Machine Learning-Based Identification of Craniosynostosis From Nonsynostotic Cranial Deformities

Area of Science:

• Medical Imaging and Diagnostics
• Computational Biology and Bioinformatics
• Pediatric Surgery

Background:

• Craniosynostosis, premature cranial suture fusion, affects 5.2/10,000 births, potentially impairing neurodevelopment.
• Current screening lacks efficiency, leading to delayed treatment, unnecessary referrals, and increased healthcare costs.
• Early differentiation of synostotic from nonsynostotic conditions is crucial for timely intervention.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for early identification of single-suture craniosynostosis.
• To differentiate between synostotic and nonsynostotic head shapes using head volume analysis.
• To identify the specific fused suture (metopic, sagittal, coronal, lambdoid) in affected infants.

Main Methods:

• Head volumes were analyzed using photogrammetry or CT scans.
• 3D point data underwent augmentation for data imbalance.
• Statistical shape modeling, Principal Component Analysis, and an XGBoost ML model were employed.
• Three models were trained and validated using 10-fold cross-validation.

Main Results:

• Model-1 achieved 97.8% accuracy in identifying craniosynostosis versus nonsynostotic conditions.
• Model-2 demonstrated 86.1% sensitivity and 96.8% specificity in identifying the fused suture.
• Model-3 showed 76.8% sensitivity and 97.8% specificity in classifying head shapes.

Conclusions:

• An ML approach reliably differentiates synostotic from nonsynostotic head shapes.
• The ML model accurately distinguishes between different types of single-suture craniosynostosis.
• This technology can assist pediatric providers in early and accurate diagnosis of craniosynostosis.