Tongue Manifestation Feature Classification and Detection for Chronic Gastritis Based on the YOLO Framework

In response to the demand for modernization and objectification of Traditional Chinese Medicine (TCM) tongue diagnosis, this study delves into the integration of YOLO Framework to automate the classification and detection of tongue manifestation features. YOLOv5, YOLOv5 with integrated Squeeze-and-Excitation (SE) attention mechanism (YOLOv5_SE), YOLOv8, and YOLOv8 with integrated SE attention mechanism (YOLOv8_SE), are selected to extract features from the tongue manifestations of chronic gastritis patients. In tongue diagnosis, four key tongue manifestation features are focused on, such as tongue color, coating color and thickness, and tongue shapes like fissured, teeth-marked, and spotted. Experimental results show that the YOLOv8 model demonstrates the highest accuracy in the classification task of tongue manifestation features, reaching 95.9%, while the accuracy of YOLOv5_SE is the lowest at 84.1 %. Compared to their original models, the accuracy rates of YOLOv5_SE and YOLOv8_SE with the integrated SE attention mechanism slightly decrease by 0.7% and 1.0 %, respectively. Valuable insights and directions are provided by this study in the construction of dataset, model optimization, and algorithm improvement in the field of multimodal tongue manifestation feature classification and detection using the YOLO Framework.