Rapid three‐dimensional magnetic resonance imaging of vocal tract shaping using compressed sensing

Conventional 3D magnetic resonance imaging (MRI) of vocal tract during sustained sound production takes prohibitively long and requires repetitions of the same articulation with pauses of the scan. This can produce image misregistration artifacts possibly due to different positioning of the jaw, tongue, and lips at each repetition. Within a scan time limit of 6–10 seconds, a duration practical for sustained sound production, it is desirable to accelerate high‐resolution 3D imaging by undersampling 3D k‐space. In this study, compressed sensing theory was adopted to achieve the acceleration by incoherent undersampling of 3D k‐space and minimum L1‐norm reconstruction. The increase in transform sparsity was exploited by constraining image phase in which rapid phase variation is expected in air‐tissue boundaries due to large magnetic susceptibility difference. Vocal tract volume of interest was imaged during sustained sound production of English consonants /s/, /sh/, /r/, and /l/. Vocal tract area functions we...