CNNT-Net: a novel deep learning-based framework for estimating high-quality fiber orientation distributions from...

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Deep Learning

Background:

• Accurate reconstruction of fiber orientation distribution (FOD) images is crucial for non-invasive white matter tractogram analysis.
• Existing methods face challenges in accurately estimating complex fiber configurations.

Purpose of the Study:

• To introduce a novel deep learning-based network for high-quality FOD image reconstruction.
• To enhance the accuracy of FOD estimation in white matter tractography.

Main Methods:

• A deep learning network integrating convolutional neural networks (CNNs) and Transformers.
• Utilizes angular correlation coefficient (ACC) thresholding and a voxel selection module.
• Employs attention mechanisms within the Transformer to refine FOD estimation for complex fiber structures.

Main Results:

• Demonstrated promising qualitative and quantitative outcomes on the Human Connectome Project (HCP) dataset.
• Showcased the robustness of the proposed method in producing high-quality FOD images.
• Achieved enhanced FOD estimation accuracy, particularly for complex fiber configurations.

Conclusions:

• The developed deep learning network offers a viable solution for high-quality FOD imaging.
• Presents a reliable method for non-invasive white matter tractogram analysis.
• Enhances the clinical applicability of advanced neuroimaging techniques.