Hyperspectral signal processing and analysis applied to histopathological diagnosis

Abstract

The anomalies typification in human tissues, including neoplasms, is usually performed by histological examinations on slides containing biopsies. Research on digital biopsy images has grown rapidly, leveraging the development and improvement of image processing methods specially developed or adapted for this category of images. The hyperspectral signals, obtained using infrared equipment, are characterized by presenting for each pixel of the image a spectrum of absorbance values for the different frequencies, which is sensitive to the characteristics of the underlying tissue. This project aims to investigate and validate the hypothesis that infrared hyperspectral signals can be used to recognize patterns that allow the identification of regions of the slide containing healthy tissue or with abnormal characteristics, especially neoplasms. In order to do this, datasets of hyperspectral signals representing absorbance over a wide range of infrared frequencies, obtained from slides containing samples of healthy tissue and abnormal tissue from different organs, such as thyroid, skin and breast, will be analyzed using deep learning techniques, implemented as classifiers, with which it will be possible to characterize the regions of the slide according to the possible anomalies present. Thus, the proposed method will allow the automatic analysis of biopsies, to make the diagnostic process more accurate and effective. Besides, images obtained from biopsy slides stained with Hematoxylin and Eosin will be processed by computer vision techniques in comparison with the hyperspectral signals approach, in order to detect patterns that are not identifiable by the currently used visual method. (AU)