CBCT-based volumetric assessment of the maxillary sinus in temporomandibular disorder: Integration of morphometric analysis and classical machine learning classification

TL;DR

This study uses CBCT scans and machine learning to find that people with temporomandibular disorders have smaller maxillary sinuses compared to those without the disorder.

Contribution

The integration of morphometric analysis and classical machine learning to assess maxillary sinus characteristics in TMD patients is novel.

Findings

Maxillary sinus volume and surface area were significantly smaller in the TMD group compared to controls.

Logistic Regression was the most successful machine learning classifier for predicting TMD status.

Including gender as a feature improved machine learning model performance.

Abstract

This study aims to determine the maxillary sinus volume and surface area values and their relationship in individuals with and without temporomandibular disorders (TMDs) using Cone-Beam Computed Tomography and Machine Learning.

This retrospective study was performed on 127 subjects, 66 in the control group (41 females, 25 males; mean age 28.35 ± 9.9years) and 61 in the TMD group (54 females, 7 males; mean age 35 ± 12.6years) using dento-maxillo-facial CBCT images. Images were acquired as DICOM files and imported into 3D Slicer (version 5.6.2). The volume and surface area of the maxillary sinus were automatically calculated by the 3D Slicer programme. In addition, automatic prediction was performed using classical machine learning techniques on the dataset obtained in the study.

Maxillary sinus volume was 30.85 ± 10.14 cm3 in the control group and 26.97 ± 10.33 cm3 in the TMD group. Maxillary sinus volume and surface area were significantly smaller in the TMD group compared to controls. No significant differences were observed between age decades in either group. Furthermore, the results obtained in machine learning showed that gender selection generally improved the results, and the most successful classifier was the Logistic Regression algorithm.

This study demonstrates that TMDs were associated with smaller sinus volume. Furthermore, a machine learning-based model can be used to discriminate temporomandibular dysfunction even when the size of the dataset is small.

Figures