Pulse2AI: An Adaptive Framework to Standardize and Process Pulsatile Wearable Sensor Data for Clinical Applications

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Wearable devices generate vast amounts of pulsatile signal data.
• Existing preprocessing methods lack adaptability and task-agnostic capabilities.
• High-quality, machine-learning-ready datasets are crucial for advancing remote health monitoring.

Purpose of the Study:

• To introduce Pulse2AI, a reproducible data preprocessing framework for pulsatile signals.
• To generate high-quality, machine-learning-ready datasets from raw wearable recordings.
• To create a framework adaptable to multiple pulsatile signal modalities and independent of downstream medical tasks.

Main Methods:

• Developed an end-to-end data preprocessing framework named Pulse2AI.
• Designed the framework to be agnostic to specific downstream medical applications.
• Validated the framework's adaptability across various pulsatile signal types.

Main Results:

• Pulse2AI preprocessing improved systolic blood pressure estimation accuracy by 29.58% (RMSE from 11.41 to 8.03 mmHg).
• Diastolic blood pressure estimation accuracy improved by 26.01% (RMSE from 7.93 to 5.87 mmHg).
• Respiration rate estimation performance increased by 19.69% (MAE from 1.47 to 1.18 BrPM).

Conclusions:

• Pulse2AI effectively transforms pulsatile signals into machine learning-ready datasets for diverse remote health monitoring.
• The framework demonstrates efficacy in medical applications, enhancing blood pressure and respiration rate estimation.
• Pulse2AI bridges the gap between remote sensing/IoMT and ML-ready datasets for medical modeling.