Comprehensive analysis of the upper airway across different skeletal malocclusions

In this study, we aimed to find out the standard volumetric space and variations in themeasurements of different landmarks in adults with different skeletal relations of the maxillaand the mandible by means of CBCT analyzing software. We also aimed to develop multiplemachine learning models to predict skeletal malocclusions with an acceptable level of accuracyusing airway and cephalometric landmark values obtained from analyzing different CBCTimages. The present retrospective study was conducted using CBCT records of 90 adult patientsof each skeletal class, i.e., Class I, Class II, and Class III. The boundaries were defined for the3 major regions: the nasopharynx, the oropharynx, and the hypopharynx. Various measurementswere recorded across these regions, as well as selective cephalometric landmarks, in an effortto establish a correlation. For the second part of the study, 300 samples of skeletal anatomicaldata were retrospectively obtained and recorded in Digital Imaging and Communications inMedicine (DICOM) file format. The DICOM files were used to reconstruct 3D models usingv3DSlicer (slicer.org) by thresholding airway regions to build up 3D polygon models of airwayregions for each sample. The 3D models were measured for different landmarks, includingmeasurements across the nasopharynx, oropharynx, and hypopharynx. Male and femalesubjects were combined as one data set to develop supervised learning models. Thesemeasurements were utilized to build 7 artificial intelligence-based supervised learning models.The regression analysis results demonstrated that Class III individuals exhibit a reduced airwayvolume in the nasopharynx compared to other groups, while Class II individuals display adiminished airway volume in the hypopharynx. On comparing Class I and II malocclusions onhypopharyngeal volume (cm3) across genders, it was seen that for females (Table 5), there wasa significant mean difference in hypopharynx between Class I (13.18) and Class II (10.84); P =0.002. Also, for males, there was a significant mean difference in hypopharynx between class I(13.29) and class II (10.71); P = 0.001. The supervised learning model with the best accuracywas Random Forest, with a value of 0.74. All the other models were lower in terms of theiraccuracy. The recall scores for Class I, II, and III malocclusions were 0.71, 0.69, and 0.77,respectively, representing the total number of actual positive cases predicted correctly,making the model's sensitivity high. The significant findings of this study were the presenceof a reduced nasopharyngeal volume in class III malocclusions while class II individualsdisplayed a diminished hypopharyngeal volume, making these critical areas to consider duringthe diagnostic and orthodontic treatment planning stage. This study also revealed a consistentcorrelation between Sella length and depth across various facial skeletal profiles, with class IIpatients exhibiting a distinct pattern and class III showing an average relationship. The authorsconcluded that the Sella turcica morphology may be utilized as an indicator for predictingskeletal malocclusions. In our study, we observed that the Random Forest model was the mostaccurate model for predicting the skeletal malocclusion based on various airway andcephalometric landmarks considered in our study.