Machine learning as new approach for predicting of maxillary sinus volume, a sexual dimorphic study

Abstract

The maxillary sinus is the most prominent in humans. Maxillary Sinus Volume (MSV) has grown in popularity as a tool to predict the success of various dental procedures and oral surgeries and determine a person's gender in cases such as forensic investigations when only partial skulls are available. Because it is an irregularly shaped cavity that may be difficult to measure manually, robust imaging techniques such as cone-beam computed tomography (CBCT) used in conjunction with machine learning (ML) algorithms may offer quick and vigorous ways to make accurate predictions using sinus data. In this retrospective study, we used data from 150 patients with normal maxillary sinuses to train and evaluate a Python ML algorithm for its ability to predict MSV from basic patient demographics (age, gender) and sinus length measurements in three directions (anteroposterior, mediolateral, and superoinferior). The model found sinus length measurements had significantly higher predictive values than either age or gender and could predict MSVs from these length measurements with almost linear accuracy indicated by R-squared values ranging from 0.97 to 0.98% for the right and left sinuses.