Finetuning and Quantization of EEG-Based Foundational BioSignal Models on ECG and PPG Data for Blood Pressure Estimation

2025-02-10

arXiv.org

AI Summary

This study investigates the cross-modal transferability of EEG pre-trained foundation models to blood pressure estimation from ECG/PPG time-series signals, proposing a fine-tuning paradigm that obviates additional large-scale pre-training. Methodologically: (1) EEG foundation models are directly adapted to ECG/PPG inputs to construct end-to-end systolic/diastolic blood pressure regression models; (2) a lightweight dynamic INT8 quantization strategy is introduced to preserve accuracy under edge-device constraints. The core contribution lies in the first empirical validation of EEG models’ representational transfer capability to cardiovascular signals, achieving joint optimization of accuracy and efficiency. On MIMIC-III and VitalDB, the method achieves diastolic and systolic BP MAEs of 1.57 mmHg and 2.72 mmHg—improving upon state-of-the-art by ~1.5×—while compressing model size by 3.5× to 3.83 MB, enabling real-time inference on wearable devices.

Technology Category

Machine Learning: Learning on the Edge & Model CompressionCognitive Modeling & Cognitive Systems: Other Foundations of Cognitive Modeling & SystemsComputer Vision: Multi-modal Vision

Application Category

User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendationGraph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphsSystems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Blood pressure (BP) is a key indicator of cardiovascular health. As hypertension remains a global cause of morbidity and mortality, accurate, continuous, and non-invasive BP monitoring is therefore of paramount importance. Photoplethysmography (PPG) and electrocardiography (ECG) can potentially enable continuous BP monitoring, yet training accurate and robust machine learning (ML) models remains challenging due to variability in data quality and patient-specific factors. Recently, multiple research groups explored Electroencephalographic (EEG)--based foundation models and demonstrated their exceptional ability to learn rich temporal resolution. Considering the morphological similarities between different biosignals, the question arises of whether a model pre-trained on one modality can effectively be exploited to improve the accuracy of a different signal type. In this work, we take an initial step towards generalized biosignal foundation models by investigating whether model representations learned from abundant EEG data can effectively be transferred to ECG/PPG data solely with fine-tuning, without the need for large-scale additional pre-training, for the BP estimation task. Evaluations on the MIMIC-III and VitalDB datasets demonstrate that our approach achieves near state-of-the-art accuracy for diastolic BP (mean absolute error of 1.57 mmHg) and surpasses by 1.5x the accuracy of prior works for systolic BP (mean absolute error 2.72 mmHg). Additionally, we perform dynamic INT8 quantization, reducing the smallest model size by over 3.5x (from 13.73 MB down to 3.83 MB) while preserving performance, thereby enabling unobtrusive, real-time BP monitoring on resource-constrained wearable devices.

Problem

• Transferring EEG foundation models to ECG/PPG data
• Enabling accurate non-invasive blood pressure estimation
• Optimizing model size for wearable device deployment

Innovation

• Fine-tuning EEG foundation models for ECG and PPG data
• Applying INT8 quantization to reduce model size
• Achieving accurate blood pressure estimation with small models