End-to-End Deep Learning for Subject-Independent Cuffless Blood Pressure Estimation from Raw PPG Signals Arky Astasari, Pratondo Busono, Agus Kartono, Erus Rustami
IPB University
National Research and Innovation Agency
Abstract
This study introduces a novel deep learning framework for systolic and diastolic BP estimation directly from raw PPG signals. Unlike traditional feature-engineering approaches, our method leverages end-to-end representation learning using hybrid architectures that combine convolutional neural networks (CNNs) for local morphological feature extraction and recurrent neural networks (RNNs) with attention mechanisms for capturing temporal dependencies and inter-beat variability. To enhance robustness, structured signal preprocessing and data augmentation strategies are integrated, while domain adaptation techniques are employed to mitigate inter-subject variability.
The proposed deep learning framework achieved a mean absolute error of 5.6 mmHg for systolic and 3.9 mmHg for diastolic BP under subject-independent evaluation, meeting international standards for clinical applicability.
Keywords: Please Just Try to Submit Thphotoplethysmography, cuffless blood pressure, machine learning, deep learning, , wearable health monitoringis Sample Abstract
