Diffusion MRI super-resolution reconstruction via sub-pixel convolution generative adversarial network.

Area of Science:

• Medical Imaging
• Neuroscience
• Computer Vision

Background:

• Diffusion magnetic resonance imaging (dMRI) is crucial for visualizing neural pathways.
• Long acquisition times in dMRI limit its clinical applicability and patient comfort.
• Current reconstruction methods struggle with high-dimensional data and may overlook critical features.

Purpose of the Study:

• To develop a super-resolution reconstruction network for dMRI to shorten sampling times.
• To improve the accuracy and efficiency of dMRI data reconstruction.
• To address limitations in traditional loss functions for high-dimensional dMRI data.

Main Methods:

• A novel dMRI super-resolution reconstruction network utilizing 3D convolution kernels.
• Integration of adversarial learning to enhance reconstruction fidelity.
• Incorporation of an attention mechanism to focus on important feature maps.

Main Results:

• The proposed network demonstrated superior performance in peak signal-to-noise ratio and structural similarity compared to traditional methods.
• Orientation distribution function (ODF) visualization showed improved accuracy.
• Quantitative and qualitative results confirmed the effectiveness of the proposed approach.

Conclusions:

• The developed dMRI super-resolution network effectively reduces sampling time while maintaining high data quality.
• Adversarial learning and attention mechanisms are feasible and beneficial for dMRI reconstruction.
• This approach offers a promising solution for faster and more accurate dMRI acquisition.