Using near-infrared spectroscopy and a random forest regressor to estimate intracranial pressure

Intracranial pressure (ICP) measurements are important for patient treatment but are invasive and prone to complications. Noninvasive ICP monitoring methods exist, but they suffer from poor accuracy, lack of generalizability, or high cost.

We previously showed that cerebral blood flow (CBF) cardiac waveforms measured with diffuse correlation spectroscopy can be used for noninvasive ICP monitoring. Here we extend the approach to cardiac waveforms measured with near-infrared spectroscopy (NIRS).

Changes in hemoglobin concentrations were measured in eight nonhuman primates, in addition to invasive ICP, arterial blood pressure, and CBF changes. Features of average cardiac waveforms in hemoglobin and CBF signals were used to train a random forest (RF) regressor.

The RF regressor achieves a cross-validated ICP estimation of $0.937r^2$ , $2.703\text{-}\mathrm{mm}{\mathrm{Hg}}^2$ mean squared error (MSE), and 95% confidence interval (CI) of $[-3.0643.160]\mathrm{mmHg}$ on oxyhemoglobin concentration changes; $0.946r^2$ , $2.301\text{-}{\mathrm{mmHg}}^2$ MSE, and 95% CI of $[-2.8412.866]\mathrm{mmHg}$ on total hemoglobin concentration changes; and $0.963r^2$ , $1.688{\mathrm{mmHg}}^2$ MSE, and 95% CI of $[-2.4502.397]\mathrm{mmHg}$ on CBF changes.

This study provides a proof of concept for the use of NIRS in noninvasive ICP estimation.