Panoptic segmentation with highly imbalanced semantic labels

TL;DR

This paper presents a panoptic segmentation approach tailored for highly imbalanced semantic labels in colon nuclei images, combining a specialized weighted loss with a state-of-the-art instance segmentation model within a Hovernet-like architecture.

Contribution

It introduces a novel weighted loss function for semantic segmentation of imbalanced cell types and integrates it with an advanced nuclei instance segmentation model in a unified framework.

Findings

Effective handling of class imbalance in semantic segmentation

Improved nuclei instance segmentation accuracy

Successful participation in ISBI 2022 challenge

Abstract

We describe here the panoptic segmentation method we devised for our participation in the CoNIC: Colon Nuclei Identification and Counting Challenge at ISBI 2022. Key features of our method are a weighted loss specifically engineered for semantic segmentation of highly imbalanced cell types, and a state-of-the art nuclei instance segmentation model, which we combine in a Hovernet-like architecture.