filimapatrick/African-Brain-MRI-Benchmark

# African Clinical Brain MRI Benchmark for Robust AI Evaluation Medical AI has achieved remarkable success in brain MRI analysis. However, most publicly available benchmarks are derived from homogeneous research datasets acquired using standardized imaging protocols, high-field MRI scanners, and carefully controlled acquisition environments. These benchmarks often fail to represent the severe imaging variability encountered in routine clinical practice, particularly within resource-limited settings such as low- and middle-income countries (LMICs). This repository introduces the **African Clinical Brain MRI Benchmark (Afri-Brain-Bench)**, a standardized benchmark ecosystem designed to evaluate the robustness, generalizability, calibration, and explainability of machine learning and deep learning models under realistic clinical imaging conditions in Africa. Rather than merely maximizing predictive accuracy on clean, homogeneous datasets, this benchmark shifts the focus to establishing rigorous, reproducible evaluation protocols for developing clinically resilient AI systems capable of handling real-world domain shifts. --- ## Table of Contents - Overview - Motivation & Clinical Domain Shift - Benchmark Goals - Research Questions - Benchmark Contributions - Clinical Dataset Details - Institutional Profiles - Disease Distribution - Dataset Characteristics - Benchmark Tasks - Standardized Preprocessing Pipeline - Baseline Models - Benchmark Metrics - Evaluation Rules - Explainability Benchmark - Statistical Evaluation Framework - Project Structure - Expected Figures & Tables - Expected Outcomes - Limitations & Future Work - Citation --- ## Overview Artificial intelligence systems in neuroimaging are notoriously sensitive to domain shift. While models trained on research repositories (e.g., ADNI, UK Biobank) achieve high performance locally, their accuracy drops precipitously when deployed in clinical environments with heterogeneous scanners, varying acquisi …