Digital Tools in TCM Diagnosis: Tongue and Pulse Devices and AI

Why people are building digital tools

Tongue and pulse examination in traditional Chinese medicine (TCM) depend on a practitioner's skill and judgment, and studies have found that agreement between practitioners is often limited. [1] One reason researchers turn to cameras, sensors and AI is to make these findings measurable and consistent. This page describes research, not products to buy. We do not recommend any device, app or service, and none replaces a clinician.

Tongue imaging and AI

• Standardizing the image. A small pilot (76 patients, two observers) used a fixed camera setup to standardize tongue photos. Agreement between observers was generally low, and results varied by observer and feature. [2]
• Deep learning overview. A 2026 systematic review describes four main methods and three main task types in deep-learning tongue diagnosis, and presents itself as a foundation for building clinically trustworthy and deployable frameworks, not as proof that they exist yet. [3]
• A concrete example: stomach conditions. A 2026 systematic review of nine studies on AI tongue-image analysis for gastric cancer, precancerous lesions and related disorders found internally validated AUCs from 0.61 to 0.99. Only three studies tested the models on outside data, and performance generally dropped there. The authors call it promising but early-stage, a possible triage aid before endoscopy and not a replacement for it. [4]

Label: promising but limited for specific screening questions; insufficient evidence that AI tongue reading can establish a TCM pattern that predicts treatment response (we found no study showing this).

Pulse devices

A 2013 study used a tactile sensor with 12 points to compare two ways of taking the pulse and found similar signals at a given depth, as part of an effort to standardize pulse-taking. [5] A 2020 feasibility study with 80 people used a smartphone with two sensors and a neural network, and reported more than 90% prediction accuracy for a TCM "wiry" pulse. [6] That is a proof of concept: it predicts a traditional label, and says nothing about whether that label guides better treatment.

AI language models

• A 2026 scoping review of 27 studies on TCM-focused large language models found uses in knowledge questions (10 studies) and diagnostic help (13). Models handle basic syndrome-differentiation reasoning but are limited in complex, individualized diagnosis, and clinical evaluation is a challenge. [7]
• A 2025 method study tested models on 200 medical cases; the authors report their fine-tuned model reached 85.7% and 81.2% on two tasks, and that existing general and TCM-specific models still have significant limitations. [8]

Matching an expert's answer on a test set is not the same as helping real patients. Chatbots can sound confident and still be wrong. Do not use an AI answer to start, change or stop a medicine or herb.

Questions to ask about any digital TCM tool

• What exactly does it claim to detect: a disease, or a TCM pattern?
• Was it tested on people unlike its training data (external validation)?
• Who funded the study, and was it independent?
• What happens to your photos or data?
• Does it say to see a doctor for serious symptoms?

Related reading

Tongue Diagnosis in Chinese Medicine, Pulse Diagnosis in Chinese Medicine, Online TCM Quizzes and Self-Diagnosis, How to Evaluate a Therapy Claim.