City Research Online

Abstract

Traumatic brain injury (TBI) is as an alteration in brain function pathology by a sudden trauma, causing damage to the brain. Symptoms vary from mild to severe, depending on the extent of the damage to the brain. Intracranial pressure (ICP) monitoring is a “gold standard” for severe TBI patients, measuring pressure inside the skull. An ICP crisis is a sustained ICP value above a threshold of 20-25 mmHg. Effective ICP monitoring and intervention at defined thresholds can reduce mortality and secondary brain injury. The gold standard for ICP monitoring involves invasive neurosurgical intervention to implant a pressure sensor into the brain, which is expensive, carries risk, requires expertise and is only accessible within a hospital setting. Consequently, timely and effective monitoring is restricted, exacerbating adverse outcomes, and excluding millions of mild and moderate TBI cases from continuous and quantitative assessment.

There is a nascent body of research investigating non-invasive ICP monitoring aiming to reduce the barrier to entry to efficacious monitoring and intervention for patients and healthcare systems. A growing body of research is exploring the use of Photoplethysmograph (PPG) for the non-invasive estimation of ICP. This thesis makes novel contributions to knowledge by evaluating the relationship between PPG derived features and variations in ICP using the largest clinically collected, labelled PPG dataset, to-date. The novel PPG data, produced by an in-house, Near-infrared spectroscopy, reflectance, non-invasive optical ICP sensor, was collected from 40 TBI patients at The Royal London Hospital. Data is categorised based on “proximal” and “distal” photodiodes, hypothesised to correspond to extracerebral and cerebral data, respectively.

The research makes novel contributions to the field via the testing of three main hypotheses: (i) PPG feature alterations correlate with ICP changes, (ii) distal PPG data shows stronger correlations with ICP than proximal data, and (iii) PPG features can estimate ICP non-invasively.

A total of 141 features were extracted for each one-minute window of PPG data, including the original waveform and its first and second derivatives. Spearman’s correlation and the Kruskal-Wallis test evaluated the first two hypotheses. Results indicated significant correlations between PPG features and ICP levels, with 77.30% and 79.43% of features significantly correlated (p < 0.05) for distal and proximal datasets, respectively. Group analysis revealed significant changes across ICP groups (0-10, 10-20, 20-39 mmHg) in 81.56% and 75.89% of features. The mean absolute correlation of all features and significantly correlated features was 25.76% and 24.24% higher for distal than for proximal features, supporting the potential of PPG-based ICP monitoring.

To test the third hypothesis, five classical machine learning models were implemented, optimised, and assessed across six key metrics using a leave-one-patient-out cross-validation approach. Distal models outperformed proximal ones, with the best model, a Random Forest, achieving a mean RMSE of 5.030 mmHg, MAE of 4.067 mmHg, and Bland-Altman limits of agreement around 8.5 mmHg, and a low correlation coefficient of -0.007.

There is a need for the development of a continuous noninvasive, easy-to-use, inexpensive monitoring device. Such a device would reduce the barrier to entry for ICP monitoring across all severities, providing timely diagnosis whilst serving the currently undeserved majority of TBI cases. This work provides a credible foundation for further research in this domain.

Publication Type:	Thesis (Doctoral)
Subjects:	R Medicine > R Medicine (General) R Medicine > RA Public aspects of medicine > RA0421 Public health. Hygiene. Preventive Medicine T Technology > T Technology (General)
Departments:	School of Science & Technology > Engineering School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads