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

Many current neural networks for medical imaging generalise poorly to dataunseen during training. Such behaviour can be caused by networks overfittingeasy-to-learn, or statistically dominant, features while disregarding otherpotentially informative features. For example, indistinguishable differences inthe sharpness of the images from two different scanners can degrade theperformance of the network significantly. All neural networks intended forclinical practice need to be robust to variation in data caused by differencesin imaging equipment, sample preparation and patient populations.  To address these challenges, we evaluate the utility of spectral decouplingas an implicit bias mitigation method. Spectral decoupling encourages theneural network to learn more features by simply regularising the networks'unnormalised prediction scores with an L2 penalty, thus having no addedcomputational costs.  We show that spectral decoupling allows training neural networks on datasetswith strong spurious correlations and increases networks' robustness for datadistribution shifts. To validate our findings, we train networks with andwithout spectral decoupling to detect prostate cancer tissue slides andCOVID-19 in chest radiographs. Networks trained with spectral decouplingachieve up to 9.5 percent point higher performance on external datasets.  Our results show that spectral decoupling helps with generalisation issuesassociated with neural networks, and can be used to complement or replacecomputationally expensive explicit bias mitigation methods, such as stainnormalization in histological images. We recommend using spectral decoupling asan implicit bias mitigation method in any neural network intended for clinicaluse.

Spectral decoupling allows training transferable neural networks in medical imaging