Hyperspectral Signal Analysis for Thyroid Neoplasm Typification On Infrared Spectrum

摘要

Thyroid abnormalities typification, including neoplasms, is usually performed by histological examinations on slides containing biopsy tissue. Until 2040, the World Health Organization (WHO) foresees 193,727 new cases of thyroid cancer around the world. Early diagnosis can lead doctors to prescribe a less aggressive treatment, providing a better prognosis and so provide patients an improvement in life quality. 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 image. The hyperspectral signals, obtained by infrared equipment, are characterized by presenting for each pixel of the image a spectrum of absorbance values for different frequencies, which is sensitive to the biochemical characteristics of the underlying tissue. The goal of this paper is to investigate if it's possible to characterize cancerous, normal, and inflammatory thyroid tissue by analyzing its radiation absorbance level over the hyperspectral point of view. For that, histological slides containing samples of thyroid biopsies were exposed to different infrared radiation in order to collect the material absorbance spectra. These signals were then used on different types of analysis, such as absorbance-level distribution analysis, feature selection analysis, and pattern recognition analysis using traditional supervised machine learning algorithms. Besides it's a complex task, hyperspectral signals showed themselves a promising tool to characterize different tissue over the infrared spectrum.