biryomumeishojoshua/AML-Detection-LMICs-Uganda

Detection of Acute Myeloblastic Leukemia from blood smear images using Transfer Learning (ResNet-50/EfficientNet), targeting low- and middle-income country (LMIC) settings with a focus on Uganda. # AML Detection in LMICs: A Focus on Uganda > Automatic detection of Acute Myeloblastic Leukemia (AML) from blood smear images using > Transfer Learning, designed for low- and middle-income country (LMIC) healthcare settings. --- ## Overview Leukemia accounts for 28% of rising cancer cases in LMICs, with AML making up 60% of adult leukemia cases. In Uganda, 2.67 people per 100,000 die from leukemia annually, a disease that is curable when diagnosed early. Current diagnosis relies on manual interpretation of blood smears, which is slow, specialist-dependent, and prone to misdiagnosis, especially given symptom overlap with common infections like malaria. This project develops and trains an AI model capable of detecting AML from blood smear images, leveraging existing light microscopes and laboratory personnel already present in LMIC settings. --- ## Objectives - Develop and train a deep learning model using existing labelled datasets - Test and evaluate model performance on relevant metrics (accuracy, sensitivity, specificity) - Design a user-friendly graphical interface for use by laboratory personnel --- ## Dataset | Source | Description | |---|---| | Munich AML Morphology Dataset | ~18,000 expert-labelled single-cell blood smear images (Matek et al., 2019) | | Local Ugandan Blood Smear Samples | Collected from Ugandan health facilities for LMIC validation | --- ## AI Method **Architecture:** Transfer Learning using ResNet-50 or EfficientNet (pretrained on ImageNet) **Explainability:** Grad-CAM (Gradient-weighted Class Activation Mapping) to highlight regions of interest in each blood smear, making model decisions interpretable for lab staff. **Supporting techniques:** - Data augmentation (rotation, flipping, colour jitter) to handle limited local data - Binary classification: AML vs. normal --- ## Project Structure ``` AML-Detection-LMICs-Uganda/ │ ├── data/ # Dataset references and preprocessing scripts ├── models/ …