Texture analysis and classification in ultrasound medical images for determining echo pattern characteristics

Ultrasound is one of the imaging modalities commonly used for detecting mass abnormalities of nodule. The observation of ultrasound images is conducted by the radiologists, which tend to be subjective. Therefore, the use of computer aided diagnosis (CADx) system based on image processing can assist the radiologists to give more objective decision-making for detecting the mass abnormalities of nodule. This study proposes an approach to identify echo pattern characteristic of nodule by analysing some extracted texture features. A total of 343 ultrasound images consisting of 191 solid and 152 cystic nodules are used in this study. Three classifiers, namely Naïve Bayes, support vector machine (SVM) and multilayer perceptron (MLP) classifier are involved to measure the performance of proposed approach. Generally, MLP classifier achieves the best performance in classifying nodule with the accuracy of 93.00%, Kappa of 0.86 and AUC of 0.974. These results show that the proposed approach successfully identifies echo pattern characteristic of cystic and solid nodules on the ultrasound images.