Efficient Synthesis of Neuronal Image Stack from 3D Morphology using the Tri-Plane Approach

In contemporary brain connectome research, neuronal reconstruction is a key task for revealing the functional connections and working mechanisms of neural networks. With the increasing demand for neuronal morphological data, automated methods have been explored. However, the advancement of deep learning technology is severely limited by the high computational costs in 3d image stack. To overcome these challenges, this article introduces a strategy using a tri-plane representation method to represent three-dimensional images. This method effectively encodes the features of neuronal images by projecting the three-dimensional images onto three mutually perpendicular two-dimensional planes, thereby reducing the computational resources required for data processing. Additionally, the tri-plane representation allows for the direct use of two-dimensional generation methods, providing good performance for automatic segmentation and reconstruction tasks. Also, we tested the effects of our generated images on various tasks, and the results indicate that our method effectively captured the image features.