A multi-dimensional lightweight attention-enhanced model for medical image segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical image segmentation is vital for analyzing lesions and anatomical structures.
• Current deep learning models, like Convolutional Neural Networks (CNNs), face limitations in modeling global dependencies due to fixed kernel sizes and restricted receptive fields.
• While Vision Transformers offer global perception, they often incur high computational costs, limiting their use in resource-constrained medical settings.

Purpose of the Study:

• To develop a multi-dimensional, lightweight, attention-enhanced model for medical image segmentation.
• To address the limitations of existing methods in capturing global dependencies and computational efficiency.
• To provide a practical and valuable tool for clinical medical imaging analysis.

Main Methods:

• The proposed model integrates omni-dimensional dynamic convolution and mask attention.
• Omni-dimensional dynamic convolution enables adaptive modeling across spatial, channel, and kernel-number dimensions, expanding the effective receptive field.
• Mask attention utilizes binary masks to suppress background noise and enhance focus on critical image boundaries, achieving lower computational cost than global self-attention.

Main Results:

• The model was evaluated on three public benchmark datasets for medical image segmentation.
• It demonstrated superior or comparable segmentation accuracy and inference efficiency against current mainstream methods.
• The results highlight the model's effectiveness in improving segmentation performance.

Conclusions:

• The developed model offers a computationally efficient and accurate solution for medical image segmentation.
• Its ability to expand the effective receptive field and focus on critical boundaries makes it highly applicable.
• The findings suggest significant potential for practical clinical adoption in medical imaging analysis.