Source: MIT News - Artificial intelligence
MIT researchers are encoding bilateral symmetry into AI models to detect fetal abnormalities from ultrasound images. Current systems miss asymmetries that signal developmental problems—a concrete failure mode the new approach addresses. The human body's symmetry is fundamental; breaking it is often pathological. By making this domain knowledge explicit in the model's architecture rather than relying on pattern matching alone, the system becomes more reliable and more interpretable to clinicians. Prenatal diagnosis errors lead to delayed interventions or false alarms. Even modest accuracy gains could change outcomes for thousands of pregnancies annually.