Analytical Comparison of Maxillary Sinus Segmentation Performance in Panoramic Radiographs Utilizing Various YOLO Versions

Abstract

Objective: In this study, we aimed to evaluate the success of the last three versions of YOLO algorithms, YOLOv5, YOLOv7 and YOLOv8, with segmentation feature in the segmentation of the maxillary sinus in panoramic radiography. Methods: In this study, a total of 376 participants aged 18 years and above, who had undergone panoramic radiography as part of routine examination at Gaziantep University Faculty of Dentistry, Department of Oral and Maxillofacial Radiology, were included. Polygonal labeling was performed on the obtained images using Roboflow software. The obtained panoramic radiography images were randomly divided into three groups training group (70%), validation group (15%) and test group (15%). Results: In the evaluation of the test data for maxillary sinus segmentation, sensitivity, precision, and F1 scores are 0.92, 1.0, 0.96 for YOLOv5, 1.0, 1.0, 1.0 for YOLOv7 and 1.0, 1.0, 1.0 for YOLOv8, respectively. Conclusion: These models have exhibited significant success rates in maxillary sinus segmentation, with YOLOv7 and YOLOv8, the latest iterations, displaying particularly commendable outcomes. This study emphasizes the immense potential and influence of artificial intelligence in medical practices to improve the diagnosis and treatment processes of patients.