AI diagnoses diseases through tongue images

Researchers from Iraq and Australia have developed a groundbreaking AI model capable of diagnosing a variety of diseases simply by analysing tongue images. This innovative technology, resulting from a collaboration between Middle Technical University (MTU) in Baghdad and the University of South Australia (UniSA), has achieved a remarkable 98 percent accuracy in identifying health conditions based on tongue color.

The concept of diagnosing diseases through tongue examination is deeply rooted in traditional Chinese medicine, a practice that has been used for over 2,000 years. This ancient technique involves assessing tongue color and texture to determine health issues. The research team, led by Ali Al-Naji, an adjunct associate professor at both MTU and UniSA, has modernised this practice by integrating it with artificial intelligence to make it more applicable in today’s medical landscape.

Professor Al-Naji explains that the AI model can identify various conditions from tongue colors and textures. For example, a yellow tongue may indicate diabetes, while a purple tongue with a thick greasy coating could suggest cancer. An unusually shaped red tongue might signal an acute stroke, and severe COVID-19 cases are often associated with a deep red tongue. Additionally, a white tongue may be a sign of anemia, while an indigo or violet tongue could point to vascular or gastrointestinal issues, or even asthma.

The AI was trained on a dataset of 5,260 labeled tongue images, enabling it to detect subtle differences in color and texture linked to specific health problems. The accuracy of the model was further validated with 60 tongue images from patients at two Middle Eastern teaching hospitals. The AI successfully identified medical conditions in nearly all cases.

Studies have suggested that AI-powered tongue analysis could become a secure, efficient, and user-friendly method for disease screening. Researchers envision integrating this technology into a smartphone app for instant health assessments. Despite its potential, challenges such as data privacy and camera reflection interference need to be addressed before widespread adoption.