Abstract:Convolutional neural networks (CNNs) are the most widely used deep learning-based architecture for medical image analysis. However, the limited receptive field of existing CNN-based methods can result in inaccurate segmentation, especially in cases of partial similarity or local environmental changes. To address this issue, transformer architectures have been introduced as an effective solution. However, self-attention mechanism used in the transformer considers all features through weighted averaging, which limits its ability to capture local dependencies. While this may not be a significant issue for internal regions of lesions, accurately delineating small lesion boundaries can be challenging. This is because local correlation is often more critical than distant correlation in medical image segmentation tasks. Furthermore, the precise segmentation of breast tumors in ultrasound images remains challenging due to unclear boundaries. In this study, we propose an angular-margin contrastive learning model that integrates information from gaussian transformer and CNN decoder. Our model uses a shallow CNN encoder, followed by a CNN decoder and a gaussian transformer decoder. Features learned from both branches are then fused for joint prediction. To maintain the U-shape structure between the decoder and encoder, each decoding feature map is skip-connected to its corresponding encoder feature. We introduce a gaussian weighting in traditional multi-head attention to enhance the local correlation modeling ability of self-attention. Additionally, we propose an angular-margin contrastive loss to encourage the model to focus on boundary features and improve segmentation performance. The proposed method is validated on an ABUS dataset, and experimental results demonstrate its state-of-the-art performance, with a Dice similarity coefficient and HD95 of 0.813 and 2.207 mm, respectively. It is worth noting that our method considers not only global semantics but also local correlation and boundary features, thereby improving segmentation accuracy.

Boundary-guided feature integration network with hierarchical transformer for medical image segmentation

Hybrid transformer-CNN with boundary-awareness network for 3D medical image segmentation

Low-Contrast Medical Image Segmentation via Transformer and Boundary Perception

BMCS-Net: A Bi-directional multi-scale cascaded segmentation network based on transformer-guided feature Aggregation for medical images

HCA-former: Hybrid Convolution Attention Transformer for 3D Medical Image Segmentation

Integrating prior knowledge into a bibranch pyramid network for medical image segmentation

TSCA-Net: Transformer based spatial-channel attention segmentation network for medical images

HD-Former: A hierarchical dependency Transformer for medical image segmentation

Dual encoder network with transformer-CNN for multi-organ segmentation

Gaussian Transformer and CNN Segmentation Method Based on Contrastive Learning of Boundary

CTC-Net: A Novel Coupled Feature-Enhanced Transformer and Inverted Convolution Network for Medical Image Segmentation

FDB-Net: Fusion Double Branch Network Combining CNN and Transformer for Medical Image Segmentation

A Feature Fusion Module Based on Complementary Attention for Medical Image Segmentation

DECTNet: Dual Encoder Network combined convolution and Transformer architecture for medical image segmentation

Sub-pixel multi-scale fusion network for medical image segmentation

Medical Image Segmentation Using Boundary-Enhanced Guided Packet Rotation Dual Attention Decoder Network

UCTNet: Uncertainty-guided CNN-Transformer hybrid networks for medical image segmentation

Feature interaction network based on hierarchical decoupled convolution for 3D medical image segmentation

TEC-Net: Vision Transformer Embrace Convolutional Neural Networks for Medical Image Segmentation

HCT-Unet: multi-target medical image segmentation via a hybrid CNN-transformer Unet incorporating multi-axis gated multi-layer perceptron

Boundary-aware context neural network for medical image segmentation