Optimizing a two-layer method for hybrid diffuse correlation spectroscopy and frequency-domain diffuse optical spectroscopy cerebral measurements in adults

Abstract

Significance: The sensitivity to extracerebral tissues is a well-known confounder of diffuse optics. Two-layer (2L) head models can separate cerebral signals from extracerebral artifacts, but they also carry the risk of crosstalk between fitting parameters. Aim:We aim to implement a constrained 2L head model for hybrid diffuse correlation spectroscopy (DCS) and frequency-domain diffuse optical spectroscopy (FD-DOS) data and to characterize errors in cerebral blood flow and tissue absorption with the proposed model. Approach:The algorithm uses the analytical solution of a 2L cylinder and an a priori extracerebral layer thickness to fit multidistance FD-DOS (0.8 to 4 cm) and DCS (0.8 and 2.5 cm) data, assuming homogeneous tissue reduced scattering. We characterized the algorithm's accuracy for simulated data with noise generated using a 2L slab and realistic adult head geometries and for in vitro phantom data.Results: Our algorithm recovered the cerebral flow index with 6.3 [2.8, 13.2]% and 34 [30,42]% (median absolute percent error [interquartile range]) for slab and head geometries, respectively. Corresponding errors in the cerebral absorption coefficient were 5.0 [3.0, 7.9]% and 4.6 [2.4, 7.2]% for the slab and head geometries and 8 [5, 12]% for our phantom experiment. Our results were minimally sensitive to second-layer scattering changes and were robust to cross-talk between fitting parameters. Conclusions:In adults, the constrained 2L algorithm promises to improve FD-DOS/DCS accuracy compared with the conventional semi-infinite approach.

Comparison of diffuse correlation spectroscopy analytical models for measuring cerebral blood flow in adults

“…[10][11][12] Regardless, it appears that there may be crosstalk between the extracerebral layer thickness and BSR for the three-layer model as well as, to a lesser extent, the two-layer model. 19 We found that CBFi estimated with the homogenous model was significantly positively correlated with TCD blood flow velocities in the MCA obtained on the same day. Given that the TCD and DCS measurements were often acquired hours apart, a lack of correlation between DCS and TCD would be difficult to interpret given the patients' possible hemodynamic instability.…”

“…13 This trend was independent of the SNR for the integration times that we investigated; however, the pass rate will likely decrease at higher sampling frequencies in which the SNR is significantly reduced because these models have been shown to be highly sensitive to noise. 12,19 One striking, although not surprising, finding was that the pass rate of the multilayer model fitting was significantly influenced by the extracerebral layer thickness. This trend was most prominent in the three-layer model, although it was also seen in the two-layer model.…”

“…The reason for these large outliers is less clear; they could be due to errors in the layer thickness or assumed optical properties or to head curvature 10 – 12 Regardless, it appears that there may be crosstalk between the extracerebral layer thickness and BSR for the three-layer model as well as, to a lesser extent, the two-layer model 19 …”

“…Given the relatively poorer performance of the multilayer model fitting in terms of pass rates and CV, we investigated ways to improve the fitting stability of these models by adding constraints on SBFi. We tried both fixing SBFi to BFihomogenoeus(ρ=1 cm) and constraining SBFi to ±20% of BFihomogenoeus(ρ=1 cm), motivated by Forti et al 19 . In general, neither of these constraints improved the multilayer model performance in terms of the pass rate or CV.…”

Diffuse Correlation Spectroscopy: A Review of Recent Advances in Parallelisation and Depth Discrimination Techniques

“…One method to account for extracerebral tissue is to switch from the traditional semiinfinite geometry model, which is commonly used in DCS, to a multi-layered (typically two or three-layered) geometry model [85][86][87]. Although this is an effective technique, these models rely on accurate prior knowledge of the optical properties and thickness of each layer.…”