Implementing Teeth in Three-Dimensional Vocal Tract Models: Evaluating Methods Using CT, Intraoral Scanning, and ZTE MRI

Introduction
MRI derived 3D vocal tract models have many applications, but conventional T1 weighted and T2 weighted sequences are not sensitive to optimally image short T2 species such as bony structures, and teeth. Traser et al. demonstrated the importance of teeth in 3D vocal tract models, particularly when acoustic evaluation of models with side branches or side cavities is considered. Teeth can be directly imaged with computed tomography (CT), though this modality is rare in vocal tract modeling due to the risks associated with ionizing radiation. Intraoral laser scans of teeth (e.g., iTero and Trios) and laser scanned conventional dental impressions can be digitally inserted into MRI-modeled vocal tracts to improve the acoustic accuracy of models, particularly regarding fR2–51,. Zero Echo-Time (ZTE) MRI sequences have recently been developed that are capable of achieving CT-like images of the skull and C-spine. It is suggested that this method may have advantages over CT in the high resolution scanning of teeth. To date, the utility of ZTE in modeling the vocal tract airspace with teeth has not been investigated.

Research questions:
1) Can ZTE sequences be optimized to produce vocal tract segmentations (including teeth) that better represent in-vivo singing or speaking than conventional T1 or T2 weighted MRI?
2) How do the three methods for imaging teeth (CT, intraoral scanning, and ZTE) compare when incorporated into vocal tract models?
3) How do the transfer functions of the modeled vocal tracts compare to the recordings of in vivo speaking?
Methods

One professional singer will undergo ultra-low-dose CT scans (Siemens SOMATOM Force) and multiple high-resolution T1 weighted MRI (3T) scans to capture two vowels. Audio will be captured during scanning using the OptoAcoustics audio recording and denoising system to establish an acoustic ground truth. Intraoral scans will be performed to capture a three dimensional view of dentition. ZTE MRI scans will be collected using low flip angles (1 degree) to minimize T1 weighting and enhance bone contrast, high receiver bandwidth (100 kHz/pixel) to reduce chemical shift, high resolution (0.7 mm3) to reduce partial voluming, 4 number of averages to improve signal to noise. A recently proposed post-processing pipeline will be adapted to extract pseudo-CT like images from the “raw” ZTE images.
Vocal tract models will be generated: CT /∧/, CT /α/, MRI+intraoral scanned dentition /∧/, MRI+intraoral scanned dentition /α/, ZTE MR /∧/, ZTE MR /α/

Resulting models will be segmented and printed in 3D as described in previous studies. Transfer functions will be computed in silico and measured acoustically from 3D printed segmentations. Comparisons will be drawn to the in-vivo audio.