Multichannel detection of evoked responses using critical values corrected by a parametric bootstrap: Frequency-domain cholesky correction

•Multivariate objective response detection (MORD) methods improve evoked response detection by using multiple channels.•MORD techniques are sensitive to channel correlation, which may lead to an inflated false positive rate.•A parametric bootstrap method is proposed to obtain critical values of MORD techniques based on frequency domain correlation.•The method was evaluated with simulated data and a database with electroencephalographic signals during auditory stimulation.•It led to higher detection rates than the single-channel methods while keeping false positive rate close to nominal values. Multivariate Objective Response Detection (MORD) techniques aim to detect evoked responses in multichannel electroencephalographic (EEG) recordings. They provide enhanced statistical power, allowing the detection of small signals in shorter or noisy recordings. However, the correlation between the signals in multichannel recordings can lead to false positive rates greater than the nominal significance level of the tests. To address this, we propose a parametric bootstrap approach that adjusts the critical values based on the correlation between EEG channels in the time domain, a method called time-domain Cholesky correction (TDCC). In that first approach, we assumed that correlation (or, more precisely, coherence) is constant across all frequency bands. However, this is unlikely to hold true, as signal-to-noise ratios (where the signal is the evoked response and noise all other signal components) may vary across frequencies. Thus, in the current work, we propose an alternative parametric bootstrap method for estimating the critical values of MORD techniques based on the correlation in the frequency domain (FDCC, frequency-domain Cholesky-corrected critical values). The proposed methods are evaluated using simulated data and an auditory steady-state response (ASSR) database in the 40 Hz range. The proposed method controlled the false positive rate well, with increased sensitivity compared to single-channel methods. When compared to TDCC, FDCC achieved similar performance but with the advantage of being, on average, 27 times faster in terms of computational time required to estimate the critical values.