Medical Detection & AI Applications Worktree

This worktree combines multiple medical detection and AI applications, integrating deep learning models, computer vision techniques, and large language models (LLMs) to create comprehensive medical imaging analysis systems. The project encompasses various medical domains including cancer detection, tissue segmentation, and diagnostic assistance.

🎯 Project Overview

This repository consolidates several medical AI applications that leverage state-of-the-art deep learning architectures for medical image analysis. The worktree demonstrates the integration of:

Medical Detection Systems: Object detection and classification models for identifying medical conditions
Medical AI Applications: Advanced AI systems combining computer vision with natural language processing for diagnostic assistance
Segmentation Models: Pixel-level segmentation for precise tissue and anomaly identification
LLM Integration: Large language models for intelligent medical report generation and consultation

📁 Project Structure

Medical_Detection/
├── Brain_Cancer/                          # Brain tumor classification system
│   ├── train_model.py                     # Training script with ResNet50
│   ├── predict_single.py                  # Single image prediction
│   └── best_brain_cancer_model.pth        # Trained classification model
│
├── Brain_Cancer_Segmentation/              # Brain cancer pixel-level segmentation
│   ├── brain_cancer_prediction.py         # DeepLabV3 segmentation predictions
│   ├── models/                            # Trained segmentation models
│   └── results/                           # Segmentation visualization results
│
├── Breast_Cancer_Detection/                # YOLO-based mammography detection
│   ├── main.py                            # YOLO prediction pipeline
│   ├── Yolo_Detection_Mamografi_*.pt      # YOLO detection model
│   └── pred/                              # Detection results and reports
│
├── Breast_Cancer_GPT/                      # Breast cancer detection with LLM integration
│   ├── main.py                            # YOLO + LLM prediction system
│   ├── app.py                             # Streamlit web interface
│   ├── llm_local.py                       # Local LLM integration
│   └── pred/                              # AI-generated reports and predictions
│
├── Breast_Full_Mri_efficientnet_b3/        # MRI-based breast cancer classification
│   ├── train.py                           # EfficientNet-B3 training
│   ├── predict.py                         # MRI prediction system
│   ├── model.py                           # Model architecture definitions
│   └── KNOWLEDGE_DISTILLATION.md          # Knowledge distillation guide
│
├── Tooth_Decay/                            # Dental decay and restoration segmentation
│   ├── predict.py                         # DeepLabV3 tooth segmentation
│   ├── models/                            # Trained dental models
│   └── dataset/                           # Dental image datasets
│
├── Skin_Cancer/                            # Skin cancer detection with interpretable AI
│   ├── src/                               # Source code modules
│   │   ├── main.py                        # Main entry point
│   │   ├── skin_cancer_cnn.py             # CNN with GradCAM implementation
│   │   ├── skin_cancer_yolo.py            # YOLO approach (demonstration)
│   │   ├── skin_cancer_visualizer.py      # Visualization tools
│   │   └── professional_gradcam.py       # Professional GradCAM visualization
│   ├── models/                            # Trained models
│   ├── outputs/                           # Generated visualizations and reports
│   └── data/                              # HAM10000 dataset files
│
├── Bone _Fracture/                         # YOLO-based bone fracture detection (X-ray)
│   ├── train.py                            # Ultralytics YOLO training script
│   ├── data.yaml                           # Dataset configuration (classes, paths)
│   ├── models/                             # Trained weights (e.g., best.pt)
│   └── results/                            # Training metrics, curves, confusion matrices
│
├── Bone_Fracture_Segmentation/             # YOLOv8 instance segmentation (fracture masks)
│   ├── train.py                            # Train YOLOv8-seg model
│   ├── validate.py                         # Validation metrics & plots
│   ├── predict.py                          # Run inference on target/ images
│   └── analysis.py                         # Dataset analysis & statistics
│
├── Bone_FractureV2/                        # Dataset merging + 17-class fracture detection
│   ├── merge_datasets.py                   # Merge datasets / remap classes
│   ├── check.py                            # Visualize YOLO annotations
│   └── Merged/                             # Unified dataset (data.yaml + splits)
│
└── Bone_Death_Tissue_Segmentation/          # Bone necrosis tissue segmentation
    ├── simple_test.py                     # Tissue segmentation testing
    └── DIA*.pth                           # Trained segmentation models

