Classification of Chest X-Ray images: Pneumonia, Turberculosis and Normal

Get Started

1. Install the required packages

   pip install -r requirements.txt

2. Start the system

   streamlit run app.py

Configuration

Nvidia GeForce RTX 4060 (8GB)

Dataset

Dataset	Samples	Classes
PneumoniaMNIST	5,856	2
Tuberculosis Chest X-Ray Images	4200	2

Preprocessing

Model

Backbone	ACC
ResNet50	0.94
EfficientNet-B0	0.90
ConvNeXt-Tiny	0.93
Vision Transformer	0.93

RestNet50:

• Performance on every class

  	precision	recall	f1-score	support
  Normal	1.00	0.90	0.954	937
  Pneumonia	0.81	1.00	0.90	390
  Tuberculosis	1.00	0.98	0.99	137

• Overall Performance

  	precision	recall	f1-score	support
  micro avg	0.94	0.96	0.94	1464
  weighted avg	0.95	0.94	0.94	1464

EfficientNet-B0

• Performance on every class

  	precision	recall	f1-score	support
  Normal	1.00	0.85	0.92	937
  Pneumonia	0.73	1.00	0.85	390
  Tuberculosis	1.00	0.99	0.99	137

• Overall Performance

  	precision	recall	f1-score	support
  macro avg	0.91	0.94	0.92	1464
  weighted avg	0.93	0.90	0.91	1464

ConvNeXt-Tiny

• Performance on every class

  	precision	recall	f1-score	support
  Normal	1.00	0.89	0.94	937
  Pneumonia	0.80	1.00	0.89	390
  Tuberculosis	0.99	1.00	0.99	137

• Overall Performance

  	precision	recall	f1-score	support
  macro avg	0.93	0.96	0.94	1464
  weighted avg	0.95	0.93	0.93	1464

Vision Transformer

• Performance on every class

  	precision	recall	f1-score	support
  Normal	0.98	0.92	0.95	937
  Pneumonia	0.83	0.98	0.90	390
  Tuberculosis	1.00	0.88	0.94	137

• Overall Performance

  	precision	recall	f1-score	support
  macro avg	0.94	0.95	0.93	1464
  weighted avg	0.94	0.93	0.93	1464

Training Scripts

You can compare different model using the training scripts in JupyterNotebook/

JupyterNotebook
  |-- ensemble.ipynb
  |-- Pneumonia.ipynb
  |-- TB_Chest.ipynb

Results

• ResNet50

• EfficientNet-B0

• ConvNeXt-Tiny

• Vision Transformer