Multi-Input Model for TENS Stimulation Intensity Classification and Prediction Using ECG, EMG and Personal Characteristics

초록

With the increasing use of personal electrical-stimulation devices, personalized TENS stimulation intensity selection that reflects individual physiological characteristics has become increasingly important. This study proposes a multiinput model for personalized TENS intensity prediction by integrating Electrocardiogram (ECG), Electromyogram (EMG), and demographic characteristics (age, sex, BMI). We collected data from 37-40 participants across four stimulation frequencies, generating 194,353 windowed segments. Time-domain and frequency-domain features extracted from ECG and EMG were combined with individual characteristics to construct an ordinal logistic regression-based classification model and a HistGradientBoosting Regressor-based regression model. Window-level predictions were aggregated at the file level to improve stability. Results demonstrated an average Top-2 accuracy of 0.746 (window level) and 0.782 (file level), indicating the model can effectively recommend the top two candidate stimulation intensities - aligned with clinical practice where therapists typically test multiple levels before final adjustment. This multi-modal approach provides a practical decision-support framework for individualized TENS therapy. © 2026 IEEE.

키워드