COVID-19 Image Classification using Deep Learning Techniques

Introduction

This repository contains the code and data for a research project focused on using deep learning techniques to classify COVID-19 X-ray and CT-Scan images. The goal of the project was to create a model that could accurately differentiate COVID-19 cases from normal cases and other negative cases. The research paper is available in the paper folder.

This is a project with a working website integrated with a CNN model to make predictions whether a patient is covid-19 positive or not. (http://13.233.123.254:8080/)

Requirements

Python 3.8
TensorFlow 2.2.0
NumPy
OpenCV
Matplotlib
Flask

Dataset

The datasets used for this project are:

COVID-19 CT-SCAN Database. This dataset contains CT-scan images of COVID-19 cases, and normal cases.

Architecture

Architecture of Binary Classifier

architecture

Code

The code folder contains the code for training and evaluating the model. There is one notebook:

covid-19 analysis using transfer learning.ipynb: Preprocesses the dataset, prepares it for training and evaluates the models.

Results

The trained model achieved an accuracy of 96.15% on the test set. The confusion matrix is shown below: | | COVID-19 | Normal | |-----------|----------|--------| | COVID-19 | 0.96 | 0.04 | | Normal | 0.04 | 0.96 |

Conclusion

The results of this research project show that deep learning techniques can be effective in classifying COVID-19 cases from normal cases and other pneumonia cases using X-ray images.

The DenseNet201 model with modified head converged to classifiy the prositive and the negative cases of COVID19 with 96% accuracy, 96% recall (macro avg) and 96% precision (macro avg).