Volumetric MRI acquisition and processing

Abstract

To provide fast and potentially more accurate three-dimensional anatomic information on European Portuguese (EP) sounds, a direct acquisition of volumetric MRI information was attempted, through a 3-D spoiled fast gradient echo sequence. The viability of the used MRI acquisition protocol relies on the possibility of obtaining consistent information. For that, two different image processing approaches were attempted: First, a so-called 2.5-D method consisting in a first segmentation on the sagittal plane, multiplanar reslicing perpendicularly to the sound propagation in the vocal tract, and new segmentation in the 45 planes obtained; second, a direct 3-D level set segmentation. A method was applied to a comprehensive single speaker corpus contemplating: Oral and nasal vowels, nasal consonants, unvoiced fricatives, and laterals. Method advantages are: Reconstruction of slices with good detail in any direction, useful to obtain slices perfectly orthogonal to the vocal tract centerline; higher SNR when compared with 2-D imaging; easier 3-D visualization; and acquisition two times faster, with positive impact in the amount of material acquired in one session with less subject effort. 3-D information was especially important for the study of nasal [P. Martins et al., InterSpeech 2007, accepted] and lateral sounds, some of them with specific characteristics for EP. [Work supported by FCT, Portuguese Research Agency, by Project HERON POSC/PLP/57680/2004.]