheng Zhu, Rong Zhang, and Dixiu Xue. Lesion
detection of endoscopy images based on convolutional neural network features. In Im-age and Signal
Processing (CISP), 2015 8th International Congress on, pages 372–376. IEEE, 2015.

Artificial Intelligence Techniques for Cancer Detection and Classification

Introduction: Cancer is the name given to a group of related diseases. In all types of cancer, several body
tissues start to divide without stopping and spread around cells. Cancer can start almost any place in the
human body, which is composed of approximately trillions of cells. Human tissues normally grow and
divide to form new tissues as the human body needs them. When cells age or become damaged, they
die and are replaced with new cells. However, when cancer develops, this orderly process breaks down.
As cells become increasingly abnormal, older or destroyed cells survive when they should die, and new
cells form when they are not needed (Dalerba, Cho, & Clarke, 2007). These extra cells can divide without
stopping and may form growths called tumors. Numerous types of cancers form solid tumors, which are
composed of cell masses. Cancers of the blood, such as leukemia’s, often do not form solid tumors.
Cancer tumors are malignant in which they can spread into, or invade, adjacent cells. A number of
cancer tissues from these tumors can break off and go to distant areas in the body. New tumors can
spread to areas away from the primary cancer growth through the blood or the lymph system. Benign
tumors can often be removed, and in numerous instances, they do not go back and spread to other
body parts. Cancer is a leading cause of disease worldwide. According to estimates from the
International Agency for Research on Cancers, 14.1 million new cancer cases occurred and 8.2 million
people died from cancer worldwide in 2012.(National Cancer Institute 2012)

Computer-aided diagnosis (CAD) in medicine Manual classification of images is a challenging and time-
consuming task. This task is highly susceptible to interobserver variability and human errors. Therefore,
manual classification results in extremely poor critical outcomes, thus markedly increasing the workload
of radiologists because of their significant shortage. In addition, medical care costs that are relevant to
imaging rapidly increase (Cheng, Cai, Chen, Hu, & Lou, 2003). New methods for diagnosis are therefore
required. At present, CAD is one of numerous major research topics in diagnostic radiology and medical
imaging(Murino, Puppo, Sona, Cristani, & Sansone, 2015). The CAD approach helps medical doctors to
diagnose diseases with a higher degree of efficiency, while minimizing examination time and cost, as
well as avoiding unnecessary biopsy procedures. To date, CAD is a more suitable method for primary
diagnosis of cancer, with the application of computed tomography (CT), X-ray, magnetic resonance
imaging (MRI), or mammogram images (Doi, 2007). CAD is an effective intermediate between input
images and the radiologist. The output from CAD is not considered as an end result; nevertheless, the
result is used as reference with regard to additional testing in the related field. The CAD assists medical
doctors in early and more precise cancer detection. CAD may be developed in relation to more than one
branch of knowledge, together with basic aspects of various areas, such as artificial intelligence (AI),
image analysis, medical information processing and management, digital image processing, and pattern
recognition (Murino et al., 2015). The CAD system is more reliable and efficient. The essential
parameters of this system include specificity, sensitivity, and absolute detection rate. • Specificity is also
called the true negative rate. It measures the actual ratio of negatives that are correctly identified
(Duncan et al., 2008). • Sensitivity is also called the true positive rate. It measures the actual ratio of
positives that are correctly identified.(Fawcett, 2006) The CAD system is often useful for breast, brain,
lung, and other cancer types. Almost any part of the human body can be affected by cancer, which can
spread to another body region. The CAD system has advanced considerably and achieved a high level of
accuracy.

Artificial Intelligence (AI) AI techniques are approaches that are utilized to produce and develop
computer software programs. AI is an application that can re-create human perception. This application
normally requires obtaining input to endow AI with analysis or dilemma solving, as well as the ability to
categorize and identify objects. This paper describes various AI techniques, such as support vector
machine (SVM) neural network, fuzzy models, artificial neural network (ANN), and K-nearest neighbor
(K-NN). All methods are presented in (Fig. 1)

Research Methodology Various intelligent techniques are utilized by researchers to help classify and
segment medical image data to identify abnormalities within different areas of the body. This type of
study is confined to the use of most of these techniques for classification and segmentation of medical
image data.

MEDICAL IMAGING Medical imaging has developed into a crucial part of earlier diagnosis,
detection, and treatment method of cancer through the years. Medical imaging is usually the first step
to avoiding the spread of cancer via earlier detection and, in numerous cases, assists in the treatment or
total elimination of cancer. CT imaging, MRI, mammography, ultrasound (US) imaging, X-ray imaging,
and so on, are typical imaging modalities used for fighting cancer, all of which are highlighted in (Fig. 2)
Figure 2. Different Types of Medical Image

Classification of cancers Different types of AI algorithms are used to detect and classify different types of
cancers. These techniques showed fluctuating accuracy across different years. This varying trend could
be due to numerous factors, including network structure. In designing architecture for specific
applications, the following selected parameters vary: network type, numbers of layers, number of nodes
in hidden layers, activation function between layers, and the size of the dataset used (Dhokia, Kumar,
Vichare, Newman, & Allen, 2008), (Peng, Jianmin, & Wu, 2009). Network generalization indicates how
these networks are able to work with different data to decrease performance error to the lowest value.

Breast Cancer Breast cancer is a malignant tumor that starts in the tissues of the breast. This cancer
can expand directly into surrounding areas or maybe distribute to distant parts of the body. The disease
occurs almost exclusively in females, but men can also develop this type of cancer (G. Schaefer et al.,
2007). With recent functions involving examinations, considerable interest with regard to the
utilization of computational strategies to assist detection and diagnosis of breast cancer among most
cancers is concentrated on mammography. The particular mammography technique is a simple yet
effective tool for prognosis that involves breast cancer at an earlier stage (Cheng, Shan, Ju, Guo, &
Zhang, 2010). The Wisconsin Breast Cancer data (WBCD) source was offered by Dr. William H. Walberg,
and the numerous AI techniques that are used by researchers and applied on WBCD database for
prognosis, detection, and classification breast cancer are discussed in the succeeding paragraphs. This
database, which is easily obtainable in the UCI database repository, contains 699 instances. (Lichman,
(2013))

Neural Network Different neural network algorithms use both supervised and unsupervised learning
techniques for the diagnosis, prognosis, prediction, and classification of breast cancer through the years.

