Tumor grading model employing geometric analysis of histopathological images with characteristic nuclei dictionary

V Brindha,P Jayashree,P Karthik,P Manikandan
DOI: https://doi.org/10.1016/j.compbiomed.2022.106008
Abstract:Histopathological study has been shown to improve diagnosis of various disease classifications effectively as any disease condition is correlated to characteristic set of changes in the tissue structure. This study aims at developing an automated neural network system for grading brain tumors (Glioblastoma Multiforme) from histopathological images within the Whole Slide Images (WSI) of hematoxylin and eosin (H&E) stains with significant accuracy. Hematoxylin channels are extracted from the histopathological image patches using color de-convolution. Cell nuclei are precisely segmented using three level Otsu thresholding. From each segmented image, nuclei boundaries are extracted to extract nucleus level features based on their shape and size. Geometric features including ellipse eccentricities, nucleus perimeter, area, and polygon edge counts are extracted using geometric algorithms to define the nuclei boundaries of the segmented image. These features are collected for a large number of nuclei and the nuclei are clustered using the K-Means algorithm in order to create a dictionary. One of the major contributions involves the creation of dictionary of a fixed number of representative cell nuclei to speed up patch level classification. This optimal dictionary is used for clustering extracted cell nuclei and a fixed length histogram of counts on different types of nuclei is obtained. The proposed system has been tested with a total of 239600 TCGA patches of GBM and 206000 patches of LGG collected from GDC data portal and it showed good diagnosis performance with auto-classification accuracy of 97.2% compared to other state-of-art methods. Our results on segmentation and classification are encouraging, with better attainment with regard to precision and accuracy in contrast with previous models. The auto grading proposed system will act as a potential guide for pathologists to make more accurate decisions.
What problem does this paper attempt to address?