Computer-Assisted Analysis of Ultrasonographic Features of Calcified Thyroid Nodules

Abstract

Purpose To quantify US characteristics of calcified thyroid nodules in terms of morphology and calcification by means of computer-assisted analysis (CAA), and to determine computerized US features of calcified thyroid nodules that are useful in the differentiation between malignant and benign lesions.

Material and Methods This study was institutional review board-approved, and patient informed consent was waived. Between June 2004 and May 2008, digital US images of 99 pathologically-proven thyroid nodules with calcifications (malignant: benign = 78: 21) from 125 patients (41 men and 84 women; mean age, 51 years; range, 29-81 years) were evaluated. We designed and implemented a CAA scheme to quantitatively analyze the US features of the calcified thyroid nodules based in two categories: morphology and calcification. In terms of morphology, we evaluated shape (feret, circulation, orientation, eccentricity, ellipitical point) and textural features (moment, contrast, homogeneity, skewness, kurtosis, entropy, diffuse histogram variation). Calcification features were assessed based on the number of calcifications, cluster ratio, area of calcifications (sum, minimum and maximum values), calcification distance, the ratio of calcification distance, intensity (minimum, maximum and mean values), skewness and kurtosis. We performed Student?s t?test and discriminant analysis and generated receiver operating characteristic (ROC) curves and the corresponding areas under the curve (AUC) in order to compare morphology, calcification and combined features.

Results Among the 12 morphology features, feret, circulation, eccentricity, elliptical point, moment, contrast, homogeneity, kurtosis, entropy and diffuse histogram variation showed significant differences (p < 0.05). Among the 12 calcification features, number of calcifications, sum of the area of calcification, calcification distance, the ratio of calcification distance, maximum intensity, skewness and kurtosis showed significant differences (p< 0.05). The AUC values for morphology, calcification and combined features were 0.81 (95% CI, 0.75-0.86), 0.79 (95% CI, 0.73-0.84), and 0.86 (95% CI, 0.81-0.91), respectively. The combined features showed a significantly higher AUC value than morphology or calcification features alone (p < 0.05).

Conclusion Use of CAA for the classification of morphology and calcification features of thyroid US images was feasible. Combined morphology and calcification features from CAA had a significantly higher diagnostic performance in differentiating malignant from benign calcified thyroid nodules when compared to the use of morphology or calcification features alone.