A New Way to Decode Brain Signals

This piece, by Onno Berkan, was published on 03/14/25. The original text, by Zhang et al., was submitted to NeurIPS 2023.

This Columbia study introduces a new method for decoding and interpreting brain activity signals without relying on spike sorting. Their method breaks down invasive EEG signals to assign activation patterns to individual neurons. Traditional approaches require this, which is time-consuming and prone to errors. Instead, this new method directly analyzes key features of the brain signals, such as their location and amplitude, to understand how they relate to behavior.

The researchers developed a probabilistic model to learn the relationships between these neural signal features and behaviors, such as a mouse's actions on a decision-making task. They used complex math to group similar neural signals and track how these groups change their activity patterns based on behavior. This approach is particularly useful for analyzing recordings from modern high-density brain probes, which can simultaneously record from many cells in the brain. Crucially, the model works in real-time, which makes it highly applicable in experimental contexts.

The team tested their method extensively across different scenarios. They compared it to traditional spike-sorting approaches using datasets from different brain regions, different types of recording devices, and various animal behaviors. Their method consistently performed better than conventional approaches, especially in challenging recording conditions where traditional spike sorting is complex to perform accurately.

One key advantage of this new approach is that it can utilize more available neural information. Traditional methods often discard signals that do not meet strict quality criteria, potentially losing valuable information. The new method can work with all detected signals, leading to better behavioral predictions. The researchers demonstrated this across multiple brain regions, including the thalamus, hippocampus, visual cortex, and cerebellum.

The method proved particularly valuable in dense brain regions where traditional spike sorting is typically challenging, such as the cerebellum. It also worked well across different types of recording devices, including the standard Neuropixels probes and modified versions designed for different species.

As for future directions, the researchers suggest their method could be improved by incorporating more advanced machine learning techniques and better accounting for coordinated firing patterns between different groups of neurons. This work represents a significant step forward in neuroscience research, offering a more reliable way to decode brain activity without depending on error-prone spike sorting procedures.

The data and code for this model are available here.

Want to submit a piece? Or trying to write a piece and struggling? Check out the guides here!

Thank you for reading. Reminder: Byte Sized is open to everyone! Feel free to submit your piece. Please read the guides first though.

All submissions to berkan@usc.edu with the header “Byte Sized Submission” in Word Doc format please. Thank you!