Automated method for extracting response latencies of subject vocalizations in event-related fMRI experiments☆

Abstract

For functional magnetic resonance imaging studies of the neural substrates of language, the ability to have subjects performing overt verbal responses while in the scanner environment is important for several reasons. Most directly, overt responses allow the investigator to measure the accuracy and reaction time of the behavior. One problem, however, is that magnetic resonance gradient noise obscures the audio recordings made of voice responses, making it difficult to discern subject responses and to calculate reaction times. ASSERT ($\text{A}$daptive $\text{S}$pectral $\text{S}$ubtraction for $\text{E}$xtracting $\text{R}$esponse $\text{T}$imes), an algorithm for removing MR gradient noise from audio recordings of subject responses, is described here. The signal processing improves intelligibility of the responses and also allows automated extraction of reaction times. The ASSERT-derived response times were comparable to manually measured times with a mean difference of −8.75 ms (standard deviation of difference = 26.2 ms). These results support the use of ASSERT for the purpose of extracting response latencies and scoring overt verbal responses.

Introduction

The use of functional magnetic resonance imaging (fMRI) to study the neural substrates of language has proven to be effective (see e.g., Ojemann et al., 1998, Bookheimer, 2002, Phelps et al., 1997, Turkeltaub et al., 2002, Binder et al., 1997, but many studies have avoided the use of overt verbal responding, despite the common use of this response modality in behavioral and cognitive language studies. Two major problems have been identified in the use of verbal responses.

First, movement and susceptibility artifacts are expected to be produced that affect the statistical quality of images (Barch et al., 1999). This problem has been addressed through the use of event-related designs and special pulse sequences Barch et al., 1999, Birn et al., 1999, Eden et al., 1999, Palmer et al., 2001, de Zubicaray et al., 2001, Huang et al., 2002.

The second problem is that echo planar imaging bold oxygenation-level dependent (EPI BOLD) sequences generate gradient noise that obscures the subject's spoken responses (Munhall et al., 2001). As a consequence, a simple microphone threshold is often insufficient to detect the response onset, making it difficult to measure the voice reaction time. Furthermore, the noise hinders assessment of the content of the subject's response, which is critical to some event-related experiments (e.g., Schlaggar et al., 2002). Making this problem more difficult, the EPI sequence gradient noise spectrum overlaps the human speech spectrum, so simple filtration is ineffective in removing the noise without also degrading the speech signal. Thus, response times were previously extracted manually by measuring the time from stimulus presentation to voice response onset from the recorded audio signal in a sound editing program Palmer et al., 2001, Schlaggar et al., 2002. This process is labor-intensive and somewhat subjective, which may introduce some user variability.

ASSERT ($\text{A}$daptive $\text{S}$pectral $\text{S}$ubtraction for $\text{E}$xtracting $\text{R}$esponse $\text{T}$imes) was developed to address this second problem by removing much of the MR gradient noise from the auditory recordings, both improving the voice response intelligibility and facilitating automated extraction of reaction time latencies based on a simple signal amplitude measurement.