Color Space-based HoVer-Net for Nuclei Instance Segmentation and Classification

TL;DR

This paper introduces a novel nuclei segmentation and classification method using a color space-based HoVer-Net with ConvNeXt encoder, multi-channel color features, UFL loss, and SAM optimization, outperforming current SOTA.

Contribution

It combines a transformer-based encoder, multi-channel color space features, and advanced loss and optimization techniques to improve nuclei segmentation and classification accuracy.

Findings

Outperforms HoVer-Net by 12.489% mPQ+ on CoNiC Challenge 2022

Effectively handles small objects and class imbalance

Enhances feature extraction with color space-based approach

Abstract

Nuclei segmentation and classification is the first and most crucial step that is utilized for many different microscopy medical analysis applications. However, it suffers from many issues such as the segmentation of small objects, imbalance, and fine-grained differences between types of nuclei. In this paper, multiple different contributions were done tackling these problems present. Firstly, the recently released "ConvNeXt" was used as the encoder for HoVer-Net model since it leverages the key components of transformers that make them perform well. Secondly, to enhance the visual differences between nuclei, a multi-channel color space-based approach is used to aid the model in extracting distinguishing features. Thirdly, Unified Focal loss (UFL) was used to tackle the background-foreground imbalance. Finally, Sharpness-Aware Minimization (SAM) was used to ensure generalizability of the model. Overall, we were able to outperform the current state-of-the-art (SOTA), HoVer-Net, on the preliminary test set of the CoNiC Challenge 2022 by 12.489% mPQ+.