SlideShare ist ein Scribd-Unternehmen logo
1 von 24
Downloaden Sie, um offline zu lesen
Classification of Breast Masses Using
Convolutional Neural Network as
Feature Extractor and Classifier
Pinaki Ranjan Sarkara
, Deepak Mishraa
and Gorthi R.K.S.S Manyamb
a,b: Indian Institute of Space Science and Technology, Trivandrum
Indian Institute of Technology, Tirupati
Second International Conference on Computer Vision and
Image Processing & Workshop on Multimedia, IIT-Roorkee
10 September, 2017
Outline
1 Introduction
Motivation
Objective
Difficulties
2 Background Theories
Deep Learning
3 Approach
Databases & Pre-processing
Extraction of Region of Interests
Overall Architecture for Mass Classification
Feature Extraction & Classification
4 Results
Results on different databases
Contribution
5 Scope of Future Works
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 2/24
Introduction Motivation
Motivation
Motivation
Breast is the most common site of cancer among women in
India (27% of total). Studies show that nearly 48.45%
patients died in the year 2012ab.
Due to very small non-palpable micro-calcification
clusters/masses, approximately 1% to 20% of breast cancer is
missed by radiologists.
Due to the difficulties of Radiologists to detect
micro-calcification clusters a Computer Aided Diagnosis
(CAD) system is much needed.
a
Trends of breast cancer in India, http://www.breastcancerindia.net/statistics/trends.html.
b
Siegel R.L. Miller; K.D. Jemal; A, “Cancer statistics” 2017,
CA: A Cancer Journal for Clinicians 67(1) 730 (2017).
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 3/24
Introduction Objective
Objective
Objectives
Automatic mass classification from breast mammograms.
Low False Positive Rate and High True Positive Rate in
classification.
Figure 1: Objective in this work
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 4/24
Introduction Difficulties
Difficulties
Figure 2: The levels 1, 2 and 3 of ROIs represents normal, benign and malignant
classes respectively1. From these figures, it is understood that classification between
them is very difficult.
1
S. Beura, B. Majhi, and R. Dash,
“Mammogram classification using two dimensional discrete wavelet transform and gray-level co-occurrence matrix for dete
” Neurocomputing, vol. 154, pp. 114, 2015.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 5/24
Background Theories Deep Learning
Deep Learning
“Deep Learning is an algorithm which has no theoretical limitations of
what it can learn; the more data you give and the more computational
time you provide, the better it is.”
- Geoffrey Hinton, Google
Deep learning maybe loosely defined as an attempt to train a
hierarchy of feature detectors with each layer learning a higher
representation of the preceding layer.
Deep learning discovers intricate structure in large data sets by
using the backpropagation algorithm to indicate how a machine
should change its internal parameters that are used to compute the
representation in each layer from the representation in the previous
layer2
.
2
LeCun, Y., Bengio, Y. and Hinton, G., 2015. Deep learning. Nature, 521(7553), pp.436-444..
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 6/24
Background Theories Deep Learning
Successful Architectures in DL
Many variants of deep learning architectures are being proposed
and some of them are proved to be successful such as:
Convolutional Neural Network (CNN)3
Deep Boltzmann Machine (DBM)4
Deep Belief Networks (DBN)5
Stacked Denoising Auto-encoders (SDAE)6
3
A. Krizhevsky, “Imagenet classification with deep convolutional neural networks”.
4
R. Salakhutdinov and G. E. Hinton, “Deep boltzmann machines, ” in AISTATS, vol. 1, p. 3, 2009.
5
G. E. Hinton, “Deep belief networks, ” Scholarpedia, vol. 4, no. 5, p. 5947, 2009.
6
P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, and P.-A. Manzagol,
“Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion”.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 7/24
Background Theories Deep Learning
Convolutional Neural Network
Figure 3: The CNN architecture is composed hierarchical units and each unit extracts
different level of features. Combining more units will produce deeper network along
with more semantic features.7
7
P.R.Sarkar, Deepak Mishra. and Gorthi R.K.S.S. Manyam,
“Classification of Breast Masses Using Convolutional Neural Network as Feature Extractor and Classifier”. CVIP-2017.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 8/24
Approach Databases & Pre-processing
Databases & Pre-processing
We validated the proposed method using two publicly available
databases i.e. Mammographic Image Analysis Society (MIAS)8
and
CBIS-DDSM9
.
Within 2620 scanned cases based on the magnitude of abnormality,
the abnormal class is divided into two more classes, benign and
malignant. We have taken total 273 benign cases and 273
malignant cases from CBIS-DDSM.
From 322 cases of mini-MIAS database 64 benign and 51 malignant
cases are there in the abnormal class.
Each ROIs are resized to 84 × 84 to reduce the number of network
parameters also we used four rotations rot = {0o
, 90o
, 180o
, 270o
}
and increased the training data three times.
8
J. Suckling, J. Parker, D. Dance, S. Astley, I. Hutt, C. Boggis, I. Ricketts, E. Stamatakis, N. Cerneaz,
S. Kok, and others, “The mammographic image analysis society digital mammogram database,
” in Exerpta Medica. International Congress Series, vol. 1069, pp. 375378, 1994..
9
R. S. Lee and D. R. Francisco Gimenez, Assaf Hoogi, “Curated Breast Imaging Subset of DDSM,
” in The Cancer Imaging Archive, 2016..
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 9/24
Approach Extraction of Region of Interests
Extraction of Region of Interests
In this work, we evaluated our model on pre-segmented ROIs.
CBIS-DDSM database offers segmented ROIs for the
development of CAD systems.
A cropping operation has been applied on MIAS mammogram
images to extract the regions of interests (ROIs) which
contain the abnormalities, excluding the unwanted portions of
the image.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 10/24
Approach Overall Architecture for Mass Classification
Overall Architecture for Mass Classification
Figure 4: Overall architecture of a complete breast mass classification framework. The
detection and localization network detects the mass in the mammogram then a
bounding-box regression is applied to get the mass ROI. The ROI is given to our
network to extract deep features for the classification task.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 11/24