Supervised techniques One supervised technique is the multi-layer perceptron (MLP). MLP can use
different variant training methods, such backpropagation, scaled conjugate gradient, gradient descent,
and Bayesian Regulation (BR) technique. Table 1 shows that (Jung, Thapa, & Wang, 2005) used MLP with
backpropagation training algorithms on Ultrasound (US) images, and (Swathi, Rizwana, Babu, Kumar, &
Sarma) employed MLP model and (BPNN). Both studies obtained high accuracies above 95% for
detecting and classifying breast cancer at 96.1%, 95.74%, and 99.28%, respectively. However, (George,
Elbagoury, Zayed, & Roushdy, 2012) and (Raad, Kalakech, & Ayache, 2012) obtained a low performance.
George and Elbagoury used microscopic images collected from hospitals and obtained 80.66% sensitivity
and 76.28% specificity. Raad and Kalakech et al. obtained 88% accuracy. (Azar & El-Said, 2013),
(Ubaidillah, Sallehuddin, & Ali, 2013), and (Seema Singh 2014) used MLP with SCG learning algorithm,
and they all obtained high accuracy of more than 95%, with accuracies of 96.34%, 98.54%, and 97.47%,
respectively. (Janghel, Shukla, Tiwari, & Kala, 2010) employed MLP with gradient descent and obtained a
low accuracy of 51.88%. By contrast, (B. K. Singh, Verma, & Thoke, 2015) reported increased accuracy
(84.6%) with the use of the same technique on US images. Increased accuracy (97.51%) was also
obtained by (Seema Singh 2014) using MLP with different training methods, such as BR technique. (Al-
Timemy, Al-Naima, & Qaeeb, 2009), (George et al., 2012), (Swathi et al.) and (Azar & El-Said, 2013)
presented a PNN as a supervised classification technique. This technique exhibited a constantly
increasing accuracy of more than 95% from 2009 to 2013. (Janghel et al., 2010) used radial base function
and obtained a low accuracy of 49.79%. (Swathi et al.), (Azar & El-Said, 2013), and (Raad et al., 2012)
employed the same algorithm (RBF) and obtained the highest increasing accuracy of 96.18%, 96.05%,
and 97%, respectively. (Swathi et al.) examined general regression neural network and resulted in a high
accuracy of 98.18% for breast cancer diagnosis.

Unsupervised Technique (Jung et al., 2005) applied linear vector quantization on US images, and
(Janghel et al., 2010) used the same algorithm on different datasets; both studies obtained high
accuracies (91.5%, 95.82%). By contrast, (George et al., 2012) obtained a low accuracy by employing the
same algorithm on microscopic images collected from hospitals. A low accuracy (65%) was also obtained
by (Seema Singh 2014) with LVQ. (Jung et al., 2005) used US images, and (Seema Singh 2014) employed
microscopic images. Both studies applied self-organization feature map technique, which yielded good
accuracies of 87.9% and 84.45%. (S. Singh, Saini, & Singh, 2012) adopted ARNN technique and obtained
an accuracy of 82.64%, and (Janghel et al., 2010) used a competitive learning network and obtained a
low accuracy of 74.48%.

ANFIS (Übeyli, 2009) and (Fatima & Amine, 2012) used hybrid ANFIS technique with learning algorithms
(backpropagation and least squares) and obtained high accuracies of 99.08% and 98.25%. (Fatima &
Amine, 2012) also used ANFIS with backpropagation algorithm and obtained a low accuracy of 64.91%.

Fuzzy Logic (R. Jain & Abraham, 2004) applied rule generation based on homogeneous fuzzy sets and
used another rule mean standard deviation of attribute values. (Al-Daoud, 2010) proposed a modified
fuzzy c-means RBF network, and (Soria et al., 2013) adopted fuzzy algorithm linguistic rule. (GÖRGEL,
SERTBAŞ, & UÇAN, 2012) applied fuzzy subtractive ANFIS. (Onan, 2015) presented a hybrid intelligent
fuzzy-rough nearest neighbor algorithm. All of these researchers used different fuzzy rules on
mammogram data and obtained a high accuracy (>90%) in detecting and classifying breast cancer. (R.
Jain & Abraham, 2004) used rule generation based on a histogram of attribute values, and (Gerald
Schaefer et al., 2007) proposed a system for breast cancer detection on thermography by using a fuzzy
classification system (Miranda & Felipe, 2014) presenting fuzzy omega algorithm method. Jain,
Schaefer, and Miranda obtained good accuracy that exceeded 80% by using different fuzzy rules. (R.
Jain & Abraham, 2004) applied a generation rule that depended on fuzzy partition of overlapping areas
and obtained a low accuracy of 62.57%.

Other AI Classifier Techniques (George et al., 2012) and (GÖRGEL et al., 2012) applied SVM and
obtained more than 80% accuracy by using mammogram data collected from hospitals. (Ubaidillah et
al., 2013) used the same technique and obtained a high accuracy of 99.51%. (Jung et al., 2005) used K-
means algorithms to classify breast cancer, yielding an accuracy of 87.4%. Different changes were
observed in accuracy over the years depending on numerous influencing factors, such as the used
technique, dataset, network architecture, learning rate, epoch, and number of samples for training and
testing. Numerous AI techniques can be used to diagnose breast cancer, yielding different accuracies by
using different sources of data, such as mammograms, thermography, US, and microscopy. Effort has
been expended to explore various AI techniques for detecting breast cancer. A comparative analysis is
carried out to present all detection and classification methods, as shown in Table 1.

Results After evaluating the performance of different classifiers with respect to classifier accuracy,
we will consider the highest classifier accuracy from each technique in all types of cancer.

Breast Cancer (Fig.3) shows the highest accuracy values obtained from each technique for the detection
and classification of breast cancer. As shown in (Fig.3), FL using mammograms provides the best
classifier technique for the detection and classification of breast cancer, yielding an accuracy of 99.73%
(R. Jain & Abraham, 2004). FL can classify cancer cases with a high accuracy rate and is a powerful tool
for decision making according to FL approval. In addition to adequate interpretability of extracted rules,
the mathematical methods in fuzzy reasoning are simple, and FL can be modified by adding or even
removing rules caused by the flexibility of FL methods. Despite the possibility of imprecise, altered noisy
input information, FLS is easy to construct and understand (Cheng & Cui, 2004). The disadvantages of FL
are generally the numerous probable fuzzy rules that exponentially increase with the dimensionality of
pattern space (R. Jain & Abraham, 2004) and the lack of systematic method for fuzzy system design.
However, the method is easy to understand. The approach is simple and suitable for problems that do
not require high accuracy. The advantages of using breast screening mammograms are as follows. The
approach minimizes the number of women who die of breast cancer. Breast screening can discover
extremely small breast cancers before being sensed or observed and results in considerably better
potential for survival as breast cancer can be found earlier with a screening mammogram. The leading
threats for mammograms include non-ideal situations, in which normal breast cells can hide a breast
cancer and thus will not appear on the mammogram. This finding is known as a false negative (Cheng et
al., 2006). As shown in (Fig.3), the K-NN applied on US images exhibited the lowest accuracy of 87.40%
(Jung et al., 2005). The advantages for K-NN include extremely rapid training, ease and simplicity of
understanding effective to noisy training data, and efficiency with usually large training data. The
disadvantages include calculation complexity and memory space restriction. In addition, K-NN, a
supervised learning lazy algorithm, runs slowly and is easily misled by unimportant attributes (Z. X.
Huang, 1998). In general, the results show the ability of different AI techniques to detect and classify
breast cancer.
Figure .3. Performance Comparison of Different AI Techniques for Breast Cancer

