Automated identification of Retinogeniculate Visual Pathway based on Multiscale Point Cloud Fusion Model Network...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• The Retinogeniculate Visual Pathway (RGVP) transmits visual signals from the retina to the lateral geniculate nucleus.
• Accurate RGVP identification is vital for visual cognition research and neurosurgical planning.
• Current automated methods often neglect inter-subject RGVP trajectory variability.

Purpose of the Study:

• To propose a novel deep learning network, MSPF-Net, for automated RGVP identification.
• To address the challenge of inter-subject RGVP trajectory variability in automated tractography.
• To improve the accuracy and robustness of RGVP identification using diffusion MRI.

Main Methods:

• Developed a Multiscale Point Cloud Fusion Network (MSPF-Net) integrating feature fusion, local, and global extraction modules.
• Utilized diffusion magnetic resonance image tractography for RGVP identification.
• Evaluated MSPF-Net on HCP and CHCP datasets, quantifying streamline pairs to validate variability mitigation.

Main Results:

• MSPF-Net achieved higher recognition accuracy compared to state-of-the-art methods.
• The model effectively mitigated the issue of inter-subject RGVP trajectory variability.
• Demonstrated generalizability and robustness across different datasets.

Conclusions:

• MSPF-Net offers a promising approach for accurate and fully automated RGVP identification.
• The study highlights the potential of deep learning in overcoming challenges in neuroimaging analysis.
• Addressing inter-subject variability is crucial for advancing automated tractography methods.