Abstract
Dynamic MRI studies of the upper airway during speaking, singing or swallowing are complicated by the need for high temporal resolution and the presence of air-tissue interfaces that may give rise to image artifacts such as signal void and geometric distortions. This work exploits a recently developed real-time MRI technique to address these challenges for monitoring speech production at 3 T. The method combines a short-echo time radial FLASH MRI sequence (pulse repetition time/echo time = 2.22/1.44 ms; flip angle 5°) with pronounced undersampling (15 radial spokes per image) and image reconstruction by regularized nonlinear inversion. The resulting serial images at 1.5 mm in-plane resolution and 33.3 ms acquisition time are free of motion or susceptibility artifacts. This application focuses on a dynamic visualization of the main articulators during natural speech production (Standard Modern German). Respective real-time MRI movies at 30 frames per second clearly demonstrate the spatiotemporal coordination of lips, tongue, velum, and larynx for generating vowels, consonants, and coarticulations. The quantitative results for individual phonetic events are in agreement with previous non-MRI findings. Magn Reson Med, 2012. © 2012 Wiley Periodicals, Inc.
Since the earliest studies of human articulation in the late 18th century, methods and applications have continuously been expanded. Nowadays, articulatory processes during speaking may be analyzed by a wide range of techniques including laryngoscopy and laryngography, electropalatography, electromyography, electromagnetic articulography, or imaging based on X-ray videofluoroscopy or sonography (1). However, while former approaches require an invasive procedure, which eventually disturbs the natural articulation of vowels and consonants, videofluoroscopy involves the exposure to radiation and sonography is restricted to specific image orientations.
Although MRI promises noninvasive examinations as well as more flexible and detailed insights into the vocal tract, insufficient temporal resolution, and limited image quality still emerge as two major obstacles for studying speech production. For example, because long image acquisition times require either repetitively generated (2, 3) or continuant (4) speech sounds, such strategies preclude a proper visualization of the lips and tongue for plosive consonants that are characterized by rapid movements within 50–100 ms (5). In addition, motion-induced image blurring may result from data sharing when using long acquisitions with sliding-window reconstructions to achieve high frame rates (6, 7). It has therefore been concluded that the true temporal resolution for MRI should be comparable to that of standard videofluoroscopy, ideally no longer than 40 ms per image (8). Other problems are the occurrence of focal signal losses and geometric distortions that depend on the sensitivity of a chosen MRI acquisition technique to the unavoidable susceptibility differences near air-tissue interfaces in the upper airway (9, 10).
Most recently, we have described a real-time MRI technique using highly undersampled radial FLASH with image reconstruction by regularized nonlinear inversion (11–13). Preliminary applications to cardiovascular function (14), quantitative blood flow (15), and normal swallowing (16) allow for serial real-time images (i.e., movies) with 1.5 to 2.0 mm in-plane resolution and 20–30 ms acquisition time. Accordingly, the purpose of this study was to exploit this real-time MRI method to address the challenges posed by dynamic MRI of speaking and to investigate the key articulatory configurations during natural speech production including vowels, consonants, and coarticulation effects in vowels preceded by consonants.
