Work

Efficient-Transformers Optimization for Continuous Blood Pressure Estimation Using Bio-Sensors and Wearable Sensors

Year: 2026
Type: article
Abstract: Blood pressure is a key indicator for detecting and preventing cardiovascular disease, which remains one of the leading causes of mortality worldwide. In this work, we focus on building a research pipeline to evaluate and compare deep learning models for cuffless BP estimation by using photo- plethysmography (PPG) and electrocardiography (ECG) signals retrieved from open datasets such as MIMIC-III and MIMIC- IV. The methodology is based on a standardized and repro- ducible framework that includes signal preprocessing (filtering, segmentation, normalization) and inter-subject data splitting to ensure fair evaluation. The system supports PPG-only, ECG- only and multimodal inputs, and enables consistent training and benchmarking of models such as LSTM, BP-Net, and several Transformer variants, including optimized and hybrid architec- tures. In addition, transfer learning approaches are explored to improve model generalization and reduce dependence on large labeled datasets. Model performance is evaluated using standard metrics such as MAE, RMSE, mean error with standard deviation, and BHS grading. Overall, this project represents the design, implementation, and evaluation of a reproducible research framework for cuffless BP estimation, highlighting key challenges such as generalization, bias, and efficiency in wearable health applications.
Language: English
Open Access status: green
OpenAlex ID: W7197725155
FWCI:
Cites: 10
Cited by:
Related to: 10