Objective assessment of familiarity in music using imagery and EEG-based machine learning.

Musical memory engages perceptual, mnemonic, and predictive processes that enable listeners to recognize and internally continue familiar melodies. When expected musical events are briefly omitted, these predictive mechanisms often trigger spontaneous musical imagery, providing a robust paradigm for examining the neural dynamics of familiarity in healthy individuals. In this study, we introduce a passive EEG-based machine-learning approach to evaluate musical familiarity using imagery. Twenty participants listened to familiar and unfamiliar songs embedded with brief silent gaps designed to elicit spontaneous musical imagery. EEG signals recorded during these silences were analyzed using both spectral features and Riemannian geometry-based tangent space mapping (TSM). Machine learning classifiers were trained independently for each subject to distinguish neural responses associated with familiar versus unfamiliar music. Our results show that models using TSM features outperformed those based on conventional spectral features, with logistic regression achieving the highest accuracy (76.5% ± 8.0%). Analysis of feature importance revealed that discriminative signals predominantly originated from auditory cortices, particularly in the right hemisphere, and reflected both local activity and functional connectivity. These findings demonstrate that machine learning approaches capture subtle, subject-specific neural patterns associated with musical imagery and memory retrieval. Replication in larger and more diverse populations will be necessary to establish the robustness and generalizability of the approach.