Noninvasive optical estimation of CSF thickness for brain-atrophy monitoring

Abstract

Dementia disorders are increasingly becoming sources of a broad range of problems, strongly interfering with the normal daily tasks of a growing number of individuals. Such neurodegenerative diseases are often accompanied with progressive brain atrophy that, at late stages, leads to drastically reduced brain dimensions. Currently, this structural change could be followed with X-ray computed tomography (XCT) or magnetic resonance imaging (MRI), but they share numerous disadvantages in terms of usability, invasiveness and costs. In this work, we aim to retrieve information concerning the brain-atrophy stage and its evolution, proposing a novel approach based on non-invasive time-resolved near infra-red (tr-NIR) measurements. For this purpose, we created a set of virtual human-head atlases in which we eroded the brain as it would happen in a clinical brain-atrophy progression. These realistic meshes were used to simulate a longitudinal tr-NIR study, investigating the effects of an increased amount of cerebral spinal fluid (CSF) in the photon diffusion. The analysis of late photons in the time-resolved reflectance curve-obtained via accurate Monte Carlo simulations-exhibited peculiar slope-changes upon CSF layer increase. The visibility of the effect under several measurement conditions suggested good sensitivity to CSF variation, even in the case of real measurement and under different geometrical models. The robustness of the results might promote the technique as a potential indicator of the dementia progression, relying only on fast and non-invasive optical observations.

Fig. 1

Infographic showing a combined view of the cylindrical 4-layer mesh volumes used in the simulations and the locations of source and detectors on their top layer. Preserving the Skin & Skull and the Gray Matter thickness, a linearly increasing CSF layer is inserted in between, representing a hypothetical Alzheimer’s progression (or, in general, any brain-atrophy dementia-related disorder).

Fig. 2

The brain erosion process in voxel coordinates. On the top, it is possible to notice how, using cubic or diamond structuring elements results in a developing of artifact as soon as the brain shrinks. The erosion modeled with a nearly-spherical element is more realistic (bottom), leading to rounded erosion and not leaving shape-artifacts. With the * operator we define the dilation, inverse of the erosion operation, which once applied to the eroded brain would return the original version. For graphical purposes we describe a 2D structuring element (right column), while in our work we have effectively used its 3D counterpart.

Fig. 3

Informative about the Alzheimer’s model creation. From the top graphs in panels A-C, it is possible to notice how the shrinkage of the brain is linear with respect the increasing thickness of the resulting CSF layer. Both the GM and the WM were shrunk independently with the same criteria, which results to preserve their average distance (and so the average thickness of the GM). D) The bottom infographics shows a tomographic cut of the models and their respective name used throughout the text.

Fig. 4

The pulsed laser source impinging the human head from its right hemisphere. The detector positions are shown in orange and their distance is $10 mm$ with respect each other. In this figure, the right lobe of the brain is from the AD model Stage 6 while the left is the original Colin27 model [31]. To help the visualization of the modeled disease progression, we label the z-coordinate with jet color bar.

Fig. 5

IRF used to reproduce realistic laboratory measurements. The main plot shows the measured IRF of a common detector system and on the nested plot its extracted and normalized version (corresponding to the green region in the main graph) that we used to reproduce the effect of a realistic measurement.

Fig. 6

Distribution of time of flight (DTOF) at four interfiber source-detector (S-D) distances for the cylindrical layered phantoms at various CSF thicknesses. On the left column, the raw data considering an ideal response of the detector, on the right the corresponding curves convolved with a typical instrument response function (IRF). In the first plot are marked the temporal windows considered as early-photons (green) and late-photons (red) gating.

Fig. 7

CSF fingerprints in the cylindrical model. On the left panel A), the peak response delays as a function of the S-D separation and it resulted to be equal for all the CSF thicknesses investigated. On panel B, the corresponding response delay after the convolution with the IRF that introduces a uniform temporal peak-delay. On the right panel C), the changing in the slope of the late photons due to increased CSF thickness.

Fig. 8

DTOF curves at various Source-Detector separations for the human-head models. Compared to that of cylindrical models, the measurements are noisier due to the impossibility of averaging many detector responses for the lack of symmetry in complex geometries. The left column shows the raw data sets, on the right the corresponding convolution with a realistic IRF.

Fig. 9

Features of the response curves for the Human Head models. A) There is no evident peak shift in the response curves at early detection times as a function of the CSF. The peaks are so close to each other that in the plot all the curves are overlaying each other. B) Even the convolution with the IRF, which shifts up the peak response in time, does not introduce any further visible effect in the measurements. C) Slope variation for late-photons detection as a function of the CSF thickness. In the plot is possible to notice how increased thicknesses of the transparent layer surrounding the brain will affect the reflectance curve, giving useful hints about the possibility to estimate its average thickness.

Fig. 10

Statistical noise of the MC simulations until 4000 ps. Coefficient of noise-variation of the standard deviation error with respect the average DTOF curve for the cylindrical model A) and for the human head B). The noise is always below 5% until 4000 ps and does not depend upon CSF variations.