Table 1. Summary of Breast Cancer Detection and Classification Works

Lung Cancer Lung cancer can include out-of-control development associated with unnatural tissues,
which start in a single or even the two lungs. The unusual cells generally do not grow into healthy lung
cells, instead separating rapidly and forming cancers. Major types of lung cancer usually involve non-
small cell and small cell lung cancer. These cancers depend on how tissues appear under a microscope.
Non-small cell lung cancer is more widespread compared with small-cell lung cancer (C. Zhou et al.,
2011). Lung cancer is probably the type that commonly leads to extremely high death rate. The most
effective method of protection against lung cancer is early prediction and diagnosis. Detection of lung
cancer at an early stage is a complicated issue because of the construction on cancer tissues, in which
almost all cells are overlapped. Besides being a crucial element in image processing, efficient
identification of lung cancer at an initial stage is usually important. Numerous techniques, including CT,
chest radiography (X-ray), sputum cytology, and MRI scan, are used to diagnose lung cancer. However,
several of these techniques are usually expensive. Many of these methods are not only time consuming
butalso often detect lung cancer in the advanced phases, relatively decreasing the patient's chance of
survival. New technologies to aid the diagnosis of lung cancer at the initial stage is necessary. Image
processing and data mining methods provide good quality tools for enhancing manual analysis.

Neural Networks (Z.-H. Zhou, Jiang, Yang, & Chen, 2002) proposed a lung cancer diagnosis system
named LCDS to identify lung cancer cells through the images of needle biopsy specimens. In this system,
neural ensemble-based detection (NED) with two-level ensemble architecture is used to accomplish its
task. The NED achieves high overall identification rate and low rate falsenegative identification, whereas
LCD needs to improve performance in dealing with overlapped cells.(Pereira, Alexandre, Mendonça, &
Campilho, 2006) examined a classification approach for the lung nodule in X-ray chest radiographs. The
multi classifier (MLP) approach and the obtained results can be used for reducing false-positive nodules.
(Nehemiah & Kannan, 2006) proposed a system that uses image processing and feedforward neural
networks to detect and classify lung nodules into non-cancerous and cancerous nodules and he success
in that . (Taher & Sammouda, 2011) presented Hopfield neural network (HNN) and fuzzy C-mean (FCM)
for detection of lung cancer in the early stages as a segmentation method. One thousand sputum color
images are used for the approach. HNN is more successful in extracting the nuclei and cytoplasm regions
than FCM, which detects only one part. However, FCM is not sensitive to intensity variations. For solving
the problem of applied threshold and detecting nuclei and cytoplasm regions, overall thresholding
exhibits a high accuracy of 98%. Bayesian classification and HNN algorithm (Taher, Werghi, & AlAhmad,
2012) are presented for extracting and segmenting the sputum cells for early lung cancer diagnosis, and
88.62% accuracy was obtained. (Ada¹, 2013) used a two-method preprocessing and a feed-forward
BPNN classifier to detect and classify lung cancer at the early stages. The neural network achieved
correct classification, with the highest rate of 96.4%. (Kuruvilla & Gunavathi, 2014) presented
feedforward and feedforward BPNN as a classification method for lung nodule detection through CT
images. Feedforward back-propagation results in superior classification and is based on the rear-
correction learning rule. Feedforward backpropagation obtained an accuracy of 93.3% and a minimum
mean square error of 0.0942. In addition, the parameter skewness and training function yield the
maximum accuracy. (Gorynski et al., 2014) presented an MLP for the detection and diagnosis of lung
cancer at an early stage. Neural networks achieved high accuracy of more than 95% in detecting and
classifying lung nodules.

FL (Hashemi, Pilevar, & Rafeh, 2013) proposed a model for lung cancer diagnosis at the early stages
through CT images. The model yielded 95% sensitivity. Segmentation and decision-making were
conducted using FIS, and an ANN was used to test the result. (Memarian, Alirezaie, & Babyn, 2006)
proposed an iterative linear discriminant analysis as a novel classification method and used this method
in addition to FCM clustering for successful false-positive reduction. The sensitivity of this approach was
80.80%.

ANFIS (Tariq, Akram, & Javed, 2013)proposed a hybrid neuro-fuzzy classifier as an effective
computerized system for lung nodule detection using CT scan images. The approach yielded 95%
accuracy.
Other classifiers (Song, Zhukov, Markov, Qian, & Tockman, 2012) proposed a system for classifying lung
cancer by using SVM and linear discriminant analysis with 10-fold cross validation and obtained an
accuracy of 85%. (Madero Orozco, Vergara Villegas, de Jesus Ochoa Dominguez, & Cruz Sanchez, 2013)
developed simple and effective methods for classifying lung cancer without a segmentation stage to
classify the used lung cancer SVM. These methods resulted in 84% accuracy. (Sivakumar &
Chandrasekar, 2013) presented an effective scheme for detecting lung cancer by performing nodule
segmentation. Classification SVM was used for this approach, and an accuracy of 80.36% was obtained.
Based on the above results, SVM is used to classify lung tissues through CT images and yields good
accuracy above 80%. (Kanakatte, Mani, Srinivasan, & Gubbi, 2008) presented an automated process that
uses K-NN and SVM classifier to analyze performance for volume and tumor delineation features from
PET lung images and obtained a high accuracy of 97%. SVM yields high accuracy with PET lung images
compared with CT images. Comparative analysis was carried out for presenting all detection and
classification methods, as shown in Table 2.

