Tongue Image Segmentation Using YOLOv11-based Object Detection and Mobile Segment Anything Model

Tongue image segmentation plays a crucial role in automated tongue diagnosis systems in Traditional Chinese Medicine (TCM), as it directly affects subsequent feature extraction and classification accuracy. Consequently, a myriad of deep learning-based models have been developed. However, most existing methods still rely on large-scale networks with heavy computational requirements and often require manual prompt inputs, which limit their efficiency and practicality in real-world applications. To address these issues, we propose a novel tongue segmentation framework that integrates YOLO-based object detection with the Mobile Segment Anything Model (MobileSAM). Specifically, YOLOv11 is employed to automatically localize the tongue region and generate high-confidence bounding boxes as spatial prompts, enabling MobileSAM to perform precise segmentation without human intervention. The proposed method realizes an end-to-end segmentation process that is both efficient and scalable. Extensive experiments on diverse tongue image datasets demonstrate that the proposed framework achieves accurate and robust segmentation while maintaining low computational cost, highlighting its strong generalization capability and practical value for clinical and research applications.