🔬 Applications & Technologies

1. Brain Cancer Detection & Segmentation

Brain_Cancer/ - Multi-class classification system

Technology: ResNet50 transfer learning
Task: Classify brain images into No Tumor, Benign, or Malignant
Features:
- Advanced data augmentation
- Class balancing with weighted loss
- Comprehensive evaluation metrics (ROC curves, confusion matrices)
- Real-time single image prediction

Brain_Cancer_Segmentation/ - Pixel-level cancer segmentation

Technology: DeepLabV3-ResNet50
Task: Precise segmentation of brain cancer regions
Features:
- Binary segmentation (background vs. cancer)
- Multiple visualization methods (masks, overlays, probability maps)
- Comprehensive statistics and region analysis
- Batch processing capabilities

2. Breast Cancer Detection Systems

Breast_Cancer_Detection/ - YOLO object detection

Technology: YOLO (You Only Look Once) object detection
Task: Detect and localize breast abnormalities in mammography images
Features:
- Automatic batch processing
- Bounding box visualization
- JSON results export
- Summary report generation

Breast_Cancer_GPT/ - AI-powered diagnostic assistant

Technology: YOLO detection + Large Language Models (LLMs)
Task: Combine detection with intelligent report generation
Features:
- YOLO-based detection pipeline
- LLM integration for medical report generation
- Streamlit web interface
- Doctor's assistant chat functionality
- Context-aware medical consultations

Breast_Full_Mri_efficientnet_b3/ - MRI classification system

Technology: EfficientNet-B3 with knowledge distillation support
Task: Classify breast MRI images as Benign or Malignant
Features:
- EfficientNet architecture for optimal accuracy/speed trade-off
- Knowledge distillation (teacher-student learning)
- Comprehensive training pipeline
- Model evaluation and metrics

3. Skin Cancer Detection

Skin_Cancer/ - Dermoscopic lesion classification with interpretable AI

Technology: EfficientNet-B0 with GradCAM visualization
Dataset: HAM10000 (10,015 dermoscopic images)
Task: Multi-class classification of 7 skin lesion types
Classes:
- Malignant: Melanoma (mel), Basal cell carcinoma (bcc)
- Benign: Melanocytic nevus (nv), Benign keratosis (bkl), Actinic keratosis (akiec), Vascular lesion (vasc), Dermatofibroma (df)
Features:
- Interpretable predictions with GradCAM heatmaps
- Shows which parts of the image influence the prediction
- Professional medical-grade visualization
- Comprehensive analysis reports (HTML)
- Dataset distribution analysis
- Model performance comparison
- Attention visualization for clinical interpretability
Approaches:
- CNN with GradCAM (Recommended): High accuracy with interpretable heatmaps
- YOLO (Demonstration): Included for educational purposes

4. Dental & Orthopedic Applications

Tooth_Decay/ - Dental restoration segmentation

Technology: DeepLabV3-ResNet50
Task: Multi-class segmentation of dental conditions
Classes: Background, Dolgu (Filling), Kanal (Root Canal), Çürük (Decay), Protez (Prosthesis)
Features:
- Multi-class semantic segmentation
- Color-coded visualization
- Detailed statistics per category
- Composite result visualization

Bone _Fracture/ - Bone fracture detection (object detection)

Technology: Ultralytics YOLO
Task: Detect and classify fracture findings in X-ray images
Classes: 7 classes (see Bone _Fracture/data.yaml)
Features:
- Config-driven training (data.yaml)
- Saved training curves/metrics and confusion matrices under Bone _Fracture/results/
- Trained weights stored under Bone _Fracture/models/

