Ensemble-LearningCOVID19

This repository contains all the scripts required for data preparation, model training, and the development of the GUI App.

About

COVID-19 continues to have a devastating impact on people’s lives all across the world. To combat this condition, it is critical to test affected patients in a quick, lowcost, and effective manner. Radiological examination, with Chest X-ray being the most readily available and least priced alternative, is one of the most promising stages toward accomplishing this goal. A Deep Convolutional Neural Network (CNN)-based method is presented in this project report to help detect COVID-19 positive patients utilising Chest X-Ray images. The proposed study uses many state-of-the-art CNN models that are well-known for image classification tasks, including DenseNet201, Resnet50V2, and InceptionV3. All have been independently trained to make independent forecasts. The models are then merged to predict a category value using a new weighted average ensemble technique. Publicly available Chest X-ray images of COVID positive and COVID negative cases are used to test the efficacy of the solution. COVID-19 is a new condition, so there is not a lot of information about it. The sorted dataset gathered from Kaggle and used, contains 5200 images of COVID-19 positive patients and COVID-19 negative subjects. The suggested ensemble model outperformed state-of-the-art CNN models with a classification accuracy of 97.33%. In addition, a web-based graphical user interface (GUI) application for public use was created and also Grad-CAM visualization was used to draw attention to key areas in the image where the predictions were made. The web-based tool can be utilised by any medical or non-medical professionals on any computer/device to detect COVID positive patients leveraging Chest X-Ray images in seconds.

Dependencies

• Python 3.7.11
• Tensorflow 2.8.0
• Flask
• Numpy
• Pandas
• Scikit-Learn

Strcture

The project has four directories:

• DataPreparation Script : Contains NPY_creator.py script for making npy arrays out of the images

• Model Training Script :

  • train_model.py - Script for training the models
  • testing.py - Script for getting the performance metrics
  • training.py - Contains functions for ensembling and function for measuring the performance of the ensembler

• GUI :

  • gradcam.py - Script for the GRADCAM implementation
  • main.py - Main application written in python using Flask
  • utils.py - Flask implementation to run the app on a browser

Data Preparation

1. The Image data are kept into separate directories as COVID_19 +ve and COVID_19 -ve.
2. These images are split into separate diretories as train and test.

Thus the directory structure is as:

Images
   |
   ----Train
   |       |
   |       ---- COVID_19 +ve
   |       |
   |       ---- COVID_19 -ve
   |
   -----Test
           |
           ---- COVID_19 +ve
           |
           ---- COVID_19 -ve
   
3. Then run the NPY_creator.py script and the enter the paths according to the prompt.

Training the Model

1. Run the train_model.py script and the enter the paths that the program asks for.
2. The models will be saved in the directory.

Performance

After training the model, the accuracy, confusion matrix will be printed in the console.

The Application

Run the main.py script and copy and paste the link generated in the console on a browser.
Or simply see it on: http://localhost:5000/