Evaluation of Underlying Lymphocytic Thyroiditis With Histogram Analysis Using Grayscale Ultrasound Images

• Published in: Journal of ultrasound in medicine 2016-03, Vol.35 (3), p.519-526
• Main Authors: Kim, Ga Ram, Kim, Eun-Kyung, Kim, Soo Jin, Ha, Eun Ju, Yoo, Jaeheung, Lee, Hye Sun, Hong, Jung Hwa, Yoon, Jung Hyun, Moon, Hee Jung, Kwak, Jin Young
• Format: Article
• Language: English
• Subjects: Adolescent; Adult; Aged; Aged, 80 and over; Algorithms; Data Interpretation, Statistical; Female; head and neck ultrasound; histogram analysis; Humans; Image Enhancement - methods; Image Interpretation, Computer-Assisted - methods; lymphocytic thyroiditis; Machine Learning; Male; Middle Aged; Observer Variation; Reproducibility of Results; Sensitivity and Specificity; thyroid gland; Thyroiditis, Autoimmune - diagnostic imaging; Ultrasonography - methods; ultrasound; Young Adult
• ISSN: 0278-4297; 1550-9613
• DOI: 10.7863/ultra.15.04014

Objectives The purpose of this study was to evaluate diagnostic performance of histogram analysis using grayscale ultrasound (US) images in the diagnosis of lymphocytic thyroiditis. Methods Three radiologists reviewed a total of 505 US images and classified the images according to the presence/existence of lymphocytic thyroiditis. After 2 months, each reviewer repeated the process with the same 505 images in a randomly mixed order. The intraobserver and interobserver variability was analyzed with a generalized κ value. Four histogram parameters (mean value, standard deviation, skewness, and kurtosis) were obtained, and an index was calculated from principal component analysis. Diagnostic performances were compared. Results Of 505 patients, 125 (24.8%) had lymphocytic thyroiditis, and 380 (75.2%) had normal thyroid parenchyma on pathologic analysis. The κ value for intraobserver variance ranged from −0.002 to 0.781, and the overall κ values for interobserver variance were 0.570 and 0.214 in the first and second tests, respectively. The sensitivity, specificity, accuracy, positive predictive value, and negative predictive value for the 3 reviewers versus the principal component analysis index were 28.0% to 83.2%, 43.7% to 82.6%, 53.5% to 79.0%, 24.6% to 56.2%, and 75.2% to 88.9% versus 58.4%, 72.4%, 68.9%, 41.0%, and 84.1%. Conclusions Histogram analysis of grayscale US images provided confirmable and quantitative information about lymphocytic thyroiditis and was comparable with performers’ assessments in diagnostic performance.