Gossamer: Scaling Image Processing and Reconstruction to Whole Brains

Area of Science:

• Neuroscience
• Computer Science
• Bioinformatics

Background:

• Neuronal reconstruction is crucial for understanding brain function, connectomics, and pathology.
• Current methods are limited by scalability, memory usage, and topological accuracy, hindering scientific progress.

Purpose of the Study:

• To develop a more efficient and scalable method for neuronal reconstruction.
• To overcome the limitations of existing disk-bound, dense-access, and single-threaded approaches.

Main Methods:

• Abstracting vision tasks into ordered specializations of search to reduce memory requirements.
• Designing a data-intensive parallel solution optimized for neuronal shape, topology, and connectivity.
• Implementing an in-memory processing approach on a single server.

Main Results:

• Achieved a memory reduction of 4 orders of magnitude.
• Enabled processing of 1 mouse brain in-memory on a single server.
• Demonstrated 67x scale increase, 870x less memory usage, and 78% higher automated yield compared to the previous state-of-the-art (APP2).

Conclusions:

• The developed method significantly enhances the performance and scalability of neuronal reconstruction.
• This advancement facilitates more comprehensive studies of brain structure and function.
• The approach overcomes key I/O and load-balancing challenges in large-scale neuroimaging data processing.