FADA: Fetal Abnormality Detection Algorithm 

Overview

Across the MENA region, prenatal care is routinely constrained by older equipment, inconsistent imaging conditions, and too few specialist sonographers to meet demand. For pregnant women in resource-limited settings, that gap carries real consequences.

FADA is an open-source, explainable AI platform built to close that gap. Trained on a harmonized corpus of 20,000 de-identified fetal ultrasound images, it integrates deep-learning anatomical segmentation, GAN-based image enhancement, and vision-language models that generate plain-language clinical explanations. Crucially, FADA was designed for the conditions that actually exist in the region: older machines, heterogeneous image quality, and limited access to specialist review.

The results are already demonstrating what context-sensitive AI can achieve. Segmentation models reached 0.89 Dice agreement with expert annotations; image quality improved by roughly 14% in peak signal-to-noise ratio; and leading clinical explanations were rated 4.6 out of 5 by expert sonographers for anatomical and clinical accuracy. Now advancing toward its third iteration, FADA is moving from validated prototype to open, deployable tool with expert annotation, a dedicated vision-language model, and full open data and code release on the roadmap.