Deep Learning-Based Segmentation of the Submandibular Salivary Gland on Ultrasonography.

Authors

Abstract

To develop and evaluate a deep learning model for automatic detection and segmentation of the submandibular salivary gland on ultrasonography using the YOLOv8x architecture. A cross-sectional analytical study. Place and Duration of the Study: Department of Oral and Maxillofacial Radiology, Faculty of Dentistry, Marmara University, Istanbul, Turkiye, from March to September 2025. A total of 414 anonymised ultrasound images of normal submandibular glands were retrospectively analysed. Polygonal annotations were performed and verified by expert radiologists. The dataset was divided into training (80%), validation (10%), and test (10%) sets. A YOLOv8x-seg model was trained for 800 epochs using augmentation techniques including noise injection, contrast adjustment, and blurring. Model performance was evaluated using precision, sensitivity, F1-score, Dice coefficient, and intersection-over-union (IoU). The model achieved a sensitivity of 93%, a precision of 90%, and an F1-score of 0.92. Segmentation performance showed a Dice coefficient of 0.87 and an IoU of 0.78, indicating high agreement with expert annotations. Deep learning using YOLOv8x enables accurate and efficient segmentation of the submandibular gland in ultrasound images. This approach may improve diagnostic consistency and workflow efficiency. Future studies should validate these findings using larger, multi-centre datasets including pathological cases. Artificial intelligence, Deep learning, YOLOv8x, Ultrasonography, Submandibular salivary gland, Automatic detection.

Tags

Topics