ALL Detection System

Automated Acute Lymphoblastic Leukemia (ALL) detection from peripheral blood smear images, powered by deep learning and on-device AI.

How It Works

1. Cell Localization — K-Means clustering in L*a*b* color space identifies WBC nuclei, followed by Euclidean Distance Transform and marker-controlled Watershed to split clumped cells into individual centroids
2. Precision Segmentation — Centroids are passed as point prompts to the Segment Anything Model (SAM ViT-Base), which traces precise cell boundaries in a single batched forward pass
3. Crop Standardization — Cells are extracted into CNMC-matched 128×128 crops with scale normalization (~75% fill ratio), mask centroid centering, and background masking
4. Classification — Each cell is classified as ALL (blast) or HEM (healthy) by a MobileNetV3-Large TFLite model with 4-orientation test-time augmentation
5. AI Summary — An on-device LLM (Phi-3 via Ollama) generates a clinical-style interpretation of the results
6. Visualization — Results are displayed in a real-time dashboard with cell cards, SAM contour annotations, and confidence scores

Quick Start

Desktop (Windows/Linux/macOS)

pip install -r requirements.txt
cd src
python ui/classification_demo.py

Raspberry Pi 5

cd linux_training
./install_deps.sh    # Installs system dependencies automatically
# Note: Models should be trained on desktop first and deployed here

The desktop handles pipeline training; Pi handles inference via src/ui/classification_demo.py.

Project Structure

src/
├── ui/                          # Desktop GUI (CustomTkinter)
│   ├── classification_demo.py   # Main application (Watershed → SAM → Classify)
│   └── theme.py                 # Design system
├── detection/                   # Core detection pipeline
│   ├── demo_pipeline.py         # Production pipeline (Watershed → SAM → TFLite)
│   └── generate_cell_crops_sam.py  # Watershed centroid finder + SAM crop generator
├── utils/                       # Visualization, preprocessing utilities
└── config.py                    # Master config for image resolution & normalization

training_scripts/        # Model training scripts (PyTorch)
paper/                   # LaTeX figures and paper generation scripts
data/                    # Reference images, processed crops
tests/                   # Pytest suite
_archive/                # Legacy models, legacy tests, and old evaluation scripts

Datasets

Dataset	Purpose	Details
C-NMC 2019	Training & Validation	12,528 pre-cropped cell images, strict patient-level 3-fold CV split
ALL-IDB1	End-to-end Testing	Full blood smear images for pipeline evaluation
ALL-IDB2	Segmentation Benchmark	260 pre-cropped cells for cross-dataset testing

Key Results

Metric	Value
AUC (ROC)	0.934
Sensitivity @ t=0.50	95.3% (with 4x TTA)
Specificity @ t=0.50	80.2%
F1 Score	0.890
Model Size (TFLite)	3.38 MB
Inference Latency	22 ms/cell

Hardware Targets

• Primary: Raspberry Pi 5 (8GB) + 1080p display
• Secondary: Android devices (via companion app)

Tech Stack

Component	Technology
Cell Localization	K-Means (L*a*b*) + Watershed + SAM (ViT-Base)
Classification	MobileNetV3-Large (TFLite) distilled via Online KD from EffB4
Optimization Pipeline	Albumentations, Differential Learning Rates (AdamW), Weighted Sampler
AI Summary	Phi-3 Mini (3.8B) via Ollama
Desktop GUI	CustomTkinter (Python)
Terminal UI	Rich Dashboard (Real-time GPU monitoring)
Segmentation	OpenCV, scikit-image, HuggingFace Transformers

Requirements

See requirements.txt for Python dependencies.

System Requirements:

• Python 3.10+
• Ollama (for AI summaries)
• 4GB+ RAM (8GB recommended for LLM + SAM)

Citations & References

Datasets

• C-NMC 2019 (ISBI Challenge): Gupta, A., & Gupta, R. (2019). ALL Challenge dataset of ISBI 2019 [Data set]. The Cancer Imaging Archive. https://doi.org/10.7937/tcia.2019.dc64i46r
• ALL-IDB: Labati, R. D., Piuri, V., & Scotti, F. (2011). ALL-IDB: The Acute Lymphoblastic Leukemia Image Database for Image Processing. Proceedings of the IEEE International Conference on Image Processing (ICIP), pp. 2045–2048.

Models & Architectures

• MobileNetV3: Howard, A., et al. (2019). Searching for MobileNetV3. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 1314–1324.
• EfficientNet: Tan, M., & Le, Q. V. (2019). EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks. Proceedings of the 36th International Conference on Machine Learning (ICML), pp. 6105–6114.
• Segment Anything (SAM): Kirillov, A., et al. (2023). Segment Anything. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 4015–4026.
• Knowledge Distillation: Hinton, G., Vinyals, O., & Dean, J. (2015). Distilling the Knowledge in a Neural Network. arXiv preprint arXiv:1503.02531.

Frameworks & Libraries

• PyTorch: Paszke, A., et al. (2019). PyTorch: An Imperative Style, High-Performance Deep Learning Library. Advances in Neural Information Processing Systems (NeurIPS), 32.
• timm: Wightman, R. (2019). PyTorch Image Models (timm). GitHub. https://github.com/huggingface/pytorch-image-models
• Albumentations: Buslaev, A., et al. (2020). Albumentations: Fast and Flexible Image Augmentations. Information, 11(2), 125.
• scikit-learn: Pedregosa, F., et al. (2011). Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research, 12, pp. 2825–2830.
• OpenCV: Bradski, G. (2000). The OpenCV Library. Dr. Dobb's Journal of Software Tools.
• Ollama / Phi-3: Microsoft. (2024). Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone. arXiv preprint arXiv:2404.14219.

Methods

• Watershed Segmentation: Meyer, F. (1994). Topographic distance and watershed lines. Signal Processing, 38(1), 113–125.
• Macenko Stain Normalization: Macenko, M., et al. (2009). A Method for Normalizing Histology Slides for Quantitative Analysis. Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI), pp. 1107–1110.
• Test-Time Augmentation (TTA): Shanmugam, D., et al. (2021). Better Aggregation in Test-Time Augmentation. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 1214–1223.