Regression analysis for peak designation in pulsatile pressure signals

Area of Science:

• Neuroscience and Biomedical Engineering
• Medical Signal Processing

Background:

• Automatic analysis of intracranial pressure (ICP) pulses is crucial for forecasting cerebrovascular and intracranial pathophysiological variations.
• Existing ICP pulse analysis frameworks extract morphological features but rely on Gaussian priors for peak designation, limiting versatility.

Purpose of the Study:

• To enhance an existing ICP pulse analysis framework by integrating machine learning techniques.
• To replace Gaussian priors with versatile regression models for improved ICP sub-peak designation accuracy.

Main Methods:

• Utilized machine learning regression models, including kernel spectral regression, to designate ICP sub-peak locations.
• Evaluated performance on a database of 700 hours of ICP recordings from 64 neurosurgical patients.
• Compared the novel regression-based approach against the original Gaussian prior algorithm.

Main Results:

• The regression-based framework achieved an average peak designation accuracy of 99%.
• This represents a significant improvement over the original algorithm's accuracy of 93%.
• Kernel spectral regression demonstrated superior performance among the evaluated regression methods.

Conclusions:

• Machine learning, specifically versatile regression models, offers a significant advancement in ICP pulse analysis.
• The proposed regression-based framework provides more accurate identification of ICP pulse features.
• This improved accuracy holds promise for enhanced forecasting of critical neurological conditions.