Development of an Active Cerebrovascular Autoregulation Model Using Representation Learning: A Proof of Concept Study...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Intensive Care Medicine

Background:

• Monitoring cerebrovascular autoregulation (CA) in intensive care units (ICUs) presents significant challenges.
• Existing methods like the pressure reactivity index (PRx) have limitations in capturing dynamic CA states.
• Advances in representation learning offer new possibilities for analyzing complex physiological signals.

Purpose of the Study:

• To develop a proof-of-concept active CA model utilizing representation learning.
• To leverage the full complexity of arterial blood pressure (ABP) and intracranial pressure (ICP) signals.
• To outperform the PRx in monitoring CA dynamics.

Main Methods:

• Utilized a porcine cranial window CA dataset (n=20).
• Preprocessed and downsampled ABP and ICP signals to 20 Hz.
• Optimized a neural network using 300-second ABP/ICP segments reflecting active CA to reconstruct its input, comparing reconstruction errors between active and inactive CA states.

Main Results:

• The model demonstrated excellent reconstruction quality for active CA segments, with deteriorating quality for inactive CA segments.
• Reconstruction errors increased with cerebral blood flow deviation from baseline.
• The model showed improved discriminative ability compared to PRx, capturing differential CA behavior and incorporating both low and high-frequency signal information.

Conclusions:

• An active CA model can be effectively constructed using advanced representation learning on complex ABP and ICP signal segments.
• Both lower and higher frequencies within ABP and ICP signals contain relevant CA state information.
• The developed model offers superior discriminative ability for CA states compared to PRx.