Result We take the highest classifier accuracy from each technique. Figure 4 shows the accuracy for
lung cancer detection and classification as follows. (Fig.4) shows that the most accurate result (i.e.,
96.04%) for CT scan is achieved using the feedforward backpropagation (Ada¹, 2013). Feedforward
neural networks that are capable of classifying cancer cases with high accuracy rate have become an
effective tool. Computation time is fixed, and extremely high computation speed results from the
parallel structure. Moreover, the approach is fault-tolerant because of the distributed nature of network
knowledge. General solutions can be learned from presented training data. Neural networks eliminate
the requirement to produce an explicit model of a process. Moreover, these networks can easily model
parts of a process that cannot be modeled or even usually unidentified. A neural network could learn
from incomplete and noisy data, but the solution may be less precise. The capability to generalize
situations is not trained to the network previously (G. B. Huang, Chen, & Siew, 2006). Similar with any
method, neural networks present specific disadvantages. As neural networks find a general
approximation of a solution, smaller error is generally associated with neural network results. The
complete nature of neural networks is not completely understood; therefore, it is proposed that future
research should consider an experimental method to assess the issue of efficiency. At present, no neural
network computers available at a reasonable cost are available. Moreover, neural network errors vary
based on the architecture. Finally, neural networks require long training periods.The use of CT images
for detecting lung cancer offers numerous advantages. Small nodules in lung cancer can be simply
identified using CT tests, and chest CT is particularly efficient for diagnosing lung cancer in the early
stages, or the nearly curable phase. Issues identified with CT screening include false-positive scans,
diagnosis of benign nodule resections, influence of radiation, and cost. As low dose CT has been proven
to minimize mortality, the issues of CT screening are recognized as manageable, with the possible
exception of cost (Gould et al., 2013). With PET images, greater accuracy of 97% is accomplished
through the use of the SVM classifier (Kanakatte et al., 2008). Numerous advantages of SVMs include
efficiency with high-dimensional spaces, maintained efficiency even when the number of dimensions
exceeds the sample number, and memory efficiency owing to the use of a subset of training points in
the decision function (called support vector). Versatility: various kernel functions can be specified for the
decision function. Different kernels, such as polynomial or linear kernels, can be utilized in SVM models,
but custom kernels can also be used for specification. Several advantages for using SVM in detecting and
classifying lung cancer are available, but certain disadvantages are also present. When the amount of
features is markedly higher than the number of samples, the technique will probably provide weak
performance. SVMs do not immediately offer probability estimates; SVMs are computed using a costly
fivefold cross-validation. PET imaging is unique, displaying the actual chemical function, involving
tissues and organs. Through various other imaging methods, such as MRI and CT display structure,
improved accurate attenuation correction and sensitivity can be supplied by the PET imaging modality.
The drawbacks of the approach include the use of ionizing radiation, high cost and rarity, care in using
radio-nuclides, and poor resolution. ZeroR classifier, which possesses the lowest accuracy of 53.30%,
can be the simplest classification technique that depends upon the target and disregards all predicators.
The ZeroR classifier simply predicts almost all classification classed. Despite the lack of predictability
power in ZeroR, the classifier is helpful in determining a baseline efficiency, such as a benchmark
intended for other classification techniques.

Figure .4. Performance Comparison of different AI Techniques for Lung Cancer

Table 2. Summary of Lung Cancer Detection and Classification Works

Brain Cancer The central nervous system includes the brain and spinal cord. Brain tumor is probably the
major driving force behind death occurrence from cancer. Brain tumors are also known as gliomas and
meningioma. Two essential types of brain tumor include brain cancer and tumors that start in the brain.
Cancer cells can spread and enter healthy cells in the brain and spinal cord but not often spread to other
body parts. Secondary brain tumor is a more common type. The cancer begins within a different part of
the human body, like lung cancer or breast cancer, and spreads to the brain. This tumor is known as a
metastatic brain tumor. Brain cancer is likely curable and treatable if detected at the initial stages of the
disease. Without treatment, brain tumors can spread and cause death. Various methods are used for
obtaining images of human brain. These types of methods include X-ray, CT, electroencephalogram
(EEG) signal, and MRI. These methods are used for diagnosis.

ANN Supervised (Ibrahim, Osman, & Mohamed, 2013; S. Jain & Mishra; Mahajani), and (Kharat,
Kulkarni, & Nagori, 2012) presented a system for the diagnosis and classification of brain tumor from
MRI images by using a backpropagation network and obtained different accuracy rates. Jain and
Mahajani obtained accuracies of 77.56% and 72.5%, whereas Ibrahim obtained 96.33%. The differences
in accuracy could be due to the different dataset used. Therefore, different features are extracted, such
as the number of images that Ibrahim used was more than that of Jain and Mahjani. In addition, the
neural network architecture and learning algorithms are different.(Karameh & Dahleh, 2000) developed
a system to identify brain cancer by using EEG signals with a multi SOM. (Al-Naami, Mallouh, & Hafez,
2014)proposed multi-layer feedforward neural networks algorithms using MRI images and obtained
accuracy rate of 86.9%. (S. Jain & Mishra, 2013), (Sapra, Singh, & Khurana, 2013), (SubbaRao, Revanth, &
UdayKumar, 2013), and (Othman & Basri, 2011) presented a system as a diagnostic tool to identify brain
tumor from MRI imaging by using PNN. Jain and Sapra obtained a high accuracy rate (98.08% and 100%).
(Al-Naami et al., 2014) proposed neural network methods for classifying brain tumor from MRI images,
and both techniques obtained a high accuracy rate. NARX neural network obtained 99.1% accuracy, and
Elman obtained 98.1% accuracy. (Mahmood & Abd-Alsalam, 2014) presented a classification technique
for identifying brain tumor from MRI images. A classification process was initially performed using a FIS,
followed by feedforward neuralnetwork. The hybrid method yielded a high accuracy of 95.66%.

Unsupervised (Hemanth, Anitha, & Balas, 2015) and (Goswami & Bhaiya, 2013) presented an
unsupervised learning-based neural network to classify brain tumor from MRI human brain images. Both
presented an accuracy of more than 95%. Hemanth proposed a hybrid classification system that
combines practical swarm optimization and Kohonen neural network. The system achieved 95%
accuracy. Goswami used SOM, and 98.6% change was achieved using this method.

ANFIS Numerous studies have adopted the ANFIS method (Darvishi & AlAni, 2007; Joshi, Rana, & Misra,
2010) for detecting and classifying brain cancer. (Hemanth, Vijila, & Anitha, 1995) and (Al-Naami et al.,
2014) proposed the ANFIS system for brain tumor classification from MRI images by using
backpropagation and least-squares technique. For training algorithms, the accuracies were 93.3% and
89.7%, respectively; these results were obtained using different datasets, ANN architecture, and number
of training. Testing samples of all these factors could explain the decreasing accuracy in 2014.
(Malakooti, Mousavi, & Taba, 2013), (Deshmukh & Khule, 2014), and (Kumar & Kumar, 2008)) proved
that ANFIS with backpropagation learning algorithm exhibited good accuracy on segmented MRI image
tumors. Malakooti obtained an accuracy rate of 89.1%. Identical ANFIS techniques and data image type
(MR) are used for the detection and classification of brain tumor, but the use of different training
algorithms (gradient descent and backpropagation) yielded better accuracy of more than 90%.(Sharma,
2012), (Bhardwaj & Siddhu) and (Basri, Othman, & Husain, 2013) proposed Gradient Descent and
Backpropagation algorithms and obtained accuracy rates of 98.67%, 94%, and 94.67%, respectively.

