Performance of Artificial Neural Network in Differentiation between Malignant and Benign Nodules in the Thyroid on Ultrasonography

摘要

The Purpose of our study is to determine the utility of artificial neural network in differentiation between malignant and benign nodules in the thyroid on ultrasonography. We evaluated 109 thyroid nodules. The pathological diagnoses of malignant nodules are papillary cancer (48), follicular cancer (1) and benign nodules are nodular hyperplasia (26), adenomatous goiter (17), benign follicular tumor (4), acute thyroiditis (2), oncocytic adenoma (1) and others (10). The area under the receiver operating characteristic curve (Az) of ANN was 0.9492 (SE: 0.0195). Az of 10 year-experienced radiologist was 0.83 and Az of 4 year-experienced radiologist was 0.75.