Artificial intelligence in the prenatal diagnosis of congenital heart defects

Abstract

Objective: To analyze the potential of artificial intelligence as a support tool for the early prenatal diagnosis of congenital heart defects. Methods: This qualitative and descriptive integrative review was conducted in PubMed, SciELO, LILACS, and Google Scholar. The descriptors “congenital heart defects,” “early diagnosis,” “prenatal diagnosis,” and “artificial intelligence” were combined using AND and OR. A total of 214 studies were identified, of which 25 comprised the final sample. Articles published between 2015 and 2026, in Portuguese and English, addressing artificial intelligence applied to obstetric ultrasound and fetal echocardiography were included. Results: Machine learning and deep learning techniques, especially Convolutional Neural Networks, showed high ability to detect fetal cardiac abnormalities, increase diagnostic sensitivity, reduce interobserver variability, and standardize image interpretation. Improvements in image quality and prenatal screening were also reported, particularly in regions with limited access to specialists. Limitations remain regarding external validation, algorithm heterogeneity, and dependence on image quality. Final considerations: Artificial intelligence is a promising strategy to strengthen prenatal diagnosis of congenital heart defects, although its clinical implementation still depends on methodological standardization and robust validation of algorithms.