Summary: A new computational model reveals how information might flow between deep brain networks and how clusters of neurons can spontaneously self-optimize over time.
Source: USC
Researchers in the Cyber-Physical Systems Group at the USC Viterbi School of Engineering, together with collaborators at the University of Illinois at Urbana-Champaign, have developed a computational model describing how information can move between neural networks deep in the brain and how neuronal clusters reorganize and self-optimize as they evolve.
Described in the paper “Network Science Characteristics of Brain-Derived Neuronal Cultures Deciphered From Quantitative Phase Imaging Data,” this work is reported to be the first to observe self-optimization in in vitro neuronal networks and challenges assumptions made by several classical models used in neuroscience.
The study’s outcomes point to new directions in biologically inspired artificial intelligence, offer methods that could improve early detection of brain tumors, and may suggest approaches relevant to neurodegenerative conditions such as Parkinson’s disease.
The team mapped the structure and temporal evolution of neuronal networks derived from mouse and rat brain tissue to identify connectivity patterns and information flow. Corresponding author Paul Bogdan, an associate professor of Electrical and Computer Engineering, framed the results by likening brain decision-making to a probabilistic process: rather than storing every possible option, the brain appears to form dynamic models of uncertainty by leveraging rich interconnections among neurons.
These rapidly adjusting clusters allow the brain to estimate degrees of uncertainty, form rough probabilistic descriptions, and discount unlikely conditions. “We observed that the brain’s networks have an extraordinary capacity to minimize latency, maximize throughput and improve robustness — all in a distributed way without a central controller,” said Bogdan, who holds the Jack Munushian Early Career Chair in the Ming Hsieh Department of Electrical Engineering. “Neuronal groups negotiate their connections to quickly enhance network performance, but the rules that govern those connections were previously unclear.”
To capture these dynamics, the researchers combined multifractal analysis with quantitative phase imaging (QPI), an advanced, label-free microscopy technique developed by Gabriel Popescu at the University of Illinois. Using high-resolution QPI data, they reconstructed neuronal cultures at both microscale (individual neurons) and mesoscale (clusters of neuronal culture networks), then evaluated network metrics such as degree, closeness, and betweenness centrality, node-degree distributions, clustering coefficients, and multifractal spectra to characterize evolving topology and information flow.
HEALTH APPLICATIONS
One promising application of this work is earlier, less invasive detection of brain abnormalities. By establishing detailed topological baselines for healthy neuronal connectivity and dynamics during cognitive tasks, clinicians and researchers could detect subtle departures from normal network behavior — changes that might indicate the early stages of tumors or other pathologies before they become visible on conventional scans.
“Cancer often spreads in small clusters of cells that remain undetectable by techniques like fMRI until later stages,” said co-author Chenzhong Yin, a Ph.D. student in the Cyber-Physical Systems Group. “By training AI on this type of microscopic, dynamic connectivity data, we can detect and potentially predict pathological changes earlier by spotting abnormal interactions between neurons.”
The researchers are refining their imaging and modeling tools with the goal of monitoring complex neuronal networks in living brains. Live imaging could reveal how networks grow and shrink, how memory and cognition develop, whether drugs restore or disrupt network connectivity, and how learning reorganizes neuronal topology. Those observations could directly inform therapeutic strategies for disorders such as Parkinson’s disease, where communication between hemispheres and specific circuits degrades over time.
USE FOR ARTIFICIAL INTELLIGENCE
Beyond medical applications, these findings could influence the design of more adaptable artificial neural networks. Current deep learning systems suffer from catastrophic forgetting: when trained sequentially on multiple tasks, they often overwrite earlier knowledge as they adapt to new tasks. Biological brains, by contrast, can learn new skills without wholly losing prior abilities.
Bogdan and colleagues suggest that understanding how biological networks self-optimize — balancing latency, throughput and robustness across decentralized connections — could enable AI systems to support continual learning and inductive inference without proportionally scaling up network size. Replicating those dynamic topological rules in artificial architectures may reduce catastrophic forgetting and improve multitask performance.
Funding: The research team includes Chenzhong Yin, Xiongye Xiao, Valeriu Balaban, Mikhail E. Kandel, Young Jae Lee, Gabriel Popescu, and Paul Bogdan. The work was supported by the National Science Foundation (NSF) and the Defense Advanced Research Projects Agency (DARPA).
About this neuroscience research news
Source: USC
Contact: Amy Liberson – USC
Image: The image is in the public domain.
Original Research: “Network science characteristics of brain-derived neuronal cultures deciphered from quantitative phase imaging data” by Chenzhong Yin et al., published in Scientific Reports. The study is open access.
Abstract
Network science characteristics of brain-derived neuronal cultures deciphered from quantitative phase imaging data
Understanding how neurons form and dissolve connections to support communication in neuronal cultures sheds light on the emergence of learning, cognition and creative behavior. Building on prior observations of self-organizing criticality in neuronal cultures, this study demonstrates that in vitro brain-derived neuronal networks also exhibit a self-optimization process. Using label-free quantitative phase microscopy and multiscale analysis, the authors reconstructed microscale and mesoscale network topologies from mouse-derived cultures and quantified node importance and information flow. By examining centrality measures, degree distributions, clustering coefficients, and multifractal spectra, the study shows that these networks evolve toward configurations that optimize information flow, robustness, and self-organization. The topological signatures observed differ from classical complex network models, with implications for modeling neuronal systems and designing biologically inspired artificial intelligence.