Texture analysis for classification of thyroid ultrasound images

Ultrasonography (USG) is one of the best imaging modalities for early detection of malignancy in the thyroid gland. Computer Aided Diagnosis (CAD) plays an important role in classifying thyroid nodules. CAD is used as an objective consideration, this is due to the high subjectivity in physician's interpretation of ultrasound images that can lead to difference scanning results. This research proposes a classification of thyroid ultrasound images by using some texture features into two classes. The dataset consists of 39 ultrasound images which grouped into 25 cystic cases and 14 solid cases. An initial step of image pre-processing is conducted to enhance the detection capability. Afterwards, followed by some methods of morphological operation, that is active contours without Edges (ACWE) and histogram equalization. The feature extraction is developed based on texture analysis by using Gray Level Co-occurrence Matrix (GLCM), Histogram and Gray Level Run Length Matrix (GLRLM). Finally, Multilayer Perceptron (MLP) is used to classify cystic nodule from solid nodule. The result shows that the proposed method achieves the accuracy of 89.74%, sensitivity of 88.89%, specificity of 91.67%, positive predictive value (PPV) of 96.00% and negative predictive value (NPV) of 78.57%. This indicates that the proposed method is excellent in classifying thyroid ultrasound images.