Abstract:The understanding of long-range pixel–pixel dependencies plays a vital role in image segmentation. The use of a CNN plus an attention mechanism still has room for improvement, since existing transformer-based architectures require many thousands of annotated training samples to model long-range spatial dependencies. This paper presents a smooth attention branch (SAB), a novel architecture that simplifies the understanding of long-range pixel–pixel dependencies for biomedical image segmentation in small datasets. The SAB is essentially a modified attention operation that implements a subnetwork via reshaped feature maps instead of directly calculating a softmax value over the attention score for each input. The SAB fuses multilayer attentive feature maps to learn visual attention in multilevel features. We also introduce position blurring and inner cropping specifically for small-scale datasets to prevent overfitting. Furthermore, we redesign the skip pathway for the reduction of the semantic gap between every captured feature of the contracting and expansive path. We evaluate the architecture of U-Net with the SAB (SAB-Net) by comparing it with the original U-Net and widely used transformer-based models across multiple biomedical image segmentation tasks related to the Brain MRI, Heart MRI, Liver CT, Spleen CT, and Colonoscopy datasets. Our training set was made of random 100 images of the original training set, since our goal was to adopt attention mechanisms for biomedical image segmentation tasks with small-scale labeled data. An ablation study conducted on the brain MRI test set demonstrated that every proposed method achieved an improvement in biomedical image segmentation. Integrating the proposed methods helped the resulting models consistently achieve outstanding performance on the above five biomedical segmentation tasks. In particular, the proposed method with U-Net improved its segmentation performance over that of the original U-Net by 13.76% on the Brain MRI dataset. We proposed several novel methods to address the need for modeling long-range pixel–pixel dependencies in small-scale biomedical image segmentation. The experimental results illustrated that each method could improve the medical image segmentation accuracy to various degrees. Moreover, SAB-Net, which integrated all proposed methods, consistently achieved outstanding performance on the five biomedical segmentation tasks.

BCAU-Net: A Novel Architecture with Binary Channel Attention Module for MRI Brain Segmentation.

Split-Attention U-Net: A Fully Convolutional Network for Robust Multi-Label Segmentation from Brain MRI

BUA-Net: Boundary and uncertainty-aware attention network for lumbar multi-region magnetic resonance imaging segmentation

MRI Brain Tumour Segmentation Using Multiscale Attention U-Net

Attention Based Multi-Scale Nested Network for Biomedical Image Segmentation

DAUnet: A U-shaped network combining deep supervision and attention for brain tumor segmentation

GAU-Net: U-Net Based on Global Attention Mechanism for Brain Tumor Segmentation

DCSAU-Net: A deeper and more compact split-attention U-Net for medical image segmentation

Category guided attention network for brain tumor segmentation in MRI

Learning U-Net Based Multi-Scale Features in Encoding-Decoding for MR Image Brain Tissue Segmentation.

CAU-Net: A Deep Learning Method for Deep Gray Matter Nuclei Segmentation.

Attention U-Net Based on Bi-ConvLSTM and Its Optimization for Smart Healthcare

ALU-Net: an Attention-Based Mechanism Lightweight U-Net for Multimodal Brain Tumor Segmentation

HMRNet: High and Multi-Resolution Network With Bidirectional Feature Calibration for Brain Structure Segmentation in Radiotherapy

BCU-Net: Bridging ConvNeXt and U-Net for medical image segmentation

DA-ResUNet: a Novel Method for Brain Tumor Segmentation Based on U-Net with Residual Block and CBAM

Attention Mechanism Trained with Small Datasets for Biomedical Image Segmentation

Deep Convolutional Neural Network with a Multi-Scale Attention Feature Fusion Module for Segmentation of Multimodal Brain Tumor.

APRNet: A 3D Anisotropic Pyramidal Reversible Network with Multi-Modal Cross-Dimension Attention for Brain Tissue Segmentation in MR Images.

A Novel End-to-end Brain Tumor Segmentation Method Using Improved Fully Convolutional Networks.

ARGA-Unet: Advanced U-net Segmentation Model Using Residual Grouped Convolution and Attention Mechanism for Brain Tumor MRI Image Segmentation