Separable-HoverNet and Instance-YOLO for Colon Nuclei Identification and Counting
This work addresses the problem of nuclei identification and counting in computational pathology for colon cancer analysis, representing an incremental improvement with specific benchmark results.
The authors tackled the challenge of automatically recognizing and quantifying different types of nuclei in colon histology images, which is difficult due to intraclass variability, by combining Separable-HoverNet and Instance-YOLOv5, achieving mPQ+ 0.389 on segmentation and classification and r2 0.599 on cellular composition in the ISBI 2022 CoNIC Challenge.
Nuclear segmentation, classification and quantification within Haematoxylin & Eosin stained histology images enables the extraction of interpretable cell-based features that can be used in downstream explainable models in computational pathology (CPath). However, automatic recognition of different nuclei is faced with a major challenge in that there are several different types of nuclei, some of them exhibiting large intraclass variability. In this work, we propose an approach that combine Separable-HoverNet and Instance-YOLOv5 to indentify colon nuclei small and unbalanced. Our approach can achieve mPQ+ 0.389 on the Segmentation and Classification-Preliminary Test Dataset and r2 0.599 on the Cellular Composition-Preliminary Test Dataset on ISBI 2022 CoNIC Challenge.