Hybrid CT+MRI Vocal Tract Modeling in Living Participants

Objective

MRI based vocal tract models have many applications in voice science and education, but they do not adequately capture bony structures, and spatial resolution is often low to minimize scanning time. Computed tomography (CT) is well suited for vocal tract imaging, but is infrequently used due to the risk of ionizing radiation. A method for high-resolution CT+MR image set combination was successfully demonstrated in a cadaveric study (Meyer et al., 2022). Briefly, a hybrid CT+MRI dataset was created by merging isotropic 1mm3 MRI volumetric data with low-dose 1mm3 CT data via a rigid body deformation scheme. The vocal tract segmentations of the hybrid dataset showed improved accuracy in modeling airspace of structures that are typically challenging to image with MRI or CT alone (eg. teeth, valleculae, and pyriform sinuses).
Replicating this study with living participants will address the following two research questions:
1) Does the enhanced spatial accuracy translate into acoustic characteristics (both model-computed and 3D printed) that better represent in-vivo singing or speaking?
2) Can deformable registration algorithms be successfully employed to merge one CT scan into multiple vocal tract configurations represented in MR?

Methods

Two professional classical singers will each undergo a single ultra-low-dose CT scan (Siemens SOMATOM Force) and multiple high-resolution MRI (GE SIGNA Premier 3T) of six sung vowels. CT and MRI scans will be blended to include bony structures in the vocal tract, and clearly delineated tissue-air boundaries. The resulting images will be segmented and printed in 3D. Vocal tract transfer functions will be generated from model-computed and 3D printed segmentations. These will be compared to acoustic characteristics measured from audio samples of the participants recorded during MR scanning as an assessment of the accuracy of the various techniques.

Results

Vocal tract models using combined CT+MRI are predicted to have more accurate spatial resolution and improved acoustic properties than has previously been achieved. Resulting segmentations will be shared via a CC license. Results of this study will facilitate computational and physical modeling of speech and singing with enhanced accuracy of the airway configuration for sound production.