Scientists are using artificial intelligence and exascale computing power to advance connectomics research, which aims to understand how individual neurons connect in the brain. As part of the Aurora Early Science Program, researchers from Argonne National Laboratory, Harvard University, and other institutions are working together to create detailed maps of brains composed neuron by neuron.
Thomas Uram, data sciences and workflows team lead at Argonne Leadership Computing Facility, and Nicola Ferrier, an Argonne computer scientist, are co-leading a large-scale connectomics project that uses the facility’s new Intel-Hewlett Packard Enterprise exascale system, Aurora.
The project leverages imaging, supercomputing, and artificial intelligence innovations to improve our understanding of how the brain’s neurons are arranged and connected. With access to ultra-high-resolution images of brain tissue, researchers can identify brain structure and function at the sub-cellular level, which could lead to breakthroughs in fields beyond neuroscience.
Reconstructing Brains with Exascale Computing: Advancing Connectomics Research
The human brain is an enormously complex and not well understood organ, comprising 80 billion neurons each connected to as many as 10,000 other neurons. To advance our understanding of the brain’s structure and function at the sub-micrometer level, researchers are leveraging exascale computing to reconstruct three-dimensional (3D) models of neurons from high-resolution electron microscopy images.
High-Throughput Electron Microscopy and Image Alignment
The connectomics research team is working with Harvard University researchers who have pioneered fast parallel electron microscopy. The tissue samples require significant preparation for connectomic analysis, involving thin slicing and imaging on the electron microscope. Each slice is sectioned into tiles, which are then matched up to ensure their features correspond across all tiles in the section, a process referred to as stitching.
To achieve accurate tracing of fine structure, the researchers employ template- and feature-matching techniques for coarse and fine-grained alignment using the Finite-Element Assisted Brain Assembly System (FEABAS) application. This approach produces optimal linear and local non-linear image transformations to align the 2D image content between sections with a high degree of precision.
Tracing Neurons with Machine Learning
After stitching and alignment are complete, the researchers use various artificial intelligence (AI) methods to speed up data processing and analysis. They employ machine learning to find and trace objects – neurons, in effect – within the stack of images they’ve built. A convolutional neural network model trained to identify neuron bodies and membranes reconstructs the 3D shapes of neurons.
The Flood Filling Network code, developed at Google and adapted to run on Argonne systems, traces individual neurons over long distances, enabling analysis at the synaptic level. Deep learning models for connectomic reconstruction have been trained on Aurora using as many as 512 nodes, demonstrating performance increases of up to 40 percent throughout the project’s lifetime.
Scalability and Performance on Exascale Machines
Collaborating with Intel, the researchers have been working to run their model in a variety of configurations on Aurora and other Argonne systems. This work has been highly productive, and Intel has been very helpful in learning how to run the model efficiently, both for training the model and using the trained model for segmentation.
Once the team has produced a trained model, they use it to produce a segmentation of the larger volume of brain sample tissue. The larger volume vastly exceeds the scale of the training data, making segmentation the task for which the computing power of Aurora is most necessary – physically it would be on the order of a cubic millimeter of tissue, which at the desired imaging resolution represents approximately a petabyte of data.
Reconstructions with these models have been run on Aurora using as many as 1,024 nodes (with multiple inference processes for each graphics processing unit) to produce a segmentation of a teravoxel of data. Projecting from these runs to the full machine, the researchers anticipate soon being able to segment a petavoxel dataset within a few days on Aurora.
Future Directions and Applications
As imaging technology advances, computing will need to achieve high performance on post-exascale machines to avoid becoming a bottleneck. The techniques developed to study neural structure have helped ensure that computing would scale from cubic millimeters of brain tissue at the start, to a cubic centimeter of mouse brain beyond that, and to larger volumes of human brain tissue in the future.
The applications of this research are far-reaching, with potential implications for our understanding of neurological disorders, development of new treatments, and advancements in artificial intelligence. The collaboration between researchers from Harvard University and Argonne National Laboratory demonstrates the power of interdisciplinary approaches to tackling complex scientific challenges.
External Link: Click Here For More
