Computational models for functional near-infrared spectroscopy and imaging

Veesa, Joshua Deepak ORCID: 0000-0002-0763-7755 (2021). Computational models for functional near-infrared spectroscopy and imaging. University of Birmingham. Ph.D.

Preview

Veesa2021PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (4MB) | Preview

Abstract

Functional near-infrared spectroscopy (fNIRS) is a neuro-monitoring tool that is non-invasive, non-ionising, cost efficient, and portable. Its application for the traumatic brain injury patients is a well suggested approach due to its role in being able to continuously monitor key biomarkers such as the tissue oxygenation and blood haemoglobin level to understand the flow of blood supply to the tissue in the brain to assess injury in patients. In light of the great potential that fNIRS has to offer in neuro-monitoring in critical care, it is hindered by the inconsistency seldom seen in multiple research works that can be attributed to the assumptions made on tissue scattering properties to decouple their dependency along with absorption properties that can provide information about the key biomarkers useful in neuro-monitoring. These inconsistencies can also be attributed to the application of an inaccurate model to represent photon migration in underlying the biological tissue, or it can also be attributed to the unavoidable contamination of the measured fNIRS data by the superficial (skin and scalp) tissue, which is intended to probe the brain tissue, due to the typical placing of measurement probes on the head. The possibility to overcome these challenges in fNIRS methodology is examined in this thesis, and the proposed methods to overcome these are derived theoretically and validated on numerical simulation and experimental data to demonstrate better performance as compared to existing methods.
A spectrally constrained approach is designed to efficiently circumvent the coupling of absorption and scattering properties to directly yield more accurate estimates of oxygenation levels for the cerebral tissue showing an average improvement of 6.6% as compared to a
conventional and widely used approach of spatially resolved spectroscopy, in estimating the tissue oxygenation level. The uncertainty factor in the knowledge of scattering coefficient of the tissue, which is a key limitation in the conventional approach, is shown to be removed in the proposed spectrally constrained approach, therefore maintaining the methodology of subject and tissue-type independence.
With the demonstration of better performance on spectral constrained approach, the role of more spectral information i.e., broadband intensity data, to allow recovery of more information is also explored and is demonstrated that when the data is measured on a complex tissue such as the human head, an often used simple semi-infinite model based layered recovery can lead to uncertain results, whereas, by using an appropriate model accounting for the tissue-boundary structure and geometry, the tissue oxygenation levels are recovered with an error of 4.2%, and brain depth with an error of 11.8%. The algorithm is finally used together with human subject data, to demonstrate the robustness in application and repeatability in the recovered parameters that adhere well to expected published parameters.
Finally, the signal regression of fNIRS data to reduce superficial signal contamination which is well defined for a continuous wave (CW) fNIRS system is expanded to another data-types, namely phase data as used in frequency-domain (FD) fNIRS systems, by proposing a new approach for FD fNIRS that utilizes a short-separation intensity signal directly to regress both intensity and phase measurements. This is shown to provide a better regression of superficial signal contamination from both intensity and phase data-types. Intensity-based phase regression is shown to achieve a better suppression of superficial signal contamination by 68% whereas for phase-based phase regression the suppression is only by 13%. Phase-based phase regression is
also shown to generate false-positives in the image reconstruction of haemodynamic activations from the cortex, which is not desirable and therefore this work provides a better methodology for minimizing the superficial signal contamination for FD fNIRS.
All the parameter recovery models and signal processing methods presented in this work, in addition to their better performance that is shown, carry an additional and most prominent advantage of being able to be applied to all existing NIRS systems without any additional instrumentation or measurement for the purpose of providing a more accurate and robust neuro-monitoring tool.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):