Computer-Aided Malignancy Risk Assessment of Nodules in Thyroid US Images Utilizing Boundary Descriptors

Abstract

This paper investigates the diagnostic potential of boundary descriptors calculated on delineations of thyroid nodules in ultrasound images. The utilized set of boundary descriptors includes compactness, chain code histogram and fractal dimension. Experiments were conducted on thyroid US images, so as to evaluate the discriminating capability of each boundary descriptor for the classification of thyroid nodules in terms of their malignancy risk. The receiver operating characteristic (ROC) is used for the evaluation of the experimental results. It is derived that thyroid nodules can be classified according to their malignancy risk, by utilizing their compactness, fractal dimension and chain code histogram, obtaining an area under curve (AUC) which reaches 0.93.