Researchers have developed a new clinical AI model that can detect an individual’s medical condition in real time by analyzing the colour of their tongue with a remarkable 98% accuracy.
According to a news article published in the New York Times, the senior research author Ali Al-Naji, who teaches at Iraq’s Middle Technical University in Baghdad and the University of South Australia, explained the rationale behind analyzing tongues using the clinical AI model to detect medical conditions.
While the tongue is usually analyzed for signs of common illnesses such as anaemia, research reveals the medical approach can also be used to detect many other illnesses.
For instance, cancer patients have a purple tongue coated with a thick grease-like layer, diabetes patients have a yellow tongue, and stroke patients are usually abnormally shaped and red, Al-Naji explained.
He added that those with severe cases of COVID-19 are more likely to have deep-red tongues, while those with asthma, gastrointestinal, and vascular issues have violet or indigo-coloured tongues.
The AI model mimics the approach used by traditional Chinese doctors to detect diseases based on the color of the patient’s tongue.
Clinical AI Model: The Study
The AI model was trained on more than 5,200 images of tongue colours with corresponding medical conditions during the study. The researchers then tested the model’s effectiveness with 60 images from two hospitals in the Middle East.
Based on the study findings, the AI model was able to determine patients’ medical conditions in almost all cases, and the study findings were published in the journal Technologies.
According to Javaan Chahl, the study co-author and a professor at the University of South Australia, the technology will be further developed for use on a smartphone app to detect a wide range of illnesses, including stroke, anaemia, asthma, diabetes, COVID-19, gallbladder, and liver problems.
While Chahal touted computerized tongue analysis as a secure, efficient, affordable, and user-friendly disease screening method, he also pointed out the current challenges the model will have to overcome to reach commercialization.
Such hurdles include patients’ reluctance to provide the needed data and the reflections captured by the camera misleading the AI algorithm.
