CBCT image based radiomic analysis for condylar resorption after orthognathic surgery

Area of Science:

• Oral and Maxillofacial Surgery
• Radiology
• Biomedical Engineering

Background:

• Condylar resorption (CR) is a significant complication following orthognathic surgery (OS).
• Identifying predictive features of CR is crucial for patient outcomes.
• Current methods for feature selection may not fully capture dynamic changes post-surgery.

Purpose of the Study:

• To introduce and validate a novel D-value method for selecting features highly correlated with CR after OS.
• To identify specific radiomic features that change significantly during CR.
• To assess the utility of the D-value method in feature specification compared to baseline values.

Main Methods:

• Cone beam computed tomography (CBCT) images from 145 orthognathic surgery patients (T1: baseline, T2: postoperative) were analyzed.
• Patients were categorized into CR and control groups.
• MVEL-Net and Pyradiomics were used for condyle segmentation and feature extraction.
• A D-value (absolute difference between T1 and T2 features) was calculated to identify dynamic changes.

Main Results:

• 82 features showed statistical differences at baseline (T1) between CR and control groups.
• The D-value method reduced the feature set to 32 features (F_d), with only 3 not present in the initial T1 set.
• This indicates the D-value effectively highlights features dynamically changing with CR.

Conclusions:

• The proposed D-value method is effective for selecting specific features related to CR after OS.
• This approach enhances feature specification compared to using only baseline (T1) values.
• The identified features hold potential for developing robust prediction models for CR post-orthognathic surgery.