Approach Overall Architecture for Mass Classification
Detailed Parameters of Each Layers
Table 1: Detailed parameters of each layer
Name Filter size Depth of Filter Dropout
Conv1 11 32
ReLU1 1
Pooling1 3
Conv2 11 64
ReLU2 1
Dropout1 0.5
Pooling2 3
Conv3 7 96
ReLU3 1
Dropout2 0.15
Pooling3 2
Conv4 5 128
ReLU4 1
Dropout3 0.25
Pooling4 2
Fc5 1 5000
ReLU5 1
Dropout4 0.5
Fc6 1 1000
ReLU6 1
Fc7 1 2
Softmax 1
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 12/24
Approach Overall Architecture for Mass Classification
Training process
Here we try to minimize an objective function (in this work,
categorical cross entropy) through gradient descent algorithm.
Studies show that stochastic gradient descent (SGD) is very
good at optimizing the objective functions for DL. We have
used SGD for this work.
Parameters are updated using backpropagation algorithm.
After getting satisfactory traing accuracy and loss value, we
ensure that our network is well trained.
The trained network is used for testing new inputs and can be
fine-tuned on another databases like Inbreast, BCDR etc.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 13/24
Approach Feature Extraction & Classification
Feature Extraction & Classification
CNN as classifier
Softmax is used for producing probability scores for each class.
Softmax works on the whole data cloud where SVM works only on
the data near to support vector.
SVM as classifier
We have used our trained CNN network as a deep feature extractor
as well as a classifier.
The deep feature is a 1000 × 1 dimensional feature vector and used
SVM with Radial Basis Function (RBF) kernel.
A grid search method is employed to find the best hyperparameters
w.r.t RBF kernel within a range.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 14/24
Results Results on different databases
Result in MIAS Database
(a) Training and validation accuracy in
mini-MIAS database
(b) Training and validation loss in
mini-MIAS database
Figure 5: Training results in mini-MIAS database
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 15/24
Results Results on different databases
Result in DDSM Database
(a) Training and validation accuracy in
CBIS-DDSM database
(b) Training and validation loss in
CBIS-DDSM database
Figure 6: Training results in CBIS-DDSM database
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 16/24
Results Results on different databases
ROC curve analysis
0.0 0.2 0.4 0.6 0.8 1.0
False Positive Rate
0.0
0.2
0.4
0.6
0.8
1.0
TruePositiveRate
SVM as classifier, AUC = 0.9768
CNN as classifier, AUC = 0.9921
(a) ROC curve of classification in
mini-MIAS database
0.0 0.2 0.4 0.6 0.8 1.0
False Positive Rate
0.0
0.2
0.4
0.6
0.8
1.0
TruePositiveRate
SVM as classifier, AUC = 0.9922
CNN as classifier, AUC = 0.9993
(b) ROC curve of classification in
CBIS-DDSM database
Figure 7: Receiver Operating Characteristics curve analysis
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 17/24
Results Results on different databases
Evaluation metrics in different databases
k-fold Accuracy Sensitivity Specificity
(MIAS) (%) (%) (%)
2 97.701 95.556 100
3 97.637 98.4375 99.82
4 99.422 98.901 100
5 99.074 98.261 100
k-fold Accuracy Sensitivity Specificity
(DDSM) (%) (%) (%)
2 98.627 97.757 99.532
3 98.171 96.375 100
4 98.970 99.078 98.863
5 99.267 98.564 100
Table 2: Classification results using k-fold cross-validation (CNN as classifier)
k-fold Accuracy Sensitivity Specificity
(MIAS) (%) (%) (%)
2 100 100 100
3 100 100 100
4 99.422 98.901 100
5 99.074 98.261 100
k-fold Accuracy Sensitivity Specificity
(DDSM) (%) (%) (%)
2 99.314 98.643 100
3 99.238 98.507 100
4 98.856 97.738 100
5 98.260 96.975 99.975
Table 3: Classification results using k-fold cross-validation (SVM as classifier)
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 18/24
Results Results on different databases
Merits of using CNN for CADs
Why this method?
CNNs are very good at learning intricate strictures from the
high-dimensional data.
It extracts semantic features which simulates the real
diagnosis process of the mammograms.
Our main idea was to show that, the CNN works as the
feature extractor as well as the classifier for this problem.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 19/24
Results Results on different databases
Contribution
References Techniques Database Classification
performance
Xie et. al.10 Gray level features, MIAS 96.0%
textural features DDSM 95.7%
Jiao et. al.11 High & medium level, DDSM 96.7%
deep features
Arevalo et. al.12 CNN, SVM DDSM 96.7%
Beura et. al.13 2D-DWT, GLCM MIAS 98.0%
DDSM 98.8%
Ours Segmented ROIs, MIAS 99.081%
CNN DDSM 99.267%
Table 4: Comparison of classification performances
10
W. Xie, Y. Li, and Y. Ma,
“Breast mass classification in digital mammography based on extreme learning machine, ” Neurocomputing,
vol. 173, pp. 930941, 2016..
11
Z. Jiao, X. Gao, Y. Wang, and J. Li, “A deep feature based framework for breast masses classification,
” Neurocomputing, vol. 197, pp. 221231, July 2016..
12
J. Arevalo, F. A. Gonz alez, R. Ramos-Poll an, J. L. Oliveira, and M. A. G. Lopez,
Convolutional neural networks for mammography mass lesion classification,
in Engineering in Medicine and Biology Society (EMBC),
2015 37th Annual International Conference of the IEEE, pp. 797800, IEEE, 2015..
13
S. Beura, B. Majhi, and R. Dash,
“Mammogram classification using two dimensional discrete wavelet transform and gray-level co-occurrence matrix for dete
” Neurocomputing, vol. 154, pp. 114, 2015..
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 20/24
Scope of Future Works
Some more contributions in this work
Towards end-to-end CAD system
Though in this work, we have used pre-segmented ROIs for
training ans testing of mammograms, later we used a fully
convolutional deep hierarchical saliency map prediction
network.
Here, we treat the masses as salient objects and predicts the
suspicious regions through this trained network.
This leads us to propose an end-to-end breast mass
classification framework.
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 21/24
Scope of Future Works
Scope of Future Works
End-to-end Breast Mass Classification Framework
The main objective of an end-to-end computer aided diagnosis
system is to improve the classification accuracy while minimizing
the false positives. One can improve our novel architecture for
better simulating the diagnosis procedure followed by radiologists
P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 22/24
Questions?
sarkar0499pinaki@gmail.com
Thank you.

