Diagnosis and Screening of Velocardiofacial Syndrome by Evaluating Facial Photographs Using a Deep Learning-Based...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Genetics

Background:

• Early detection of rare genetic diseases like velocardiofacial syndrome (VCFS) is crucial but challenging due to rarity and limited physician experience.
• Deep learning (DL) offers a promising avenue for efficient and accurate diagnosis of such conditions.
• This study explores DL for developing a face recognition model to diagnose VCFS.

Purpose of the Study:

• To investigate the effectiveness of a deep learning algorithm for diagnosing VCFS using facial photographs.
• To develop and evaluate a face recognition model for VCFS detection.
• To assess the potential of AI in identifying rare genetic disorders.

Main Methods:

• Trained a multitask cascaded convolutional neural networks (MTCNN) model using publicly available labeled facial datasets.
• Evaluated the binary classification performance for VCFS diagnosis using the most efficient face recognition model.
• Utilized a dataset of 98 VCFS patients and 91 controls, randomly divided into training and testing sets.

Main Results:

• The face recognition model achieved high accuracy, between 94% and 99%, depending on the training data.
• The VCFS diagnostic model's accuracy ranged from 81% to 88% with varied angles and 95% with frontal photographs.
• Gradient-weighted class activation mapping highlighted perinasal and periorbital areas, consistent with VCFS facial phenotypes.

Conclusions:

• A MTCNN-based model can effectively detect VCFS from facial photographs alone.
• The high accuracy demonstrates the potential of deep learning in aiding early diagnosis of rare genetic diseases.
• This AI-driven approach can facilitate timely interventions and improve patient care for VCFS.