X-Ray Image Analysis for Body Part Classifier & Bone Fracture Detection

Project Description

Reading X-ray images is a crucial step for patients' treatment, but it is also very time-consuming for radiologists. Bone fractures commonly go undetected, leading to complications and delays in patients' care. The increased interest in computer-aided diagnosis can reduce radiologists' burden and improve their detection of bone fractures.

This project addresses two primary problems:

Problem 1: Image Classification

Prediction of the specific body part pictured in an X-ray.

• Classes: 22 classes focused on cases where there is one body part - others termed "mixed".
• Models: Custom CNN and EfficientNet.

Problem 2: Object Detection

Combination of classification and regression problem, focused on identifying the presence and location of fractures in X-rays.

• Models: Faster R-CNN and YOLO.

Datasets

We leverage two datasets for this project:

1. FracAtlas
2. UNIFESP X-ray Bodypart Classification

Please download and unzip these datasets into the data directory.

Models Used

• EfficientNet: For image classification.
• YOLO: For object detection.

Demonstration

We use Gradio to deploy the models as a web-based example.

To use it, run deployment.py

Installation

To install the required packages, run:

pip install -r requirements.txt