A fused attention-based hybrid model for semi-supervised medical image segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Extensive data annotation poses a challenge for medical image segmentation.
• Semi-supervised learning offers a solution by utilizing both labeled and unlabeled data.
• Existing methods struggle with complex local structures and global context, leading to inconsistent segmentation.

Purpose of the Study:

• To propose a novel fused transformer-based hybrid architecture for semi-supervised medical image segmentation.
• To enhance the accurate capture of complex local structures and global contextual information.
• To improve segmentation consistency and reduce the need for extensive data annotation.

Main Methods:

• Developed a hybrid architecture with a parallel backbone integrating Deformation Convolution Blocks (DCB) and Fused Transformer Blocks (FTB).
• Incorporated a Ghost Layer Perceptron (GLP) for computational efficiency within the transformer.
• Utilized consistency loss and unsupervised contrastive learning for robust feature discrimination on unlabeled data.

Main Results:

• The proposed model achieved comparable or superior accuracy to state-of-the-art methods across four medical imaging datasets.
• Demonstrated substantial reductions in model parameters and computational cost.
• Showcased improved generalization across different medical imaging modalities.

Conclusions:

• The fused transformer-based hybrid architecture offers a practical and efficient solution for semi-supervised medical image segmentation.
• The model effectively addresses limitations in capturing local and global information, leading to more consistent segmentation.
• This approach holds significant potential for real-world clinical applications due to its performance and efficiency.