By Abby Hagen (CAS’23)
Since her undergraduate studies, Statistics Professor Emily Stephen has been fascinated by the ways that statistics can unfurl the secrets of the brain. Her work with speech perception began at BU as a PhD student studying with Professor of Speech, Language, and Hearing Sciences Frank Guenther at Sargent College; Now, she collaborates with neuroscience colleagues at BU to decode and understand speech across spatial scales.
“Academically, I consider myself a computational neuroscientist,” said Stephen. “I study the brain and I use data science and statistical methods to do that. Statistics is what allows you to kind of skim over the details that you don’t know, and use it for a really valuable purpose.”
Though well adept at analyzing and employing the data in statistical models, Stephen isn’t a neuroscientist by practice, relying on collaboration with neuroscientists at BU to supply the data she needs to create statistical models of the brain’s underlying dynamics.
“I’m a statistician,” Stephen said. “I don’t have a lab, but everything I do requires data. So everything I do requires making friends with someone who collects data and helping them. I’ve been able to do that with people at BU.”
Interdepartmental collaborations at BU—like Stephen’s relationship with statistics and neuroscience— allow faculty to innovate and elevate each other, achieving greater advancements in their fields than they could alone.
“Interdisciplinary work is challenging and I’m still learning how to do it,” said Stephen. “It’s easy to get going on things at BU, which I was a little bit worried about. It’s not a concern amongst departments here.”
Stephen primarily studies how the brain processes auditory speech and how the brain encodes the different parts of speech when we hear it.
Understanding speech is more work for our brain than one might think. “There’s different aspects of speech that need to be split apart and then put back together in order to understand a sentence if you’re listening to it,” said Stephen.
According to Stephen, one part of your brain may be responsible for encoding the “puh” sound the letter ‘P’ makes in the word ‘pumpernickel’, and another part of the brain may be responsible for more temporal aspects of encoding speech, like the time it takes the brain to register the many syllables of ‘pumpernickel’ versus ‘pie’.
These are nuanced phenomena, and answering them from a purely neuroscientific perspective leaves you with a blurry image. Stephen sees statistics as a way to get that image to a higher resolution without employing invasive medical procedures.
Most neuroscience recordings are non-invasive. Electroencephalogram (EEG) recordings, which detects the electrical activity in your brain, are taken by placing electrodes on the scalp. While these recordings produce reliable markers of brain states like epilepsy or seizures, they leave gaps in the understanding of what the brain is actually doing to produce these observable markers.
“One of the problems is that these non-invasive recordings, even though they produce very reliable effects of the condition, we don’t always know where they come from in terms of the underlying cellular dynamics, what the actual neurons are doing in the brain that causes you to observe these kind of reliable effects in the higher spatial scale outside the brain,” says Stephen.
Stephen uses statistics to map and model these underlying cellular dynamics of auditory speech perception based on the data neuroscientists can get from these noninvasive brain readings.
“It’s really difficult to study speech because it’s in humans,” said Stephen. “We can’t cut open their brains to record them whenever we want.”
Currently, powerful speech research comes from epilepsy patients who have electrodes implanted to record directly from the brain, limiting the crop of volunteers; Finding ways to get the same data non-invasively would allow healthy people to volunteer for studies without having to undergo surgery.
There are hundreds of thousands of neurons that contribute to the effects we can see from EEG, but most of the current theoretical models for neurons are being done with small sets of hundreds to a thousand neurons at maximum.
“It’s important to be able to describe interactions between small subsets of neurons, but you’re always going to miss higher level processing if you don’t ever look at and if you don’t have a way of describing the higher level dynamics,”says Stephen.
Stephen believes that to understand the brain fully, we have to be able to visualize how the individual neuron interacts with and affects the whole brain.
Think about the brain like a community:
“If you want to talk about important functions or behaviors of a community, you can’t look at one person. You have to look at everyone,” Stephen said, “It’s not that the community doesn’t impact the one person or that the one person impacts the community, it’s just that even the concept of a community requires more people.”
So, if we know that different aspects of speech are registered by different parts of the brain, we ideally have to be able to look at and analyze the activity of all the neurons at once to get a full appreciation for how speech is encoded.
Stephen’s ultimate goal is to simultaneously model small and higher level brain dynamics to get a complete picture of how neurons are interacting spatially by building intermediate models, noting, “There are important processing or things in the brain that I believe wouldn’t even make sense to talk about on the scale of a thousand neurons, but that become extremely important when you’re looking at a million neurons.”
Stephen’s statistical models will facilitate a more precise and detailed image of the emergent effects associated with small scale brain activity, enhancing the information extracted from the reliable effects of noninvasive brain recordings like EEG.
Through her interdisciplinary approach to research at BU, Stephen is actively modeling the effectiveness of looking to the greater community to understand the complicated nature of the brain.
“This is statistics applied to neuroscience, but it’s still really statistics. I’m not collecting any necessarily new data, but I’m going to be able to develop models that will be really valuable for neuroscience.”