Multi-Model Approach for Tongue Image Classification in Traditional Thai Medicine

Abstract

Nowadays, complementary medicine is gaining widespread acceptance and is widely accepted, particularly within traditional Thai medicine (TTM). Tongue inspection is a primary method for diagnosing health conditions, as it reflects organ functionality. However, diagnostic results can vary depending on the expertise of TTM practitioners. In this work, we propose methods that incorporate transfer learning (TL) from deep learning (DL), machine learning (ML), and statistical models, using various tongue features. We introduced a collected dataset for evaluation. Experimental results demonstrated that the DenseNet121 model, trained on tongue images pre-processed with histogram equalisation (HE), achieved the best performance, with accuracy, sensitivity, and specificity of 0.89, 0.83, and 0.92, respectively. Model ensembling and paired t-tests were used to analyse the results. Finally, we identified the best approach and models for potential clinical use to assist in the pre-diagnostic analysis of tongue images for TTM practitioners and general users via our web application at http:// bioservices.sci.psu.ac.th/.