Computerized analysis of calcification of thyroid nodules as visualized by ultrasonography Woo Jung Choi, Jeong Seon Park, Kwang Gi Kim, Soo-Yeon Kim, Hye Ryoung Koo, Young-Jun Lee European Journal of Radiology Volume 84, Issue 10, Pages 1949-1953 (October 2015) DOI: 10.1016/j.ejrad.2015.06.021 Copyright © 2015 Elsevier Ireland Ltd Terms and Conditions
Fig. 1 Application of a region of interest (ROI). Thyroid US showing a hypoechoic nodule (left). The lesion boundary was manually drawn by a radiologist (right). European Journal of Radiology 2015 84, 1949-1953DOI: (10.1016/j.ejrad.2015.06.021) Copyright © 2015 Elsevier Ireland Ltd Terms and Conditions
Fig. 2 Example of automatic segmentation of a calcification detected by Otsu’s algorithm. A. The US image shows a hypoechoic nodule with one rim calcification (white spot) within the manually drawn boundary. B. The calcification particle was detected as a tiny spot. C. Dilation was performed to connect the tiny spots of the calcification particles, and they were determined to be true positive rim calcifications. European Journal of Radiology 2015 84, 1949-1953DOI: (10.1016/j.ejrad.2015.06.021) Copyright © 2015 Elsevier Ireland Ltd Terms and Conditions
Fig. 3 Schematic diagram of the ratio of calcification distance, which is the calcification distance divided by the radius, which suggests the eccentricity of the calcification. European Journal of Radiology 2015 84, 1949-1953DOI: (10.1016/j.ejrad.2015.06.021) Copyright © 2015 Elsevier Ireland Ltd Terms and Conditions
Fig. 4 Receiver operating characteristic (ROC) curve of the values of the four features and of the neural network classifier (all features). The neural network utilizing all four features exhibited the best performance, with an Az value of 0.84. European Journal of Radiology 2015 84, 1949-1953DOI: (10.1016/j.ejrad.2015.06.021) Copyright © 2015 Elsevier Ireland Ltd Terms and Conditions