Domain generalization via feature disentanglement with reconstruction for pathology image segmentation

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In pathology, the learned model may suffer from performance degradation due to stain variations between the training and testing datasets. To address this challenge, this paper proposes a feature disentanglement approach to learn the domain-invariant features to achieve domain generalization. The adaptive instance normalization (AdaIN)-based reconstruction is introduced to preserve the important semantic information. The generalization ability of the proposed method is further improved by using a contrastive loss function based on color augmentation to attract the domain-invariant features and repel the domain-specific features in the feature disentanglement process. The proposed method is evaluated on liver tumor and liver lipid droplet segmentation tasks. The results demonstrate that the proposed method can be applied to unseen datasets scanned by different scanners without significant performance degradation.

Original languageEnglish
Title of host publicationProceedings - 2023 IEEE Conference on Artificial Intelligence, CAI 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages152-153
Number of pages2
ISBN (Electronic)9798350339840
DOIs
Publication statusPublished - 2023
Event2023 IEEE Conference on Artificial Intelligence, CAI 2023 - Santa Clara, United States
Duration: 2023 Jun 52023 Jun 6

Publication series

NameProceedings - 2023 IEEE Conference on Artificial Intelligence, CAI 2023

Conference

Conference2023 IEEE Conference on Artificial Intelligence, CAI 2023
Country/TerritoryUnited States
CitySanta Clara
Period23-06-0523-06-06

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Computer Vision and Pattern Recognition
  • Modelling and Simulation
  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'Domain generalization via feature disentanglement with reconstruction for pathology image segmentation'. Together they form a unique fingerprint.

Cite this