Incorporating explicit architectural bias to emphasize temporal differences between baseline and follow-up scans significantly enhances the performance of longitudinal multiple sclerosis lesion segmentation compared to state-of-the-art single timepoint and existing longitudinal methods.
A Monte Carlo-guided interpolation consistency-based framework for segmenting 2D MR images of the prostate region, which improves the generalization ability of the Segment Anything Model (SAM) through semi-supervised learning.
One-Prompt Segmentation combines the strengths of one-shot and interactive segmentation methods to enable zero-shot generalization across diverse medical imaging tasks, requiring only a single prompted sample during inference.
SaLIP, a unified framework that leverages the combined capabilities of the Segment Anything Model (SAM) and Contrastive Language-Image Pre-Training (CLIP) to perform zero-shot organ segmentation in medical images, without relying on domain expertise or annotated data for prompt engineering.
Skeleton Recall Loss is a novel loss function that effectively preserves the connectivity of thin tubular structures in segmentation tasks, while being computationally efficient and compatible with multi-class problems.
This paper introduces the Semi-Mamba-UNet, a novel framework that integrates a purely visual mamba-based U-Shape Encoder-Decoder architecture with a conventional CNN-based UNet into a Semi-Supervised Learning (SSL) framework, leveraging both networks to simultaneously generate pseudo labels and cross supervise each other on the pixel level.
A new U-Net variant, U-Net v2, is introduced that features a novel and straightforward design of skip connections to explicitly integrate semantic information from higher-level features and finer details from lower-level features into feature maps at each level, leading to improved medical image segmentation performance.
The core message of this paper is to introduce a novel spatially agile transformer UNet architecture, termed AgileFormer, that systematically incorporates deformable patch embedding, spatially dynamic self-attention, and multi-scale deformable positional encoding to effectively capture diverse target objects in medical image segmentation tasks.
The proposed UltraLight Vision Mamba UNet (UltraLight VM-UNet) is a highly efficient model for skin lesion segmentation, achieving excellent performance with only 0.049M parameters and 0.060 GFLOPs, which is significantly lower than existing lightweight Vision Mamba models.
Medical image segmentation foundation models can be effectively deployed in a federated learning framework, maintaining performance and privacy.