On-Chip Spiking Neural Networks for PPG-Based Blood Pressure Estimation

Blood pressure (BP) is one of the most critical physiological parameters for assessing cardiovascular health, but current measurement techniques either rely on bulky cuff-based devices or computationally expensive machine learning models that are unsuitable for long-term wearable monitoring. Photoplethysmography (PPG), already available in consumer devices such as smartwatches, provides a low-cost and non-invasive signal that carries useful information about both systolic and diastolic blood pressure [1]. However, extracting this information robustly in real time under tight memory and energy constraints remains an unsolved challenge.

This project explores a neuromorphic approach to PPG-based BP estimation, inspired by the way biological neural systems process temporal signals efficiently. By leveraging spiking neural networks (SNNs) implemented on ultra-low-power hardware, we aim to design a system that encodes PPG-derived features into spike trains and estimates BP values through collective spike-based computation. Building on our previous work on neuromorphic heart rate monitoring [2], the proposed project extends the concept to more complex cardiovascular biomarkers, requiring the integration of temporal dynamics, feature encoding, and statistical inference into the same neuromorphic substrate.

The envisioned system will be validate d on publicly available PPG–BP datasets and benchmarked against state-of-the-art deep learning approaches, but with a strong emphasis on embedded feasibility: memory footprint, inference latency, and energy per prediction. The final outcome will be a proof-of-concept wearable-ready neuromorphic BP estimator that runs directly on a resource-constrained microcontroller or neuromorphic chip, eliminating the need for cloud-based processing and enabling continuous, private, and energy-efficient monitoring.

 

Expected Outcomes

• - Design and implement a compact SNN architecture for BP estimation from PPG signals, optimized for resource-constrained platforms.

• - Deploy the network on a neuromorphic chip to validate feasibility in real-world wearable scenarios.

• - Evaluate accuracy, memory footprint, latency, and energy efficiency against state-of-the-art digital machine learning methods on public PPG-BP datasets.

 

Requirements

• - Proficiency in Python

• - Curiosity and creativity in applying bio-inspired principles to real-world healthcare challenges.

 

References:
[1] González, Sergio, Wan-Ting Hsieh, and Trista Pei-Chun Chen. "A benchmark for machine-learning based non-invasive blood pressure estimation using photoplethysmogram." Scientific Data 10, no. 1 (2023): 149.

[2] De Luca, Chiara, Mirco Tincani, Giacomo Indiveri, and Elisa Donati. "A neuromorphic multi-scale approach for real-time heart rate and state detection." npj Unconventional Computing 2, no. 1 (2025): 6.

Contact:
Elisa Donati – elisa@ini.uzh.ch