Daiichi Sankyo Healthcare: AI face analysis links skin signs to health

The starting point for this work is a practical gap the company describes: blood tests, questionnaires, and wearables have advanced, but test values and scores alone are hard for people to feel as their own condition, so they do not necessarily lead to self-care or behavior change. At the same time, research estimating internal and mind-body states from face images has been progressing, but it was not sufficiently clear whether the information AI captures can be understood and owned by the people themselves.

Daiichi Sankyo Healthcare's approach was to treat the face not only as an object of AI analysis but also as everyday visible information the person sees in the mirror. The study combined two methods: deep learning that tested whether high and low groups for indicators such as inflammation, aging changes, oxidation, hormones, stress, fatigue, and insomnia could be distinguished, and statistical analysis linking eight extracted visible features — skin brightness, cheek redness, yellowing, under-eye shadows, eye-corner wrinkles, fine lines at the eye corners, fine skin texture, and nasolabial fold depth — to blood test values and mind-body data. When the two sets of results were cross-checked, some of the indicators flagged by deep learning also showed associations with visible features.

The stakes hinge on whether larger participant numbers and more advanced analysis methods actually improve the accuracy of what face images can yield, and on whether the company can build features people find easy to understand. The framing is self-care support rather than diagnosis, so the near-term value is likely to be judged by how well these signals translate into awareness and action for ordinary users, not by clinical claims.