Conditional deep generative normative modeling for structural and developmental anomaly detection in the fetal brain.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Developmental Biology

Background:

• Fetal brain development is complex; disruptions can cause neurological disorders.
• Early detection of fetal brain anomalies is crucial for timely intervention.
• Current methods for anomaly detection in fetal neuroimaging require improvement.

Purpose of the Study:

• To propose a novel deep generative anomaly detection framework, CCVAEGAN, for identifying structural brain anomalies in fetal MRI.
• To enhance anomaly detection accuracy across different developmental stages and diagnoses.
• To improve the efficiency of clinical workflows for early fetal brain anomaly diagnosis.

Main Methods:

• Developed a conditional cyclic variational autoencoding generative adversarial network (CCVAEGAN).
• Leveraged covariate conditioning on gestational age and cyclic consistency training.
• Utilized multi-site fetal brain MRI data from typically developing and clinically abnormal fetuses.

Main Results:

• CCVAEGAN achieved superior image generation quality and anomaly detection accuracy compared to existing models.
• Demonstrated near-perfect Area Under the Receiver Operating Characteristic Curve (AUROC) values (>0.99) for anomaly detection.
• External validation confirmed the framework's generalizability and robust performance across different data variations.

Conclusions:

• CCVAEGAN is a powerful tool for automated, objective anomaly screening in fetal brain MRI.
• The framework shows significant potential to enhance early diagnosis and clinical workflows.
• The approach may be universally applicable to other medical imaging modalities and organs.