Real-World Performance of Computer-aided Diagnosis System for Thyroid Nodules on Ultrasonography
- Presentation Speakers / Moderators
Purpose: To retrospectively evaluate the diagnostic performance of a commercially available computer-aided diagnosis (CAD) system for detecting thyroid cancers.
Materials and Methods: A total of 106 consecutive patients with 218 thyroid nodules (?5 mm), who underwent US-guided FNA or US examination prior to scheduled surgery, were enrolled. An experienced radiologist performed US examinations and applied the CAD systems (S-Detect 1 and S-Detect 2 for thyroid) integrated into a US machine. We compared the diagnostic performances of the radiologist, the CAD systems, and the radiologist assisted by the CAD system in detecting thyroid cancers.
Results: The sensitivity, specificity, PPV, NPV, and accuracy of the CAD systems were 80.2%, 82.6%, 75.0%, 86.3%, and 81.7%, respectively, for the S-Detect 1, and 81.4%, 68.2%, 62.5%, 84.9%, and 73.4%, respectively, for the S-Detect 2. The sensitivities of the CAD systems were not significantly different from that of the radiologist (P =0.454 and P =0.629); while the specificities and accuracies were significantly lower from those of the radiologist (P =0.001 and P <0.001; P =0.003 and P <0.001). The radiologist assisted by the CAD systems improved the diagnostic sensitivities, but specificities and accuracies were reduced compared to the radiologist alone. The interobserver agreement was moderate to substantial for the final diagnosis and each US descriptor, except fair for the calcifications.
Conclusions: The current CAD systems has a limited role as a decision-making supporter alongside radiologists in the thyroid cancer diagnosis. One of the main limitations is its inaccuracy to recognize calcifications, so this feature must be established by the radiologist.
Two Approaches to Level 5 Neck Dissection
- Jeff Blumberg