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

The application of 3D ViTs to medical image segmentation has seen remarkablestrides, somewhat overshadowing the budding advancements in ConvolutionalNeural Network (CNN)-based models. Large kernel depthwise convolution hasemerged as a promising technique, showcasing capabilities akin to hierarchicaltransformers and facilitating an expansive effective receptive field (ERF)vital for dense predictions. Despite this, existing core operators, rangingfrom global-local attention to large kernel convolution, exhibit inherenttrade-offs and limitations (e.g., global-local range trade-off, aggregatingattentional features). We hypothesize that deformable convolution can be anexploratory alternative to combine all advantages from the previous operators,providing long-range dependency, adaptive spatial aggregation and computationalefficiency as a foundation backbone. In this work, we introduce 3DDeformUX-Net, a pioneering volumetric CNN model that adeptly navigates theshortcomings traditionally associated with ViTs and large kernel convolution.Specifically, we revisit volumetric deformable convolution in depth-wisesetting to adapt long-range dependency with computational efficiency. Inspiredby the concepts of structural re-parameterization for convolution kernelweights, we further generate the deformable tri-planar offsets by adapting aparallel branch (starting from $1\times1\times1$ convolution), providingadaptive spatial aggregation across all channels. Our empirical evaluationsreveal that the 3D DeformUX-Net consistently outperforms existingstate-of-the-art ViTs and large kernel convolution models across fourchallenging public datasets, spanning various scales from organs (KiTS: 0.680to 0.720, MSD Pancreas: 0.676 to 0.717, AMOS: 0.871 to 0.902) to vessels (e.g.,MSD hepatic vessels: 0.635 to 0.671) in mean Dice.

DeformUX-Net: Exploring a 3D Foundation Backbone for Medical Image Segmentation with Depthwise Deformable Convolution