Project Details

Disease Predictor is a unified web-based application capable of predicting four major diseases — Pneumonia, Malaria, Liver Disease, and Cardiovascular Disease — using machine learning and deep learning models.

The Flask Web Application

• The Web Application has been built with Flask in the backend and HTML and Bootstrap for the frontend.

• Respective images (.jpg, .jpeg, .png) can be uploaded to get predictions for Pneumonia and Malaria.

• Functionality for preventing upload of file of any format other than .jpg, .jpeg and .png has been included.

• Sample images are included in the Sample images folder.

Execution

• The prerequisites for running the Flask Application are included in the requirements.txt file.

• To run the application:

export FLASK_APP=run.py

export FLASK_ENV=development

flask run

The Models

Pneumonia Model

• The Chest X-Ray Images (Pneumonia) dataset from Kaggle has been used to train this model.

• This model has been trained to identify 2 classes (Positive or negative for Pneumonia) using transfer learning on the InceptionV3 model till layer ‘mixed7’, followed by a Dense layer with 512 nodes (RelU), and a sigmoid layer with 1 output node using Keras with Tensorflow backend.

• It was trained using the RMSprop optimizer with a batch size of 64. Input size of the images were (150, 150, 3). The images were rescaled before training. (trainPneumonia.ipynb)

• The final trained model resulted in an accuracy of 85.2% on the test set with 777 images.

• Each image is resized to 150x150 and then normalized before feeding into the network to make a prediction.

Malaria Model

• The Malaria Cell Images Dataset dataset from Kaggle has been used to train this model.

• It was trained using the public Kaggle notebook Detecting Malaria (val accuracy > 97%).