Differential diagnosis of thyroid nodules with ultrasound elastography based on support vector machines

Abstract

A fine needle aspiration (FNA) biopsy is the standard procedure of choice for differentiating between benign and malignant thyroid nodules, and -300,000 thyroid FNA biopsies are performed in the U.S. each year. However, FNA is invasive, costly and uncomfortable for patients. Furthermore, a large percentage (-70%) of these FNAs turn out to be benign. In this paper, we present a non-invasive and automatic approach for differentiating benign and malignant thyroid nodules with ultrasound elastography based on support vector machines (SVM) with biased penalties. We used the elastography data of 98 thyroid nodules (16 malignant and 82 benign) from 92 subjects previously acquired with a clinical ultrasound machine, Hitachi Hi Vision 5500. We conducted the leave-one-out cross-validation (LOOCV) in evaluating the performance of our classification method. Our goal was to obtain the maximum geometric mean (MGM) of sensitivity and specificity. The results show that our method was able to get MGM of 90.1% with the sensitivity of 93.8% and the specificity of 86.6%.