Noninvasive Intracranial Pressure Estimation Using Subspace System Identification and Bespoke Machine Learning Algorithms: A Learning-to-Rank Approach

TL;DR

A bespoke machine learning algorithm that integrates system identification and ranking-constrained optimization to estimate mean ICP from noninvasive signals is developed, demonstrating the feasibility of the proposed noninvasive ICP estimation approach.

Abstract

Objective: Accurate noninvasive estimation of intracranial pressure (ICP) remains a major challenge in critical care. We developed a bespoke machine learning algorithm that integrates system identification and ranking-constrained optimization to estimate mean ICP from noninvasive signals. Methods: A machine learning framework was proposed to obtain accurate mean ICP values using arbitrary noninvasive signals. The subspace system identification algorithm is employed to identify cerebral hemodynamics models for ICP simulation using arterial blood pressure (ABP), cerebral blood velocity (CBv), and R-wave to R-wave interval (R-R interval) signals in a comprehensive database. A mapping function to describe the relationship between the features of noninvasive signals and the estimation errors is learned using innovative ranking constraints through convex optimization. Patients across multiple clinical settings were randomly split into testing and training datasets for performance evaluation of the mapping function. Results: The results indicate that about 31.88% of testing entries achieved estimation errors within 2 mmHg and 34.07% of testing entries between 2 mmHg to 6 mmHg from the nonlinear mapping with constraints. Conclusion: Our results demonstrate the feasibility of the proposed noninvasive ICP estimation approach. Significance: Further validation and technical refinement are required before clinical deployment, but this work lays the foundation for safe and broadly accessible ICP monitoring in patients with acute brain injury and related conditions.

Figures (9)

• Figure 1: Left: A dynamical system view of ICP and noninvasive signals. Right: A dynamical system view of ICP, ABP, CBv, and R-R interval.
• Figure 2: Block diagram of the nICP calibration framework. The contents within the blue dashed line represent the offline training procedure. The green area represents the online training procedure.
• Figure 3: Summary of various algorithms to learn the mapping function. Left: Linear mapping function learning without constraints. Middle: Linear mapping function learning with constraints. Right: Nonlinear mapping with kernelization & ranking constraints.
• Figure 4: Data processing and feature extraction procedures for nICP estimation. Left: Data selection procedure from the database. Middle: Signal segmentation for signals with and without ECG. Right: Feature extraction for CBv pulse.
• Figure 5: Bland–Altman plots for testing entries with comparisons from different algorithms. (a): Bland-Altman plot for nICP results comparions between the linear mapping without constraints and linear mapping with constraints. (b): Bland-Altman plot for nICP results comparions between the linear mapping with constraints and nonlinear mapping. Values on the left y-axis represent the results from data points in blue. Values on the right y-axis in the images represent the results from data point in orange. 'LM Wo Constraints' stands for the linear mapping without constraints. 'LM With Constraints' stands for the linear mapping with constraints. 'NM With Gaussian Kernel' stands for nonlinear mapping with Gaussian kernel.