PLS regression approach for oxygen saturation in spectroscopic fundus images

Abstract

Two-dimensional distribution of oxygen saturation levels across the human retina is predicted by measuring the multispectral images. A liquid crystal tunable filter is used to scan wavelength ranging from 500 to 700 nm to acquire the multispectral images with the spectral resolution of 20 nm. After preprocessing reflected spectra at every pixel to cancel common offsets and amplifications, the partial least squares regression is adopted to estimate the oxygen saturation levels at each pixel point. All the resulted response values compose the oxygen saturation level map. The effects of preprocessing types on the predicted oxygen saturation levels are also discussed.

Retinal Oximetry with 510–600 nm Light Based on Partial Least-Squares Regression Technique

“…The oxygen saturation of retinal blood is calculated from the difference in the spectra of oxyhemoglobin and deoxyhemoglobin. 9,10) Figure 1(a) shows these two spectra in the range from 400 to 800 nm. In the wavelength range below 450 nm, both oxyhemoglobin and deoxyhemoglobin have strong absorption.…”