Tongue Tech: AI diagnoses diseases by tongue color with 96% accuracy

Artificial intelligence systems can now diagnose diseases by analyzing tongue color with over 96% accuracy, bridging ancient medical wisdom with cutting-edge machine learning technology. This breakthrough represents a fascinating convergence where traditional Chinese medicine meets modern healthcare innovation, potentially offering a non-invasive, rapid diagnostic tool for conditions ranging from diabetes to COVID-19.

The technology stems from a practice thousands of years old. Traditional Chinese Medicine (TCM) practitioners have long examined patients’ tongues as part of comprehensive health assessments, studying color, shape, and coating to detect illness. What was once entirely dependent on human observation and interpretation is now being standardized and automated through sophisticated AI systems.

The science behind tongue diagnosis

In traditional Chinese medicine, tongue color reflects the condition of blood and qi—a concept often translated as “vital energy”—making it a primary health indicator. However, this examination method has always been highly subjective, relying entirely on individual practitioners’ color perception and experience.

Western medicine has been more cautious about tongue analysis. While dentists and hygienists routinely examine tongues during oral cancer screenings, no standardized clinical system exists for monitoring tongue features as diagnostic indicators. Frank Scannapieco, a periodontist and microbiologist at the University at Buffalo, notes that although defined tongue lesions can indicate certain cancers, and some studies have linked tongue appearance to diseases like breast cancer and psoriasis, the practice lacks systematic application.

The World Health Organization officially recognized TCM diagnoses in 2022 by adding them to the International Classification of Diseases, the global standard for health information classification. However, the medical community remains divided on TCM’s scientific validity, with many researchers citing insufficient evidence-based medicine support.

How AI transforms tongue analysis

Recent advances in computing technology have enabled researchers to apply machine learning to tongue diagnosis, potentially eliminating human subjectivity while maintaining diagnostic accuracy. A 2024 study published in Technologies demonstrated how AI models could classify tongue colors and predict associated medical conditions with remarkable precision.

The key breakthrough involved solving a fundamental challenge: lighting conditions. Previous tongue-imaging studies suffered from perception bias caused by varying light environments, making color analysis unreliable. Javaan Chahl, a roboticist and joint chair of sensor systems at the University of South Australia, explains that “color is very subjective” without controlled lighting conditions.

Chahl’s team developed a standardized lighting system within a kiosk setup. Patients place their heads inside a box illuminated by LED lights that emit stable, controllable wavelengths. This eliminates lighting variables that previously made color analysis inconsistent.

The researchers trained their machine learning models using 5,260 images—both real tongue photographs from internet sources and additional color-gradient images. These models learned to recognize seven specific colors (red, yellow, green, blue, gray, white, and pink) at different saturation levels across various lighting conditions.

Research results and diagnostic accuracy

The AI system achieved impressive diagnostic accuracy when tested on real patients. Using 60 tongue images taken at two hospitals in Iraq during 2022 and 2023, the system correctly identified 58 out of 60 cases when compared against patients’ medical records—a 96.6% accuracy rate.

The research revealed specific tongue color patterns associated with different health conditions:

• Healthy tongues typically appear pink with a thin white coating
• Iron deficiency manifests as a whiter-looking tongue
• Diabetes often produces a bluish-yellow tongue coating
• Certain cancers may cause a purple tongue with a thick, fatty layer
• COVID-19 severity correlates with tongue color intensity: faint pink for mild cases, crimson for moderate infections, and deep red for serious cases

These findings suggest that tongue color changes may serve as early indicators for various medical conditions, potentially enabling faster diagnosis and treatment.

Commercial applications and market potential

The technology has already attracted commercial interest, though scaling remains challenging. The biggest limitation involves data collection—implementing kiosk systems in large hospitals requires extensive coordination and patient consent for accessing medical records.

Consumer applications are emerging despite regulatory constraints. Earlier this year, researchers from the University of Missouri launched BenCao, a GPT-based AI application that allows users to upload tongue images and receive personalized health guidance based on TCM principles. The app is marketed strictly as a “wellness” tool rather than a clinical diagnostic system, providing food and lifestyle recommendations instead of medical diagnoses.

This cautious approach reflects the regulatory reality of medical AI applications. Providing actual medical diagnoses requires far more rigorous validation and regulatory approval than wellness guidance.

Limitations and challenges ahead

Several significant obstacles must be overcome before AI tongue diagnosis becomes mainstream medical practice. Dong Xu, whose research at the University of Missouri focuses on computational biology and bioinformatics, emphasizes that tongue analysis represents only one component of complete medical diagnosis. The technology cannot stand alone for making accurate clinical decisions.

Standardization presents another major hurdle. Image labeling isn’t widely standardized for tongue analysis research, making it difficult to reproduce findings across different studies. Scannapieco highlights that broad AI-based tongue analysis would require massive investment and enormous databases of images paired with medical histories.

The technology also faces inherent limitations. Many diseases show no changes in tongue appearance, meaning this diagnostic tool would serve as just one element in comprehensive medical assessment rather than a standalone solution.

Data collection remains the primary bottleneck for scaling research. Gathering usable data requires extensive coordination with hospitals, healthcare providers, and patients—a complex process that limits rapid expansion.

Future development and research directions

Current research is expanding beyond basic color analysis. Ali Al-Naji, now a medical engineering professor at the Middle Technical University in Iraq and co-author of the 2024 study, is working to narrow diagnostic focus to the tongue’s center and tip for more precise analysis.

His team is also using a new dataset of 750 internet images to examine tongue shape and oral conditions such as ulcers and cracks using YOLO (You Only Look Once), a deep-learning algorithm designed for real-time object detection. Eventually, researchers hope to analyze entire faces rather than just tongues.

The BenCao app developers aim to bridge the gap between AI and clinical practice by collaborating with physicians to compare machine learning diagnoses with human doctor assessments, identifying performance gaps and improvement opportunities.

Market implications

This technology represents a compelling intersection of ancient wisdom and modern innovation, potentially offering healthcare providers a rapid, non-invasive diagnostic tool. While significant challenges remain—from regulatory approval to data standardization—the high accuracy rates and growing commercial interest suggest a promising future.

For healthcare organizations, AI tongue diagnosis could complement existing diagnostic tools, potentially reducing costs and improving early detection capabilities. However, successful implementation will require substantial investment in standardized equipment, training, and data collection infrastructure.

The technology’s evolution from traditional practice to AI application demonstrates how machine learning can enhance rather than replace human expertise, creating new possibilities for accessible, efficient healthcare delivery.