Estimation of frontal sinus volume with morphometric measurements and machine learning algorithms

In our study, designed to predict frontal sinus volumes using morphometric measurements taken from the face, we utilized paranasal computed tomography (CT) images of 256 individuals aged between 20 and 55 years (median: 32, IQR: 15) who presented at the Ear, Nose, and Throat Clinic of Bolu Abant Izzet Baysal University Education and Research Hospital. These individuals had no history of trauma, congenital anomalies, or surgery. The CT images were obtained using a 1.5 Tesla scanner. For the calculation of frontal sinus volume, stereological methods were employed in the transverse plane. In the frontal plane, measurements were taken for frontal sinus height and width, orbital height and width, maxillary sinus height and width, nasal cavity height and width, as well as interzygomatic, intermastoidal, and mastoangular distances. Parameters such as Orbital Width, Maxillary Height, and Interangular Width, which did not conform to a normal distribution and showed low correlation with sinus volume, were excluded. Machine learning algorithms were then applied for regression analysis. At the end of the regression analysis, frontal sinus volume was explained by the parameters of Frontal Height, Frontal Width, and Interzygomatic Distance. The Random Forest model exhibited the highest performance, while the K Nearest Neighbor model was found to be the least costly.