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

Bálint Tóth, Dominik Senti, Thorir Mar Ingolfsson, Jeffrey Zweidler, Alexandre Elsig, Luca Benini, Yawei Li
July 2025
PDF Project Preprint (arXiv) DOI

Abstract

This work explores whether EEG-pretrained foundation models can transfer to other biosignals for cuffless blood-pressure estimation. An EEG foundation model is fine-tuned on photoplethysmography (PPG) and electrocardiography (ECG) signals from the MIMIC-III and VitalDB datasets, achieving near state-of-the-art diastolic blood-pressure estimation (mean absolute error ≈ 1.57 mmHg) and surpassing prior methods for systolic BP with a mean absolute error of 2.72 mmHg. Dynamic INT8 quantization compresses the smallest model from 13.73 MB to 3.83 MB without loss of accuracy, demonstrating feasibility for deployment on constrained devices.

Type
Conference paper

Key Highlights

  • Demonstrates transfer of EEG-based foundation models to ECG/PPG for cuffless blood-pressure estimation.
  • Achieves 1.57 mmHg MAE on diastolic BP and 2.72 mmHg on systolic BP across MIMIC-III and VitalDB.
  • Dynamic INT8 quantization reduces model size from 13.73 MB to 3.83 MB with negligible accuracy loss.
blood pressure estimation Foundation Models transfer learning ECG PPG Quantization
Thorir Mar Ingolfsson
Thorir Mar Ingolfsson
Postdoctoral Researcher

I develop efficient machine learning systems for biomedical wearables that operate under extreme resource constraints. My work bridges foundation models, neural architecture design, and edge deployment to enable real-time biosignal analysis on microwatt-scale devices.

Related