The Identification of Beckwith-Wiedemann Syndrome Through Swap Disentangled Variational Autoencoder.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in diagnostics
• Syndromology

Background:

• Subtle maxillofacial morphology changes in congenital syndromes present diagnostic challenges.
• Artificial intelligence (AI) offers potential for diagnosis through advanced shape analysis.

Purpose of the Study:

• To apply the Swap Disentangled Variational Autoencoder (SD-VAE) for diagnosing Beckwith-Wiedemann syndrome (BWS).
• To evaluate the diagnostic accuracy of SD-VAE in distinguishing BWS patients from controls using 3D head scans.

Main Methods:

• Trained SD-VAE on 72 3D head scans (stereophotogrammetry) and CT scans from 56 BWS patients.
• Pre-processed scans with 68 anatomic landmarks for uniformity and comparison.
• Visualized and classified SD-VAE outputs in 2D space to assess diagnostic performance per facial region.

Main Results:

• Achieved perfect diagnostic accuracy for BWS on the test set.
• Identified chin, cheeks, zygoma, eyes, jaw, and supraorbital region as most characteristic.
• Demonstrated high accuracy in distinguishing BWS-specific features from the general population.

Conclusions:

• SD-VAE is a promising tool for quantifying BWS features from 3D head meshes.
• The AI model can aid in the future referral and diagnosis of Beckwith-Wiedemann syndrome.
• AI-driven shape analysis provides high diagnostic accuracy for syndromic facial morphology.