Automated analysis of histological images by computational algorithms

Abstract

The study of cellular tissues provides an incontestable source of information and comprehension about thehuman body and the surrounding environment. Accessing this information is, therefore, crucial to determineand diagnose a wide variety of pathologies detectable only at a microscopic scale. Hence, histology plays animportant role in the clinical diagnosis of pathologies involving abnormal cellular conformation. Inhistological images, semi- or automated segmentation algorithms are able to separate and identify cellularstructures according to morphological differences. The segmentation is usually the first task incomputational vision systems and, concerning histopathology, for the automated analysis of histologicalimages. Since the histological samples are thin, the volumetric features are almost unnoticeable,corresponding to losses of valuable information, mainly topographical and volumetric data, critical for acorrect analysis. Hence, the combination of segmentation and 3D reconstruction algorithms applied tohistological image datasets provides more information about the analyzed pathology and microscopicstructures, highlighting abnormal areas [1].In order to provide insights on pathological volumetric data, the present work focused on developing anautomatic computational solution for performing the 3D surface reconstruction of relevant tissue structurespresented in 2D histological slices. A state of the art technique, called stain deconvolution, was implementedto achieve color image segmentation providing an accurate segmentation of two different stains present inthe histological data: Hematoxylin and Eosin tissues. To register, i.e. align, the image slices presented in theinput datasets, an intensity based registration method was implemented, being the alignment performedbetween each slice in the input dataset and the reference slice (middle slice of the dataset). The datasetchosen for the previous alignment operation was the set of images obtained through the stain deconvolutionmethod for the hematoxylin stain. The transformation matrix obtained for each slice was then applied to theeosin stained images. The 3D reconstruction was implemented based on the Marching Cubes algorithm.Thus, combining algorithms of image segmentation and registration with of 3D surface reconstruction, itwas possible to obtain a volumetric representation of the pertinent tissue structures from the input imagedatasets. The experiments conducted revealed accurate and fast surface reconstructions of the differentstained tissues under study, highlighting the interesting structures and their volumetric interactions with thesurrounding healthy tissues