Weitere ähnliche Inhalte

Was ist angesagt?

An Introduction to Deep Learning
An Introduction to Deep LearningAn Introduction to Deep Learning
An Introduction to Deep LearningPoo Kuan Hoong
 
Breast cancer diagnosis machine learning ppt
Breast cancer diagnosis machine learning pptBreast cancer diagnosis machine learning ppt
Breast cancer diagnosis machine learning pptAnkitGupta1476
 
Prospects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical ImagingProspects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical ImagingGodswll Egegwu
 
Segmentation of thermograms breast cancer tarek-to-slid share
Segmentation of thermograms breast cancer tarek-to-slid shareSegmentation of thermograms breast cancer tarek-to-slid share
Segmentation of thermograms breast cancer tarek-to-slid shareTarek Gaber
 
Pneumonia Detection using CNN
Pneumonia Detection using CNNPneumonia Detection using CNN
Pneumonia Detection using CNNYashIyengar
 
Image Classification using deep learning
Image Classification using deep learning Image Classification using deep learning
Image Classification using deep learning Asma-AH
 
2.6 support vector machines and associative classifiers revised
2.6 support vector machines and associative classifiers revised2.6 support vector machines and associative classifiers revised
2.6 support vector machines and associative classifiers revisedKrish_ver2
 
Deep learning for medical imaging
Deep learning for medical imagingDeep learning for medical imaging
Deep learning for medical imaginggeetachauhan
 
IRJET- Breast Cancer Prediction using Supervised Machine Learning Algorithms
IRJET- Breast Cancer Prediction using Supervised Machine Learning AlgorithmsIRJET- Breast Cancer Prediction using Supervised Machine Learning Algorithms
IRJET- Breast Cancer Prediction using Supervised Machine Learning AlgorithmsIRJET Journal
 
Final thesis presentation
Final thesis presentationFinal thesis presentation
Final thesis presentationPawan Singh
 
Predictive Analysis of Breast Cancer Detection using Classification Algorithm
Predictive Analysis of Breast Cancer Detection using Classification AlgorithmPredictive Analysis of Breast Cancer Detection using Classification Algorithm
Predictive Analysis of Breast Cancer Detection using Classification AlgorithmSushanti Acharya
 
Deep Learning - CNN and RNN
Deep Learning - CNN and RNNDeep Learning - CNN and RNN
Deep Learning - CNN and RNNAshray Bhandare
 
Feature selection concepts and methods
Feature selection concepts and methodsFeature selection concepts and methods
Feature selection concepts and methodsReza Ramezani
 
Deep Learning in Computer Vision
Deep Learning in Computer VisionDeep Learning in Computer Vision
Deep Learning in Computer VisionSungjoon Choi
 
Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)Gaurav Mittal
 
Deep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural NetworksDeep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural NetworksChristian Perone
 

Was ist angesagt? (20)

An Introduction to Deep Learning
An Introduction to Deep LearningAn Introduction to Deep Learning
An Introduction to Deep Learning
 
Breast cancer diagnosis machine learning ppt
Breast cancer diagnosis machine learning pptBreast cancer diagnosis machine learning ppt
Breast cancer diagnosis machine learning ppt
 
Prospects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical ImagingProspects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical Imaging
 
Segmentation of thermograms breast cancer tarek-to-slid share
Segmentation of thermograms breast cancer tarek-to-slid shareSegmentation of thermograms breast cancer tarek-to-slid share
Segmentation of thermograms breast cancer tarek-to-slid share
 
Pneumonia Detection using CNN
Pneumonia Detection using CNNPneumonia Detection using CNN
Pneumonia Detection using CNN
 
Image Classification using deep learning
Image Classification using deep learning Image Classification using deep learning
Image Classification using deep learning
 
1.Introduction to deep learning
1.Introduction to deep learning1.Introduction to deep learning
1.Introduction to deep learning
 
2.6 support vector machines and associative classifiers revised
2.6 support vector machines and associative classifiers revised2.6 support vector machines and associative classifiers revised
2.6 support vector machines and associative classifiers revised
 
Deep learning for medical imaging
Deep learning for medical imagingDeep learning for medical imaging
Deep learning for medical imaging
 
PhD Defense
PhD DefensePhD Defense
PhD Defense
 
Deep Learning for Computer Vision: Medical Imaging (UPC 2016)
Deep Learning for Computer Vision: Medical Imaging (UPC 2016)Deep Learning for Computer Vision: Medical Imaging (UPC 2016)
Deep Learning for Computer Vision: Medical Imaging (UPC 2016)
 
