DeepSlice

Harry Carey and colleagues out of Simon McMullan’s lab at Macquarie University in collaboration with colleagues at the University of Oslo, OpenBench, and University of Sydney, published their open-source tool, DeepSlice, last September.

There is a great need for technology that registers brain slice images to coordinate references. Current techniques require many hours of human processing time, in-depth knowledge of programming and data-processing, and large, powerful computers that many do not have access to. DeepSlice solves these problems by automating the process of orienting histological slice imaging to reference atlases.

DeepSlice is a CNN trained on a large scale dataset of mouse histological data from the Allen Institute for Brain Science as well as the reference atlas, the Allen Common Coordinate Framework (CCF), which the dataset had been aligned to already. Performance of DeepSlice was improved by including synthetic images to the training dataset and post-processing for predictions. While the automation of this task has many benefits to use, there are a few downfalls. Mainly, there is a general trend of underperformance when tissue has obscured or low-contrast landmark anatomy, or in tissue with abnormal tissue/deformities. This, however, can be mitigated over time by adding more abnormal tissue to the training dataset. Overall, DeepSlice has shortened the amount of time needed to perform this task immensely — shortening what once took several hours into just a few seconds.

This research tool was created by your colleagues. Please acknowledge the Principal Investigator, cite the article in which the tool was described, and include an RRID in the Materials and Methods of your future publications. RRID:SCR_023854

Special thanks to Abby St. Jean, a neuroscience undergraduate at American University, for providing this project summary.