Skip to main content
Network of neurons
March 11, 2024

Reverse engineering the brain’s connections from fuzzy data

Written By: Tom Ziemer


By better understanding the connections among different parts of the brain—from individual neurons all the way up to whole regions—neuroscientists could glean insights about how the brain works and what changes occur in various diseases.

But mapping those connections is extraordinarily difficult; devices like electrode arrays that are capable of sensing neural activity are invasive and can’t cover a large enough area of the brain. Using software analysis of noninvasive imaging to deduce and predict neural connectivity is an intriguing possibility, but technologies such as MRI can’t capture data at the sheer speed of neural activity that occurs in single milliseconds.

University of Wisconsin-Madison neuroengineers have developed and tested a workaround. Their method reverse engineers networks of neural connectivity while leveraging data that lacks the brain’s actual level of granularity in timescale. The researchers, led by Aviad Hai, an assistant professor of biomedical engineering, detail their work in a paper in the Journal of Neuroscience Methods.

Adam Vareberg
Adam Vareberg

“One of the huge goals of neuroscience is to be able to map the brain so that we understand how neurons connect to each other, how populations of neurons connect to other populations of neurons, and even more broadly, how regions of the brain interact,” says Adam Vareberg, a PhD student in Hai’s lab and first author of the paper. “And then, understanding how in different disease states that connectivity can change at every level. And if we can understand a healthy model versus a diseased model, then we can intervene to try to change that state.”

The researchers modified a machine learning algorithm called “the perceptron” to infer the connectivity of a network of neurons from data with low temporal resolution (longer timescales, which fail to discretely capture each electrical impulse, or “action potential,” from neurons). The group also used its method to predict subsequent neuron spikes in a sequence.

“Any neural readout that has a lower resolution, temporally, than what you would hope for could use this to then decipher network connectivity and predict activity from that suboptimal data,” says Vareberg.

He says the group’s results provide a roadmap for others, particularly academic researchers, to build upon. It’s another tool in the Hai lab’s arsenal for yielding a better view of the brain.

“We’re developing devices and methods for interfacing with the brain on the micro and nanoscale, and many of our approaches rely on MRI or other imaging modalities,” says Vareberg, who’s working on developing wireless stimulation devices for therapeutic effect. “So if we can combine these novel sensors with something that is able to supplement the readouts that they get, that’s the hope. We’re enhancing the hardware with the software.”

Aviad Hai is a Vilas Early Career Assistant Professor, part of the Wisconsin Institute for Translational Neuroengineering and an affiliate of the Department of Electrical and Computer Engineering. Additional UW-Madison authors include Ilhan Bok, a PhD student in electrical and computer engineering; Jenna Eizadi (BSBME ’21), a former undergraduate researcher in the Hai lab; and Xiaoxuan Ren, a PhD student in electrical and computer engineering.

Funding for this research came from the National Institutes of Health (grants K01EB027184 and DP2NS122605), as well as the U.S. Office of Naval Research (award numbers N00014-23-1-2006 and N00014-22-1-2371) and the Wisconsin Alumni Research Foundation (WARF).

Top image courtesy Pixabay.