ROLE OF ARTIFICIAL INTELLIGENCE IN THE AUTOMATIC DETECTION OF FETAL MALFORMATIONS BY ULTRASOUND

Abstract

Anatomical ultrasound remains the main screening tool for fetal malformations, although its sensitivity varies widely according to operator experience and image acquisition quality. This critical narrative review examines the role of artificial intelligence systems in the automatic detection of fetal anomalies by ultrasound, aiming to separate what has already been demonstrated in clinical validation from what remains a technical promise. The literature search was conducted in 2026, with PubMed/MEDLINE as the primary database, complemented by reference list screening and direct consultation of official regulatory documents. The analysis follows the diagnostic chain: image acquisition and quality control, performance by anatomical territory, the effect of assistance on the examiner, generalization, and conditions for adoption. Some models achieve accuracy close to that of specialists in circumscribed tasks, particularly in cardiac and neurosonographic screening, and improve the performance of less experienced operators; heterogeneity across studies, small human comparator groups, and insufficient external validation preclude generalizing this performance. Dependence on image quality, scarcity of prospective external validation, and risk of deference to incorrect suggestions persist. The plausible gain lies in standardization and in the recovery of cases missed through acquisition failure, not in replacing the examiner.