gms | German Medical Science

Withholding read-out pulses quenches peak sound levels during echo-planar imaging (EPI). Thereby, read-out omissions from EPI are suited to reliably evoke auditory blood oxygenation level dependent (BOLD) signal fluctuations. Here, we re-examine a large set of auditory BOLD data recorded with read-out omissions from 60 normal hearing subjects model-free. Using probabilistic independent component analysis (PICA) and its tensorial extension to multiple subjects, we demonstrate that independent component and statistical maps from the general linear model (GLM), thresholded by mixture modelling or clustering, may attain similar sensitivities to detect auditory responses on the single-subject as well as group level. In contrast to model-based analysis, however, tensor-PICA isolated lateralized auditory responses, motion-associated edge artefacts, and other non-auditory stimulus-correlated as well as uncorrelated signals. Moreover, data-driven PICA is not affected by irregular autocorrelations or slight uncertainties of perceptual or stimulus timing. In terms of the measures of response strength, its subject (S-) mode was more sensitive to physiological and physical (e.g. scanner / coil) influences than effect size estimates derived from the GLM analysis. To further emphasize the clinical potential of read-out omissions from EPI, a male toddler with postmeningitic cochlear damage and severe binaural sensorineural hearing loss (SNHL) is reported on. In this patient, although studied under anaesthesia, PICA of our novel EPI modification established residual audition prior to cochlear implantation (CI).

Keywords: Auditory Functional Magnetic Resonance Imaging (FMRI), Probabilistic Independent Component Analysis (PICA), General Linear Model (GLM), Mixture Modelling (MM), SensoriNeural Hearing Loss (SNHL).