English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Performance of convolutional neural networks (CNNs) in image analysis tasksis often marred in the presence of acquisition-related distribution shiftsbetween training and test images. Recently, it has been proposed to tackle thisproblem by fine-tuning trained CNNs for each test image. Suchtest-time-adaptation (TTA) is a promising and practical strategy for improvingrobustness to distribution shifts as it requires neither data sharing betweeninstitutions nor annotating additional data. Previous TTA methods use a helpermodel to increase similarity between outputs and/or features extracted from atest image with those of the training images. Such helpers, which are typicallymodeled using CNNs, can be task-specific and themselves vulnerable todistribution shifts in their inputs. To overcome these problems, we propose tocarry out TTA by matching the feature distributions of test and trainingimages, as modelled by a field-of-experts (FoE) prior. FoEs model complicatedprobability distributions as products of many simpler expert distributions. Weuse 1D marginal distributions of a trained task CNN's features as experts inthe FoE model. Further, we compute principal components of patches of the taskCNN's features, and consider the distributions of PCA loadings as additionalexperts. We validate the method on 5 MRI segmentation tasks (healthy tissues in4 anatomical regions and lesions in 1 one anatomy), using data from 17 clinics,and on a MRI registration task, using data from 3 clinics. We find that theproposed FoE-based TTA is generically applicable in multiple tasks, andoutperforms all previous TTA methods for lesion segmentation. For healthytissue segmentation, the proposed method outperforms other task-agnosticmethods, but a previous TTA method which is specifically designed forsegmentation performs the best for most of the tested datasets. Our code ispublicly available.

A Field of Experts Prior for Adapting Neural Networks at Test Time