Thyroid Nodule Classification Based on Characteristic of Margin using Geometric and Statistical Features

Abstract

Ultrasound is a digital imaging modality used to assess thyroid nodules. However, ultrasound still has some deficiencies to the result of diagnosis. Ultrasound produce operator dependent result, it means the result of the analysis of ultrasound is highly dependent on the ability, expertise, and skills of the operators who perform the examination. To solve this problem, it is necessary to develop computerization system that can help radiologist in making decisions of diagnosis. This system works by analyzing the characteristics of thyroid nodules. One of these characteristics of margin. Previous research has discussed the classification of margin characteristics to define the diagnosis of thyroid nodules with the classification of two classes: smooth and irregular. This study try to classify the thyroid nodule in three: smooth, ill define, and irregular. To solve the problem, a total of 99 images are used. The proposed method are started with removing the artefacts and noises using adaptive median filtering and speckle reducing bilateral filtering. The result of this step is segmented using active contour and morphological operation to fine the concern area of nodule. Segmented area is used to classify thyroid nodule in three classes using MLP. Experiment result show the performance of method with the accuracy of 90.91%, sensitivity of 90.71%, specificity of 93.46%, PPV of 90.84%, and NPV of 93.49%. These results show that proposed method has good performance to classify thyroid nodule based on characteristics of margin.