Application of Texture Analysis in the Differential Diagnosis of Benign and Malignant Thyroid Nodules: Comparison With Gray-Scale Ultrasound and Elastography

Abstract

OBJECTIVE: The purposes of this study were to investigate the optimal subset for texture analysis by use of a histogram and cooccurrence matrix in the differential diagnosis of benign and malignant thyroid nodules and to compare the results with those of gray-scale ultrasound and elastography.

MATERIALS AND METHODS: From a retrospective search of an institutional database between June and November 2009, 633 solid nodules 5 mm or larger from 613 patients who underwent gray-scale ultrasound and elastography and subsequent ultrasound-guided fine-needle aspiration were included in this study. Each nodule was categorized as probably benign or suspicious of being malignant according to findings at gray-scale ultrasound and elastography. Histogram parameters (mean, SD, skewness, kurtosis, and entropy) and cooccurrence matrix parameters (contrast, correlation, uniformity, homogeneity, and entropy) were extracted from gray-scale ultrasound and elastographic images. The diagnostic performances of gray-scale ultrasound, elastography, and texture analysis for differentiating thyroid nodules were evaluated.

RESULTS: Gray-scale ultrasound had the best diagnostic performance with an ROC AUC (Az) of 0.809 among all parameters. Elastography had significantly poorer performance (Az = 0.646) than gray-scale ultrasound (p < 0.001). Mean extracted from gray-scale ultrasound had the highest Az (0.675) among all histogram and cooccurrence matrix parameters extracted from gray-scale ultrasound and elastographic images. However, mean and the combination of mean and gray-scale ultrasound had poorer performance than gray-scale ultrasound alone.

CONCLUSION: Using texture analysis does not improve diagnostic performance in the evaluation of thyroid cancers.