Performance of Lacunarity Features for Classifying Thyroid Nodule using Thyroid Ultrasound Images
Eka Legya Frannita, Hanung Adi Nugroho, Anan Nugroho, Zulfanahri, Igi Ardiyanto · 2018
Diagnosis of thyroid cancer can be analysis using nodule characteristic. The diagnosis analysis of thyroid nodule is depended on how long the radiologist gets their experience. To reduce radiologist dependency, a computerized system is necessary to build. This study focuses on classifying the thyroid nodule by using texture features into three classes. This research uses 97 thyroid ultrasound images. The first step of proposed method is pre-processing used to enhance the detection capability. After that, morphological operation and active contour are applied to find the correct nodules. The segmented area is extracted using histogram, GLCM, GLRLM, and lacunarity. The extracted value is used to classify the data using Multilayer Perceptron (MLP). The result shows the highest performance is felt in MLP method using lacunarty features. It is about 98.97% accuracy, 98.92% sensitivity, 99.47% specificity, 99.05% PPV, and 99.50% NPV. It means that lacunarity feature has good performance to classify three classes.