MGDC-UNet: Multi-group Deformable Convolution for Medical Image Segmentation
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Deformable Convolution, Convolutional Neural Network, Vision Transformer, Medical Image Segmentation, CT, MRI
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We propose MGDC-UNet, a novel multi-group deformable convolution model to address limitations in traditional CNNs and ViTs for 3D medical image segmentation.
Abstract: Recently, there has been growing interest in developing Vision Transformer (ViT) or Convolutional Neural Network (CNN) methods for 3D medical image segmentation, which necessitates both large receptive fields and adaptations to varying spatial geometries. Previous works in both CNNs and ViTs demonstrated limitations in capturing the complex spatial and semantic structure of 3D medical images. In this paper, we introduce MGDC-UNet, a multi-group deformable convolution network for 3D volumetric medical image segmentation. Our MGDC-UNet employs deformable convolution operators with learnable spatial offsets to improve attention on semantically important regions. Our approach leverages stable spatial distribution across subjects to enhance semantic learning. We also incorporate transformer components to augment feature learning and reduce inductive biases inherent in traditional CNNs. MGDC-UNet demonstrated superior performance accuracy on three challenging segmentation tasks using public datasets: 1). brain tumor segmentation (BraTS21), 2). CT multi-organ segmentation (FLARE21) and 3). cross-modality MR/CT segmentation (AMOS22). Our network also compared favorably with existing methods in terms of computational efficiency.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6320
Loading