Presentation time: Sunday Dec 8th at 12:00PM
Poster number: P54/B55
The technology presented in this poster is a part of our latest software. Request a demo to try it out: https://www.aivia-software.com/demo
Abstract
The human brain is composed of approximately 100 billion neurons that collaborate to interpret our senses and control our thoughts and actions. Neuron type, localization and connectivity all contribute to the complex processes required for every organism with a nervous system to function. Thus, systematically reconstructing the 3D morphology of neurons is essential to further advance our understanding of how the brain functions. In the last five years, the BigNeuron project [1] has ran several comparative tests using 30 distinct algorithms for automated 3D neuron reconstruction. The tests were primarily ran using small data sets containing a single neuron or a few sparsely distributed neurons. While these tests were useful at showcasing the pros and cons of each algorithm, many real‐world challenges remain when attempting to reconstruct neurons in large and densely populated data sets. We have developed an enhanced voxel scooping algorithm and deployed it in four distinct modes: 1) fully automatic with soma detection, 2) fully automatic without soma detection, 3) semi‐automatic path prediction and 4) semi‐automatic point‐to‐point. The last two modes can be used in parallel. To further facilitate the creation of neurons using the semi‐automatic modes we have created a dynamic camera mode which follows new traces as they are validated, and we have added an auto‐clipping mode which gives users the option to render the data in the region the user is working on. We validated our novel approach on a large data set (1200x1200x2613) depicting a mouse brain section which had been optically cleared using PEGASOS [2] and imaged with a Leica SP8 confocal microscope or the Cleared Tissue LightSheet (CTLS) system [3]. Initial tests show that the fully automated version takes 48 mins to complete producing 395 neurons, 815 dendrites, and 325,541 microns in total path length. Thus, the automatic approach can trace approximately 400 mm per hour. The same data set would take an estimated 12 workdays (assuming each day a human expert would manually trace neurons for an average of 5 hours). The estimation is based on 5 hours of human manual tracing which resulted in 33 completed neurons (24,156 microns in total path length), representing a throughput of 4.8 mm per hour. In comparison, when using the semi‐automated modes mentioned above, we could trace at a pace of 46 mm per hour. We demonstrated that the automated method was 8.7x faster than the semi‐automated method. Moreover, the automated and semi‐automated modes are 83x and 9.5x faster in comparison to manual tracing, respectively. We are now testing this approach on even bigger and denser optically cleared data sets imaged with the CTLS.
Authors
B. Graff, H. Sasaki, S. McElroy, Y. Yi, C. Chou Huang, H. Zhao, S. J. Lee, L. A. G. Lucas
Y. Yi and H. Zhao are a part of the Department of Restorative Sciences, School of Dentistry, at the Texas A&M University, Dallas, TX
Other authors are part of DRVision Technologies LLC, Bellevue, WA
References
Yang J., et al. FMST: an automatic neuron tracing method based on fast marching and minimum spanning tree. Neuroinformatics. Apr 2017. 17(2):185-196.
Dian J., et al. Tissue clearing of both hard and soft tissue organs with the PEGASOS method. Cell Research. Aug 2019. 28:803-818.
Wang D., et al. Tiling light sheet selective plane illumination microscopy using discontinuous light sheets. Opt. Express. Nov 2019. 27(23):34472-34483.