Patro331/sickle-cell-detection

Deep Learning for Automated Sickle Cell Detection in Peripheral Blood Smears using African Clinical Microscopy Data # Deep Learning for Automated Sickle Cell Detection in Peripheral Blood Smears An explainable deep learning project for automated sickle cell disease detection using African clinical microscopy data, built as part of MSB7216: Deep Learning for Health Data at Makerere University. **Author:** Okidi Patrovas Gabriel | 2025/HD07/26020U **Institution:** Makerere University, Kampala, Uganda **Course:** MSB7216: Deep Learning for Health Data ## Live Demo Launch the Screening Tool: huggingface.co Upload a blood smear image and get an instant prediction with Grad-CAM explainability showing exactly which regions of the image the model focused on. ## Project Overview Sickle cell disease affects an estimated 13.3% of the Ugandan population, with prevalence reaching 19.8% in Kampala alone. Diagnosis currently requires trained laboratory personnel, specialist equipment, and manual microscopic examination — resources that are largely inaccessible in rural and low-resource settings across Uganda. This project builds an explainable deep learning model for automated sickle cell detection from blood smear images, designed specifically for low-resource, mobile-first point-of-care screening. This is the first deep learning study to apply transfer learning and explainability to the Tushabe (2024) Ugandan clinical mobile-phone microscopy dataset. ## Datasets **Primary Dataset — Positive Class:** Tushabe et al. (2024) Ugandan Sickle Cell Microscopy Dataset. 422 unlabelled blood smear images collected from patients in Soroti and Kumi districts, Uganda, captured using mobile phone cameras placed on basic microscopes under real clinical conditions. Published in Acta Scientific Microbiology, Volume 7, Issue 12. Available at kaggle.com **Supplementary Dataset — Negative Class:** BCCD Dataset (Shenggan et al.). 364 normal peripheral blood smear images used to supplement the ne …