Designing a use-error robust machine learning model for quantitative analysis of diffuse reflectance spectra
Allison Scarbrough,Keke Chen,Bing Yu
DOI: https://doi.org/10.1117/1.jbo.29.1.015001
IF: 3.758
2024-01-12
Journal of Biomedical Optics
Abstract:SignificanceMachine learning (ML)-enabled diffuse reflectance spectroscopy (DRS) is increasingly used as an alternative to the computation-intensive inverse Monte Carlo (MCI) simulation to predict tissue's optical properties, including the absorption coefficient, μa and reduced scattering coefficient, μs′.AimWe aim to develop a use-error-robust ML algorithm for optical property prediction from DRS spectra.ApproachWe developed a wavelength-independent regressor (WIR) to predict optical properties from DRS data. For validation, we generated 1520 simulated DRS spectra with the forward Monte Carlo model, where μa=0.44 to 2.45 cm−1, and μs′=6.53 to 9.58 cm−1. We introduced common use-errors, such as wavelength miscalibrations and intensity fluctuations. Finally, we collected 882 experimental DRS images from 170 tissue-mimicking phantoms and compared performances of the WIR model, a dense neural network, and the MCI model.ResultsWhen compounding all use-errors on simulated data, the WIR model best balanced accuracy and speed, yielding errors of 1.75% for μa and 1.53% for μs′, compared to the MCI's 50.9% for μa and 24.6% for μs′. Regarding experimental data, WIR model had mean errors of 13.2% and 6.1% for μa and μs′, respectively. The errors for MCI were about eight times higher.ConclusionsThe WIR model presents reliable use-error-robust optical property predictions from DRS data.
radiology, nuclear medicine & medical imaging,biochemical research methods,optics