Virtual mouse brain histology from multi-contrast MRI via deep learning

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Histology

Background:

• Magnetic Resonance Imaging (MRI) non-invasively maps brain structure and function.
• Directly linking MRI signals to specific cellular structures for histopathological inference remains a challenge.

Purpose of the Study:

• To develop a method for estimating histological staining intensity directly from MRI signals.
• To create detailed 3D maps of brain tissue components like axons and myelin.
• To optimize multi-contrast MRI acquisition protocols.

Main Methods:

• Utilized deep convolutional neural networks trained on co-registered multi-contrast MRI and histological data from mouse brains.
• Developed a voxel-wise estimation of histological staining intensity from MRI signals.
• Analyzed the contribution of different MRI contrasts within the neural networks.

Main Results:

• Generated 3D maps of axons and myelin with high fidelity to histological staining.
• Achieved enhanced sensitivity and specificity compared to conventional MRI markers.
• Demonstrated the ability to optimize multi-contrast MRI acquisition parameters.

Conclusions:

• Deep convolutional neural networks can translate MRI signals into virtual histology.
• The developed method offers a novel approach for neurobiologists to interpret MRI data.
• This work provides a foundation for validating new MRI techniques and enhancing understanding of brain micro-environments.