Computer-aided thyroid nodule diagnosis demonstrates sensitivity similar to radiologists

Issue: February 2017

Sensitivity of an ultrasound computer-aided system for diagnosis of thyroid nodules was similar to that of a radiologist, although specificity and accuracy were lower, according to study findings.

Perspective from

Jung Hwan Baek, MD, PhD, professor in the department of radiology at the University of Ulsan College of Medicine in Korea, and colleagues evaluated 89 adults (mean age, 43.5 years) with thyroid nodules (57.8% benign; 42.2% malignant). Researchers sought to determine the effectiveness of an ultrasound computer-aided diagnosis system for its diagnosis of malignant thyroid nodules and categorization of nodule characteristics compared with a radiologist review of ultrasound images.

Jung Hwan Baek

Diagnostic specificity for thyroid malignancy was higher with an experienced radiologist (94.9%) compared with the computer-aided diagnosis system (74.6%; P = .002), but there were no significant differences for diagnosis sensitivity. The radiologist produced a higher area under the receiver operating characteristic curve for diagnosis of thyroid malignancy compared with the computer-aided diagnosis system (P = .021).

There was substantial agreement between the radiologist and computer-aided diagnosis system for classifications of the ultrasound characteristics, including composition (kappa = 0.659), orientation (kappa = 0.74), echogenicity (kappa = 0.733) and spongiform appearance (kappa = 0.658).

“Sensitivity of the [ultrasound computer-aided diagnosis] system showed no statistically difference compared with an experienced radiologist,” Baek told Endocrine Today. “However, radiologists showed better diagnostic performance in terms of specificity. Therefore the [ultrasound computer-aided diagnosis system] can be used as a screening tool for thyroid nodules to decide on a further management plan. The result of the [ultrasound computer-aided diagnosis] system should be validated by an experienced thyroid specialists.” – by Amber Cox

Disclosure: The study was supported by a grant and provision of equipment from Samsung Medison Co.

Perspective

Megan Haymart

In a novel study assessing a computer-aided diagnosis system using artificial intelligence, Choi and colleagues challenge our perceptions of the role of humans in determining the relationship between sonographic patterns of thyroid nodules and risk of malignancy. The authors show that compared with an experienced radiologist a computer-aided diagnosis system has a similar sensitivity, but a lower specificity, for predicting benign versus malignant thyroid nodules. This study was limited by the fact that indeterminate cytologic results were excluded, the computer-aided diagnosis system could not evaluate calcifications, and its results were compared to those of just one radiologist. In addition, in an ideal world a computer-aided diagnosis system would perform better than a radiologist — with accuracy that would entirely eliminate the need for fine needle aspiration or unnecessary surgery. Regardless, this study highlights the potential for the future, a future that may include computer-aided diagnosis system instead of radiologists.

Megan Haymart, MD
Assistant Professor of Medicine, Metabolism, Endocrinology and Diabetes and Heme/Onc, University of Michigan

Disclosure: Haymart reports no relevant financial disclosures.

Disclosures:

Published by:

Sources/Disclosures

Collapse

Source:

Choi YJ, et al. Thyroid. 2017;doi:10.1089/thy.2016.0372.