Applying deep learning techniques to segmentize and classify tongue regions for traditional and complementary medicine (TCM) diagnosis

Yeap, Chun Hong (2025) Applying deep learning techniques to segmentize and classify tongue regions for traditional and complementary medicine (TCM) diagnosis. Final Year Project, UTAR.

PDF Download (3529Kb)

Abstract

Tongue diagnosis is a fundamental component of Traditional and Complementary Medicine (TCM), yet manual inspection remains subjective and inconsistent. This study proposes a deep learning framework to enhance tongue image analysis through segmentation, classification, and explainable artificial intelligence (XAI). A Mobile U-Net model was proposed and developed for efficient and accurate tongue region segmentation. Classification tasks were conducted for both binary (stained vs. non-stained) and multi-class pathological coatings, covering clinically relevant categories. Lightweight architectures, including the proposed Efficient-ResNet, achieved competitive accuracy with minimal computational cost, demonstrating strong potential for deployment in resource-constrained environments. Grad-CAM was integrated to provide visual explanations of model decisions, improving transparency and clinical trust. Experimental results show that ResNet50 and LECA-EfficientNetV2-S achieved the highest accuracy of 99% in binary classification, while EfficientNetV2-B3 and -S excelled in multi-class tasks. Efficient-ResNet maintained strong accuracy (98.5%) with only 0.31M parameters. The findings highlight the framework’s balance of efficiency, accuracy, and interpretability, offering a practical solution to standardize and modernize tongue diagnosis in TCM for both clinical and telemedicine applications.

Actions (login required)