Attention U-Net: Learning Where to Look for the Pancreas
By Ozan Oktay, Jo Schlemper, Loic Le Folgoc, M. J. Lee, Mattias Paul Heinrich, Kazunari Misawa, Kensaku Mori, Steven G. McDonagh, Nils Y. Hammerla, Bernhard Kainz and 2 more. 2018
7,627 citations recorded by the source corpus.
We propose a novel attention gate (AG) model for medical imaging that automatically learns to focus on target structures of varying shapes and sizes. Models trained with AGs implicitly learn to suppress irrelevant regions in an input image while highlighting salient features useful for a specific task. This enables us to eliminate the necessity of using explicit external tissue/organ localisation modules of cascaded convolutional neural networks (CNNs). AGs can be easily integrated into standard CNN architectures such as the U-Net model with minimal computational overhead while increasing the model sensitivity and prediction accuracy. The proposed Attention U-Net architecture is evaluated on two large CT abdominal datasets for multi-class image segmentation. Experimental results show that AGs consistently improve the prediction performance of U-Net across different datasets and training sizes while preserving computational efficiency. The code for the proposed architecture is publicly available.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Dacheng Tao
- Fahad Shahbaz Khan
- Ming-Hsuan Yang
- Yunhe Wang
- Alan L. Yuille
- Salman Hameed Khan
- Qi Tian
- Kai Han
In Computer Vision and Pattern Recognition
- Dacheng Tao
- Luc Van Gool
- Ming-Hsuan Yang
- Ling Shao
- Alan L. Yuille
- Xiaogang Wang
- Fahad Shahbaz Khan
- Chunhua Shen
In Radiology, Nuclear Medicine and Imaging
- Radu Marculescu
- Md Mostafijur Rahman
- Vishal M. Patel
- Jeya Maria Jose Valanarasu
- Yunhe Wang
- Kai Han
- Suncheng Xiang
- Jiacheng Ruan
Citations in the corpus
- Scalable Neural Architecture Search for 3D Medical Image Segmentation (2019, methodology): “3D U-Net [1] has been popularly used for segmenting high-resolution 3D medical images (see [4, 8, 11]) since both semantic as well as spatial information can be efficiently exploited through skip connections between an encoder and a decoder.”
- EMCAD: Efficient Multi-Scale Convolutional Attention Decoding for Medical Image Segmentation (2024, methodology): “Attention mechanisms [12, 17, 20, 41, 57] have also been integrated into these models to enhance feature maps and improve pixel-level classification.”
- VM-UNet: Vision Mamba UNet for Medical Image Segmentation (2024, neutral): “Attention-UNet [30] employs attention gates to control the importance of features, thereby enabling the model to focus more on target areas.”
- ResViT: Residual Vision Transformers for Multimodal Medical Image Synthesis (2021, methodology): “The baseline methods included convolutional models (task-specific models: pGAN [24], pix2pix [43], medSynth [32]; unified models: MM-GAN [41], pGANuni), attention-augmented convolutional models (A-UNet [50], SAGAN [85]), and transformer models (task-specific: TransUNet [51], PTNet [57]; unified: Tra”
- Inf-Net: Automatic COVID-19 Lung Infection Segmentation From CT Images (2020, neutral): “WITH 638 SLICES (285 NON-INFECTED AND 353 INFECTED SLICES). THE BEST TWO RESULTS ARE SHOWN IN RED ANDBLUEFONTS. Model Dice Sen. Spec. Prec. MAE U-Net [58] 0.308 0.678 0.836 0.265 0.214 Attention-UNet [76] 0.466 0.723 0.930 0.390 0.095 Gated-UNet [77] 0.447 0.674 0.956 0.375 0.066 Dense-UNet [78] 0.4”
- Attention mechanisms in computer vision: A survey (2021, background): “[19] proposed a simple and yet effective mechanism, the attention gate (AG), to focus on targeted regions while suppressing feature activations in irrelevant regions.”
- Learning with Limited Annotations: A Survey on Deep Semi-Supervised Learning for Medical Image Segmentation (2022, background): “skip connections [8], incorporating residual/dense convolution blocks [9], [10], attention mechanisms [11], [12], etc.”
- A Survey on Vision Transformer (2020, background): “PSANet [151], OCNet [139], DANet [38] and CFNet [147] are the first works that introduce the self-attention module to semantic segmentation task, which consider and augment the relation and similarity [146, 74, 46, 89, 130] between the contextual pixels.”
- Unsupervised Medical Image Translation With Adversarial Diffusion Models (2022, methodology): “AttGAN comprised two convolutional attention UNet generators and two patch discriminators [71].”
- MALUNet: A Multi-Attention and Light-weight UNet for Skin Lesion Segmentation (2022, methodology): “Due to its simplicity and scalability advantages, plenty of improved models are proposed, such as UNet++ [2], AttentionUNet [3], 3D-UNet [4], V-Net [5], UNet3+ [6] and so on.”
- TransFuse: Fusing Transformers and CNNs for Medical Image Segmentation (2021, methodology): “Finally, the multi-level fused feature maps are combined using gated skip-connection [12] and generate the segmentation.”