Deep Learning Based Cross-Modality Histological Brain Section Registration in Multiple Species Using Synthetic Images

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Large-scale brain mapping requires integrating histological imaging data into anatomical frameworks.
• Registration techniques align diverse imaging datasets to reference atlases.
• Deep learning methods for registration typically need large annotated datasets, which are often unavailable.

Purpose of the Study:

• To develop a framework for training convolutional neural networks for anatomical registration using simulated data.
• To demonstrate the applicability of this framework across different species (mouse and marmoset).
• To validate the accuracy of the developed method against existing techniques.

Main Methods:

• A convolutional neural network was trained using a framework based entirely on simulated histological imaging data.
• The framework was applied to datasets from both mouse and marmoset brains.
• Accuracy was validated using Dice and Hausdorff distance metrics for anatomical region alignment.

Main Results:

• The framework successfully trained a convolutional neural network for anatomical registration using only simulated data.
• The approach demonstrated cross-species applicability, performing accurately for both mouse and marmoset brain mapping.
• Validation showed comparable or improved accuracy in anatomical region registration compared to an alternative method.

Conclusions:

• Training deep learning models for brain anatomical registration is feasible using simulated data, overcoming the need for large annotated datasets.
• This simulated data-driven framework accelerates large-scale and high-throughput brain anatomy studies.
• The method offers a scalable solution for integrating histological imaging data into standardized anatomical reference frameworks.