FL Numerous studies have been conducted on the detection and classification of brain cancer through
FL. (Padmapriya & Maragatham, 2013), (Dasgupta, 2012), (Anandgaonkar & Sable, 2013), and (Fazel
Zarandi, Zarinbal, & Izadi, 2011) agreed that FL is an effective tool.

Other Classifiers In this paper ,(Zacharaki et al., 2009) used SVM for classifying brain tumor from
mixedperfusion (MRI), and the accuracy was 87%. (Dhanalakshmi & Rajamani, 2013) showed a fuzzy
support vector (fuzzy SVM) to enhance diagnosis using CT scan brain images. The results show that the
method achieves good accuracy of up to 88%. A new step-wise procedure was proposed by
(Karuppathal & Palanisamy, 2006) for detecting and classifying brain cancer in MRI image. In this
procedure, a fuzzy-KNN classifier was used with the MRI images, and the resulting accuracy from this
approach was 99%. In a different research,(Mahajani) developed a system for classifying and detecting
brain tumor from MRI images with KNN and 70% accuracy was achieved. (Hiran & Doshi, 2013; Joshi et
al., 2010; Kathalkar, Kawitkar, & Chopade, 2013), (Madhusudhanareddy & Prabha, 2013; Nalbalwar,
Majhi, Patil, & Gonge) proposed a system for detecting and classifying brain tumor from MRI images
using ANN algorithm for various affected people. Different image processing methods, such as image
segmentation, histogram equalization image enhancement, morphological operation, and feature
extraction, are used (Joshi et al., 2010). A comparative analysis was carried out to present all detection
and classification methods, as shown in Table 3.

Result We take the highest classifier accuracy from each technique. (Fig.5) shows the accuracy for the
detection and classification of brain cancer as follows. As shown in (Fig.5), the most accurate result, that
is, 100%, is achieved using PPN, which (Sapra et al., 2013) approved to be a powerful tool. This
technique can classify brain cancer cases with high accuracy rate. Several advantages for using PNN
include the rapid training process, inherently parallel structure, insensitivity to outliers, and the ability to
compute nonlinear decision boundaries, and the possibility of addition or removal of training samples
without extensive retraining. Contrary to those advantages, the disadvantages for PNN include large
memory requirements, a representative training set, and slow execution of network when classifying
new cases. MRI has developed into a widespread, high-quality medical imaging. Especially within brain
imaging, MRI is helpful for scanning and detecting abnormalities within soft tissue structures. MRI
provides an unparalleled view inside the human body and is utilized mainly in medical settings to
provide high quality images inside the human body. The amount of details is considerable with the use
of another imaging modality. Any type of radiation is not involved in the MRI. Several disadvantages of
using MRI include the scanning noise because of operator performance, which can cause serious
inaccuracies in classification, and the high cost of MRI scanners. K-NN applied on MRI images presented
the lowest accuracy of 70%. K-NN classification rule (Mahajani) is one of the most well-known and
widely used nonparametric pattern classification method. K-NN is a simple supervised classifier that
provides good efficiency for optimal values of K. Moreover, in K-NN, training is extremely fast and any
learning task is easy.

In addition, the approach is effective for noisy training data and efficient with large training data.
Disadvantages of this approach include computation complexity and memory limitation. In addition, as a
supervised learning lazy algorithm, K-NN runs slowly and is easily misled by irrelevant attributes.

Figure .5. Performance Comparison of different AI Techniques for Brain Cancer Table 3. Summary of
Brain Cancer Detection and Classification Works
Conclusion: Accurate cancer classification is significant in saving the lives of many humans. Despite the
use of known diagnostic tools, many researchers are currently interested in using AI classification
techniques to classify cancer. This type of study was conducted to compare performance associated
with AI classification techniques with cancer classification data, such as ANN techniques, ANFIS, FL, and
SVM neural network. The techniques are efficient tools for classifying cancer data. We found that
mammograms yielded the most accurate results for breast cancer, achieving 99.73% with the use of FL
simple grid classifier. In lung cancer, PET images presented higher accuracy of 97%, which was achieved
using an SVM classifier, and CT scan obtained the most accurate result of 96.04%, which was achieved
using feedforward backpropagation. For brain cancer, MRI scan yielded the most accurate result of
100%, which was achieved using a PPN. Numerous AI techniques can be used to diagnose cancers and
obtain different accuracies using different types of data. We found the best classification algorithm and
the best medical image type with the highest accuracy for the detection and classification of breast,
lung, and brain cancers.

ANFIS

Backpropagation and a least Squares Technique

93.30% NA NA NA MRI

(Hemanth, Vijila, & Anitha, 1995)

1995
89.70% 95.60% NA 79.50% MRI

(Al-Naami et al., 2014)

2014

Gradient Descent and Backpropagation Algorithms

98.67% 96.60% NA 95% MRI

(Sharma, 2012)

2012

94.00% 96% NA 93% MRI (Bhardwaj & Siddhu) 2013

94.67% 96% NA 93.30% MRI

(Basri, Othman, & Husain, 2013)

2013

Backpropagation Algorithm 89.10% NA NA NA MRI

(Malakooti, Mousavi, & Taba, 2013)

2013

Other Classifier

Fuzzy K-NN 99% 99.00% NA 98.97% MRI

(Karuppathal & Palanisamy, 2006)

2006

SVM 87.00% 89.00% NA 79%

(MRI) and Perfusion MRI

(Zacharaki et al., 2009) 2009

Fuzzy SVM 87.90% 86.60% NA 89.20% CT

(Dhanalakshmi & Rajamani, 2013)

2013

K-NN 70.00% NA NA NA MRI (Mahajani) 2013 ANFIS, adaptive neuro-fuzzy inference system
SVM ,Support Vector MachineK-NN , K- Nearest Neighbour

