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

Well-labeled datasets of chest radiographs (CXRs) are difficult to acquiredue to the high cost of annotation. Thus, it is desirable to learn a robust andtransferable representation in an unsupervised manner to benefit tasks thatlack labeled data. Unlike natural images, medical images have their own domainprior; e.g., we observe that many pulmonary diseases, such as the COVID-19,manifest as changes in the lung tissue texture rather than the anatomicalstructure. Therefore, we hypothesize that studying only the texture without theinfluence of structure variations would be advantageous for downstreamprognostic and predictive modeling tasks. In this paper, we propose agenerative framework, the Lung Swapping Autoencoder (LSAE), that learnsfactorized representations of a CXR to disentangle the texture factor from thestructure factor. Specifically, by adversarial training, the LSAE is optimizedto generate a hybrid image that preserves the lung shape in one image butinherits the lung texture of another. To demonstrate the effectiveness of thedisentangled texture representation, we evaluate the texture encoder $Enc^t$ inLSAE on ChestX-ray14 (N=112,120), and our own multi-institutional COVID-19outcome prediction dataset, COVOC (N=340 (Subset-1) + 53 (Subset-2)). On bothdatasets, we reach or surpass the state-of-the-art by finetuning $Enc^t$ inLSAE that is 77% smaller than a baseline Inception v3. Additionally, insemi-and-self supervised settings with a similar model budget, $Enc^t$ in LSAEis also competitive with the state-of-the-art MoCo. By "re-mixing" the textureand shape factors, we generate meaningful hybrid images that can augment thetraining set. This data augmentation method can further improve COVOCprediction performance. The improvement is consistent even when we directlyevaluate the Subset-1 trained model on Subset-2 without any fine-tuning.

Lung Swapping Autoencoder: Learning a Disentangled Structure-texture Representation of Chest Radiographs