Detecting COVID-19 From Chest X-Ray Images using CNN

Last Updated : 15 Mar, 2024

A Django Based Web Application built for the purpose of detecting the presence of COVID-19 from Chest X-Ray images with multiple machine learning models trained on pre-built architectures. Three different machine learning models were used to build this project namely Xception, ResNet50, and VGG16. The Deep Learning model was trained on a publicly available dataset, the SARS-COV-2-Ct-Scan Dataset. The purpose of this project is to apply Convolutional Neural Network (CNN) Architectures in solving problems of the pandemic on a preliminary stage.

Tools and Technologies Used

Some important libraries and technologies used are listed below

Programming Language: Python
Web Framework: Django
Machine Learning Framework: Tensorflow
Frontend Dev: HTML, CSS (BootStrap)
Essential Libraries: keras, sklearn, venv, seaborn, matplotlib

A detailed list of all the libraries can be found here.

Step by Step Implementation

Deep Learning Part

1) Convert Dataset into Dataframe

Convert the data into a pandas dataframe with the corresponding columns

File [Image File]
DiseaseID [Serial Number]
DiseaseType [COVID, non-COVID]

2) Read and Preprocess the Dataframe

Read the Images
Convert the Images into the standard size of (64 x 64)
Create numpy arrays for input/output X_Train & Y_Train
Normalize the RGB values by dividing with 255.

3) Split the Dataset into Train/Validation

Split into Train, Validation datasets
Select split percentage & random state accordingly

4) Define Model Architecture

We are going to import three different architectures listed below :
- VGG16
- ResNet50
- Xception
Structure of Model Architecture
- Conv2D of Input Shape (3,3)
- ResNet50/Xception/VGG16 Architecture
- Add a GlobalAveragePooling2D()
- Add a Dropout Layer
- Final DenseNet Layer with relu activation
- For Multiple Output add a Softmax layer
Use an 'adam' optimizer, hyperparameters can be tuned accordingly
The following code proposes a sample code for building a model

5) Train the Model

Call build_model() function
Use an annealer, a callback that monitors a quantity and if no improvement is seen for a 'patience' number of epochs, the learning rate is reduced.
Use an ImageDataGenerator to carry out real time image data augmentation
Train the model on x_train, y_train
Save the model weights in .hdf5 format and model graph in .json format

Building the Web App

Create a Django Project with an application initialized inside it which would be using the saved model weights to predict uploaded Chest X-Ray Images
Create a basic Static Page with a form to send the image file to the backend

Inside the views.py folder, handle the uploaded image. Load the model files and send the response back to the front end.
The response would contain the following details
- Model Prediction
- Confidence Score
- Prediction Duration (in s)
Add styling to the frontend using CSS (Bootstrap) accordingly

Note: Loading multiple models and using model.predict() takes a lot of time and it'd be much more in the absence of GPU services in the Cloud instance. For scaling this application to a higher server load consider using TensorFlow Serving

Demo

A Demo Version of the project built and tested on localhost is demonstrated in the video below

👁 Image

COV-CNN Demo

Applications in Real Life & Future Work

The project built in the previous lines cannot be directly applied, however, a lot of such applications can be built on a similar tangent for serving the purpose of preliminary medical diagnosis based on inputs of patients saving a lot of screening stage costs to the medical industry. The machine learning pipeline presented in the project can be taken up a notch by making it dynamic in nature. By adding more training data dynamically to the model and train it on them to improve its accuracy. The ML Model can be converted into a REST API making the application more robust in nature and scalable. A MySQL Database could be used for storing patient data with diagnosis details and other parameters. I've presented a use-sketch diagram for illustrating the architecture of the application which could be built in the near future to be applied in the medical industry.

👁 Image