Spectral super-resolution spectroscopy for biomedical applications

Abstract

Reflectance spectroscopy and hyperspectral (or multispectral) imaging that can acquire a matrix of intensity as a function of the position and the wavelength of light (also known as a hypercube) are extensively used to quantify biochemical composition, structure, and vasculature in biological tissue. However, these methods often rely on bulky and costly optical components, which limit the development of compact, rapid, and cost-effective technologies. Fortuitously, several different research communities have demonstrated that it is possible to mathematically reconstruct hyperspectral (with high spectral resolution) or multispectral data from RGB images taken by a conventional camera (three-color sensor). However, these methods, such as compressive (compressed) sensing and deep learning, are often limited for extracting detailed biological spectral profiles and require an extremely large amount of training data. We have recently developed a spectral super-resolution framework that enables us to virtually transform the built-in camera (RGB sensor) of a smartphone into a hyperspectral imager for accurate and precise spectroscopic analyses, without a need for any hardware modifications or accessories. Super-resolution means high-resolution reconstruction of digital images acquired with lowresolution systems. We have extended this concept to the frequency domain for hyperspectral imaging, which has numerous biomedical applications. As an example, our mobile version of spectral super-resolution combines imaging of peripheral tissue and spectroscopic quantification of blood hemoglobin levels in a noninvasive manner. Spectral superresolution spectroscopy can also serve as an example that data-driven technologies can minimize hardware complexity, facilitating the tempo of clinical translation.