A team of researchers from Middle Technical University (MTU) in Baghdad, Iraq, and the University of South Australia (UniSA), Adelaide, Australia, has reportedly trained machine learning algorithms to spot diseases through the color analysis of the human tongue.

According to the team, the algorithms achieved a 98% rate of accuracy in predicting assorted diseases when the artificial intelligence (AI) algorithm analyzed the color of the human tongue.

A researcher demonstrates how a camera captures images of the tongue and the AI model analyzes the organ's color for disease diagnostics. Source: Middle Technical UniversityA researcher demonstrates how a camera captures images of the tongue and the AI model analyzes the organ's color for disease diagnostics. Source: Middle Technical University

Specifically, the AI model diagnosed diabetes, stroke, anemia, asthma, liver and gall bladder conditions, COVID-19 and various vascular and gastrointestinal issues, the researchers reported

Noting that the color, shape and thickness of the tongue can reveal a host of health conditions, the researchers explained that, in general, patients with diabetes tend to present with a yellow tongue, while cancer patients present with a purple tongue in addition to a thick greasy coating. Meanwhile, acute stroke patients tend to present with an unusually shaped red tongue.

Further, the team noted that a white tongue can suggest anemia while people with severe cases of COVID-19 tend to present with a deep red tongue. Similarly, an indigo or violet-colored tongue tends to point to vascular and gastrointestinal issues or asthma.

As such, the team of researchers trained the computer vision systems along with an accompanying imaging system using 5,260 images distinguished with seen classes of colors — red, yellow, green, blue, gray, white and pink. Six machine-learning algorithms trained the AI algorithms to predict tongue color under any lighting conditions.

During testing of the system, 60 images of patients’ tongues who were experiencing specific health conditions were sourced from teaching hospitals in the Middle East. According to the team, the AI model matched the tongue color to the diseases diagnosed in those patients with a 98% rate of accuracy.

The researchers suggest that this method could one day mean faster detection of diseases, with on-the-spot diagnostics. Eventually, the team believes a smartphone might be used for this purpose.

An article detailing the team’s findings, "Tongue Disease Prediction Based on Machine Learning Algorithms,” appears in the journal Technologies.

To contact the author of this article, email mdonlon@globalspec.com