PhysOnline: An Open Source Machine Learning Pipeline for Real-Time Analysis of Streaming Physiological Waveform

Area of Science:

• Biomedical Engineering
• Data Science
• Machine Learning

Background:

• Real-time analysis of streaming physiological data is crucial for precision medicine.
• A lack of open-source systems hinders this workflow.
• Existing methods often struggle with scalability and real-time processing.

Purpose of the Study:

• To present PhysOnline, an open-source pipeline for real-time physiological data analysis.
• To enable online feature extraction and machine learning on streaming data.
• To demonstrate the pipeline's capability for early detection of abnormal conditions.

Main Methods:

• Developed PhysOnline using the Apache Spark platform for scalable, horizontal deployment.
• Integrated real-time feature extraction including pattern recognition and statistical methods.
• Applied the pipeline to online classification of electrocardiography (ECG) data.

Main Results:

• PhysOnline successfully ingests and processes streaming physiological data.
• The pipeline extracts temporal and morphological features from waveform signals.
• Demonstrated prediction of Paroxysmal Atrial Fibrillation (PAF) at least 45 minutes before an episode.

Conclusions:

• PhysOnline provides a scalable, open-source solution for real-time physiological data analysis.
• The pipeline facilitates early detection of abnormal conditions through machine learning.
• Applicable to various time-series data domains requiring pattern matching and temporal analysis.