This post is also available in:
עברית (Hebrew)
The AI revolution we are currently part of is touching almost every field. In the past few years, the healthcare industry has been working hard to utilize the power of AI to its advantage, and new discoveries are made regularly.
Recently, a machine-learning-based algorithm that can diagnose diseases based on tongue color with 98% accuracy was developed by engineers from the Middle Technical University (MTU) in Baghdad and the University of South Australia (UniSA) in Adelaide. This innovative algorithm was able to diagnose a wide range of diseases, such as diabetes, stroke, anemia, asthma, liver and gallbladder conditions, COVID-19, as well as a variety of vascular and gastrointestinal problems.
The scientists explain that inspecting the tongue for signs of disease is a practice that has been part of traditional Chinese medicine for centuries, and many health conditions can be revealed when examining its color, shape, and thickness.
The AI model was trained with 5260 images, with multiple machine-learning algorithms used to train it to predict tongue color in various lighting conditions, utilizing varying saturations and five color space models (RGB, YcbCr, HSV, LAB, and YIQ).
When tested, the system was able to ascertain the correct illnesses for most of the 60 images of patients’ tongues given to it, taken from two hospitals in the Middle East. The cameras were positioned 20 centimeters away from the patients’ tongues.
This study is an interesting stepping stone when it comes to real-time initial patient diagnosis. Hospitals are often crowded, and waiting times are often very long. If such technology were implemented in an app that could allow patients to have an initial diagnosis from a picture they take at home on their smartphone, lines could potentially become shorter, and doctors would have to spend less time diagnosing different ailments. This reinforces the impact AI could have on the field of medicine, helping patients and doctors alike.
The study was published in the Technologies journal.
