(A313) The Role of Artificial Intelligence in Vocal Cord Ultrasound Interpretation

Introduction: Thyroidectomy carries a risk of vocal cord paralysis (VCP). Good clinical practice supports evaluation of vocal cord function before and after thyroid surgery. The widely accepted examination method is fiberoptic laryngoscopy, which requires expensive equipment, lengthy cleaning, and training. Further, it may be unaffordable or unavailable in many healthcare settings.
Vocal cord ultrasound (VCUS) has been suggested as an alternative to laryngoscopy, but is operator-dependent and may be difficult to interpret in some patients. In this research, we created an artificial intelligence (AI) model that can differentiate vocal cord motion from non-motion, by computer analysis of ultrasound images.

Methods: De-identified VCUS images will be obtained from healthy volunteers with clinically normal voice function. VCUS images from larynges known to have unilateral VCP have been obtained from de-identified teaching files. A convolutional neural network (CNN) model will be trained to determine the length, angle and position of the vocal cords from still ultrasound images.
Images will only be used for CNN training if judged to be interpretable by a VCUS quality scoring scale (QSS), developed by the authors.

Results: Preliminary analysis of human interpretation of evaluable VCUS still images and video clips has revealed a moderate ability to interpret correctly with some user-user variability. It is anticipated that a CNN can be readily trained to differentiate symmetric (bilaterally mobile) from geometrically asymmetric images (unilateral VCP). Image collection and computational learning are currently ongoing.

Conclusions: VCUS is a method of vocal cord motion assessment that carries with it minimal risk, rapid deployment, ease of cleaning, and reduced cost. AI interpretation of VCUS imaging eliminates user variability and potential error. Future goals include employment of AI interpretive programming into apps typically paired with handheld, ultraportable ultrasound systems in common use.