Schedule of Events | Symposia

Using a machine learning classifier to estimate neural distinctiveness in EEG data

Poster Session D - Monday, March 31, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Noah Reardon¹, Erin Knape¹, Poortata Lalwani², David Brang¹, Molly Simmonite¹, Thad Polk¹; ¹University of Michigan, ²University of California, Irvine

Patterns of neural responses to different stimuli, such as faces vs. buildings, become less distinctive and more confusable with advancing age. This so-called neural dedifferentiation is hypothesized to contribute to age-related declines in cognitive and sensory abilities. The distinctiveness of neural activity patterns is most commonly measured using functional magnetic resonance imaging (fMRI). However, fMRI is an indirect measure of neural activity and has relatively poor temporal resolution. Electroencephalography (EEG) complements these limitations by measuring neural activity more directly with excellent temporal resolution. Here, we test whether a machine learning classifier (support vector machines, SVM) can reliably distinguish EEG responses to faces vs. buildings. EEG data was recorded from 27 healthy older adults aged 65+ as they completed a task in which they viewed images of faces and buildings. To ensure participants were alert and paying attention, they were asked to respond whenever they saw famous faces or buildings. EEG recordings were preprocessed and 600ms epochs beginning at stimulus onset were extracted from all 64 electrodes. The data were then randomly divided into training and testing datasets and SVM classifiers were trained on the training data using 20-fold cross-validation. The most accurate model across the folds for each participant was then used to classify the independent testing data. The SVM classifiers demonstrated robust above chance discrimination of house and face trials in the testing dataset. This result highlights the potential of using SVM classifiers to examine neural distinctiveness in EEG data.

Topic Area: METHODS: Neuroimaging