🩻 VisionAI: MURA Dataset Classification & Explainable X-Ray Analysis

📌 Project Overview

VisionAI: MURA Dataset Classification & Explainable X-Ray Analysis is a deep learning project designed to automatically classify musculoskeletal X-ray images as Normal or Abnormal using a transfer-learning approach based on MobileNetV2.

In addition to classification, the project includes an Explainable AI (XAI) module that generates visual heatmaps highlighting regions that contributed most to the model's prediction, making the results more interpretable for medical analysis and educational purposes.

---

📂 Dataset

The project utilizes the MURA (Musculoskeletal Radiographs) Dataset, one of the largest publicly available radiographic image datasets released by the Stanford ML Group.

About MURA

MURA contains musculoskeletal X-ray studies collected from clinical practice and is widely used for research in medical image classification and computer-aided diagnosis.

Dataset Characteristics

• Task: Binary Classification (Normal vs. Abnormal)

• Body Parts Covered:

  • Elbow
  • Finger
  • Forearm
  • Hand
  • Humerus
  • Shoulder
  • Wrist

• Image Type: Musculoskeletal Radiographs (X-rays)

• Labels: Normal / Abnormal

Dataset Link

https://www.kaggle.com/datasets/cjinny/mura-v11

---

🎯 Objectives

• Detect abnormalities in musculoskeletal X-rays.
• Leverage transfer learning for efficient model training.
• Improve model interpretability using heatmap visualization.
• Demonstrate the application of AI in medical imaging.

---

🧠 Model Architecture

The classification model is built using TensorFlow/Keras and utilizes:

MobileNetV2 Transfer Learning

MobileNetV2 was selected because it provides:

• Lightweight architecture
• Fast inference
• Strong feature extraction capability
• Reduced computational requirements

Architecture Components

• MobileNetV2 Base Model
• Global Average Pooling Layer
• Dense Layers
• Dropout Regularization
• Sigmoid Output Layer for Binary Classification

---

🔍 Explainable AI Module (explainer.py)

One of the key highlights of this project is the Explainable AI Dashboard implemented using Streamlit.

Features

• Upload an X-ray image.
• Predict whether the image is Normal or Abnormal.
• Generate Grad-CAM style heatmaps.
• Visualize regions influencing model decisions.
• Display confidence scores.
• Interactive web interface.

How It Works

The explainer.py file:

1. Loads the trained MobileNetV2 model.
2. Accepts user-uploaded X-ray images.
3. Preprocesses images to match training dimensions.
4. Generates predictions using the trained model.
5. Computes gradients from the final convolutional layer.
6. Creates a heatmap highlighting important regions.
7. Overlays the heatmap onto the original X-ray.
8. Displays classification results and confidence scores through Streamlit.

Explainability Benefits

• Improves transparency of AI predictions.
• Helps users understand model focus areas.
• Enhances trust in automated diagnosis systems.
• Demonstrates practical Explainable AI techniques in healthcare.

---

✨ Features

• ✅ Binary X-Ray Classification (Normal vs Abnormal)
• ✅ MobileNetV2 Transfer Learning
• ✅ Medical Image Analysis
• ✅ Grad-CAM Inspired Heatmap Visualization
• ✅ Explainable AI Dashboard
• ✅ Streamlit-Based User Interface
• ✅ Confidence Score Display
• ✅ Image Overlay Visualization
• ✅ Interactive X-Ray Upload System

---

🛠️ Technologies & Libraries Used

Deep Learning

• TensorFlow
• Keras
• MobileNetV2

Data Processing

• NumPy
• Pandas

Visualization

• Matplotlib

Computer Vision

• OpenCV (cv2)
• Pillow (PIL)

Web Application

• Streamlit

Utilities

• Glob
• OS

---

📁 Project Structure

MURA-Dataset-Classification-VisionAI/
│
├── classification.ipynb      # Model training notebook
├── explainer.py              # Streamlit explainability dashboard
├── models/
│   └── xray_mobilenetv2.h5   # Trained model
│
├── README.md
└── requirements.txt

🛠️ Installation & Requirements

1. Clone the Repository

git clone https://github.com/Shashank14105/MURA-Dataset-Classification-VisionAI.git
cd MURA-Dataset-Classification-VisionAI

2. Install Dependencies

pip install numpy pandas matplotlib opencv-python scikit-learn
pip install tensorflow keras
pip install streamlit pillow

Alternatively:

pip install -r requirements.txt

---

💻 Usage

1. Train the Model

To start training the classification model on the MURA dataset:

python XRayClassification.py

or run:

jupyter notebook classification.ipynb

---

2. Launch the Explainability Dashboard

streamlit run explainer.py

After launching, open the local Streamlit URL displayed in your terminal.

---

📊 Output

The system provides:

• Classification Prediction

  • Normal
  • Abnormal

• Confidence Score

• Heatmap Visualization

• Overlay Analysis View

Example:

Prediction: ABNORMAL (94.32%)

The AI detected potential abnormalities
in the highlighted regions.

---

🚀 Future Improvements

• Multi-class abnormality detection
• Severity estimation
• Support for additional radiology datasets
• Advanced explainability techniques
• Model deployment on cloud platforms
• Integration with hospital workflows

---

📚 Educational Value

This project demonstrates:

• Medical Image Classification
• Transfer Learning
• Computer Vision
• Explainable AI (XAI)
• Streamlit Deployment
• Deep Learning in Healthcare

making it an excellent portfolio project for students and researchers interested in AI for healthcare.

---

👨‍💻 Author

Shashank K

AI & Machine Learning Enthusiast VisionAI Project Series

If you found this project useful, consider giving the repository a ⭐.