Covid-Chestxray-lambda-fuzzy

Our solution for Novel COVID-19 Chestxray Repository and COVIDx [PAPER]

In this project, we have applied Choquet integral for ensemble of deep CNN models and propose a novel method for the evaluation of fuzzy measures using Coalition Game Theory, Information Theory and Lambda fuzzy approximation. Three different sets of Fuzzy Measures are calculated using three different weighting schemes along with information theory and coalition game theory. Using these three sets of fuzzy measures three Choquet Integrals are calculated and their decisions are finally combined.To the best of our knowledge,our experimental results outperform many state-of-the-art methods.

Table of Contents

• Team Members
• Journal Paper
• Installation
• Dependencies
• Directory Structure
• Method Overview
• Dataset
• Results
• Contact

Team Members

• Subhankar Sen
• Pratik Bhowal Github
• Prof. Jin Hee Yoon, faculty of the Dept. of Mathematics and Statistics at Sejong University, Seoul, South Korea Google Scholar
• Prof. Zong Woo Geem, faculty of College of IT Convergence at Gachon University, South Korea Google Scholar
• Prof. Ram Sarkar, Professor at Dept. of Computer Science Engineering, Jadavpur Univeristy Kolkata, India Google Scholar

Journal Paper

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

@article{bhowal2021choquet,
  title={Choquet Integral and Coalition Game-based Ensemble of Deep Learning Models for COVID-19 Screening from Chest X-ray Images},
  author={Bhowal, Pratik and Sen, Subhankar and Yoon, Jin Hee and Geem, Zong Woo and Sarkar, Ram},
  journal={IEEE Journal of Biomedical and Health Informatics},
  year={2021},
  publisher={IEEE}
}

Installation

1. Make sure you have python3 setup on your system
2. Clone the repo

git clone https://github.com/subhankar01/Covid-Chestxray-lambda-fuzzy

3. Install requirements

pip install -r requirements.txt

Dependencies

Our project is built using Python 3.8.6 and the following packages

numpy==1.19.5
pandas==1.1.5
matplotlib==3.2.2
seaborn==2.5.0
opencv-python==4.2.0.32
tensorflow==2.5.1

Directory Structure

+-- COVID_Xray
|   +-- covid xray dataset
|   |   +-- training set
|   |   +-- test set
|   +-- extracted features
|   +-- labels
|   |  +-- train_labels.npy
|   |  +-- val_labels.npy
|   |  +-- test_labels.npy
|   +-- augment.py
|   +-- deep CNN features (feature extraction scripts)
|   |  +-- incep_extract.py
|   |  +-- xcep_extract.py
|   |  +-- vgg16_extract.py
|   +-- classifier.py
|   +-- lambda_fuzzy_script.py
|   +-- driver.ipynb

Please note that the image labels are generated and stored in the labels folder on execution of incep_extract.py script.

Method Overview

Dataset

We have used the Novel COVID-19 Chestxray Repository for evaluation of our proposed methodology. We have also used our code to show our method performance over the popular COVIDx dataset. Information about the Novel COVID-19 Chestxray Database and its parent image repositories is provided in Table 1

Table 1: Dataset Description

Dataset	COVID-19	Pneumonia	Normal
COVID Chestxray set	521	239	218
COVID-19 Radiography Database	219	1345	1341
Actualmed COVID chestxray dataset	12	0	80
Total	752	1584	1639

Results

Table 2: Results of 3-class classification

Classifier/Ensemble	Validation Accuracy(in %)	Test Accuracy(in %)	Precision(Avg)	Recall(Avg)	AUC
VGG16	96.71	91.22	0.92	0.92	0.92
Xception	97.02	92.98	0.93	0.93	0.92
InceptionV3	97.49	93.48	0.94	0.94	0.94
Choquet Integral (Weight 1)	97.74	94.23	0.94	0.94	-
Choquet Integral (Weight 2)	98.24	94.23	0.94	0.94	-
Choquet Integral (Weight 3)	97.49	93.73	0.95	0.95	-
Ensemble	98.99	95.49	0.96	0.96	0.97

Fig 2:ROC of the 3 DCNN models and proposed ensemble method

Fig 3:Multi-labelled ROC curve of the proposed ensemble method

Fig 4:Confusion Matrix of the proposed method

Contact

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

• Subhankar Sen ([email protected])