Breast Imaging Model Development
CVAT
Image
ClassificationProject Summary The project focuses on developing AI-powered breast cancer detection and breast density classification models using mammography images from African populations, addressing the underrepresentation of Africans in existing datasets largely trained on Caucasian and Asian populations. Scope The project involves: Collecting anonymized mammogram images from African hospitals and diagnostic centers Developing AI models for: Breast density classification Lesion detection Breast cancer prediction Bias and fairness evaluation across racial groups Comparing model performance across African, Caucasian, and Asian datasets Data Labelling Tasks Medical experts and annotators will perform: BI-RADS breast density classification Annotation of masses, calcifications, distortions, and suspicious lesions Benign/malignant classification Tumor localization using bounding boxes or segmentation masks Metadata annotation such as age, ethnicity, imaging view, and device type Project Size The project may be executed in phases: Pilot phase: 5,000–20,000 mammogram images Expansion phase: 100,000+ images from multiple centers Long-term vision: Pan-African imaging dataset with 500,000+ images Quality Measures The project will adhere to strict medical AI quality standards including: Double-blind radiologist annotations Consensus review for disagreements Inter-rater agreement validation DICOM image quality checks Dataset balancing across demographics and density classes AI validation using sensitivity, specificity, ROC-AUC, and fairness metrics Full anonymization, ethical approvals, and secure data governance Expected Outcome The project aims to create more accurate, fair, and clinically reliable breast cancer AI systems for African populations while contributing to global health equity and precision medicine.