Fast estimation of adult cerebral blood content and oxygenation with hyperspectral time-resolved near-infrared spectroscopy

Abstract

Near-infrared spectroscopy (NIRS) can measure tissue blood content and oxygenation; however, its use for adult neuromonitoring is challenging due to significant contamination from their thick extracerebral layers (ECL; primarily scalp and skull). This report presents a fast method for accurate estimation of adult cerebral blood content and oxygenation from hyperspectral time resolved NIRS (trNIRS) data. A two-phase fitting method, based on a two-layer head model (ECL and brain), was developed. Phase 1 uses spectral constraints to accurately estimate the baseline blood content and oxygenation in both layers, which are then used by Phase 2 to correct for the ECL contamination of the late-arriving photons. The method was validated with in silico data from Monte-Carlo simulations of hyperspectral trNIRS in a realistic model of the adult head obtained from a high-resolution MRI. Phase 1 recovered cerebral blood oxygenation and total hemoglobin with an accuracy of 2.7 ± 2.5 and 2.8 ± 1.8%, respectively, with unknown ECL thickness, and 1.5 ± 1.4 and 1.7 ± 1.1% when the ECL thickness was known. Phase 2 recovered these parameters with an accuracy of 1.5 ± 1.5 and 3.1 ± 0.9%, respectively. Future work will include further validation in tissue-mimicking phantoms with various top layer thicknesses and in a pig model of the adult head before human applications.

HyperTRCSS: A hyperspectral time-resolved compressive sensing spectrometer for depth-sensitive monitoring of cytochrome-c-oxidase and blood oxygenation

“…A second sourcedetector pair could be added in future work, which would allow for the use of more sophisticated methods of TR data analysis, such as subtraction-based approaches 88 or approaches based on the solution to the diffusion equation for two-layer media. 36,69 Future work will include in vivo testing of HyperTRCSS; however, before that, the SNR limitations identified in this study must be addressed. Potential solutions include replacing the current hybrid PMT detector with a SPAD array and increasing the photon count rate above 1% of the laser repetition rate in combination with pulse pile-up correction.…”

“…Finally, our work did not explore analysis with data from multiple source–detector distances. A second source–detector pair could be added in future work, which would allow for the use of more sophisticated methods of TR data analysis, such as subtraction-based approaches 88 or approaches based on the solution to the diffusion equation for two-layer media 36 , 69 …”

“…(2): μa(λ)=−m(λ)c/n,where m is the asymptotic slope of the natural logarithm of the DTOF, c is the speed of light in a vacuum, and n is the refractive index of the phantom (assumed to be 1.33 60 ). The definition of the tail of the DTOF varies in the literature; some suggest that the slope can be fit after the intensity of the DTOF falls below 50% to the right of the peak intensity, 64 whereas others suggest using later windows (10% to 2% 67 or 5% to 1% 68 , 69 ) to ensure that the asymptotic assumption is met 65 . Importantly, using a later window comes at the cost of SNR since the number of detected photons is lower 70 .…”