IRJET- Breast Cancer Prediction using Supervised Machine Learning Algorithms
IRJET- Breast Cancer Prediction using Supervised Machine Learning AlgorithmsIRJET- Breast Cancer Prediction using Supervised Machine Learning Algorithms
IRJET- Breast Cancer Prediction using Supervised Machine Learning Algorithms
 
Final thesis presentation
Final thesis presentationFinal thesis presentation
Final thesis presentation
 
Predictive Analysis of Breast Cancer Detection using Classification Algorithm
Predictive Analysis of Breast Cancer Detection using Classification AlgorithmPredictive Analysis of Breast Cancer Detection using Classification Algorithm
Predictive Analysis of Breast Cancer Detection using Classification Algorithm
 
Deep Learning - CNN and RNN
Deep Learning - CNN and RNNDeep Learning - CNN and RNN
Deep Learning - CNN and RNN
 
Feature selection concepts and methods
Feature selection concepts and methodsFeature selection concepts and methods
Feature selection concepts and methods
 
Deep Learning in Computer Vision
Deep Learning in Computer VisionDeep Learning in Computer Vision
Deep Learning in Computer Vision
 
Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)
 
AlexNet
AlexNetAlexNet
AlexNet
 
Deep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural NetworksDeep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural Networks
 

Ähnlich wie Classification of Breast Masses Using Convolutional Neural Network as Feature Extractor and Classifier

Classification of Breast Cancer Diseases using Data Mining Techniques
Classification of Breast Cancer Diseases using Data Mining TechniquesClassification of Breast Cancer Diseases using Data Mining Techniques
Classification of Breast Cancer Diseases using Data Mining Techniquesinventionjournals
 
Automatic Attendance System using Deep Learning Framework
Automatic Attendance System using Deep Learning FrameworkAutomatic Attendance System using Deep Learning Framework
Automatic Attendance System using Deep Learning FrameworkPinaki Ranjan Sarkar
 
25 17 dec16 13743 28032-1-sm(edit)
25 17 dec16 13743 28032-1-sm(edit)25 17 dec16 13743 28032-1-sm(edit)
25 17 dec16 13743 28032-1-sm(edit)IAESIJEECS
 
Breast cancer detection using ensemble of convolutional neural networks
Breast cancer detection using ensemble of convolutional neural networksBreast cancer detection using ensemble of convolutional neural networks
Breast cancer detection using ensemble of convolutional neural networksIJECEIAES
 
Controlling informative features for improved accuracy and faster predictions...
Controlling informative features for improved accuracy and faster predictions...Controlling informative features for improved accuracy and faster predictions...
Controlling informative features for improved accuracy and faster predictions...Damian R. Mingle, MBA
 
Regularized Weighted Ensemble of Deep Classifiers
Regularized Weighted Ensemble of Deep Classifiers Regularized Weighted Ensemble of Deep Classifiers
Regularized Weighted Ensemble of Deep Classifiers ijcsa
 
Face Recognition for Human Identification using BRISK Feature and Normal Dist...
Face Recognition for Human Identification using BRISK Feature and Normal Dist...Face Recognition for Human Identification using BRISK Feature and Normal Dist...
Face Recognition for Human Identification using BRISK Feature and Normal Dist...ijtsrd
 
Multivariate feature descriptor based cbir model to query large image databases
Multivariate feature descriptor based cbir model to query large image databasesMultivariate feature descriptor based cbir model to query large image databases
Multivariate feature descriptor based cbir model to query large image databasesIJARIIT
 
KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER
KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCERKNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER
KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCERcscpconf
 
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...IRJET Journal
 
A HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETS
A HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETSA HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETS
A HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETSEditor IJCATR
 
Clustering and Classification of Cancer Data Using Soft Computing Technique
Clustering and Classification of Cancer Data Using Soft Computing Technique Clustering and Classification of Cancer Data Using Soft Computing Technique
Clustering and Classification of Cancer Data Using Soft Computing Technique IOSR Journals
 
IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...
IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...
IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...IRJET Journal
 
Feature Subset Selection for High Dimensional Data using Clustering Techniques
Feature Subset Selection for High Dimensional Data using Clustering TechniquesFeature Subset Selection for High Dimensional Data using Clustering Techniques
Feature Subset Selection for High Dimensional Data using Clustering TechniquesIRJET Journal
 
Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...
Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...
Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...IRJET Journal
 
K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...
K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...
K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...ijscmcj
 
APPLICATION OF CNN MODEL ON MEDICAL IMAGE
APPLICATION OF CNN MODEL ON MEDICAL IMAGEAPPLICATION OF CNN MODEL ON MEDICAL IMAGE
APPLICATION OF CNN MODEL ON MEDICAL IMAGEIRJET Journal
 
IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2
 IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2 IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2
IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2IRJET Journal
 

Ähnlich wie Classification of Breast Masses Using Convolutional Neural Network as Feature Extractor and Classifier (20)

Classification of Breast Cancer Diseases using Data Mining Techniques
Classification of Breast Cancer Diseases using Data Mining TechniquesClassification of Breast Cancer Diseases using Data Mining Techniques
Classification of Breast Cancer Diseases using Data Mining Techniques
 
Automatic Attendance System using Deep Learning Framework
Automatic Attendance System using Deep Learning FrameworkAutomatic Attendance System using Deep Learning Framework
Automatic Attendance System using Deep Learning Framework
 
25 17 dec16 13743 28032-1-sm(edit)
25 17 dec16 13743 28032-1-sm(edit)25 17 dec16 13743 28032-1-sm(edit)
25 17 dec16 13743 28032-1-sm(edit)
 
Breast cancer detection using ensemble of convolutional neural networks
Breast cancer detection using ensemble of convolutional neural networksBreast cancer detection using ensemble of convolutional neural networks
Breast cancer detection using ensemble of convolutional neural networks
 
Controlling informative features for improved accuracy and faster predictions...
Controlling informative features for improved accuracy and faster predictions...Controlling informative features for improved accuracy and faster predictions...
Controlling informative features for improved accuracy and faster predictions...
 
