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

Large-scale brain mapping increasingly relies on integrating histological imaging datasets within standardized anatomical reference frameworks. To accomplish this, registration techniques are used to align imaging datasets between different modalities or to reference atlases. Deep learning approaches have emerged a fast and scalable framework for approaching this challenge, but such methods typically require large annotated datasets that are unavailable in this setting. To address this, we developed a framework for training a convolutional neural network for this task using entirely simulated data. We show that the same approach can be used for different species (mouse and marmoset), and we provide accuracy validation in terms of Dice and Hausdorff distance between anatomical regions in comparison to an alternative method. This approach has the potential to accelerate large or high throughput studies of brain anatomy.