Robustness of diffuse reflectance spectra analysis by inverse adding doubling algorithm

Abstract

Analysing diffuse reflectance spectra to extract properties of biological tissue requires modelling of light transport within the tissue, considering its absorption, scattering, and geometrical properties. Due to the layered skin structure, skin tissue models are often divided into multiple layers with their associated optical properties. Typically, in the analysis, some model parameters defining these properties are fixed to values reported in the literature to speed up the fitting process and improve its performance. In the absence of consensus, various studies use different approaches in fixing the model parameters. This study aims to assess the effect of fixing various model parameters in the skin spectra fitting process on the accuracy and robustness of a GPU-accelerated two-layer inverse adding-doubling (IAD) algorithm. Specifically, the performance of the IAD method is determined for noiseless simulated skin spectra, simulated spectra with different levels of noise applied, and in-vivo measured reflectance spectra from hyperspectral images of human hands recorded before, during, and after the arterial occlusion. Our results suggest that fixing multiple parameters to a priori known values generally improves the robustness and accuracy of the IAD algorithm for simulated spectra. However, for in-vivo measured spectra, these values are unknown in advance and fixing optical parameters to incorrect values significantly deteriorates the overall performance. Therefore, we propose a method to improve the fitting performance by pre-estimating model parameters. Our findings could be considered in all future research involving the analysis of diffuse reflectance spectra to extract optical properties of skin tissue.

Estimating quantitative physiological and morphological tissue parameters of murine tumor models using hyperspectral imaging and optical profilometry

“…Although we have no ground truth values for estimated tissue parameters, based on our previous observations for simulated reflectance skin spectra, the mean values of fitted parameters are within AE1 STD from the actual values. We also demonstrated that the correlation between the estimated parameters depends on the selection of fixed parameters [57]. In the present study, f m and d e exhibited the highest negative correlation (Pearson correlation coefficient was p 12 ¼ À0:7538), meaning a decrease in epidermal thickness yields a higher melanin volume fraction.…”

“…The fit results also show a good agreement between the measured and fitted spectra, suggesting that the IAD algorithm is accurate and robust in extracting the tissue parameters. However, our recent publication [57] showed that different sets of model parameters could yield precisely the same reflectance spectra, meaning there is already some intrinsic uncertainty in parameter estimation.…”

“…In our recently published research work [57], we showed that the accuracy and robustness of tissue parameter extraction by the IAD algorithm improved significantly as more model parameters were fixed. For example, the standard deviation of decreased eightfold as the number of free model parameters was reduced from 11 to 5.…”

“…A GPU‐accelerated inverse adding‐doubling [56] (IAD) algorithm was implemented in MATLAB R2020b (Mathworks, MA), allowing rapid and accurate light propagation simulation in the murine skin model. A two‐layer skin model presented in Figure 2 was used, consisting of the epidermis and semi‐infinite dermis with 11 tissue properties thoroughly described in our recent publication [57]. Briefly, the absorption coefficient of the epidermis was calculated as [7]: where is the volume fraction of melanin, is the melanin absorption coefficient and is the baseline absorption of bloodless skin.…”

Daily monitoring of CT26 murine tumor model using hyperspectral imaging and optical profilometry

“…As for hyperspectral image analysis, a GPU-accelerated inverse adding-doubling (IAD) algorithm was developed in MATLAB R2020b (Mathworks, Natick, MA) for fast and accurate light propagation simulation in layered turbid media 5 . Briefly, we utilized a two-layer murine skin model consisting of an upper layer (epidermis) and a semi-infinite lower layer (dermis) with 11 model parameters describing tissue physiology (e.g., melanin, hemoglobin) and morphology (e.g., scattering power) 7 . The measured reflectance spectra were fitted using the Levenberg-Marquardt (LM) algorithm adopted for GPU to extract all model parameters.…”