Regularized Weighted Ensemble of Deep Classifiers
Regularized Weighted Ensemble of Deep Classifiers Regularized Weighted Ensemble of Deep Classifiers
Regularized Weighted Ensemble of Deep Classifiers
 
[IJET-V2I3P21] Authors: Amit Kumar Dewangan, Akhilesh Kumar Shrivas, Prem Kumar
[IJET-V2I3P21] Authors: Amit Kumar Dewangan, Akhilesh Kumar Shrivas, Prem Kumar[IJET-V2I3P21] Authors: Amit Kumar Dewangan, Akhilesh Kumar Shrivas, Prem Kumar
[IJET-V2I3P21] Authors: Amit Kumar Dewangan, Akhilesh Kumar Shrivas, Prem Kumar
 
Face Recognition for Human Identification using BRISK Feature and Normal Dist...
Face Recognition for Human Identification using BRISK Feature and Normal Dist...Face Recognition for Human Identification using BRISK Feature and Normal Dist...
Face Recognition for Human Identification using BRISK Feature and Normal Dist...
 
Multivariate feature descriptor based cbir model to query large image databases
Multivariate feature descriptor based cbir model to query large image databasesMultivariate feature descriptor based cbir model to query large image databases
Multivariate feature descriptor based cbir model to query large image databases
 
KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER
KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCERKNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER
KNOWLEDGE BASED ANALYSIS OF VARIOUS STATISTICAL TOOLS IN DETECTING BREAST CANCER
 
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
 
A HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETS
A HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETSA HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETS
A HYBRID MODEL FOR MINING MULTI DIMENSIONAL DATA SETS
 
Clustering and Classification of Cancer Data Using Soft Computing Technique
Clustering and Classification of Cancer Data Using Soft Computing Technique Clustering and Classification of Cancer Data Using Soft Computing Technique
Clustering and Classification of Cancer Data Using Soft Computing Technique
 
IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...
IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...
IRJET - Breast Cancer Prediction using Supervised Machine Learning Algorithms...
 
Feature Subset Selection for High Dimensional Data using Clustering Techniques
Feature Subset Selection for High Dimensional Data using Clustering TechniquesFeature Subset Selection for High Dimensional Data using Clustering Techniques
Feature Subset Selection for High Dimensional Data using Clustering Techniques
 
Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...
Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...
Cervical Cancer Detection: An Enhanced Approach through Transfer Learning and...
 
Updated proposal powerpoint.pptx
Updated proposal powerpoint.pptxUpdated proposal powerpoint.pptx
Updated proposal powerpoint.pptx
 
K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...
K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...
K-Medoids Clustering Using Partitioning Around Medoids for Performing Face Re...
 
APPLICATION OF CNN MODEL ON MEDICAL IMAGE
APPLICATION OF CNN MODEL ON MEDICAL IMAGEAPPLICATION OF CNN MODEL ON MEDICAL IMAGE
APPLICATION OF CNN MODEL ON MEDICAL IMAGE
 
IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2
 IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2 IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2
IRJET - Explicit Content Detection using Faster R-CNN and SSD Mobilenet V2
 

Kürzlich hochgeladen

Engineering Drawing focus on projection of planes
Engineering Drawing focus on projection of planesEngineering Drawing focus on projection of planes
Engineering Drawing focus on projection of planesRAJNEESHKUMAR341697
 
A Study of Urban Area Plan for Pabna Municipality
A Study of Urban Area Plan for Pabna MunicipalityA Study of Urban Area Plan for Pabna Municipality
A Study of Urban Area Plan for Pabna MunicipalityMorshed Ahmed Rahath
 
Moment Distribution Method For Btech Civil
Moment Distribution Method For Btech CivilMoment Distribution Method For Btech Civil
Moment Distribution Method For Btech CivilVinayVitekari
 
Tamil Call Girls Bhayandar WhatsApp +91-9930687706, Best Service
Tamil Call Girls Bhayandar WhatsApp +91-9930687706, Best ServiceTamil Call Girls Bhayandar WhatsApp +91-9930687706, Best Service
Tamil Call Girls Bhayandar WhatsApp +91-9930687706, Best Servicemeghakumariji156
 
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdfAldoGarca30
 
GEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLE
GEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLEGEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLE
GEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLEselvakumar948
 
Online food ordering system project report.pdf
Online food ordering system project report.pdfOnline food ordering system project report.pdf
Online food ordering system project report.pdfKamal Acharya
 
Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...
Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...
Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...Arindam Chakraborty, Ph.D., P.E. (CA, TX)
 
data_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfdata_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfJiananWang21
 
Computer Networks Basics of Network Devices
Computer Networks  Basics of Network DevicesComputer Networks  Basics of Network Devices
Computer Networks Basics of Network DevicesChandrakantDivate1
 
"Lesotho Leaps Forward: A Chronicle of Transformative Developments"
"Lesotho Leaps Forward: A Chronicle of Transformative Developments""Lesotho Leaps Forward: A Chronicle of Transformative Developments"
"Lesotho Leaps Forward: A Chronicle of Transformative Developments"mphochane1998
 
Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...
Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...
Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...drmkjayanthikannan
 
PE 459 LECTURE 2- natural gas basic concepts and properties
PE 459 LECTURE 2- natural gas basic concepts and propertiesPE 459 LECTURE 2- natural gas basic concepts and properties
PE 459 LECTURE 2- natural gas basic concepts and propertiessarkmank1
 
DeepFakes presentation : brief idea of DeepFakes
DeepFakes presentation : brief idea of DeepFakesDeepFakes presentation : brief idea of DeepFakes
DeepFakes presentation : brief idea of DeepFakesMayuraD1
 
Thermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - VThermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - VDineshKumar4165
 
School management system project Report.pdf
School management system project Report.pdfSchool management system project Report.pdf
School management system project Report.pdfKamal Acharya
 
Computer Lecture 01.pptxIntroduction to Computers
Computer Lecture 01.pptxIntroduction to ComputersComputer Lecture 01.pptxIntroduction to Computers
Computer Lecture 01.pptxIntroduction to ComputersMairaAshraf6
 
HAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKAR
HAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKARHAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKAR
HAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKARKOUSTAV SARKAR
 
kiln thermal load.pptx kiln tgermal load
kiln thermal load.pptx kiln tgermal loadkiln thermal load.pptx kiln tgermal load
kiln thermal load.pptx kiln tgermal loadhamedmustafa094
 
Double Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torqueDouble Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torqueBhangaleSonal
 

Kürzlich hochgeladen (20)

Engineering Drawing focus on projection of planes
Engineering Drawing focus on projection of planesEngineering Drawing focus on projection of planes
Engineering Drawing focus on projection of planes
 
A Study of Urban Area Plan for Pabna Municipality
A Study of Urban Area Plan for Pabna MunicipalityA Study of Urban Area Plan for Pabna Municipality
A Study of Urban Area Plan for Pabna Municipality
 
Moment Distribution Method For Btech Civil
Moment Distribution Method For Btech CivilMoment Distribution Method For Btech Civil
Moment Distribution Method For Btech Civil
 
Tamil Call Girls Bhayandar WhatsApp +91-9930687706, Best Service
Tamil Call Girls Bhayandar WhatsApp +91-9930687706, Best ServiceTamil Call Girls Bhayandar WhatsApp +91-9930687706, Best Service
Tamil Call Girls Bhayandar WhatsApp +91-9930687706, Best Service
 
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
 
GEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLE
GEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLEGEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLE
GEAR TRAIN- BASIC CONCEPTS AND WORKING PRINCIPLE
 
Online food ordering system project report.pdf
Online food ordering system project report.pdfOnline food ordering system project report.pdf
Online food ordering system project report.pdf
 
Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...
Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...
Navigating Complexity: The Role of Trusted Partners and VIAS3D in Dassault Sy...
 
data_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfdata_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdf
 
Computer Networks Basics of Network Devices
Computer Networks  Basics of Network DevicesComputer Networks  Basics of Network Devices
Computer Networks Basics of Network Devices
 
"Lesotho Leaps Forward: A Chronicle of Transformative Developments"
"Lesotho Leaps Forward: A Chronicle of Transformative Developments""Lesotho Leaps Forward: A Chronicle of Transformative Developments"
"Lesotho Leaps Forward: A Chronicle of Transformative Developments"
 
Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...
Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...
Unit 4_Part 1 CSE2001 Exception Handling and Function Template and Class Temp...
 
PE 459 LECTURE 2- natural gas basic concepts and properties
PE 459 LECTURE 2- natural gas basic concepts and propertiesPE 459 LECTURE 2- natural gas basic concepts and properties
PE 459 LECTURE 2- natural gas basic concepts and properties
 
DeepFakes presentation : brief idea of DeepFakes
DeepFakes presentation : brief idea of DeepFakesDeepFakes presentation : brief idea of DeepFakes
DeepFakes presentation : brief idea of DeepFakes
 
Thermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - VThermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - V
 
School management system project Report.pdf
School management system project Report.pdfSchool management system project Report.pdf
School management system project Report.pdf
 
Computer Lecture 01.pptxIntroduction to Computers
Computer Lecture 01.pptxIntroduction to ComputersComputer Lecture 01.pptxIntroduction to Computers
Computer Lecture 01.pptxIntroduction to Computers
 
HAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKAR
HAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKARHAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKAR
HAND TOOLS USED AT ELECTRONICS WORK PRESENTED BY KOUSTAV SARKAR
 
kiln thermal load.pptx kiln tgermal load
kiln thermal load.pptx kiln tgermal loadkiln thermal load.pptx kiln tgermal load
kiln thermal load.pptx kiln tgermal load
 
Double Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torqueDouble Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torque
 

Classification of Breast Masses Using Convolutional Neural Network as Feature Extractor and Classifier