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

363
References: 1. Ada¹, R. K. (2013). Early Detection and Prediction of Lung Cancer Survival using Neural
Network Classifier: IJAIEM. 2. Al-Daoud, E. (2010). Cancer diagnosis using modified fuzzy network.
Universal J. Comput. Sci. & Engg. Technol, 1(2), 73-78. 3. Al-Naami, B., Mallouh, M. A., & Hafez, E. A.
(2014). Performance Comparison of Adaptive Neural Networks and Adaptive NeuroFuzzy Inference
System in Brain Cancer Classification. JJMIE, 8(5). 4. Al-Timemy, A. H., Al-Naima, F. M., & Qaeeb, N. H.
(2009). Probabilistic neural network for breast biopsy classification. Paper presented at the
Developments in eSystems Engineering (DESE), 2009 Second International Conference on. 5.
Anandgaonkar, G. P., & Sable, G. S. (2013). Detection and Identification of Brain Tumor in Brain MR
Images Using Fuzzy CMeans Segmentation. International Journal of Research in Computer and
Communication Engineering Vol. 2 Issue, 10. 6. Azar, A. T., & El-Said, S. A. (2013). Probabilistic neural
network for breast cancer classification. Neural Computing and Applications, 23(6), 1737-1751. 7. Basri,
M. A. M., Othman, M. F., & Husain, A. R. (2013). An Approach to Brain Tumor MR Image Detection and
Classification using Neuro Fuzzy. Jurnal Teknologi, 61(2). 8. Bhardwaj, A., & Siddhu, K. K. An Approach to
Medical Image Classification Using Neuro Fuzzy Logic and ANFIS Classifier. 9. Cheng, H. D., Cai, X. P.,
Chen, X. W., Hu, L. M., & Lou, X. L. (2003). Computer-aided detection and classification of
microcalcifications in mammograms: a survey. Pattern Recognition, 36(12), 2967-2991.
doi:10.1016/s0031-3203(03)00192-4 10. Cheng, H. D., & Cui, M. (2004). Mass lesion detection with a
fuzzy neural network. Pattern Recognition, 37(6), 1189-1200. doi:10.1016/j.patcog.2003.11.002 11.
Cheng, H. D., Shan, J., Ju, W., Guo, Y. H., & Zhang, L. (2010). Automated breast cancer detection and
classification using ultrasound images: A survey. Pattern Recognition, 43(1), 299-317.
doi:10.1016/j.patcog.2009.05.012 12. Cheng, H. D., Shi, X. J., Min, R., Hu, L. M., Cai, X. R., & Du, H. N.
(2006). Approaches for automated detection and classification of masses in mammograms. Pattern
Recognition, 39(4), 646-668. doi:10.1016/j.patcog.2005.07.006 13. Dalerba, P., Cho, R. W., & Clarke, M.
F. (2007). Cancer stem cells: models and concepts. Annu. Rev. Med., 58, 267-284.

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

364

14. Darvishi, S., & Al-Ani, A. (2007). Brain-computer interface analysis using continuous wavelet
transform and adaptive neuro-fuzzy classifier. Paper presented at the Engineering in Medicine and
Biology Society, 2007. EMBS 2007. 29th Annual International Conference of the IEEE. 15. Dasgupta, A.
(2012). Demarcation of brain tumor using modified Fuzzy C-Means. International Journal of Engineering
Research and Applications, 2(4), 529-533. 16. Deshmukh, R., & Khule, R. (2014). Brain Tumor Detection
Using Artificial Neural Network Fuzzy Inference System (ANFIS). International Journal of Computer
Applications Technology and Research, 3(3), 150-154. 17. Dhanalakshmi, K., & Rajamani, V. (2013). An
Efficient Decision Support System for Diagnosing Brain Tumor Images. International Journal of Fuzzy
Systems, 15(2), 254-261. 18. Dhokia, V., Kumar, S., Vichare, P., Newman, S., & Allen, R. (2008). Surface
roughness prediction model for CNC machining of polypropylene. Proceedings of the Institution of
Mechanical Engineers, Part B: Journal of Engineering Manufacture, 222(2), 137-157. 19. Doi, K. (2007).
Computer-aided diagnosis in medical imaging: Historical review, current status and future potential.
Computerized Medical Imaging and Graphics, 31(4-5), 198-211.
doi:10.1016/j.compmedimag.2007.02.002 20. Duncan, A., Sacks, S., Melnick, G., Cleland, C. M., Pearson,
F. S., & Coen, C. (2008). Performance of the CJDATS co-occurring disorders screening instruments
(CODSIs) among minority offenders. Behavioral sciences & the law, 26(4), 351. 21. Fatima, B., & Amine,
C. M. (2012). A neuro-fuzzy inference model for breast cancer recognition. International Journal of
Computer Science & Information Technology, 4(5), 163. 22. Fawcett, T. (2006). An introduction to ROC
analysis. Pattern recognition letters, 27(8), 861-874. 23. Fazel Zarandi, M. H., Zarinbal, M., & Izadi, M.
(2011). Systematic image processing for diagnosing brain tumors: A Type-II fuzzy expert system
approach. Applied Soft Computing, 11(1), 285-294. doi:http://dx.doi.org/10.1016/j.asoc.2009.11.019 24.
George, Y. M., Elbagoury, B. M., Zayed, H. H., & Roushdy, M. I. (2012). Breast Fine Needle Tumor
Classification using Neural Networks. International Journal of Computer Science Issues(IJCSI), 9(5).

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

365

25. GÖRGEL, P., SERTBAŞ, A., & UÇAN, O. N. (2012). FEATURE EXTRACTION BASED WAVELET
TRANSFORM IN BREAST CANCER DIAGNOSIS USING FUZZY AND NON-FUZZY CLASSIFICATION. 26.
Gorynski, K., Safian, I., Gradzki, W., Marszall, M. P., Krysinski, J., Gorynski, S., . . . Bucinski, A. (2014).
Artificial neural networks approach to early lung cancer detection. Central European Journal of
Medicine, 9(5), 632-641. doi:10.2478/s11536-013-0327-6 27. Goswami, S., & Bhaiya, L. K. P. (2013).
Brain tumour detection using unsupervised learning based neural network. Paper presented at the
Communication Systems and Network Technologies (CSNT), 2013 International Conference on. 28.
Gould, M. K., Donington, J., Lynch, W. R., Mazzone, P. J., Midthun, D. E., Naidich, D. P., & Wiener, R. S.
(2013). Evaluation of individuals with pulmonary nodules: When is it lung cancer?: Diagnosis and
management of lung cancer: American College of Chest Physicians evidence-based clinical practice
guidelines. CHEST Journal, 143(5_suppl), e93S-e120S. 29. Hashemi, A., Pilevar, A. H., & Rafeh, R. (2013).
Mass Detection in Lung CT Images Using Region Growing Segmentation and Decision Making Based on
Fuzzy Inference System and Artificial Neural Network. International Journal of Image, Graphics and
Signal Processing (IJIGSP), 5(6), 16. 30. Hemanth, D. J., Anitha, J., & Balas, V. E. (2015). Performance
Improved Hybrid Intelligent System for Medical Image Classification. Paper presented at the Proceedings
of the 7th Balkan Conference on Informatics Conference. 31. Hemanth, D. J., Vijila, C. K. S., & Anitha, J.
(1995). Application of Neuro-Fuzzy Model for MR Brain Tumor Image. 32. Hiran, K. K., & Doshi, R.
(2013). An artificial neural network approach for brain tumor detection using digital image
segmentation. Brain, 2(5). 33. Huang, G. B., Chen, L., & Siew, C. K. (2006). Universal approximation using
incremental constructive feedforward networks with random hidden nodes. Ieee Transactions on Neural
Networks, 17(4), 879-892. doi:10.1109/tnn.2006.875977 34. Huang, Z. X. (1998). Extensions to the k-
means algorithm for clustering large data sets with categorical values. Data Mining and Knowledge
Discovery, 2(3), 283-304. doi:10.1023/a:1009769707641 35. Ibrahim, W. H., Osman, A. A. A., &
Mohamed, Y. I. (2013). MRI Brain Image Classification using neural networks. Paper presented at

