README.md

metadata

title: Wound Analysis V2
emoji: 🏢
colorFrom: yellow
colorTo: pink
sdk: gradio
sdk_version: 5.42.0
app_file: app.py
pinned: false

Wound Analysis LE

🩹 Project Overview

Wound Analysis LE is an advanced medical imaging tool for automated wound assessment using deep learning. It provides:

• Wound classification (type identification)
• Depth estimation (3D wound structure)
• Segmentation (precise wound area extraction)
• Severity analysis (quantitative and AI-powered clinical assessment)

The system is built for research and educational purposes, integrating state-of-the-art computer vision models and a user-friendly Gradio interface.

---

🚀 Features & Workflow

1. Wound Classification: Identifies wound type using a vision transformer model.
2. Depth Estimation: Generates depth maps and 3D visualizations from 2D images using DepthAnythingV2 (DINOv2 + DPT).
3. Segmentation: Extracts wound regions using deep learning models (Deeplabv3+, FCN, SegNet, Unet).
4. Severity Analysis: Computes wound area, depth, volume, and provides AI-powered medical assessment (Gemini AI integration).
5. Interactive Gradio App: Step-by-step workflow with visualization, overlays, and downloadable results.

---

🏗️ Model Architecture

Segmentation Models

• Deeplabv3+: Encoder-decoder with atrous convolutions for semantic segmentation.
• FCN (VGG16-16s): Fully convolutional network for pixel-wise segmentation.
• SegNet: Encoder-decoder architecture for efficient segmentation.
• Unet (multiple variants): U-shaped architecture for biomedical image segmentation.

Depth Estimation

• DepthAnythingV2: Combines DINOv2 vision transformer backbone with DPT head for monocular depth prediction.
• DINOv2: Self-supervised vision transformer for feature extraction.
• DPT: Dense Prediction Transformer for pixel-wise depth estimation.

Classification

• Vision Transformer (ViT): Used for wound type classification (via HuggingFace Transformers).

---

🛠️ Installation & Requirements

1. Clone the repository

   git clone <repo-url>
   cd Wound-Analysis-LE
   

2. Install dependencies

   pip install -r requirements.txt
   

   • Key dependencies: gradio, torch, tensorflow, opencv-python, transformers, open3d, plotly, google-generativeai, etc.
3. Download model weights
   • The app will auto-download required weights (e.g., DINOv2, segmentation models) on first run if not present.

---

💻 Usage Instructions

Run the Gradio App

python app.py

• Access the app at: http://localhost:7860

Segmentation Tool (Standalone)

python temp_files/segmentation_app.py

Workflow

1. Upload a wound image
2. Classify: Get wound type and initial AI analysis
3. Depth Estimation: Generate depth map and 3D visualization
4. Segmentation: Auto-segment wound area
5. Severity Analysis: Quantitative and AI-powered report
6. Download: Export masks, overlays, and 3D data

---

📊 Training & Evaluation

• Training scripts: See temp_files/train.py
• Metrics: Dice coefficient, precision, recall, loss (see utils/learning/metrics.py)
• Results: Training history and model checkpoints in training_history/
  • Example: Dice coefficient > 0.98 on training set (see 2025-08-07_16-25-27.json)

---

📁 Code Structure

• app.py — Main Gradio app (classification, depth, segmentation, severity)
• models/ — Segmentation model definitions (Deeplab, FCN, SegNet, Unet)
• depth_anything_v2/ — Depth estimation (DINOv2, DPT, utility layers)
• utils/ — Data loading, augmentation, metrics, postprocessing
• temp_files/ — Standalone scripts, experiments, and legacy tools
• training_history/ — Model checkpoints and training logs

---

📚 References

• Deeplabv3+ Paper
• DINOv2 (Meta AI)
• DPT: Vision Transformers for Dense Prediction
• HuggingFace Transformers
• Gradio
• Open3D
• Augmentor
• Datasets: Custom wound datasets (not included)

---

⚠️ Disclaimer

This tool is for research and educational purposes only. It does not provide medical advice or diagnosis. Always consult a medical professional for clinical decisions.