  • 1. Classification of Breast Masses Using Convolutional Neural Network as Feature Extractor and Classifier Pinaki Ranjan Sarkara , Deepak Mishraa and Gorthi R.K.S.S Manyamb a,b: Indian Institute of Space Science and Technology, Trivandrum Indian Institute of Technology, Tirupati Second International Conference on Computer Vision and Image Processing & Workshop on Multimedia, IIT-Roorkee 10 September, 2017
  • 2. Outline 1 Introduction Motivation Objective Difficulties 2 Background Theories Deep Learning 3 Approach Databases & Pre-processing Extraction of Region of Interests Overall Architecture for Mass Classification Feature Extraction & Classification 4 Results Results on different databases Contribution 5 Scope of Future Works P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 2/24
  • 3. Introduction Motivation Motivation Motivation Breast is the most common site of cancer among women in India (27% of total). Studies show that nearly 48.45% patients died in the year 2012ab. Due to very small non-palpable micro-calcification clusters/masses, approximately 1% to 20% of breast cancer is missed by radiologists. Due to the difficulties of Radiologists to detect micro-calcification clusters a Computer Aided Diagnosis (CAD) system is much needed. a Trends of breast cancer in India, http://www.breastcancerindia.net/statistics/trends.html. b Siegel R.L. Miller; K.D. Jemal; A, “Cancer statistics” 2017, CA: A Cancer Journal for Clinicians 67(1) 730 (2017). P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 3/24
  • 4. Introduction Objective Objective Objectives Automatic mass classification from breast mammograms. Low False Positive Rate and High True Positive Rate in classification. Figure 1: Objective in this work P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 4/24
  • 5. Introduction Difficulties Difficulties Figure 2: The levels 1, 2 and 3 of ROIs represents normal, benign and malignant classes respectively1. From these figures, it is understood that classification between them is very difficult. 1 S. Beura, B. Majhi, and R. Dash, “Mammogram classification using two dimensional discrete wavelet transform and gray-level co-occurrence matrix for dete ” Neurocomputing, vol. 154, pp. 114, 2015. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 5/24
  • 6. Background Theories Deep Learning Deep Learning “Deep Learning is an algorithm which has no theoretical limitations of what it can learn; the more data you give and the more computational time you provide, the better it is.” - Geoffrey Hinton, Google Deep learning maybe loosely defined as an attempt to train a hierarchy of feature detectors with each layer learning a higher representation of the preceding layer. Deep learning discovers intricate structure in large data sets by using the backpropagation algorithm to indicate how a machine should change its internal parameters that are used to compute the representation in each layer from the representation in the previous layer2 . 2 LeCun, Y., Bengio, Y. and Hinton, G., 2015. Deep learning. Nature, 521(7553), pp.436-444.. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 6/24
  • 7. Background Theories Deep Learning Successful Architectures in DL Many variants of deep learning architectures are being proposed and some of them are proved to be successful such as: Convolutional Neural Network (CNN)3 Deep Boltzmann Machine (DBM)4 Deep Belief Networks (DBN)5 Stacked Denoising Auto-encoders (SDAE)6 3 A. Krizhevsky, “Imagenet classification with deep convolutional neural networks”. 4 R. Salakhutdinov and G. E. Hinton, “Deep boltzmann machines, ” in AISTATS, vol. 1, p. 3, 2009. 5 G. E. Hinton, “Deep belief networks, ” Scholarpedia, vol. 4, no. 5, p. 5947, 2009. 6 P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, and P.-A. Manzagol, “Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion”. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 7/24
  • 8. Background Theories Deep Learning Convolutional Neural Network Figure 3: The CNN architecture is composed hierarchical units and each unit extracts different level of features. Combining more units will produce deeper network along with more semantic features.7 7 P.R.Sarkar, Deepak Mishra. and Gorthi R.K.S.S. Manyam, “Classification of Breast Masses Using Convolutional Neural Network as Feature Extractor and Classifier”. CVIP-2017. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 8/24
  • 9. Approach Databases & Pre-processing Databases & Pre-processing We validated the proposed method using two publicly available databases i.e. Mammographic Image Analysis Society (MIAS)8 and CBIS-DDSM9 . Within 2620 scanned cases based on the magnitude of abnormality, the abnormal class is divided into two more classes, benign and malignant. We have taken total 273 benign cases and 273 malignant cases from CBIS-DDSM. From 322 cases of mini-MIAS database 64 benign and 51 malignant cases are there in the abnormal class. Each ROIs are resized to 84 × 84 to reduce the number of network parameters also we used four rotations rot = {0o , 90o , 180o , 270o } and increased the training data three times. 8 J. Suckling, J. Parker, D. Dance, S. Astley, I. Hutt, C. Boggis, I. Ricketts, E. Stamatakis, N. Cerneaz, S. Kok, and others, “The mammographic image analysis society digital mammogram database, ” in Exerpta Medica. International Congress Series, vol. 1069, pp. 375378, 1994.. 9 R. S. Lee and D. R. Francisco Gimenez, Assaf Hoogi, “Curated Breast Imaging Subset of DDSM, ” in The Cancer Imaging Archive, 2016.. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 9/24
  • 10. Approach Extraction of Region of Interests Extraction of Region of Interests In this work, we evaluated our model on pre-segmented ROIs. CBIS-DDSM database offers segmented ROIs for the development of CAD systems. A cropping operation has been applied on MIAS mammogram images to extract the regions of interests (ROIs) which contain the abnormalities, excluding the unwanted portions of the image. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 10/24
  • 11. Approach Overall Architecture for Mass Classification Overall Architecture for Mass Classification Figure 4: Overall architecture of a complete breast mass classification framework. The detection and localization network detects the mass in the mammogram then a bounding-box regression is applied to get the mass ROI. The ROI is given to our network to extract deep features for the classification task. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 11/24
