Adaptation of a Multi-Site Network to a New Clinical Site Via Batch-Normalization Similarity

Shira Kasten Serlin, Jacob Goldberger, Hayit Greenspan

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

Abstract

This paper tackles the challenging problem of medical site adaptation; i.e., learning a model from multi-site source data such that it can be modified and adapted to a new site using only unlabeled data from the new site. The method is based on Domain Specific Batch Normalization architecture and uses the Batch Normalization statistics of the new site to find the most similar internal site. The similarity measure is computed in an embedded space of the BN parameters. We evaluated our method on the task of MRI prostate segmentation. Public datasets from six different institutions were used, containing distribution shifts. The experimental results show that the proposed approach outperforms other generalization and adaptation methods.

Original languageEnglish
Title of host publicationISBI 2022 - Proceedings
Subtitle of host publication2022 IEEE International Symposium on Biomedical Imaging
PublisherIEEE Computer Society
ISBN (Electronic)9781665429238
DOIs
StatePublished - 2022
Event19th IEEE International Symposium on Biomedical Imaging, ISBI 2022 - Kolkata, India
Duration: 28 Mar 202231 Mar 2022

Publication series

NameProceedings - International Symposium on Biomedical Imaging
Volume2022-March
ISSN (Print)1945-7928
ISSN (Electronic)1945-8452

Conference

Conference19th IEEE International Symposium on Biomedical Imaging, ISBI 2022
Country/TerritoryIndia
CityKolkata
Period28/03/2231/03/22

Keywords

  • batch-normalization
  • domain adaptation
  • multi-site
  • prostate segmentation

Fingerprint

Dive into the research topics of 'Adaptation of a Multi-Site Network to a New Clinical Site Via Batch-Normalization Similarity'. Together they form a unique fingerprint.

Cite this