Bone_Fracture_Segmentation/ - Bone fracture instance segmentation (masks + boxes)

Technology: Ultralytics YOLOv8-seg
Task: Detect and segment fracture regions in X-ray images
Classes: 7 classes (see Bone_Fracture_Segmentation/BoneFractureYolo8/data.yaml)
Features:
- Dataset analysis (analysis.py) with charts/CSV under output/analysis/
- Training (train.py), validation (validate.py), and inference (predict.py)
- Outputs saved under Bone_Fracture_Segmentation/output/

Bone_FractureV2/ - Multi-source dataset merging (17-class detection)

Technology: Ultralytics YOLO (detection)
Task: Merge multiple YOLO datasets and train a unified fracture detector
Features:
- Merge/remap classes via merge_datasets.py (creates Merged/data.yaml)
- Annotation visualization via check.py
- Training outputs under runs/detect/

Bone_Death_Tissue_Segmentation/ - Bone necrosis detection

Technology: Deep learning segmentation models
Task: Identify and segment necrotic bone tissue
Features:
- Tissue segmentation
- Interactive testing scripts
- Model checkpoint management

🛠️ Technologies & Frameworks

Deep Learning Frameworks

PyTorch: Primary deep learning framework
Torchvision: Pre-trained models and transforms
Ultralytics YOLO: Object detection models

Computer Vision

OpenCV: Image processing
PIL/Pillow: Image manipulation
Matplotlib: Visualization and plotting

Natural Language Processing

Local LLM Integration: Medical report generation
Streamlit: Web interface for AI assistant

Model Architectures

ResNet50: Transfer learning for classification
DeepLabV3: Semantic segmentation
EfficientNet: Efficient classification models (B0, B3)
YOLO: Real-time object detection
GradCAM: Interpretable AI visualization for medical diagnosis

🚀 Getting Started

Prerequisites

Python 3.7+
CUDA-compatible GPU (recommended for training)
8GB+ RAM (16GB+ recommended)
10GB+ disk space for models and datasets

Installation

Each subdirectory contains its own requirements.txt. Install dependencies for specific applications:

# Example: Install dependencies for Brain Cancer Detection
cd Brain_Cancer
pip install -r requirements.txt

# Example: Install dependencies for Breast Cancer Detection
cd Breast_Cancer_Detection
pip install -r requirements.txt

Common Dependencies

Most projects require:

pip install torch torchvision
pip install opencv-python pillow matplotlib
pip install numpy tqdm
pip install ultralytics  # For YOLO projects
pip install streamlit    # For web interfaces

📊 Key Features Across Applications

1. Detection & Classification

Multi-class medical condition classification
Object detection with bounding boxes
Confidence scoring and uncertainty quantification

2. Segmentation

Pixel-level precise segmentation
Multi-class tissue identification
Overlay visualizations and probability maps

3. AI Integration

LLM-powered medical report generation
Interactive diagnostic assistance
Context-aware medical consultations
Interpretable AI with GradCAM heatmaps for clinical transparency

4. Comprehensive Analysis

Statistical summaries
Visual result generation
Export capabilities (JSON, images, HTML reports)
Interpretable heatmaps showing model decision-making process

🎓 Usage Examples

Brain Cancer Classification

cd Brain_Cancer
python predict_single.py path/to/brain_scan.jpg

Breast Cancer Detection with AI Assistant

cd Breast_Cancer_GPT
streamlit run app.py
# Or run the command-line version
python main.py

Tooth Decay Segmentation

cd Tooth_Decay
python predict.py
# Processes images from target/data/ and saves to target/pred/

Bone Fracture Detection (YOLO Training)

cd "Bone _Fracture"
pip install -r requirements.txt
python train.py --model yolo12s --data data.yaml --epochs 100 --device cuda

Bone Fracture Segmentation (YOLOv8-seg)

cd Bone_Fracture_Segmentation
python analysis.py
python train.py
python validate.py
python predict.py

MRI Breast Cancer Classification

cd Breast_Full_Mri_efficientnet_b3
python train.py    # Train model
python predict.py  # Make predictions

