Introduction: Chronic rhinosinusitis in children represents a significant clinical challenge. Imaging-based diagnosis relies primarily on descriptive assessment of computed tomography scans and simplified scoring systems that do not account for the actual volume of inflammatory lesions. A more quantitative analysis of sinus involvement could improve diagnostic accuracy and reduce interobserver variability. Objective: The aim of this study was to develop a machine learning algorithm for the automated quantitative assessment of maxillary sinus volume and inflammatory involvement in children based on computed tomography scans. Materials and methods: The study includes 92 computed tomography scans of the paranasal sinuses obtained from paediatric patients, with the maxillary sinuses manually annotated by human experts. A threedimensional convolutional neural network was developed for sinus segmentation. Based on the segmentation results, sinus volume and the percentage of sinus involvement by inflammatory changes were calculated. Results: The algorithm demonstrates high segmentation performance, achieving mean sensitivity, precision, and Dice coefficient values of nearly 90%. The root-mean-square error for estimation of sinus volume and inflammatory involvement is 0.4 cm³ and 1.3 percentage points, respectively. Conclusions: The proposed approach enables rapid and precise assessment of maxillary sinus volume and the extent of inflammatory involvement in paediatric computed tomography examinations. It may serve as a valuable tool supporting clinical decision-making in the diagnosis and longitudinal monitoring of chronic rhinosinusitis in children.

paediatric chronic rhinosinusitis, maxillary sinus, volumetric analysis, computed tomography, medical image segmentation, convolutional neural network, deep learning, artificial intelligence