Conditional Deep Generative Normative Modeling for Structural and Developmental Anomaly Detection in the Fetal Brain

Motivation: Early and precise identification of fetal brain anomalies is crucial for appropriate postnatal care. However, detecting subtle abnormalities is challenging since fetal brains undergo complex and rapid development leading to high individual variability. Goal(s): To improve the detection performance of fetal brain anomalies through a novel deep learning model that provides enhanced normative references. Approach: A deep generative framework was developed with gestational age conditioning and cyclic consistency training to generate precise normative references for anomaly detection. Results: The proposed model achieved high accuracy (AUROC value of 0.992), in differentiating typical development and gross fetal brain abnormalities on MRI, with good regional localization. Impact: The proposed anomaly detection framework offers a new tool for clinicians to identify fetal brain anomalies with high accuracy, enhancing the precision of prenatal diagnostics. This can lead to earlier and more targeted interventions for neurodevelopmental abnormalities, potentially improving outcomes.