332 Estimation of Intracranial Pressure Using Non-Invasive Monitor and Machine Learning

Abstract

INTRODUCTION: Several non-invasive technologies have attempted to utilize surrogate markers of cerebral hemodynamics to infer intracranial pressure (ICP). Successful clinical application of these methods have been limited by hesitancy to rely upon surrogate data and inferences about ICP, in lieu of directly measured ICP. One newer method utilizes a mechanical extensometer to detect subtle, dynamic skull deformations can yield a non-invasive ICP waveform. This data has been validated to provide important information about intracranial dynamics, including inferences on the cerebral compliance. This method does not provide an absolute intracranial pressure and thus questions about clinical application remain. METHODS: High frequency (326 Hz) arterial blood pressure, non-invasive and invasive ICP measurements were collected from 5 patients. Samples were collected over 30 minutes. The first 15 minutes were used for training and last 15 minutes were used for testing. Facebook Prophet was used to perform signal analysis and forecasting. Switch-point analysis tools were also used to identify shifts in ICP measurements. RESULTS: Model per patient was generated from the 5 patients. Trend and seasonality analysis were then performed using Facebook Prophet modeling. Arterial blood pressure and non-invasive measurement were used as regressor to invasive intracranial pressure measurement. Mean absolute error of the testing samples of all models ranged from 0.8–2.4 mmHg. CONCLUSIONS: More closely and reliably estimating invasive intracranial pressure measurement from non-invasive measurements and arterial blood pressure is feasible using sequence analysis and machine learning. This will aid the clinical applicability of non-invasive monitoring and perhaps provide the pathway to replacing invasive measurement altogether. Further study of the applications of this clinical tool is warranted with the goal of intervening early to improve neurologic outcomes following acute brain injuries.