Tongue Image Segmentation and Constitution Identification with Deep Learning

Electronics, Год журнала: 2025, Номер 14(4), С. 733 - 733

Опубликована: Фев. 13, 2025

Traditional Chinese medicine (TCM) gathers patient information through inspection, olfaction, inquiry, and palpation, analyzing interpreting the data to make a diagnosis offer appropriate treatment. Traditionally, interpretation of this relies heavily on physician’s personal knowledge experience. However, diagnostic outcomes can vary depending clinical experience subjective judgment. This study employs AI methods focus localized tongue assessment, developing an automatic body segmentation using deep learning network “U-Net” series optimization processes applied surface images. Furthermore, “ResNet34” is utilized for identification “cold”, “neutral”, “hot” constitutions, creating system that enhances consistency reliability results related tongue. The final demonstrate accuracy reaches level junior TCM practitioners (those who have passed practitioner assessment with ≤5 years experience). framework findings serve as (1) foundational step future integration pulse electronic medical records, (2) tool personalized preventive medicine, (3) training resource students diagnose constitutions such “hot.”

Язык: Английский

Development of a tongue image-based machine learning tool for the diagnosis of acute respiratory tract infection (Preprint)

Опубликована: Март 18, 2025

Background: Tongue characteristics, widely utilized in traditional Chinese medicine for health assessment, have been shown to correlate with specific respiratory infections. With the ongoing global spread of Human adenoviruses (HAdVs), COVID-19, and other seasonal viruses, this study aims enhance convenience cost-effectiveness infection diagnoses by developing prediction models based on tongue characteristics. Method: This deep learning extract features from 280 images collected COVID-19 patients, HAdVs healthy individuals. Machine diagnostic were subsequently trained these characteristics distinguish between normal cases those indicative The key identified machine algorithms further visualized a two-dimensional space. Result: Nine significant identified: coating color (red, green, blue), presence tooth marks, crack ratio, moisture level, texture directionality, roughness, contrast. Diagnostic achieved an area under precision-recall curve exceeding 70%, receiver operating characteristic surpassing 80% general performance. SHAP value revealed that color, direction most influential features. Conclusion: Our findings demonstrate potential diagnosis identifying pathogens responsible acute tract infections at time admission. approach holds clinical implications, offering reduce clinician workloads while improving accuracy overall quality medical care.

Язык: Английский

A hybrid model using multimodal feature perception and multiple cross-attention fusion for depressive episodes detection

Information Fusion, Год журнала: 2025, Номер unknown, С. 103354 - 103354

Опубликована: Июнь 1, 2025

Язык: Английский