Neural network‐based optimization of sub‐diffuse reflectance spectroscopy for improved parameter prediction and efficient data collection

Abstract

In this study, a general and systematical investigation of sub‐diffuse reflectance spectroscopy is implemented. A Gegenbauer‐kernel phase function‐based Monte Carlo is adopted to describe photon transport more efficiently. To improve the computational efficiency and accuracy, two neural network algorithms, namely, back propagation neural network and radial basis function neural network are utilized to predict the absorption coefficient μnormala, reduced scattering coefficient μnormals′ and sub‐diffusive quantifier γ, simultaneously, at multiple source‐detector separations (SDS). The predicted results show that the three parameters can be predicated accurately by selecting five SDSs or above. Based on the simulation results, a four wavelength (520, 650, 785 and 830 nm) measurement system using five SDSs is designed by adopting phase‐lock‐in technique. Furtherly, the trained neural‐network models are utilized to extract optical properties from the phantom and in vivo experimental data. The results verify the feasibility and effectiveness of our proposed system and methods in mucosal disease diagnosis.

Application of transfer learning for rapid calibration of spatially resolved diffuse reflectance probes for extraction of tissue optical properties

“…demonstrated recovery of μa and μs′ with mean relative error ranging from 1.62% to 5.92% and 0.8% to 1.6%, respectively, with their baseline neural network and its physics-guided variants but only for a simulation dataset. An et al 24 . developed ANNs for their spectroscopy system, which predicted μa with mean absolute error ranging from 0.13 to 0.17 cm−1 and μs′ ranging from 1.43 to 1.71 cm−1.…”

“…An et al. 24 developed ANNs for their spectroscopy system, which predicted with mean absolute error ranging from 0.13 to and ranging from 1.43 to . However, their ANN prediction was limited to only four wavelengths compared to our broad-spectrum prediction ranging from 500 to 740 nm.…”

Application of Transfer Learning for Rapid Calibration of Spatially-resolved Diffuse Reflectance Probes for Extraction of Tissue Optical Properties

“…Chong et al demonstrated recovery of μ a and μ’ with mean relative error ranging from 1.62-5.92% and 0.8-1.6%, respectively, with their baseline neural network and its physics-guided variants for a simulated dataset[20]. An et al developed ANNs for their four-wavelength spectroscopy system, which predicted μ a with mean absolute error ranging from 0.13-0.17 cm -1 and μ s ′ ranging from 1.43-1.71 cm -1 [21]. Lan et al predicted μ a with Euclidean distance = 0.25 and μ s ′ with Euclidean distance = 4.14[13].…”

Neural Network-based Inverse Model for Diffuse Reflectance Spectroscopy

“…These techniques leverage the inherent capacity of ML models to learn complex patterns and relationships from large datasets, enabling accurate predictions and analysis of optical properties. In the aspect of measuring technique, compared to ours, the existing ML-based models deal with different DRS techniques, such as frequency domain DRS (FD-DRS) that estimates optical properties from measuring amplitude and phase shift of the diffusely reflected light [11][12][13][14], or spatially resolved DRS [15][16][17], or subdiffusive reflectance in which measurements are made closer than MFP and therefore more optical properties are involved (first similarity parameter W) [17,18]. The generic DRS, which is the technique we work with, compared to FD-DRS, requires less expensive devices, less complicated experimental setups, and the data interpretation is a lot easier due to the simpler mechanism.…”