A LOSO-NTD-Based Analytical Framework With Less Setting-Dependence Reveals Attentional Modulation of Gamma-Band...

Although attentional modulation of gamma-band auditory steady-state response (gamma ASSR) has been widely investigated, variability in feature-level analytical parameter settings undermines cross-study comparability and may partly account for inconsistent findings. This work proposed a leave-one-subject-out nonnegative tensor decomposition (LOSO-NTD) framework to quantify gamma ASSRs across attentional states with less setting-dependence. Within this framework, gamma ASSRs were quantified using task factor coefficients obtained by projecting individual data onto gamma ASSR core components derived from nonnegative tensor decomposition. These coefficients index pattern-level gamma ASSR activation strength, rather than local differences sensitive to specific analytical settings. Twenty healthy participants completed three tasks designed to systematically manipulate auditory attention, while EEG was recorded and subjective attention was assessed using a visual analog scale (VAS). Task factor coefficients for each condition were derived from phase- and power-based tensors using LOSO-NTD. Task 3 (visual distraction with auditory ignoring) exhibited significantly lower VAS scores and phase-based task factor coefficients than Task 1 (auditory counting) and Task 2 (passive listening) (all ${p} \lt 0.001$ ). Power-based task factor coefficients in Task 3 were likewise significantly lower than those in Task 1 ( ${p} \lt 0.001$ ) and Task 2 ( ${p} \lt 0.05$ ). Both phase- and power-based coefficients showed significant positive correlations with VAS scores (all ${p} \lt 0.001$ ). These results demonstrate that gamma ASSR activation strength increases with higher levels of auditory attention, providing compelling evidence for attentional modulation. Collectively, these findings establish LOSO-NTD as a less setting-dependent analytical framework that enables more objective and comparable findings across studies in neural computation.