Reader performance in the ultrasonographic evaluation of oropharyngeal carcinoma

Objective: To examine reader performance in evaluating oropharyngeal anatomy on ultrasonography.

Materials and methods: Ultrasound images of the oropharynx comprising normal and malignant anatomic variants were organized into slideshows. Slideshows were administered to 6 readers blinded to participant tumor status and with varying experience reading oropharyngeal sonograms. A training slideshow oriented readers to images of the oropharynx with and without malignant lesions. Readers then evaluated images in a test slideshow for tumor presence and marked orthogonal long and short dimensions of the tumor. Results were analyzed for accuracy, sensitivity, specificity, inter-reader agreement, and measurement error relative to prospectively-identified reference measurements.

Results: Eighty-seven percent of base of tongue (BOT) sonograms were identified correctly by a majority of readers. In identifying BOT tumors, median accuracy, sensitivity, specificity, and Fleiss's kappa were 79%, 73%, 85%, and 0.51, respectively. Median measurement error in the long and short axes for BOT tumors was -2.6% (range: -40% to 29%) and -2.6% (range: -56% to 156%), respectively. Eighty-four percent of palatine tonsil sonograms were identified correctly by a majority of readers. In identifying tonsil tumors, median accuracy, sensitivity, specificity, and Fleiss's kappa were 77%, 74%, 78%, and 0.41, respectively. Median measurement error in the long and short axes for tonsil tumors was 3.8% (range: -45% to 32%) and -6.5% (range: -83% to 42%), respectively.

Conclusions: Overall, US has clinically useful sensitivity for identification of oropharyngeal carcinoma among readers of diverse clinical backgrounds and experience. US may be useful for the evaluation of features such as tumor dimensions.