Skip to main navigation Skip to search Skip to main content

UNet-2022: Exploring Dynamics in Non-isomorphic Architecture

  • Jiansen Guo
  • , Hong Yu Zhou
  • , Liansheng Wang*
  • , Yizhou Yu
  • *Corresponding author for this work
  • Xiamen University
  • The University of Hong Kong

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In this paper, we first analyze the differences between the weight allocation mechanisms of the self-attention and convolution. Based on this analysis, we propose to construct a parallel non-isomorphic block that takes the advantages of self-attention and convolution with simple parallelization. We name the resulting U-shape segmentation model as UNet-2022. In experiments, UNet-2022 obviously outperforms its counterparts in a range segmentation tasks, including abdominal multi-organ segmentation, automatic cardiac diagnosis, neural structures segmentation, and skin lesion segmentation, sometimes surpassing the best performing baseline by 4%. Specifically, UNet-2022 surpasses nnUNet, the most recognized segmentation model at present, by large margins. These phenomena indicate the potential of UNet-2022 to become the model of choice for medical image segmentation.

Original languageEnglish
Title of host publicationMedical Imaging and Computer-Aided Diagnosis - Proceedings of 2022 International Conference on Medical Imaging and Computer-Aided Diagnosis MICAD 2022
EditorsRuidan Su, Yudong Zhang, Han Liu, Alejandro F Frangi
PublisherSpringer Science and Business Media Deutschland GmbH
Pages465-476
Number of pages12
ISBN (Print)9789811667749
DOIs
StatePublished - 2023
Externally publishedYes
EventInternational Conference on Medical Imaging and Computer-Aided Diagnosis, MICAD 2022 - Leicester, United Kingdom
Duration: 20 Nov 202221 Nov 2022

Publication series

NameLecture Notes in Electrical Engineering
Volume810 LNEE
ISSN (Print)1876-1100
ISSN (Electronic)1876-1119

Conference

ConferenceInternational Conference on Medical Imaging and Computer-Aided Diagnosis, MICAD 2022
Country/TerritoryUnited Kingdom
CityLeicester
Period20/11/2221/11/22

Keywords

  • Medical image segmenation
  • Transformer

Fingerprint

Dive into the research topics of 'UNet-2022: Exploring Dynamics in Non-isomorphic Architecture'. Together they form a unique fingerprint.

Cite this