Microscopy-informed structural connectivity mapping in the in vivo human brain via domain adaptation

Area of Science:

• Neuroscience
• Biomedical Imaging
• Artificial Intelligence

Background:

• Mapping human brain connectivity is challenging.
• Multimodal datasets link macroscopic imaging (diffusion MRI) with microstructural detail (microscopy).
• Existing tools struggle to leverage multimodal data for improved in vivo human connectivity estimates.

Purpose of the Study:

• Develop a deep learning model to predict high-resolution microscopy-informed fiber orientations from diffusion MRI.
• Bridge the gap between animal model data and in vivo human imaging.
• Enhance the biological accuracy of human brain connectivity mapping.

Main Methods:

• Trained a deep learning model on a macaque dataset integrating in vivo MRI, postmortem MRI, and whole-brain microscopy.
• Used microscopy-derived 3D fiber orientation maps as training targets.
• Applied domain adaptation for macaque postmortem to in vivo translation and across species (macaque to human).

Main Results:

• The model accurately predicts microscale-informed fiber architecture from diffusion MRI.
• Achieved generalization from macaque to in vivo human diffusion MRI with minimal requirements.
• Microscopy-informed tractography improved delineation of white matter and cortical-subcortical pathways in humans.

Conclusions:

• Established a general framework for transferring microstructural information from microscopy to non-invasive imaging.
• Enables biologically informed mapping of brain connectivity.
• The method enhances in vivo human diffusion MRI tractography without requiring microscopy at inference.