Abstract Introduction Over the past decade, the development of wearable devices and smartphone-based bioinformation applications has rapidly expanded, promising new horizons in health monitoring. Despite these technological leaps, however, their sustained adoption remains disproportionately confined to individuals already invested in their own health. Those less inclined to engage in routine health management stand to miss out on the potential benefits of wearable technology, highlighting the urgent need for more accessible monitoring solutions. Purpose This study seeks to narrow the gap in health monitoring accessibility by introducing a novel, contact-free system for acquiring biometric data. By obviating the need for invasive procedures or dedicated wearable accessories, this system aims to democratize the early detection and prevention of disease across diverse populations, effectively reducing barriers to participation and compliance. Methods and Results A prospective cohort clinical study was conducted in a hospital setting with 300 consenting participants, encompassing both patients and healthy individuals. We employed a high-speed spectral camera to capture RGB video data of participants’ faces and hands, while simultaneous blood pressure measurements served as ground truth for hypertension. HbA1c levels from venous blood samples and physicians’ diagnostic records were used as ground truth for diabetes mellitus. A machine learning algorithm was developed to analyze skin perfusion and spectral characteristics indicative of pathological states. Hypertension was defined as systolic blood pressure ≥130 mmHg or diastolic blood pressure ≥80 mmHg. Diabetes mellitus was identified in participants with HbA1c ≥6.5% or a prior clinical diagnosis. Algorithmic performance in detecting hypertension leveraged pulse-wave features extracted from the video data, accurately identifying American Heart Association Stage 1 hypertension with 95.0% accuracy for 30 seconds of footage and 90.3% for 5 seconds. For diabetes, the algorithm, using blood-flow patterns as markers, correctly classified the condition with 83.1% accuracy for 5 seconds of face-and-hand data and 84.6% for face-only data. Notably, it also recognized patients previously diagnosed with diabetes but now maintaining HbA1c 6.5% under treatment. Conclusions The integration of advanced AI algorithms with spectral imaging of skin and vasculature presents a promising pathway for non-contact, early detection of hypertension and diabetes. This method not only addresses the limitations of conventional health monitoring modalities but also propels the field towards more equitable, user-friendly strategies for disease prevention and management.Graphical Abstract
Uchida et al. (Sat,) studied this question.