A Deep Learning-Based Automated Framework for Subpeak Designation on Intracranial Pressure Signals

Area of Science:

• Biomedical Engineering
• Neuroscience
• Intensive Care Medicine

Background:

• Intracranial pressure (ICP) monitoring is crucial for critically ill patients.
• The P2/P1 subpeak ratio in ICP signals reflects cerebral compliance.
• Accurate P2/P1 ratio calculation is vital for understanding cerebrospinal system dynamics.

Purpose of the Study:

• To develop and evaluate a deep learning pipeline for automated P2/P1 ratio computation from raw ICP signals.
• To assess the performance of recurrent neural networks (RNNs) and convolutional neural networks (CNNs) for pulse detection and selection.
• To create a reliable tool for continuous cerebral compliance assessment.

Main Methods:

• A deep learning pipeline comprising pulse detection, selection, subpeak designation, and signal processing.
• Comparison of RNN and CNN performance for heartbeat-induced pulse detection and selection.
• Evaluation on a dataset of 4344 pulses from 10 patient recordings.

Main Results:

• The pulse selection algorithm achieved an area under the curve of 0.90.
• The subpeak designation algorithm accurately identified pulses with P2/P1 ratio > 1 with 97.3% accuracy.
• The automated framework shows promise for integration into bedside monitoring.

Conclusions:

• The developed deep learning pipeline offers an automated approach to P2/P1 ratio calculation from ICP signals.
• The system demonstrates high accuracy in pulse selection and subpeak designation.
• This tool has the potential to enhance real-time patient monitoring and management of neurological conditions.