Intracranial pressure (ICP) is an important parameter to monitor in several neuropathologies. However, because current clinically accepted methods are invasive, its monitoring is limited to patients in critical conditions. On the other hand, there are other less critical conditions for which ICP monitoring could still be useful; therefore, there is a need to develop non-invasive methods. We propose a new method to estimate ICP based on the analysis of the non-invasive measurement of pulsatile, microvascular cerebral blood flow with diffuse correlation spectroscopy. This is achieved by training a recurrent neural network using only the cerebral blood flow as the input. The method is validated using a 50% split sample method using the data from a proof-of-concept study. The study involved a population of infants (n = 6) with external hydrocephalus (initially diagnosed as benign enlargement of subarachnoid spaces) as well as a population of adults (n = 6) with traumatic brain injury. The algorithm was applied to each cohort individually to obtain a model and an ICP estimate. In both diverse cohorts, the non-invasive estimation of ICP was achieved with an accuracy of <4 mm Hg and a negligible small bias. Further, we have achieved a good correlation (Pearson's correlation coefficient >0.9) and good concordance (Lin's concordance correlation coefficient >0.9) in comparison with standard clinical, invasive ICP monitoring. This preliminary work paves the way for further investigations of this tool for the non-invasive, bedside assessment of ICP.
Keywords: diffuse optics; intracranial pressure; near-infrared spectroscopy; neural networks; non-invasive.