https://cardiab.biomedcentral.com/articles/10.1186/s12933-025-02982-4

https://www.jst.go.jp/pr/announce/20251111/index.html

A team at the Tokyo Institute of Science has developed a new AI model that can accurately detect prediabetes using data from standard electrocardiograms (ECGs) performed during health checkups, without the need for blood tests. The researchers confirmed that similar accuracy could be achieved with an equivalent ECG recorded on a wristwatch-type wearable device (I-lead ECG), demonstrating its potential for screening in everyday life.

The team developed and rigorously validated DiaCardia, an AI model that successfully identifies individuals with prediabetes solely from an ECG. This may be the first demonstration that subtle cardiac manifestations of impaired glucose homeostasis can be detected non-invasively at the prediabetic stage. From a clinical perspective, this represents a significant advancement by establishing the ECG as a viable, non-invasive screening tool for the earliest stages of glucose dysregulation. The model achieved high performance not only in an internal test dataset (AUROC: 0.851) but also, critically, in an independent external validation cohort (AUROC: 0.785) for identifying individuals meeting prediabetic criteria.

Further developing these research results may enable prediabetes to be detected anytime, anywhere, by anyone, without the need for health checkups or blood tests at the hospital, thereby contributing to diabetes prevention.