Carotid Plaque Texture Feature Extraction In 3-D Ultrasound Images And Its Application To Atorvastatin Evaluation

In the past decades, the morbidity and mortality of cardiovascular and cerebrovascular diseases have consistently ranked high around the world. And one of the important pathological basis of cardiovascular and cerebrovascular diseases is carotid atherosclerosis. As an inexpensive, convenient and safe means of detection, ultrasound technology is applied widely in the prevention and treatment of carotid atherosclerosis. Recently, many studies have focused on how to quantitatively evaluate local arterial effects of medicine treatment for carotid diseases. This paper mainly using two-dimensional(2-D) and three-dimensional(3-D) features of 3-D carotid artery ultrasound images, to detect the potential changes of carotid plaque under the atorvastatin treatment.In this thesis, the database, supplied from Robarts Research Institute, is consist of 120 3-D carotid ultrasound images of 30 patients(13 atorvastatin groups and 17 placebo groups) at baseline and after 3 months of treatment. With the 2-D and 3-D features, including the features of first order statistics, grey level co-occurrence matrix, grey level difference matrix, grey level run length matrix, local ternary pattern and Laws texture energy, the evaluation is realized by the support vector machine(SVM) and 10-fold cross-validation protocol. The data are classified by 2-D features, and the optimal feature subset is obtained by the optimization strategy of principal component analysis and feature combination. The experiment results show that the performance of optimal feature subset have good recognition ability in CCA, ECA and ICA, with the accuracy over 96%, sensitivity around 90%, specificity around 98%. Moreover, the positive predictive value, negative predictive value, Matthew's correlation coefficient, Youden's index and AUC of the ROC prove the validity of the evaluation, which shows that under the atorvastatin treatment, the texture changes of images is significant.Based on the results of 2-D feature evaluation, 3-D extended features were applied in 3-D ultrasound images. Under the same pharmacodynamic evaluation model, the gray level run length matrix features in 3-D features set show a best discriminative for the CCA data after treatment, with the classification accuracy rate 97.50%, sensitivity 88.46%, specificity 100% and AUC 0.8695, proved the classification evaluation is valid. Also it shows that gray level run length matrix features can be evaluated as one of the specific indicators of the efficacy of the atorvastatin, which provides a reference for the medicine follow-up study in the future.