Abstract

Measuring intracranial pressure (ICP) is typically a highly invasive procedure, in which a ventricular catheter or pressure sensor is placed into the brain. To improve the availability of ICP measurements in non-intensive care patients and research and to reduce the invasiveness and underlying risks of ICP sensing, we developed a non-invasive method to measure ICP with Diffuse Correlation Spectroscopy (DCS) and machine learning. ICP baseline changes were induced in non-human primates (Macaca mulatta) through adjusting the height of a saline reservoir connected to the lateral ventricle via a catheter. ICP was precisely measured with an invasive parenchymal pressure sensor. Cerebral blood flow (CBF) was measured with DCS. The DCS system was operated by a software correlator able to resolve cardiac pulse waves at a sampling rate of 100Hz. To increase signal-to-noise ratio, multiple cardiac pulse waves in CBF were averaged based on systolic peak maximum in invasively measured arterial blood pressure. We hypothesized that the cerebral blood flow pulse waves will change their shape with increasing ICP. The shape of the curve was expressed in numerical features and passed into a regression forest training algorithm. Preliminary results show successful prediction of underlying ICP baselines by the decision forest in one animal. The prediction of non-invasive ICP was achieved with a sampling rate of 1 Hz, an equivalent of about 120 averaged pulses. A larger data set for increased generalizability is the next step to push this approach further.

© 2019 SPIE/OSA