<![CDATA[One of the important features for diagnosing malignant thyroid nodules is based on their echogenicity characteristics, namely the grey intensity of each nodule. Therefore, computer-aided diagnosis (CADx) is required to select important features and classify nodules that are more likely to be malignant or benign and the form of treatment is known. This study proposes classifying thyroid nodules based on texture features from histogram, GLCM and GLRLM into 4 classes: anechoic, hyperechoic, hypoechoic and very hypoechoic. Ultrasound is the best way to filter information about the characteristics of the degree of malignancy of thyroid nodules used by doctors. The doctor's decision takes a long time. Early detection is necessary, so that doctors can provide treatment and prevention quickly. This study uses machine learning for early detection of the level of malignancy of thyroid nodules using ultrasound images with a dataset obtained from Sardjito Hospital, Yogyakarta Radiology Department. The results showed that the Linear SVM method is the best method to classify the level of malignancy of thyroid nodules based on the dataset which has been divided into 4 classes with 64 features resulting in an accuracy of 68.3%, a positive predictive value of 70% and a sensitivity value of 68%. Keywords: Ultrasound, Thyroid nodules, Classification, Early detection of disease, Feature extraction, Machine learning techniques, SVM echogenicity classification.]]>
Copyrights © 2025
