Clinical-Grade ECG Biosignals Processing Platform
We built a comprehensive biosignals processing platform centred on ECG signal analysis — combining adaptive digital filtering, QRS complex detection, and multi-domain feature extraction into a single, reproducible pipeline. The platform was designed for research collaborators and medical device developers who need algorithms that hold up under real-world signal conditions: motion artefacts, electrode noise, and varying patient physiology.
Category:
MedicalIndustry:
Medical Research / BioelectronicsClient:
Internal R&D / University PartnersYear:
2022–2024The Challenge
Raw ECG signals captured from wearable or clinical electrodes are rarely clean. Baseline wander, power-line interference, muscle noise, and electrode contact artefacts degrade signal quality and introduce systematic errors into downstream analysis. Developing algorithms that remain accurate across this variability — while remaining computationally tractable on embedded targets — required rigorous signal processing design and multi-dataset validation.
Our Solution
We designed a layered DSP pipeline: adaptive FIR/IIR pre-filters remove low-frequency baseline drift and 50/60 Hz interference while preserving the diagnostic bandwidth. A custom Pan-Tompkins-based QRS detector then localises heartbeat events with sub-millisecond precision. From detected beats we extract amplitude, duration, interval, and morphology features needed for arrhythmia screening, HRV computation, and signal quality indexing. The entire pipeline was validated on annotated public ECG databases (MIT-BIH, PTB-XL) and against clinical reference measurements from our university partners.
Key Features
- Adaptive FIR/IIR Filtering: Baseline wander removal and powerline rejection (50/60 Hz) preserving full diagnostic signal bandwidth without phase distortion.
- QRS Complex Detection: Pan-Tompkins-derived detector tuned for wearable-grade signals, delivering >99% sensitivity on MIT-BIH benchmark.
- Multi-Domain Feature Extraction: Time-domain (amplitude, RR intervals, QRS duration), frequency-domain (PSD via Welch), and nonlinear (SampEn, DFA) features for comprehensive analysis.
- Signal Quality Index (SQI): Per-beat and per-window quality scores flag corrupted segments before they contaminate HRV or arrhythmia outputs.
- Validated Against Public Databases: Algorithm performance benchmarked on MIT-BIH Arrhythmia Database and PTB-XL, enabling transparent comparison with published literature.
- Embedded-Ready Implementation: Core detection and filtering routines available as portable C for deployment on STM32 and similar MCUs with limited RAM.
Technologies
- Python
- NumPy / SciPy
- MATLAB
- FIR / IIR digital filters
- Pan-Tompkins QRS detection
- MIT-BIH / PTB-XL ECG databases
- C (embedded port)
- STM32
Results
The platform has been used across several KeySoft research and product projects — from the MonitorZ electrode evaluation device to the edge AI ECG quality classifier running on STM32. By open-sourcing the validation methodology and benchmarking against standard databases, our university partners can reproduce and extend every result. The result is a reusable signal processing foundation that accelerates both research publications and commercial product development.
Discuss a Similar Project