Premium
Quantitative Evaluation for Differentiating Malignant and Benign Thyroid Nodules Using Histogram Analysis of Grayscale Sonograms
Author(s) -
Nam Se Jin,
Yoo Jaeheung,
Lee Hye Sun,
Kim Eun-Kyung,
Moon Hee Jung,
Yoon Jung Hyun,
Kwak Jin Young
Publication year - 2016
Publication title -
journal of ultrasound in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.574
H-Index - 91
eISSN - 1550-9613
pISSN - 0278-4297
DOI - 10.7863/ultra.15.05055
Subject(s) - medicine , kurtosis , thyroid nodules , principal component analysis , standard deviation , nuclear medicine , confidence interval , histogram , skewness , cutoff , radiology , thyroid , statistics , mathematics , artificial intelligence , physics , quantum mechanics , computer science , image (mathematics)
Objectives To evaluate the diagnostic value of histogram analysis using grayscale sonograms for differentiation of malignant and benign thyroid nodules. Methods From July 2013 through October 2013, 579 nodules in 563 patients who had undergone ultrasound‐guided fine‐needle aspiration were included. For the grayscale histogram analysis, pixel echogenicity values in regions of interest were measured as 0 to 255 (0, black; 255, white) with in‐house software. Five parameters (mean, skewness, kurtosis, standard deviation, and entropy) were obtained for each thyroid nodule. With principal component analysis, an index was derived. Diagnostic performance rates for the 5 histogram parameters and the principal component analysis index were calculated. Results A total of 563 patients were included in the study (mean age ± SD, 50.3 ± 12.3 years;range, 15–79 years). Of the 579 nodules, 431 were benign, and 148 were malignant. Among the 5 parameters and the principal component analysis index, the standard deviation (75.546 ± 14.153 versus 62.761 ± 16.01; P < .001), kurtosis (3.898 ± 2.652 versus 6.251 ± 9.102; P < .001), entropy (0.16 ± 0.135 versus 0.239 ± 0.185; P < .001), and principal component analysis index (–0.386±0.774 versus 0.134 ± 0.889; P < .001) were significantly different between the malignant and benign nodules. With the calculated cutoff values, the areas under the curve were 0.681 (95% confidence interval, 0.643–0.721) for standard deviation, 0.661 (0.620–0.703) for principal component analysis index, 0.651 (0.607–0.691) for kurtosis, 0.638 (0.596–0.681) for entropy, and 0.606 (0.563–0.647) for skewness. The subjective analysis of grayscale sonograms by radiologists alone showed an area under the curve of 0.861 (0.833–0.888). Conclusions Grayscale histogram analysis was feasible for differentiating malignant and benign thyroid nodules but did not show better diagnostic performance than subjective analysis performed by radiologists. Further technical advances will be needed to objectify interpretations of thyroid grayscale sonograms.