A 3D Clinical Face Phenotype Space of Genetic Syndromes Using a Triplet-Based Singular Geometric Autoencoder

Area of Science:

• Medical Genetics
• Computer Vision
• Machine Learning

Background:

• Objective facial phenotyping is crucial for diagnosing genetic syndromes.
• Current methods may lack the precision needed for complex dysmorphic features.
• A standardized, quantitative approach to facial analysis is needed.

Purpose of the Study:

• To develop and evaluate a novel low-dimensional metric space, the clinical face phenotypic space (CFPS), for enhancing clinical diagnosis of genetic syndromes.
• To create a facial matching tool that aids in interpreting facial dysmorphisms by contextualizing them within known patterns.
• To demonstrate the utility of CFPS in classifying syndromes, generalizing to new ones, and preserving genetic disease relatedness.

Main Methods:

• Development of a triplet loss-based autoencoder using geometric deep learning (GDL) and multi-task learning (combining supervised and unsupervised approaches).
• The model comprises a GDL-based encoder, a reconstruction-focused decoder, and a singular value decomposition layer.
• Experiments were designed to test CFPS properties including classification accuracy, generalization to novel syndromes, and preservation of genetic disease relationships.

Main Results:

• The developed CFPS accurately classifies syndromes and generalizes to novel, previously unseen syndromes.
• The CFPS preserves the relatedness of genetic diseases, clustering phenotypically similar disorders that reflect functional gene relationships.
• The proposed GDL-based CFPS outperforms a linear metric learning baseline in both syndrome classification and generalization.

Conclusions:

• The clinical face phenotypic space (CFPS) offers a powerful, quantitative tool for assessing facial dysmorphism in clinical genetics.
• Its ability to accurately classify syndromes, generalize to new cases, and reflect genetic relationships makes it valuable for clinical practice.
• This novel approach can be integrated into current clinical workflows to improve diagnostic capabilities for genetic syndromes.