Machine learning workflow for edge computed arrhythmia detection in exploration class missions.

Area of Science:

• Aerospace Medicine
• Biomedical Engineering
• Machine Learning

Background:

• Deep-space missions necessitate remote health monitoring and predictive diagnostics for astronaut pathologies.
• Edge computing and Open Neural Network Exchange (ONNX) formats are crucial for on-device inference in remote environments.

Purpose of the Study:

• To develop and validate a self-optimizing machine learning pipeline for classifying cardiac arrhythmias (Normal Sinus Rhythm, Atrial Fibrillation, Atrial Flutter) on wearable edge devices.
• To assess the feasibility of deploying ONNX-optimized models for real-time point-of-care diagnostics in space missions.

Main Methods:

• Processed 742 hours of electrocardiogram (ECG) recordings, applying variable mode decomposition to remove noise.
• Extracted 17 heart rate variability and morphological ECG features using peak detection and discrete wavelet transforms.
• Utilized a self-optimizing decision tree classifier with stratified triple nested cross-validation, optimized via F1-scoring against cardiologist labels.

Main Results:

• Achieved a macro F1-score of 0.899, with high scores for Normal Sinus Rhythm (0.993) and Atrial Fibrillation (0.938), and 0.767 for Atrial Flutter.
• Identified key features including median P-wave amplitudes, PRR20, and mean heart rates.
• The ONNX-translated pipeline processed samples in 9.2 seconds, demonstrating efficient operational tachycardia detection.

Conclusions:

• The combination of a self-optimizing training scheme and ONNX deployment enables accurate, on-device classification of cardiac arrhythmias.
• This approach provides a viable solution for preventative care and real-time health monitoring in deep-space missions.