Traditional Chinese Medicine (TCM) relies heavily on tongue diagnosis as one of its most important diagnostic methods, utilizing the tongue to reflect the overall health of a patient. A new lightweight version of YOLOv8 has been developed, which enhances the detection of tongue fissures and teeth marks, potentially revolutionizing how these attributes are assessed clinically.
The newly proposed YOLOv8 model incorporates the C2f_DCNv3 module, integrating advanced Deformable Convolutions. This upgrade significantly boosts the model’s ability to accurately identify and analyze complex features, such as fine fissures and teeth marks found on the tongue, which are of great importance for TCM diagnosis. The introduction of the Squeeze-and-Excitation (SE) attention mechanism also optimizes feature weighting, allowing the model to focus on the most relevant areas of the image.
According to the study, the enhanced YOLOv8 model achieved impressive results, with a mean Average Precision (mAP) of 92.77%. This marks substantial progress compared to the original YOLOv8, providing practitioners with more reliable diagnostic tools. Notably, the model has managed to increase its mAP for teeth marks to 94.21%, and for tongue fissures, it reached 91.34%. Along with higher accuracy, the improved model also reduces computational costs by approximately one-third, allowing for increased efficiency in clinical environments.
Authors from Shanxi University of Traditional Chinese Medicine underscored the importance of accurate damage detection for effective patient diagnosis: "By analyzing tongue fissures and teeth marks, TCM practitioners can make more accurate assessments of diseases and formulate individualized treatment plans." This enhancement not only supports healthcare practitioners but also allows for the potential integration of these tools within patient self-assessment applications, fostering preventive care.
Prior methods of tongue feature analysis often fell short due to limitations arising from traditional image processing techniques. The new design of YOLOv8 tackles these challenges head-on by focusing on adaptive learning capabilities which are fundamental for processing stereotypically irregular features such as those found on the tongue.
This model’s development and the integration of the C2f_DCNv3 highlights the importance of modern computational methods within TCM diagnostics. The adaptability of the model means it can accurately detect variations, which is key to successful tongue diagnostics. The researchers concluded: "The combination of advanced machine learning techniques with TCM practices paves the way for innovative diagnostic approaches. This not only provides stronger support for TCM tongue diagnosis but also opens new application prospects for medical image analysis across various medical fields, potentially improving outcomes for many patients."
Continued exploration and enhancement of the YOLOv8 architecture could lead to new breakthroughs, as this model currently does not support video recognition yet. This leaves the door open for future developments and adjustments, indicating the vast potential for applying such technology widely, adapting it for various clinical conditions where tongue diagnostics are relevant.