subhankar01 / fuzzyBACH

Our solution for ICIAR 2018 Grand Challenge BACH dataset

Fuzzy-Ensemble-of-Deep-Learning-Models-for-Breast-Cancer-Histology-Classification

Our solution for ICIAR 2018 Grand Challenge dataset on BreAst Cancer Histology images

In the present work, we have proposed an approach for breast cancer image classification,implemented using Tensorflow and Keras, which at first uses five fine-tuned, pre-trained deep learning models for classification breast cancer histology im-ages. Then a fuzzy ensemble approach is introduced where the confidencescores of the five models are fused using Choquet integral, Coalition game theory and Information theory. The dataset used for evaluating the proposed model is the ICIAR 2018 Grand Challenge on Breast Cancer Histology (popularly known as BACH) images. We have considered both 2-class (Malignant and Benign) and 4-class (Benign, In-situ carcinoma,Invasive carcinoma, and Normal tissue). To the best of our knowledge,our experimental results outperform many state-of-the-art methods.

Table of Contents

• Team Members
• Reference Paper
• Method Overview
• Dataset
• Results
• Dependencies
• Contact

Team Members

• Subhankar Sen LinkedIn
• Pratik Bhowal LinkedIn
• Prof. Juan D. Velasquez Silva,University of Chile LinkedIn ,Google Scholar
• Associate Prof. Ram Sarkar,Jadavpur University,Kolkata LinkedIn , Google Scholar

Reference Paper

If you find this work useful for your publications, please consider citing:

@article{bhowal2021fuzzy,
  title={Fuzzy ensemble of deep learning models using choquet fuzzy integral, coalition game and information theory for breast cancer histology classification},
  author={Bhowal, Pratik and Sen, Subhankar and Silva, Juan D Velasquez and Sarkar, Ram},
  journal={Expert Systems with Applications},
  pages={116167},
  year={2021},
  publisher={Elsevier}
}

Method Overview

Fig 1:

Fig 2:Flowchart of the proposed method

Dataset

Click to access the BACH dataset

Examples of microscopic biopsy images in the dataset: (A) normal; (B) benign; (C) in situ carcinoma; and (D) invasive carcinoma

Table 1: Dataset Overview

Results

Table 2: Results of 2-class classification

Classifier/Ensemble	Validation Accuracy	Test Accuracy
VGG16	100	89
VGG19	99.8	94
Xception	100	95
Inception V3	100	94
InceptionResnetV2	99.7	93
Ensemble	-	96

Table 3: Results of 4-class classification

Classifier/Ensemble	Validation Accuracy	Test Accuracy
VGG16	97	86
VGG19	98	83
Xception	99	91
Inception V3	99	90
InceptionResnetV2	99	91
Ensemble	-	95

Dependencies

• Python3
• Tensorflow
• Keras
• OpenCV

Contact

In case of doubt or further collaboration, feel free to email us ! 😊

• Subhankar Sen ([e-mail])
• Pratik Bhowal ([e-mail])