Speed of fully automated connectomics tracing at whole-brain scale

Determine the achievable segmentation throughput of fully automated neuron-and-synapse tracing software applied to whole-brain mammalian connectomics image data, even when executed on a dedicated high-performance computing center designed specifically for reconstructing a mouse or human brain.

Background

The paper highlights that computational processing, particularly automated segmentation and tracing, is a major bottleneck for mammalian connectomics due to the vast data volumes involved. While imaging technologies are advancing, fully automated tracing of electron microscopy data is not yet possible, and expansion microscopy combined with light-sheet fluorescence microscopy has not matured sufficiently to address these issues.

Within this context, the authors emphasize that the practical speed at which automated tracing software could segment data at whole-brain scale is unknown, even assuming the availability of a high-performance computing center dedicated to the task. Quantifying this throughput is essential for planning large-scale mouse and human connectome projects.

References

Additionally, it remains an open question as to the speed at which such automated tracing software could segment image data, even assuming construction of a powerful high-performance computing (HPC) center built with the singular goal of reconstructing a mouse or human brain.

— Comparative prospects of imaging methods for whole-brain mammalian connectomics  (2405.10488 - Collins et al., 2024) in An aside on computational constraints