A pilot clinical study to estimate intracranial pressure utilising cerebral photoplethysmograms in traumatic brain injury patients

Abstract

Objective: In this research a non-invasive ICP optical sensor has been developed and evaluated in a clinical pilot study. The technology is based on infrared light interrogating brain tissue, including photodetectors used to detect the backscattered light, which is modulated by vascular pulsations in the brain vascular tissue. The hypothesis underpinning this research is based on the expected changes in the extramural arterial pressure affecting the morphology of the recorded optical signals (photoplethysmograms (PPGs)), therefore analysis of the acquired signals using a bespoke algorithm could enable the calculation of the intracranial pressure non-invasively (nICP). Methods: This pilot study is the first evaluation of the nICP probe in patients in whom the gold standard comparator of invasive ICP was available. nICP monitoring was performed for up to 48 hours in each of the 40 patients undergoing invasive ICP monitoring as part of their prescribed medical treatment. The quality of the recorded PPG signals was analysed, and time-based features were extracted off-line. Data from all ICP levels were randomly allocated into two groups to train (80%) and test (20%) a machine learning algorithm. A Bland Altman analysis and ROC curve were calculated to evaluate the accuracy of the estimated nICP value compared to the gold standard of synchronously acquired invasive ICP data. Results: The successful acquisition of cerebral PPG signals from TBI patients for the subsequent extraction of morphological features allowed the generation of a bagging tree model to estimate ICP non-invasively. The non-invasive estimation of ICP was achieved with 95% limits of agreement of 3.8 mmHg and a negligible bias. Furthermore, the model achieved a good correlation coefficient of 82.54% compared to standard clinical invasive ICP monitoring. Finally, the ROC curve analysis showed good diagnostic capability with a sensitivity of 80% and specificity of 89%. Conclusion: Clinical evaluation of this novel optical nICP sensor demonstrated for the first time that it could estimate ICP non-invasively to an acceptable and clinically useful accuracy and paves the way for further technology optimisation and larger clinical studies.