Skin Cancer Detection with GradCAM

cd Skin_Cancer
# Train CNN model with GradCAM visualization
python src/main.py --mode cnn --train --epochs 20

# Make prediction with interpretable heatmap
python src/main.py --mode cnn --predict "path/to/skin_image.jpg"

# Generate comprehensive visualizations
python src/main.py --mode visualize

📈 Model Performance & Evaluation

Each application includes comprehensive evaluation metrics:

Classification Metrics: Accuracy, Precision, Recall, F1-Score, ROC-AUC
Segmentation Metrics: IoU (Intersection over Union), Dice Score
Detection Metrics: mAP (mean Average Precision), Confidence scores
Visualization: Confusion matrices, training curves, result overlays

🔄 Workflow Integration

This worktree demonstrates a complete medical AI workflow:

Data Preparation: COCO format annotations, dataset organization
Model Training: Transfer learning, fine-tuning, knowledge distillation
Inference: Single image and batch processing
Visualization: Comprehensive result visualization
AI Enhancement: LLM integration for intelligent reporting
Deployment: Web interfaces and command-line tools

🏥 Medical Domains Covered

Oncology: Brain cancer, breast cancer detection and segmentation
Dermatology: Skin cancer detection with interpretable AI (melanoma, basal cell carcinoma, benign lesions)
Dentistry: Tooth decay and restoration identification
Orthopedics: Bone fracture detection, bone fracture segmentation, bone necrosis tissue segmentation
Radiology: Mammography and MRI analysis

⚠️ Important Notes

Medical Disclaimer

This software is for research and educational purposes only. It is NOT intended for clinical diagnosis or treatment decisions. Always consult qualified medical professionals for actual medical diagnosis and treatment.

Data Privacy

Ensure compliance with HIPAA, GDPR, and local medical data regulations
Medical images should be properly anonymized
Follow institutional data handling policies

Model Limitations

Models are trained on specific datasets and may not generalize to all populations
Performance depends on image quality and acquisition parameters
Regular model validation and updates are recommended

📝 Contributing

When contributing to this worktree:

Maintain clear separation between different medical applications
Include comprehensive documentation for each module
Follow medical data privacy guidelines
Add appropriate disclaimers and citations
Test thoroughly before integration

📚 Documentation

Each subdirectory contains detailed README files with:

Installation instructions
Usage examples
Model architecture details
Configuration options
Troubleshooting guides

🔗 Related Resources

PyTorch Documentation: https://pytorch.org/docs/
Ultralytics YOLO: https://docs.ultralytics.com/
Medical Imaging Datasets: Various public and private sources
Streamlit: https://docs.streamlit.io/

📄 License

Please refer to individual project directories for specific licensing information. Most projects follow research and educational use licenses.

🙏 Acknowledgments

This worktree combines various medical AI applications and leverages:

Pre-trained models from PyTorch and Torchvision
YOLO detection models from Ultralytics
Public medical imaging datasets
Open-source deep learning frameworks

Last Updated: 2025
Maintainer: Yahya Status: Active Development

Name		Name	Last commit message	Last commit date
BREAST_FULL_MRI_efficientnet_b3		BREAST_FULL_MRI_efficientnet_b3
Bone _Fracture		Bone _Fracture
Bone_Death_Tissue_Segmentation		Bone_Death_Tissue_Segmentation
Bone_FractureV2		Bone_FractureV2
Bone_Fracture_Segmentation		Bone_Fracture_Segmentation
Brain_Cancer		Brain_Cancer
Brain_Cancer_Segmentation		Brain_Cancer_Segmentation
Breast_Cancer_Detection		Breast_Cancer_Detection
Skin_Cancer		Skin_Cancer
Tooth_Decay		Tooth_Decay
breast_cancer_GPT		breast_cancer_GPT
.gitattributes		.gitattributes
LICENSE		LICENSE
README.md		README.md

Latest commit

History

Folders and files

Repository files navigation