Background
We designed a deep convolutional neural network (CNN) to diagnose thyroid malignancy on ultrasound (US) and compared the diagnostic performance of CNN with that of experienced radiologists.
Methods
Between May 2012 and February 2015, 589 thyroid nodules in 519 patients were diagnosed as benign or malignant by surgical excision. Experienced radiologists retrospectively reviewed the US of the thyroid nodules in a test set. CNNs were trained and tested using retrospective data of 439 and 150 US images, respectively. Diagnostic performances were compared between the two groups.
Results
Of the 589 thyroid nodules, 396 were malignant and 193 were benign. The area under the curve (AUC) for diagnosing thyroid malignancy was 0.805‐0.860 for radiologists. The AUCs for diagnosing thyroid malignancy for the three CNNs were 0.845, 0.835, and 0.850. There was no significant difference in AUC between radiologists and CNNs.
Conclusions
CNNs showed comparable diagnostic performance compared to experienced radiologists in differentiating thyroid malignancy on US.
Purpose:To verify the usefulness of the Thyroid Imaging Reporting and Data System (TI-RADS) for thyroid nodule diagnosis by less experienced physicians.Methods:From March 2012 to May 2012, ultrasonography-guided fine needle aspiration was performed in 204 thyroid nodules in 195 consecutive patients by four less experienced radiologists (<1 year in thyroid imaging). The number of suspicious ultrasonography features and the total risk score of each thyroid nodule were calculated according to the previous two models suggested by Kwak et al. The Delong method was used to compare the areas under the curve (AUCs) of the two models. Associations between the two models and the risk of malignancy were analyzed using penalized B-splines and the Cochran-Armitage trend test.Results:Among 204 thyroid nodules, 65 were malignant and 139 were benign. The probability of malignancy tended to increase as the number of suspicious ultrasonography features, and the sum of risk scores increased. There was no significant difference in the AUCs of the two models (P=0.673). The Cochran-Armitage trend test demonstrated an increased risk of malignancy as the number of suspicious ultrasonography features and the total risk score increased (P=0.001).Conclusion:Both the number of suspicious ultrasonography features and the total risk score are applicable and show comparable results in the risk stratification of thyroid nodules by less experienced radiologists in thyroid imaging.
ObjectiveTo compare automated volumetric breast density measurement (VBDM) with radiologists' evaluations based on the Breast Imaging Reporting and Data System (BI-RADS), and to identify the factors associated with technical failure of VBDM.Materials and MethodsIn this study, 1129 women aged 19-82 years who underwent mammography from December 2011 to January 2012 were included. Breast density evaluations by radiologists based on BI-RADS and by VBDM (Volpara Version 1.5.1) were compared. The agreement in interpreting breast density between radiologists and VBDM was determined based on four density grades (D1, D2, D3, and D4) and a binary classification of fatty (D1-2) vs. dense (D3-4) breast using kappa statistics. The association between technical failure of VBDM and patient age, total breast volume, fibroglandular tissue volume, history of partial mastectomy, the frequency of mass > 3 cm, and breast density was analyzed.ResultsThe agreement between breast density evaluations by radiologists and VBDM was fair (k value = 0.26) when the four density grades (D1/D2/D3/D4) were used and moderate (k value = 0.47) for the binary classification (D1-2/D3-4). Twenty-seven women (2.4%) showed failure of VBDM. Small total breast volume, history of partial mastectomy, and high breast density were significantly associated with technical failure of VBDM (p = 0.001 to 0.015).ConclusionThere is fair or moderate agreement in breast density evaluation between radiologists and VBDM. Technical failure of VBDM may be related to small total breast volume, a history of partial mastectomy, and high breast density.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.