An Anisotropic Cross-View Texture Transfer With Multi-Reference Non-Local Attention for CT Slice Interpolation.

Area of Science:

• Medical Imaging
• Deep Learning
• Image Reconstruction

Background:

• Computed Tomography (CT) is crucial for medical diagnosis.
• Anisotropic CT volumes with low inter-slice resolution hinder accurate diagnosis.
• Existing super-resolution methods inadequately address CT's anisotropic nature.

Purpose of the Study:

• To develop a novel deep learning approach for CT slice interpolation.
• To enhance the inter-slice resolution of anisotropic 3D CT volumes.
• To improve disease diagnosis by increasing CT image quality.

Main Methods:

• Proposed a cross-view texture transfer framework for CT slice interpolation.
• Utilized high-resolution in-plane texture details to guide low-resolution through-plane image reconstruction.
• Introduced a multi-reference non-local attention module for feature extraction.

Main Results:

• The proposed method significantly outperforms existing CT slice interpolation techniques.
• Demonstrated superior performance on public CT datasets, including a real-paired benchmark.
• Verified the effectiveness of the cross-view texture transfer approach.

Conclusions:

• The novel framework effectively addresses the anisotropic nature of 3D CT volumes.
• The method enhances through-plane resolution by transferring in-plane texture details.
• This approach offers improved CT image quality for better medical diagnosis.