Objective classification system for sagittal craniosynostosis based on suture segmentation.

Area of Science:

• Medical Informatics
• Computational Biology
• Neurosurgery

Background:

• Spring-assisted surgery is a minimally invasive treatment for sagittal craniosynostosis (CSO).
• Accurate patient-specific spring selection is crucial but currently relies on subjective clinical judgment of CT scans.
• An objective system is needed to stratify CSO patients and standardize spring selection.

Purpose of the Study:

• To develop a novel informatics system for automatic segmentation and classification of sagittal CSO subtypes.
• To reduce subjectivity in spring force selection for spring-assisted surgery.

Main Methods:

• A three-phase system involving preprocessing, suture segmentation using a "hemispherical projection" algorithm, and classification.
• Extraction and selection of 108 quantified suture features using a multiclass feature scoring system.
• Classification of sagittal CSO into four subtypes (anterior, central, posterior, complex) using a support vector machine, evaluated with fivefold cross-validation and ROC curves.

Main Results:

• The automated suture segmentation was found to be clinically acceptable.
• The classification system achieved a fivefold cross-validation accuracy of 72.7% for the four subtypes.
• The system demonstrated reliability with an area under the ROC curve greater than 0.8 for two-class problems.

Conclusions:

• The developed hemispherical projection algorithm effectively segments cranial vault sutures.
• The classification system provides desirable performance for stratifying sagittal CSO subtypes.
• This informatics system is expected to enhance clinical research and facilitate the widespread adoption of minimally invasive spring-assisted CSO treatment.