AI detects diseases from privately captured hand images

After analyzing existing developments, the researchers found that most AI models rely on facial images, which can raise privacy concerns. Then a team from Kobe University Yuka Omachi suggested focusing on the hands – they are already examined during diagnosis, and with acromegaly the changes are especially noticeable. To further protect personal data, the scientists used only photographs of the back of the hand and a clenched fist, avoiding shooting palm lines. The project involved 725 patients from 15 medical centers in Japan, who provided more than 11 thousand images for training and testing the algorithm.

In the journal Clinical Endocrinology and Metabolism, the researchers reported that their model demonstrated very high sensitivity and specificity in detecting acromegaly. Moreover, she surpassed experienced endocrinologists in accuracy, who were shown the same images. The scientists admitted that they were surprised by such a high performance when using only photographs of the back of the hand and a clenched fist, especially without analyzing facial features, which makes the method more convenient for screening.

In the future, the researchers intend to adapt the model to identify other diseases that can be suspected by the condition of the hands, including rheumatoid arthritis, anemia and finger deformities. According to scientists, this could open up new opportunities for medical applications of AI. At the same time, scientists emphasize that in real practice, doctors rely not only on images, but also on a complex of clinical data. Therefore, they view the new model as a tool that can complement professional experience, reduce the risk of diagnostic errors and promote earlier initiation of treatment.