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

366

the Computing, Electrical and Electronics Engineering (ICCEEE), 2013 International Conference on. 36.
Jain, R., & Abraham, A. (2004). A comparative study of fuzzy classification methods on breast cancer
data. Australasian Physics & Engineering Sciences in Medicine, 27(4), 213-218. 37. Jain, S., & Mishra, S.
ANN Approach Based On Back Propagation Network and Probabilistic Neural Network to Classify Brain
Cancer. Paper presented at the ases. 38. Jain, S., & Mishra, S. (2013). ANN Approach Based On Back
Propagation Network and Probabilistic Neural Network to Classify Brain Cancer. Paper presented at the
ases. 39. Janghel, R., Shukla, A., Tiwari, R., & Kala, R. (2010). Breast cancer diagnosis using artificial
neural network models. Paper presented at the Information Sciences and Interaction Sciences (ICIS),
2010 3rd International Conference on. 40. Joshi, D. M., Rana, N., & Misra, V. (2010). Classification of
brain cancer using artificial neural network. Paper presented at the Electronic Computer Technology
(ICECT), 2010 International Conference on. 41. Jung, I.-S., Thapa, D., & Wang, G.-N. (2005). Neural
Network Based Algorithms for Diagnosis and Classification of Breast Cancer Tumor. In Y. Hao, J. Liu, Y.
Wang, Y.-m. Cheung, H. Yin, L. Jiao, J. Ma, & Y.-C. Jiao (Eds.), Computational Intelligence and Security
(Vol. 3801, pp. 107-114): Springer Berlin Heidelberg. 42. Kanakatte, A., Mani, N., Srinivasan, B., & Gubbi,
J. (2008). Pulmonary tumor volume detection from positron emission tomography images. Paper
presented at the BioMedical Engineering and Informatics, 2008. BMEI 2008. International Conference
on. 43. Karameh, F. N., & Dahleh, M. A. (2000). Automated classification of EEG signals in brain tumor
diagnostics. Paper presented at the American Control Conference, 2000. Proceedings of the 2000. 44.
Karuppathal, R., & Palanisamy, V. (2006). FUZZY BASED AUTOMATIC DETECTION AND CLASSIFICATION
APPROACH FOR MRI-BRAIN TUMOR. 45. Kathalkar, A. A., Kawitkar, R., & Chopade, A. (2013). Artificial
Neural Network based Brain Cancer Analysis and Classification. International Journal of Computer
Applications, 66(10). 46. Kharat, K. D., Kulkarni, P. P., & Nagori, M. (2012). Brain tumor classification
using neural network based methods. International Journal of Computer Science and Informatics, 1(4).

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

367

47. Kumar, G. J., & Kumar, G. V. (2008). Biological Early Brain Cancer Detection Using Artificial Neural
Networks. Paper presented at the Artificial Intelligence and Pattern Recognition. 48. Kuruvilla, J., &
Gunavathi, K. (2014). Lung cancer classification using neural networks for CT images. Computer Methods
and Programs in Biomedicine, 113(1), 202-209. doi:http://dx.doi.org/10.1016/j.cmpb.2013.10.011 49.
Lichman, M. ((2013)). UCI Machine Learning Repository from Irvine, CA: University of California, School
of Information and Computer Science [http://archive.ics.uci.edu/ml] 50. Madero Orozco, H., Vergara
Villegas, O. O., de Jesus Ochoa Dominguez, H., & Cruz Sanchez, V. G. (2013). Lung Nodule Classification
in CT Thorax Images Using Support Vector Machines. Paper presented at the Artificial Intelligence
(MICAI), 2013 12th Mexican International Conference on. 51. Madhusudhanareddy, P., & Prabha, I. S.
(2013). Novel Approach In Brain Tumor Classification Using Artificial Neural Networks. 52. Mahajani, P.
P. P. Detection and Classification of Brain Tumor in MRI Images. 53. Mahmood, A. F., & Abd-Alsalam, A.
M. (2014). Automatic Brain MRI Slices Classification Using Hybrid Technique. Al-Rafidain Engineering
Journal, 22(3). 54. Malakooti, D. M. V., Mousavi, S. A., & Taba, D. N. H. (2013). MRI Brain Image
Segmentation Using Combined Fuzzy Logic and Neural Networks for Tumor Detection. Journal of
Academic and Applied Studies, 3(5). 55. Memarian, N., Alirezaie, J., & Babyn, P. (2006). Computerized
detection of lung nodules with an enhanced false positive reduction scheme. Paper presented at the
Image Processing, 2006 IEEE International Conference on. 56. Miranda, G. H. B., & Felipe, J. C. (2014).
Computer-aided diagnosis system based on fuzzy logic for breast cancer categorization. Computers in
biology and medicine. 57. Murino, V., Puppo, E., Sona, D., Cristani, M., & Sansone, C. (2015). New
Trends in Image Analysis and Processing--ICIAP 2015 Workshops: ICIAP 2015 International Workshops,
BioFor, CTMR, RHEUMA, ISCA, MADiMa, SBMI, and QoEM, Genoa, Italy, September 7-8, 2015,
Proceedings (Vol. 9281): Springer. 58. Nalbalwar, R., Majhi, U., Patil, R., & Gonge, S. Detection of Brain
Tumor by using ANN. image, 2(3), 7. 59. National Cancer Institute (2012). Cancer Statistics. Retrieved
from http://www.cancer.gov/about-cancer/what-is-cancer/statistics

