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

3D imaging enables accurate diagnosis by providing spatial information aboutorgan anatomy. However, using 3D images to train AI models is computationallychallenging because they consist of 10x or 100x more pixels than their 2Dcounterparts. To be trained with high-resolution 3D images, convolutionalneural networks resort to downsampling them or projecting them to 2D. Wepropose an effective alternative, a neural network that enables efficientclassification of full-resolution 3D medical images. Compared to off-the-shelfconvolutional neural networks, our network, 3D Globally-Aware Multiple InstanceClassifier (3D-GMIC), uses 77.98%-90.05% less GPU memory and 91.23%-96.02% lesscomputation. While it is trained only with image-level labels, withoutsegmentation labels, it explains its predictions by providing pixel-levelsaliency maps. On a dataset collected at NYU Langone Health, including 85,526patients with full-field 2D mammography (FFDM), synthetic 2D mammography, and3D mammography, 3D-GMIC achieves an AUC of 0.831 (95% CI: 0.769-0.887) inclassifying breasts with malignant findings using 3D mammography. This iscomparable to the performance of GMIC on FFDM (0.816, 95% CI: 0.737-0.878) andsynthetic 2D (0.826, 95% CI: 0.754-0.884), which demonstrates that 3D-GMICsuccessfully classified large 3D images despite focusing computation on asmaller percentage of its input compared to GMIC. Therefore, 3D-GMIC identifiesand utilizes extremely small regions of interest from 3D images consisting ofhundreds of millions of pixels, dramatically reducing associated computationalchallenges. 3D-GMIC generalizes well to BCS-DBT, an external dataset from DukeUniversity Hospital, achieving an AUC of 0.848 (95% CI: 0.798-0.896).

An efficient deep neural network to find small objects in large 3D images