SCANeXt: Enhancing 3D medical image segmentation with dual attention network and depth-wise convolution

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Convolutional Neural Networks (CNNs) excel at local feature extraction but struggle with global representations in 3D medical images.
• Vision transformers effectively capture long-range dependencies and global context via self-attention mechanisms.
• Existing methods face limitations in simultaneously addressing local and global feature extraction for 3D medical image segmentation.

Purpose of the Study:

• To introduce SCANeXt, a novel hybrid approach for 3D medical image segmentation.
• To leverage the strengths of both convolutional and transformer-based architectures for improved performance.
• To enhance representation learning by integrating spatial and channel attention with advanced convolutional blocks.

Main Methods:

• Proposed SCANeXt model combines dual attention (Spatial and Channel Attention) with ConvNeXt architecture.
• Introduced a novel self-attention mechanism focusing on spatial and channel relationships across feature dimensions.
• Incorporated a depth-wise convolution block inspired by ConvNeXt for multiscale feature extraction post-attention.

Main Results:

• SCANeXt demonstrated superior accuracy in 3D medical image segmentation across Synapse, BraTS, and ACDC datasets.
• Achieved a state-of-the-art Dice Similarity Score of 95.18% on the ACDC dataset.
• Significantly outperformed existing 3D medical image segmentation methods in evaluations.

Conclusions:

• SCANeXt effectively enhances representation learning for 3D medical images by combining dual attention and ConvNeXt.
• The proposed method offers a powerful new approach for accurate 3D medical image segmentation.
• SCANeXt sets a new benchmark for performance in medical image analysis tasks.