Thyroid Nodules Stratification Based on Orientation Characteristics Using Machine Learning Approach

Abstract

Patients with thyroid nodules should undergo further assessment to define malignancy stratification. One of the assessments is conducted using ultrasound. There are ten characteristics of thyroid nodules in the ultrasound examination. One of the characteristics is orientation. The orientation characteristic has its perks as it can tell the malignancy state from the growth direction to the orientation category. This study develops an approach to determine the stratification of the thyroid nodules based on orientation characteristics. The proposed approach consists of pre-processing, segmentation, feature extraction, and classification by using a support vector machine. Nine geometrical and moment features are used to describe orientation characteristics. Furthermore, the features are trained to differentiate two classes, namely are parallel and non-parallel. The testing results achieve more than 0.98 for accuracy, sensitivity, and NPV, respectively, while specificity and PPV achieve a perfect score. Looking by the results, it can be concluded that the proposed method has a good performance in differentiating orientation characteristics into a parallel and non-parallel category.