Deteksi Tuberkulosis dari Citra Chest X-Ray Menggunakan Computer Vision & Machine Learning

👉 Try the App Here!

Upload X-Ray image → Get instant TB detection results with visualization!

• ✅ Automated Lung Segmentation - K-Means clustering untuk isolasi region paru-paru
• ✅ Advanced Preprocessing - CLAHE + Gaussian Blur untuk enhancement
• ✅ Multi-Region Detection - Deteksi lung, nodule, dan cavity
• ✅ Feature Extraction - LBP, GLCM, Edge, dan Hough Line features
• ✅ ML Classification - SLDT-MSA (Stacking + Moth Search Algorithm)
• ✅ Interactive Visualization - Real-time visualization dengan Streamlit
• ✅ Morphological Analysis - Complete morphology operations analysis

• Shape Features (3): Edge Sum, Number of Lines, Corner Count
• Texture Features (2): GLCM Contrast, GLCM Homogeneity
• LBP Features (9): Local Binary Pattern Histogram (9 bins)

• Sumber: Kaggle TB Chest X-ray Database
• Total: 4.200 citra (3.500 Normal, 700 TB)
• Split: 80% Training, 20% Testing (stratified)
• Format: PNG/JPG grayscale images
• Struktur:

  data/raw/TB_Chest_Radiography_Database/
    ├── Normal/          → 3.500 citra
    └── Tuberculosis/    → 700 citra
  

tb-cxr-detection/
├── README.md
├── requirements.txt         # Dependencies untuk production
├── requirements-local.txt   # Dependencies untuk development
├── packages.txt            # System dependencies (deployment)
├── config.yaml             # Configuration file
├── .streamlit/
│   └── config.toml         # Streamlit app configuration
├── app/
│   ├── main.py            # 🎨 Streamlit UI
│   ├── pipeline.py        # Backend analysis pipeline
│   └── utils/
│       └── visualizer.py  # Visualization functions
├── src/
│   ├── __init__.py
│   ├── tes.py             # Complete pipeline (preprocessing → features)
│   ├── preprocessing/
│   ├── segmentation/
│   ├── morphology/
│   └── classification/
│       ├── train.py       # Model training (SLDT-MSA)
│       └── test_a.py      # Model evaluation
├── models/
│   └── tb_model_raw.pkl   # Trained classifier (Git LFS)
├── data/
│   ├── raw/               # Original dataset (gitignored)
│   ├── processed/
│   │   ├── features/
│   │   │   └── dataset.csv  # Extracted features
│   │   ├── images/
│   │   └── masks/
│   └── mock/              # Sample images for testing
├── notebooks/             # Jupyter notebooks (development)
└── tests/                 # Unit tests

• Python 3.10 or 3.11
• Git
• (Optional) Git LFS for model file

# 1. Clone repository
git clone https://github.com/benedictuserwinwidianto/tb-cxr-detection.git
cd tb-cxr-detection

# 2. Create virtual environment
python -m venv .venv

# Windows
.venv\Scripts\activate

# Linux/Mac
source .venv/bin/activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Verify installation
python -c "import streamlit; import cv2; import sklearn; print('✓ All dependencies installed!')"

# From project root directory
streamlit run app/main.py

# App will open at http://localhost:8501

Classifier: SLDT-MSA (Stacking Loopy Decision Tree + Moth Search Algorithm)

• Feature Selection: Moth Search Algorithm (MSA)
• Base Learners: Decision Tree + Random Forest
• Meta Learner: Decision Tree
• Optimization: Grid Search (class weight + max depth)

1. Upload chest X-Ray image (PNG/JPG)
2. Click "🔬 Analyze Image"
3. View results:
   • Prediction (Normal/TB) with confidence
   • Segmentation masks
   • Morphological operations
   • Extracted features

# Generate features from dataset
python src/tes.py

# Train classifier
python src/classification/train.py

# Test model
python src/classification/test_a.py

# Create feature branch
git checkout -b feature/module-name-yourname

# Example
git checkout -b feature/segmentation-revy

# Work → Commit → Push
git add .
git commit -m "Add segmentation module"
git push origin feature/segmentation-revy

1. Create PR dari feature branch ke main
2. Tag: @benedictuserwinwidianto + 1 teammate untuk review
3. Merge setelah mendapat 1 approval
4. Delete feature branch setelah merge

preprocess_image(image_path: str) -> np.ndarray

• Gaussian Blur (3×3)
• CLAHE (clipLimit=2.0, tileGridSize=8×8)

segment_lungs(img: np.ndarray) -> dict

• K-Means clustering (k=3)
• Adaptive threshold untuk nodule/cavity

apply_morphology(mask: np.ndarray, kernel_size: int) -> dict

• Otsu thresholding
• Erosion, Dilation, Opening, Closing

extract_lbp_features(img: np.ndarray, mask: np.ndarray) -> dict

• Edge detection (Canny)
• Line detection (Hough)
• GLCM features
• LBP histogram

• Dataset: Tawsifur Rahman et al.
• Streamlit Community
• UGM - Pengantar Citra Digital Course

Project Lead: Benedictus Erwin Widianto
📧 benedictuserwinwidianto@mail.ugm.ac.id
🔗 GitHub Issues