AI Decodes Brain Activity to Predict What Mice See

Source: Pixx|Teufel/Pixabay

A new study shows how an artificial intelligence (AI) machine-learning algorithm can decode the brain activity of mice in order to predict what they see with 95% accuracy.

“Mapping behavioral actions to neural activity is a fundamental goal of neuroscience,” wrote study co-authors Steffen Schneider, Jin Hwa Lee, and Mackenzie Weygandt Mathis at the Brain Mind Institute & Neuro X Institute, École Polytechnique Fédérale de Lausanne (EPFL). “As our ability to record large neural and behavioral data increases, there is growing interest in modelling neural dynamics during adaptive behaviors to probe neural representations.”

The EPFL researchers sought out to use AI machine learning to discover neural dynamics between behavior and brain activity by using a nonlinear dimensionality reduction method.

“So roughly 10 years ago we were at the point where we could decode very simple shapes from the brains of animals or humans,” said Mathis, the study’s principal investigator and the Bertarelli Chair of Integrative Neuroscience at EPFL, in a video. “But now we’re at the point where we can decode literally movie frame which has never been done before.”

The scientists achieved this with math tweaks. They modified the loss function of an AI algorithm that performs contrastive learning so that it can use both discrete and continuous data streams. The new deep learning algorithm called CEBRA combines brain activity and behavioral data in order to learn features.

The study used synthetic spiking data for benchmarking, multicellular electrophysiological data from the hippocampus area of rats, electrophysiological recordings from the somatosensory cortex (S1) in a rhesus monkeys, and mouse visual cortex data from the Allen Institute that was produced via two-photon calcium imaging (2PCI), which is an in vivo method of recording neural activity in the intact brains.

In AI, contrastive learning is an emerging deep learning method for computer vision tasks that uses positive and negative contrasting samples against each other in order to learn common and differing attributes.

The researchers show that they were able to successfully decode what a mouse sees while watching a movie using the AI algorithm with greater than 95% accuracy. According to the EPFL scientists, their AI algorithm significantly outperformed naïve Bayes or k-nearest neighbors algorithm (kNN) baseline decoding methods.

“CEBRA of course doesn’t just have to be about decoding the brain state for movie watching, but actually can be generalized to completely other domains,” Mathis emphasized.

The new deep learning algorithm called CEBRA can be used to advance research in fields that examine complex systems with data that involves a time-series or joint brain activity and behavior such as neuroscience, brain-computer interfaces (BCIs) or brain-machine interfaces (BMIs), ethology (animal behavior), and even gene expression.

Copyright © 2023 Cami Rosso. All rights reserved.