SwRI develops computational method to track brain cell development

A team led by Southwest Research Institute (SwRI) developed a computational method to improve techniques to track brain cell development over time. By studying neurons and their dynamics in brain networks, the researchers hope to gain knowledge to identify drug treatments for people suffering from neurological disorders.

SwRI collaborated with The University of Texas at San Antoino (UTSA) to track individual brain cells undergoing neurogenesis, the process where new neurons grow and connect to neuronal networks. UTSA researchers developed novel methods to grow human stem cells into neuronal networks; this included differentiating the cells into regions of human brain tissue that regulate sleep, temperature and mood. As the neuronal networks developed, their activity was captured by confocal microscopy. SwRI used the visual data to train algorithms to track the neurons. The project was funded by a $200,000 award from the San Antonio Medical Foundation.

"The research results are a significant step toward automatically classifying the health of growing neuronal networks," said Dr. Courtney Rouse, an SwRI computer scientist who led the project. "The algorithm can help study various neurological diseases and assist in the development and testing of associated therapies."

Conventional neuron-tracking methods frequently rely on images of labeled cells in fixed tissue, a process that can obscure cellular dynamics. The SwRI-led analysis overcomes technological gaps in prior computational methods by capturing unlabeled cells and fine structures in dense, live cultures. This allows for complex analysis over longer time spans.

Dr. Amina Qutub's UTSA lab developed experimental models of neuronal development and will use SwRI's cell-tracking methods to inform AI models in a follow-up project. Each video of the neuronal network cultures consists of timestamped images with hundreds to thousands of neurons per image. SwRI trained a U-Net machine learning algorithm to detect the shape and location of individual neurons on the images.

Neurons have root-like structures that connect to other neurons to form neuronal networks, which send and receive signals. SwRI trained the algorithm to recognize two of the main parts of a neuron - the soma, which contains the nucleus, and dendrites, which branch off the soma.

The team focused on tracking the somas because each neuron has one soma, while dendrite numbers vary. The algorithm applied a tracking number to a key point on each soma to match neurons in consecutive images based on proximity within a pixel.

When detecting somas, the algorithm achieved a 93.8 percent precision rate and 99.1 percent recall rate. When detecting dendrites, the smaller branch-like structures, the algorithm achieved an 88.3 percent precision rate and an 80.9 percent recall rate. Overall, the algorithm had an 85.7 percent probability of correctly tracking a single neuron in consecutive images.

"This project will help us understand fundamentals of how brain cells develop and communicate," said Qutub, the UTSA professor of biomedical engineering and assistant director of strategic partnerships of the MATRIX AI Consortium. "Artificial intelligence methods from this project are also helping us develop a screening tool to accelerate the discovery of biomedicine for brain health and neurological disorders."

The team plans future research to identify connections between soma and dendrites, test neurons exposed to various environmental stresses - such as low oxygen or circadian disruption - and correlate neuron electrical activity to tissue health.

We are excited that this collaboration is helping close technological gaps in computational neural research. Our algorithm accurately tracked individual soma across timeframes, a fundamental step toward classifying the health of a developing neuronal network."

Hakima Ibaroudene, manager of SwRI's Bioinformatics Section

SwRI's research builds upon its expertise in computational biomedicine. The Institute applies artificial intelligence and data analytics to detect and assess health conditions and aid medical experts in making complex decisions from disparate data sources.