SafeLens: A Statistically Controlled Nonlinear Machine Learning Framework for Characterizing Hidden Dynamics in Event-Locked Neural Activity

SafeLens detects polarity-resolved event-locked deflections in resting intracranial EEG injected with simulated cortico-cortical evoked potentials, recovering ~2.5× more than trial-average thresholding.

• Why it matters: Event-locked neural activity is often analyzed with linear, feature-limited statistics that miss subtle nonlinear, trial-varying dynamics and lack reliable statistical thresholding for nonlinear models.
• What they did: SafeLens uses a sliding-window random forest with multiscale temporal derivatives, then uses randomized feature ablation and an empirical null from rest-vs-rest training to set a channel-specific noise-floor threshold.
• The result: In eight patients, CCEPs stayed detectable above chance until 1,040 ms post-stimulation, versus 570 and 455 ms for benchmark methods, implying stricter false-positive control for nonlinear discovery.

AI summary of the abstract · may contain errors · not medical advice · read the paper.