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

TL;DR

This work demonstrates that a transformer backbone pre-trained on EEG data can be fine-tuned to estimate blood pressure from ECG and PPG signals with near-clinical accuracy, achieving DBP MAEs around $1.57$ mmHg and SBP MAEs around $2.72$ mmHg on MIMIC-III, and $DBP$ MAE of $1.92$ mmHg and $SBP$ MAE of $3.14$ mmHg on VitalDB. It also shows that dynamic INT8 post-training quantization compresses the model by approximately $3.5 imes$ (from ~13.73 MB to ~3.83 MB) with negligible performance loss, enabling real-time, edge-friendly BP monitoring on wearables. The approach leverages a two-channel ECG/PPG input to a CEReBrO transformer with an MLP head, comparing frozen vs unfrozen fine-tuning and training-from-scratch, and finds that unfrozen, larger models yield the best accuracy while maintaining efficiency. Collectively, these results support cross-biosignal foundation models as a viable path for cuffless BP estimation and highlight quantization as a key enabler for on-device deployment, with implications for scalable, continuous cardiovascular monitoring.

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.

Figures (1)

• Figure 1: The modified CEReBrO Architecture CEReBrO, supplemented with an additional MLP-head, which is utilized for the estimation task.