HoVer-UNet: Accelerating HoVerNet with UNet-based multi-class nuclei segmentation via knowledge distillation

TL;DR

HoVer-UNet is a compact, efficient UNet-based model that distills knowledge from HoVerNet for nuclei segmentation, achieving similar accuracy with significantly faster inference in histopathology analysis.

Contribution

The paper introduces HoVer-UNet, a streamlined UNet model with a Mix Vision Transformer backbone that effectively distills HoVerNet knowledge, reducing computational costs while maintaining performance.

Findings

Achieved comparable results to HoVerNet on PanNuke and Consep datasets.

Reduced inference time by threefold.

Made the model code publicly available.

Abstract

We present HoVer-UNet, an approach to distill the knowledge of the multi-branch HoVerNet framework for nuclei instance segmentation and classification in histopathology. We propose a compact, streamlined single UNet network with a Mix Vision Transformer backbone, and equip it with a custom loss function to optimally encode the distilled knowledge of HoVerNet, reducing computational requirements without compromising performances. We show that our model achieved results comparable to HoVerNet on the public PanNuke and Consep datasets with a three-fold reduction in inference time. We make the code of our model publicly available at https://github.com/DIAGNijmegen/HoVer-UNet.