  • 12. Approach Overall Architecture for Mass Classification Detailed Parameters of Each Layers Table 1: Detailed parameters of each layer Name Filter size Depth of Filter Dropout Conv1 11 32 ReLU1 1 Pooling1 3 Conv2 11 64 ReLU2 1 Dropout1 0.5 Pooling2 3 Conv3 7 96 ReLU3 1 Dropout2 0.15 Pooling3 2 Conv4 5 128 ReLU4 1 Dropout3 0.25 Pooling4 2 Fc5 1 5000 ReLU5 1 Dropout4 0.5 Fc6 1 1000 ReLU6 1 Fc7 1 2 Softmax 1 P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 12/24
  • 13. Approach Overall Architecture for Mass Classification Training process Here we try to minimize an objective function (in this work, categorical cross entropy) through gradient descent algorithm. Studies show that stochastic gradient descent (SGD) is very good at optimizing the objective functions for DL. We have used SGD for this work. Parameters are updated using backpropagation algorithm. After getting satisfactory traing accuracy and loss value, we ensure that our network is well trained. The trained network is used for testing new inputs and can be fine-tuned on another databases like Inbreast, BCDR etc. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 13/24
  • 14. Approach Feature Extraction & Classification Feature Extraction & Classification CNN as classifier Softmax is used for producing probability scores for each class. Softmax works on the whole data cloud where SVM works only on the data near to support vector. SVM as classifier We have used our trained CNN network as a deep feature extractor as well as a classifier. The deep feature is a 1000 × 1 dimensional feature vector and used SVM with Radial Basis Function (RBF) kernel. A grid search method is employed to find the best hyperparameters w.r.t RBF kernel within a range. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 14/24
  • 15. Results Results on different databases Result in MIAS Database (a) Training and validation accuracy in mini-MIAS database (b) Training and validation loss in mini-MIAS database Figure 5: Training results in mini-MIAS database P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 15/24
  • 16. Results Results on different databases Result in DDSM Database (a) Training and validation accuracy in CBIS-DDSM database (b) Training and validation loss in CBIS-DDSM database Figure 6: Training results in CBIS-DDSM database P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 16/24
  • 17. Results Results on different databases ROC curve analysis 0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate 0.0 0.2 0.4 0.6 0.8 1.0 TruePositiveRate SVM as classifier, AUC = 0.9768 CNN as classifier, AUC = 0.9921 (a) ROC curve of classification in mini-MIAS database 0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate 0.0 0.2 0.4 0.6 0.8 1.0 TruePositiveRate SVM as classifier, AUC = 0.9922 CNN as classifier, AUC = 0.9993 (b) ROC curve of classification in CBIS-DDSM database Figure 7: Receiver Operating Characteristics curve analysis P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 17/24
  • 18. Results Results on different databases Evaluation metrics in different databases k-fold Accuracy Sensitivity Specificity (MIAS) (%) (%) (%) 2 97.701 95.556 100 3 97.637 98.4375 99.82 4 99.422 98.901 100 5 99.074 98.261 100 k-fold Accuracy Sensitivity Specificity (DDSM) (%) (%) (%) 2 98.627 97.757 99.532 3 98.171 96.375 100 4 98.970 99.078 98.863 5 99.267 98.564 100 Table 2: Classification results using k-fold cross-validation (CNN as classifier) k-fold Accuracy Sensitivity Specificity (MIAS) (%) (%) (%) 2 100 100 100 3 100 100 100 4 99.422 98.901 100 5 99.074 98.261 100 k-fold Accuracy Sensitivity Specificity (DDSM) (%) (%) (%) 2 99.314 98.643 100 3 99.238 98.507 100 4 98.856 97.738 100 5 98.260 96.975 99.975 Table 3: Classification results using k-fold cross-validation (SVM as classifier) P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 18/24
  • 19. Results Results on different databases Merits of using CNN for CADs Why this method? CNNs are very good at learning intricate strictures from the high-dimensional data. It extracts semantic features which simulates the real diagnosis process of the mammograms. Our main idea was to show that, the CNN works as the feature extractor as well as the classifier for this problem. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 19/24
  • 20. Results Results on different databases Contribution References Techniques Database Classification performance Xie et. al.10 Gray level features, MIAS 96.0% textural features DDSM 95.7% Jiao et. al.11 High & medium level, DDSM 96.7% deep features Arevalo et. al.12 CNN, SVM DDSM 96.7% Beura et. al.13 2D-DWT, GLCM MIAS 98.0% DDSM 98.8% Ours Segmented ROIs, MIAS 99.081% CNN DDSM 99.267% Table 4: Comparison of classification performances 10 W. Xie, Y. Li, and Y. Ma, “Breast mass classification in digital mammography based on extreme learning machine, ” Neurocomputing, vol. 173, pp. 930941, 2016.. 11 Z. Jiao, X. Gao, Y. Wang, and J. Li, “A deep feature based framework for breast masses classification, ” Neurocomputing, vol. 197, pp. 221231, July 2016.. 12 J. Arevalo, F. A. Gonz alez, R. Ramos-Poll an, J. L. Oliveira, and M. A. G. Lopez, Convolutional neural networks for mammography mass lesion classification, in Engineering in Medicine and Biology Society (EMBC), 2015 37th Annual International Conference of the IEEE, pp. 797800, IEEE, 2015.. 13 S. Beura, B. Majhi, and R. Dash, “Mammogram classification using two dimensional discrete wavelet transform and gray-level co-occurrence matrix for dete ” Neurocomputing, vol. 154, pp. 114, 2015.. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 20/24
  • 21. Scope of Future Works Some more contributions in this work Towards end-to-end CAD system Though in this work, we have used pre-segmented ROIs for training ans testing of mammograms, later we used a fully convolutional deep hierarchical saliency map prediction network. Here, we treat the masses as salient objects and predicts the suspicious regions through this trained network. This leads us to propose an end-to-end breast mass classification framework. P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 21/24
  • 22. Scope of Future Works Scope of Future Works End-to-end Breast Mass Classification Framework The main objective of an end-to-end computer aided diagnosis system is to improve the classification accuracy while minimizing the false positives. One can improve our novel architecture for better simulating the diagnosis procedure followed by radiologists P. R. Sarkar A deep-CAD system for Breast Mass Classification 10 September, 2017 22/24