Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Sketchpad Series 

Machine Learning Decoding of Emotional Valence from Intracranial Cortical and Subcortical Electrophysiological Signals

Poster Session D - Monday, March 9, 2026, 8:00 – 10:00 am PDT, Fairview/Kitsilano Ballrooms

Max Zhu¹ (max.zhu2@ucsf.edu), Jinxiao Zhang¹, Clay Smyth^1,2, Lexin Li², Philip Starr¹, Simon Little¹; ¹University of California, San Francisco, ²University of California, Berkeley

Both cortical and subcortical brain regions are involved in emotion generation, but the extent to which neural activity in these circuits supports decoding of emotional valence remains unclear. Here, we investigated the feasibility of decoding emotional valence from intracranial recordings in patients with Parkinson’s disease. We recorded local field potentials (LFPs) from the subcortical subthalamic nucleus (STN) and sensorimotor cortex of two participants chronically implanted with intracranial EEG. Participants performed an emotion rating task in which each trial consisted of viewing an image followed by a valence rating on a continuous scale (−50 to +50). Each participant completed five sessions (300 trials per session; 1,500 images in total). Spectral power features were extracted from LFPs for each trial, spanning image-viewing and valence-rating periods. Logistic regression and support vector machine classifiers were trained to discriminate between (1) positive versus negative valence and (2) neutral versus non-neutral stimuli. Performance was evaluated with 5-fold cross-validation and permutation testing against chance. Across participants, classifiers achieved statistically significant above-chance decoding accuracy for both valence contrasts. We further found that decoding strength varied across brain regions and individuals. These findings demonstrate the feasibility of decoding subjective valence responses from intracranial STN and sensorimotor LFP activity during an emotion rating task. This supports the idea that signals from circuits commonly characterized to be motor-related can be leveraged to probe affect-related dynamics, motivating future work to characterize spatial and temporal specificity, dissociate affective representations from salience and motor factors, and explore applications of adaptive neuromodulation for affective symptoms.

Topic Area: EMOTION & SOCIAL: Emotion-cognition interactions