Copied from the ML-Capsule repository, where I had contributed as a GSSoC'24 participant.

This project implements different ways of pneumonia prediction from Chest X-Ray images. Used custom built CNN model as well as various pre-trained models. The details of the performances of the above models can be found in the Models directory. The best-fitting model is used for prediction in the web app.

*Adapted version of Paul Mooney's 'Chest X-Ray Images (Pneumonia)' dataset, where the amount of observations for training and validation purposes was redistributed to allow for a more balanced machine learning exercise.

• Data Augmentation: Used ImageDataGenerator class for augmenting images with the following params whose ranges were supplied:

  • brightness_range : Varies brightness

  • height_shift_range : Shifts the image vertically to a random value

  • width_shift_range : Shifts the image horizontally to a random value

  • rotation_range : Rotates the image to a random degree

  • zoom_range : Zooms the image to a random degree

  • preprocessing_function : Passed the preprocessing function of the respective pretrained model being used.

• Built custom CNN model

• Transfer Learning Techniques: VGG-16, ResNet50V2, MobileNet and Xception

• Model Evaluation: Used classification metrics and graphs for visualization

• For running on localhost : Prepare the environment by installing all the requirements from requirements.txt Then type the following command in the terminal : streamlit run app.py

• tensorflow~=2.15.0

• keras~=2.15.0

• matplotlib

• opencv-python

• pillow

• numpy

• scikit-learn

• seaborn

  Available in the requirements.txt file

VGG 16 was performing the best among the other pre-trained models.

VGG 16 classification metrics with graphs :

Loss and Accuracy curves

Confusion matrix	Classification report

• I have originally used Python version 3.11 for this project.

• Install the required packages from the requirements.txt file by using one of the following commands in the terminal:

• pip install -r requirements.txt (for pip environment)

• conda install --yes --file requirements.txt (for conda environment)

The model accuracy could be improved so as to become fairly acceptable for clinical trials and real-life usage. The Pneumonia class can be further classified into Bacterial Pneumonia or Viral Pneumonia.

Subhranil Nandy GSSoC 2024 contributor

Connect with me :)

• LinkedIn : subhranil-nandy

• GitHub : Subhranil2004