September 10, 2024 | A team of Iraqi and Australian researchers from Middle Technical University (MTU) in Baghdad, Iraq, and the University of South Australia (UniSA) collaborated to develop an AI model that can predict diseases based on the color of the patient’s tongue.
Reading the tongue’s complexion is a long-standing practice in traditional Chinese medicine. Human tongues have characteristics that are uniquely tied to the body’s internal organs, which means that the color of the tongue can indicate underlying health conditions. For example, a white tongue can mean a lack of iron or chill syndrome, and a purple tongue with a fatty layer can signal cancer. Redness can even be an indicator of COVID-19, with the severity depending on the shade of red (Technologies, DOI: 10.3390/technologies12070097).
The group used two different datasets to conduct the study. The first set consisted of 5,260 images of seven color classes (red, yellow, green blue, gray, white, and pink) under different lighting conditions, each labeled with a specific color that corresponds to different health conditions. The second dataset was a compilation of 60 abnormal tongue images collected from Al-Hussein Teaching Hospital in Dhi Qar, Iraq, and Mosul General Hospital in Mosul. The patients sat in front of a camera about 20 cm away and presented their tongue. The AI then analyzed the color intensity of their tongues and categorized them according to different color space models, including RGB, YCbCr, HVS, LAB, and YIQ. After the image was extracted, the algorithm compared the colors to what it learned during training and predicted if the patient has a disease or not.
The research team used different learning algorithms that were evaluated for their performance in detecting tongue colors. Out of all of them, the most accurate was the XGBoost model, which had a 98% accuracy rate (Technologies, DOI: 10.3390/technologies12070097). The AI analyzed the 60 collected images and compared them to the colors that would correspond to certain diseases.
“This validated the idea that modern technology can enhance traditional diagnostic methods, confirming the effectiveness of the proposed system,” says Al-Naji, associate adjunct professor at MTU and UniSA and senior author of this study.
However, Al-Naji mentions that collecting real-time data from hospitals, ensuring ethical compliance, managing diverse patient conditions, potential embarrassment during imaging, and temporary tongue discoloration caused by food and fruits potentially caused unexpected logistical or ethical challenges during the study. Nonetheless, the system still proved great efficacy.
The team believes the AI prediction model will work well in clinics because of its non-invasive approach, real-time analysis capabilities, and convenience. Al-Naji confirms that there are plans to develop an app for this tongue reading diagnosis method and that further trials are planned, “including expanded dataset collection and multi-center clinical validation to ensure the model’s robustness.”