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

368

60. Nehemiah, H. K., & Kannan, A. (2006). An intelligent system for lung cancer diagnosis from chest
radiographs. International Journal of Soft Computing, 1(2), 133-136. 61. Onan, A. (2015). A fuzzy-rough
nearest neighbor classifier combined with consistency-based subset evaluation and instance selection
for automated diagnosis of breast cancer. Expert Systems with Applications, 42(20), 6844-6852.
doi:http://dx.doi.org/10.1016/j.eswa.2015.05.006 62. Othman, M. F., & Basri, M. A. M. (2011).
Probabilistic neural network for brain tumor classification. Paper presented at the Intelligent Systems,
Modelling and Simulation (ISMS), 2011 Second International Conference on. 63. Padmapriya, A., &
Maragatham, K. S. C. (2013). Priority Based Apriori Algorithm For Cancer Prediction Using Fuzzy
Classification. Paper presented at the International Journal of Engineering Research and Technology. 64.
Peng, Y. H., Jianmin, J., & Wu, Z. Q. (2009). Computational Intelligence on Medical Imaging with Artificial
Neural Networks Computational Intelligence in Medical Imaging (pp. 1-26): Chapman and Hall/CRC. 65.
Pereira, C. S., Alexandre, L. A., Mendonça, A. M., & Campilho, A. (2006). A multiclassifier approach for
lung nodule classification Image Analysis and Recognition (pp. 612-623): Springer. 66. Raad, A., Kalakech,
A., & Ayache, M. (2012). Breast cancer classification using neural network approach: MLP and RBF.
Networks, 7(8), 9. 67. Sapra, P., Singh, R., & Khurana, S. (2013). Brain Tumor Detection Using Neural
Network. International Journal of Science and Modern Engineering (IJISME) ISSN, 2319-6386. 68.
Schaefer, G., Nakashima, T., Zavisek, M., Yokota, Y., Drastich, A., & Ishibuchi, H. (2007). Breast cancer
classification using statistical features and fuzzy classification of thermograms. Paper presented at the
Fuzzy Systems Conference, 2007. FUZZ-IEEE 2007. IEEE International. 69. Schaefer, G., Nakashima, T.,
Zavisek, M., Yokota, Y., Drastich, A., Ishibuchi, H., & Ieee. (2007). Breast cancer classification using
statistical features and fuzzy classification of thermograms 2007 Ieee International Conference on Fuzzy
Systems, Vols 1-4 (pp. 11011105). New York: Ieee, Electron Devices Soc & Reliability Group. 70. Seema
Singh , S. H., Harini J and Surabhi B.R. (2014). An Efficient Neural Network Based System for Diagnosis of
Breast Cancer

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

369

International Journal of Computer Science and information Technologies, 71. 5 (((3)), 4354-4360. 72.
Sharma, M. (2012). Artificial Neural Network Fuzzy Inference System (ANFIS) For Brain Tumor Detection.
arXiv preprint arXiv:1212.0059. 73. Singh, B. K., Verma, K., & Thoke, A. (2015). Adaptive Gradient
Descent Backpropagation for Classification of Breast Tumors in Ultrasound Imaging. Procedia Computer
Science, 46, 1601-1609. 74. Singh, S., Saini, S., & Singh, M. (2012). Cancer detection using adaptive
neural network. International Journal of Advancements in Research & Technology, 1(4), 93-97. 75.
Sivakumar, S., & Chandrasekar, C. (2013). Lung nodule detection using fuzzy clustering and support
vector machines. International Journal of Engineering and Technology (IJET), 5(1), 179-185. 76. Song, D.,
Zhukov, T. A., Markov, O., Qian, W., & Tockman, M. S. (2012). Prognosis of stage I lung cancer patients
through quantitative analysis of centrosomal features. Paper presented at the Biomedical Imaging (ISBI),
2012 9th IEEE International Symposium on. 77. Soria, D., Garibaldi, J. M., Green, A. R., Powe, D. G.,
Nolan, C. C., Lemetre, C., . . . Ellis, I. O. (2013). A quantifier-based fuzzy classification system for breast
cancer patients. Artificial intelligence in medicine, 58(3), 175-184. 78. SubbaRao, M., Revanth, B., &
UdayKumar, D. (2013). Mri Brain Image Classification Using Probabilistic Neural Network And Tumor
Detection Using Clustering Technique. Paper presented at the International Journal of Engineering
Research and Technology. 79. Swathi, S., Rizwana, S., Babu, G. A., Kumar, P. S., & Sarma, P. Classification
Of Neural Network Structures For Brea St Cancer Diagnosis. 80. Taher, F., & Sammouda, R. (2011). Lung
cancer detection by using artificial neural network and fuzzy clustering methods. Paper presented at the
GCC Conference and Exhibition (GCC), 2011 IEEE. 81. Taher, F., Werghi, N., & Al-Ahmad, H. (2012).
Bayesian classification and artificial neural network methods for lung cancer early diagnosis. Paper
presented at the Electronics, Circuits and Systems (ICECS), 2012 19th IEEE International Conference on.
82. Tariq, A., Akram, M. U., & Javed, M. Y. (2013). Lung Nodule Detection in CT images using neuro fuzzy
classifier. Paper presented at the Computational Intelligence in Medical Imaging (CIMI), 2013 IEEE
Fourth International Workshop on.

European Scientific Journal January 2017 edition vol.13, No.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857-
7431

370

83. Ubaidillah, S. H. S. A., Sallehuddin, R., & Ali, N. A. (2013). Cancer Detection Using Aritifical Neural
Network and Support Vector Machine: A Comparative Study. Jurnal Teknologi, 65(1). 84. Übeyli, E. D.
(2009). Adaptive neuro-fuzzy inference systems for automatic detection of breast cancer. Journal of
medical systems, 33(5), 353-358. 85. Zacharaki, E., Wang, S., Chawla, S., Yoo, D. S., Wolf, R., Melhem, E.
R., & Davatzikos, C. (2009). MRI-based classification of brain tumor type and grade using SVM-RFE.
Paper presented at the Biomedical Imaging: From Nano to Macro, 2009. ISBI'09. IEEE International
Symposium on. 86. Zhou, C., Wu, Y.-L., Chen, G., Feng, J., Liu, X.-Q., Wang, C., . . . Ren, S. (2011). Erlotinib
versus chemotherapy as first-line treatment for patients with advanced EGFR mutation-positive non-
small-cell lung cancer (OPTIMAL, CTONG-0802): a multicentre, open-label, randomised, phase 3 study.
The lancet oncology, 12(8), 735-742. 87. Zhou, Z.-H., Jiang, Y., Yang, Y.-B., & Chen, S.-F. (2002). Lung
cancer cell identification based on artificial neural network ensembles. Artificial Intelligence in Medicine,
24(1), 25-36. doi:http://dx.doi.org/10.1016/S0933-3657(01)00094-X