Classification of thyroid nodules in H-scan ultrasound images using texture and prinicipal component analysis

Abstract

H-scan ultrasound (US) imaging is a new tissue classification approach that depicts the relative size of scattering objects. The purpose of this study was to assess the ability of in vivo H-scan US imaging to discriminate benign from malignant thyroid nodules in human subjects. Image data was acquired using a SonixTouch US scanner (Analogic Ultrasound) equipped with an L14-5 transducer. To generate the H-scan US image, three parallel convolution filters were applied to the radiofrequency (RF) data sequences to measure the relative strength of the backscattered US signals. H-scan US was used to image thyroid lesions in human subjects. To examine the H-scan outputs for both benign and malignant lesions, seven texture features were derived from the spatial gray-level dependency (SGLD) matrix. These features were extracted from a region-of-interest (ROI) that was segmented from each H-scan US image. In addition, the principal component analysis (PCA) parameters were used to cluster the color texture feature image from benign and malignant lesions to further highlight differences. The results from texture analysis and PCA demonstrated significant differences between benign and malignant thyroid lesions (p<0.05). Overall, this study reveals the effectiveness of H-scan US imaging for classification of benign and malignant thyroid lesions.