Marfan Syndrome Prediction Via Graph Neural Networks on 3D Facial Cues.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Genetics

Background:

• Marfan syndrome (MFS) is a rare genetic connective tissue disorder.
• Early detection of MFS is crucial for preventing life-threatening cardiovascular complications.
• Current diagnostic methods involve complex, costly clinical assessments and genetic testing.

Purpose of the Study:

• To propose a novel 3D vision-based method for rapid, non-invasive screening of Marfan syndrome.
• To leverage AI for analyzing 3D facial morphology for MFS detection.
• To improve the efficiency and accessibility of Marfan syndrome screening.

Main Methods:

• Utilized Graph Neural Networks (GNNs) to analyze 3D facial morphology from scans.
• Employed manually or automatically extracted facial landmarks to capture craniofacial patterns.
• Incorporated a statistically derived sparse graph topology for enhanced robustness and accuracy.
• Implemented an adjustable decision threshold to prioritize a low false-negative rate.

Main Results:

• Achieved strong discriminative performance in experimental evaluations.
• Demonstrated high classification accuracy above 93% on a dataset of 440 subjects (126 MFS, 314 controls).
• Reported an Area Under the Curve (AUC) of up to 0.98, indicating excellent diagnostic capability.

Conclusions:

• The proposed 3D vision-based method shows significant potential for rapid, non-invasive Marfan syndrome screening.
• AI-powered facial analysis can effectively identify subtle craniofacial patterns associated with MFS.
• This approach could complement existing diagnostic tools, facilitating earlier intervention and improved patient outcomes.