Explore the innovative open-source device designed to disentangle neural connections within the human brain.
In late 2023, the first drug with potential to slow the progression of Alzheimer's disease was approved by the U.S. Federal Drug Administration. This breakthrough came as a much-needed step towards tackling one of the many neurological disorders that affect an eighth of the global population. Yet, there's still a long journey ahead to fully comprehend these diseases.
According to Lars Gjesteby, a technical staff member and algorithm developer from MIT Lincoln Laboratory's Human Health and Performance Systems Group, understanding the intricacies of the human brain at a cellular level is a significant challenge in neuroscience. High-resolution, networked brain atlases can aid the quest for knowledge by pinpointing differences between healthy and diseased brains. However, the lack of efficient tools to visualize and process large brain imaging datasets has slowed progress.
Gjesteby is working on a project to construct the Neuron Tracing and Active Learning Environment (NeuroTrALE), a software pipeline that combines machine learning, supercomputing, and user-friendly interfaces to combat this challenge. NeuroTrALE automates most data processing, presenting the output in an interactive interface, where researchers can edit and manipulate the data to mark, filter, and search for specific patterns.
One of NeuroTrALE's standout features is its use of active learning. Trained algorithms automatically label incoming data based on existing brain imaging data. However, unfamiliar data can lead to errors. Active learning empowers users to manually correct these mistakes, instructing the algorithm to improve its accuracy the next time it encounters similar data. This synergy between automation and manual labeling ensures precise data processing with a minimal burden on the user.
Michael Snyder, from the laboratory's Homeland Decision Support Systems Group, likens the process to untangling a ball of yarn. If two yarn strands cross, the ocular uncertainty arises as to whether one is making a 90-degree bend while the other takes a straight path. With NeuroTrALE's active learning, users trace yarn strands just a few times, teaching the algorithm to follow the correct path moving forward. Without NeuroTrALE, users would need to manually trace each yarn strand within the complex human brain repeatedly.
Since NeuroTrALE streamlines labeling, it enables researchers to analyze more data more quickly. Additionally, the axon tracing algorithms harness parallel computing, dividing computations across multiple GPUs for faster, scalable processing. The team demonstrated a 90% decrease in computing time required to process 32 gigabytes of data using NeuroTrALE compared to conventional AI methods.
Furthermore, a substantial increase in data size does not result in an equivalent growth in processing time using NeuroTrALE. For example, a 10,000% increase in dataset size resulted in only a 9% and a 22% increase in total data processing time, using two different central processing units.
Benjamin Roop, one of the project's algorithm developers, notes that manually labeling the estimated 86 billion neurons making 100 trillion connections in the human brain is impractical, given that it would take lifetimes to complete this task. NeuroTrALE has the potential to automate connectome creation for multiple individuals, enabling large-scale research on brain disease.
Originally formed as an internally funded collaboration between Lincoln Laboratory and Professor Kwanghun Chung's laboratory on MIT campus, NeuroTrALE's development aimed to help the Chung Lab analyze and extract useful information from their vast brain imaging data flowing into the MIT SuperCloud – a supercomputer managed by Lincoln Laboratory to support MIT research.
In 2022, the Chung Lab started producing results using NeuroTrALE. In one study they published in Science, they used NeuroTrALE to estimate prefrontal cortex cell density in relation to Alzheimer's disease, showing that Alzheimer's-affected brains had lower cell density in specific regions compared to healthy ones. The same team also traced where in the brain harmful neurofibers tend to become entangled in Alzheimer's-affected brain tissue.
NeuroTrALE's capabilities are currently being integrated with Google's Neuroglancer program, an open-source, web-based neuroscience data viewer. This integration allows for real-time collaboration among researchers and a dynamic, simultaneous review of annotated data. Users can create and edit various shapes like polygons, points, and lines to facilitate annotation tasks, as well as customize color displays for various annotations to distinguish neurons in dense regions.
Adam Michaleas, a high-performance computing engineer from Lincoln Laboratory's Artificial Intelligence Technology Group, emphasizes NeuroTrALE's flexibility and adaptability. It can be easily deployed on standalone, virtual, cloud, and high-performance computing environments via containers. Additionally, NeuroTrALE significantly enhances the end-user experience by providing real-time collaboration capabilities among the neuroscience community.
The NeuroTrALE project's goal is to make it a fully open-source tool accessible to any researcher. As Gjesteby suggests, this open-source approach is crucial for reaching the ultimate goal of mapping the entire human brain for research and eventual drug development by fostering data and algorithm sharing within the neuroscience community.
You can find the codebases for NeuroTrALE's axon tracing, data management, and interactive user interface publicly available via open-source licenses. For more information on using NeuroTrALE, contact Lars Gjesteby.
- The approval of the first drug to slow Alzheimer's disease progression could be a pivotal step in addressing neuroscience's quest to understand various medical-conditions like Alzheimer's disease and other neurological disorders.
- High-resolution brain atlases can aid in locating differences between healthy and diseased brains, contributing to the solution of many ongoing neurological concerns.
- The Neuron Tracing and Active Learning Environment (NeuroTrALE) is a project aiming to accelerate research in neuroscience with the help of machine learning, supercomputing, and user-friendly interfaces.
- NeuroTrALE's active learning feature enables users to manually correct mistakes in the labeling of brain imaging data, improving the accuracy of the algorithm.
- NeuroTrALE's potential application lies in the automation of connectome creation for multiple individuals, which could speed up large-scale research on brain disease.
- The NeuroTrALE project, initially funded internally, intended to support the analysis and data processing of the Chung Lab's vast brain imaging data for research purposes.
- NeuroTrALE has demonstrated a significant decrease in computing time for processing large brain imaging datasets compared to conventional AI methods.
- NeuroTrALE's capabilities are being integrated with Google's Neuroglancer program, allowing for collaboration among researchers in real-time and enhancing the end-user experience.
- To achieve the ultimate goal of mapping the entire human brain, NeuroTrALE's developers aim to make it an open-source tool accessible to any researcher, promoting data and algorithm sharing within the neuroscience community.
- The codebases for NeuroTrALE's axon tracing, data management, and interactive user interface are publicly available via open-source licenses for interested researchers to access and utilize.
