Nuclei Instance Segmentation And Type Classification In Histology Images
Cancer is a leading cause of death worldwide, accounting for nearly 10 million deaths in 2021, due to late-stage presentation and lack of early diagnosis. This disease results from transformation of normal cells into tumors. Automated detection and segmentation of every nucleus is a fundamental task in biomedical research However, it is a very challenging task because of the different types of nuclei, large intraclass variations, diverse cell morphologies and large number of clusters of crowded nuclei resulting in poor performance by deep learning models.CNN based models mostly targets coarse details of image and fails to model long range similarities pixel wise, while in medical imaging proper identification of fine details at higher resolution is necessary. Hence to address these limitations of deep learning models in classification and segmentation of nuclei we present a new state of the art network that performs mask segmentation, class identification and bbox generation tasks in an end-to-end manner.
In particular, we first exploit a model to generate a mask for each instance in the image via instance mask generation using hierarchical transformer, features output is passed to border following the polygon generator and in parallel to multi compound transformer architecture for feature based classification and Bbox generation. For training and experimentation publicly available PanNuke dataset with 19 different tissue types and CryoNuSeg with 10 different organ instances will be used. We are optimistic that the proposed transformer based model can remarkably boost the instance wise segmentation accuracy of nuclei. To our knowledge this is the first state of art network focusing on utilizing border following algorithm alongside transformer based feature classification for medical imaging and specifically nuclei instances segmentation