LKCell: Efficient Cell Nuclei Instance Segmentation with Large Convolution Kernels

The segmentation of cell nuclei in tissue images stained with the blood dyehematoxylin and eosin (H&E) is essential for various clinical applicationsand analyses. Due to the complex characteristics of cellular morphology, alarge receptive field is considered crucial for generating high-qualitysegmentation. However, previous methods face challenges in achieving a balancebetween the receptive field and computational burden. To address this issue, wepropose LKCell, a high-accuracy and efficient cell segmentation method. Itscore insight lies in unleashing the potential of large convolution kernels toachieve computationally efficient large receptive fields. Specifically, (1) Wetransfer pre-trained large convolution kernel models to the medical domain forthe first time, demonstrating their effectiveness in cell segmentation. (2) Weanalyze the redundancy of previous methods and design a new segmentationdecoder based on large convolution kernels. It achieves higher performancewhile significantly reducing the number of parameters. We evaluate our methodon the most challenging benchmark and achieve state-of-the-art results (0.5080mPQ) in cell nuclei instance segmentation with only 21.6% FLOPs compared withthe previous leading method. Our source code and models are available athttps://github.com/hustvl/LKCell.