Optical taxonomic signal

TL;DR

This paper introduces a method combining feature selection and discriminant statistics to improve optical spectroscopy for medical diagnosis by identifying features with consistent signals across detector pixels.

Contribution

It proposes a new approach to mitigate scattering issues in biological tissue spectroscopy through specific feature and signal analysis techniques.

Findings

Feature selection improves discrimination in spectroscopic data

Optical taxonomic signal quantifies discrimination efficacy

Method accounts for variables like source-detector separation

Abstract

Scattering of light by biological tissue has hindered applications of spectroscopy to medical diagnosis. We describe here a combination of feature selection techniques and several discriminant statistics that may mitigate this problem. In the particular case of spectroscopy, a useful feature should have linewidth, which in practice means that the discriminant statistic should have significant values on several contiguous pixels of the detector. We also suggest a definition for optical taxonomic signal as a measure of how efficacious a particular combination may be and how much other variables such as source-detector separation and fiber width may affect discrimination.