EEG Machine Learning for Antidepressant Response Stratification

The patent segments the unselected depressed population into distinct subgroups based on baseline neural activity patterns measured by EEG. The SELSER model identifies specific phenotypes (e.g., high alpha power, low theta power patterns) that predict differential response to antidepressants versus placebo. This segmentation allows treatment to be tailored to specific neural phenotypes rather than applied uniformly, resolving the contradiction between broad coverage and effective treatment.

The patent implements preliminary classification of patients into antidepressant-responsive and placebo-responsive phenotypes using baseline EEG measurements and the SELSER machine learning model before treatment initiation. This preliminary action enables precise treatment selection upfront, allowing uniformly simple implementation of the classification protocol followed by targeted treatment assignment, thus maintaining ease of operation while achieving high treatment precision.

The SELSER model extracts and focuses on specific, biologically relevant neural features from EEG data that are most predictive of treatment response, such as alpha band power, theta band power, and their interactions. Rather than analyzing all possible brain signal features, the model selectively extracts the most informative subset, thereby maintaining high prediction accuracy while reducing model complexity and improving interpretability.