The proposed Medical Visual Prompting (MVP) framework leverages pre-training and prompting concepts to enable efficient and versatile lesion segmentation across diverse medical imaging tasks, outperforming task-specific methods while simplifying the model.
A novel hybrid CNN-Transformer architecture, PAG-TransYnet, that seamlessly integrates Pyramid features, CNN features, and Transformer features using a Dual-Attention Gate mechanism to achieve state-of-the-art performance across diverse medical imaging segmentation tasks.
PAM-UNet, a novel architecture that combines mobile convolution blocks and a Progressive Luong Attention (PLA) mechanism, achieves state-of-the-art biomedical image segmentation performance while maintaining low computational cost.
ASSNet, a novel transformer-based architecture, effectively integrates local and global features to achieve state-of-the-art performance in segmenting small tumors and miniature organs across diverse medical imaging datasets.
SMAFormer, a novel Transformer-based architecture, effectively integrates synergistic multi-attention mechanisms and a multi-scale segmentation modulator to achieve state-of-the-art performance in diverse medical image segmentation tasks.
The proposed PMR-Net can effectively extract and fuse multi-scale local and global features to accurately segment objects of different sizes in medical images.
BRAU-Net++ is a hybrid CNN-Transformer network that effectively integrates the merits of convolutional neural networks and transformers to achieve accurate and robust medical image segmentation.
DMC-Net, a novel convolutional neural network architecture, achieves superior pancreas segmentation accuracy in CT images by dynamically integrating multi-scale and multi-resolution features while maintaining computational efficiency.
Integrating frozen transformer blocks from pre-trained large language models (LLMs) into vision transformer (ViT) architectures significantly improves the performance and accuracy of medical image segmentation tasks.
Med-TTT, a novel deep learning model, effectively segments medical images by integrating Vision-TTT layers, multi-resolution fusion, and frequency domain information, achieving high accuracy while maintaining computational efficiency.