Presentation Information
[B-19-13]Estimation of RMSSD from Fingernail Images
◎△Yuya Matsuka1, Momona Kikuzawa1, Hideaki Kimura1 (1. Department of Computer Science , Graduate School of Engineering, Chubu University)
Keywords:
RMSSD,HRV,strees,nail
RMSSD (Root Mean Square of Successive Differences) is a heart rate variability index reflecting autonomic nervous system activity and widely used for stress evaluation. Conventional measurement requires dedicated contact sensors, placing a burden on users during prolonged wear. In this study, leveraging the property that fingernail regions are rich in capillaries and blood flow changes appear as color variations, we propose a non-contact RMSSD estimation method from webcam-captured fingernail images using deep learning. We constructed a ResNet18-based regression model, replacing the final fully connected layer with a one-dimensional output layer to directly predict RMSSD. In the experiment, each set comprised 30 images of 10 fingers captured from three angles, trained with RMSSD values derived from ECG signals via the Pan-Tompkins algorithm as ground truth. The dataset was stratified into 36 training and 10 test sets. The best model achieved a mean absolute error (MAE) of 9.9 ms, with relatively high accuracy in the 20–60 ms range. Outside this range, accuracy decreased due to data scarcity and regression-to-the-mean effects.