Applying the Logistic Regression Model to Predict the Stenosis in Carotid Artery Using the Sequential Color Doppler Ultrasound Image Processing

Early detection of stenosis in carotid artery is essential because it directly affects the patients'' clinical management and is of prognostic value. Therefore, estimating mechanical properties of this artery in normal and atherosclerosis cases is important as far as medical treatment is concerned. We applied a logistic regression model to predict carotid artery stenosis in a group of patients based on the quantitative features extracted from the processing of the conventional color Doppler ultrasound images.
Our database includes 128 patient records consisting 10 quantitative features. The database is then randomly divided into the training and validation samples including 98 and 30 patient records respectively. The training and validation samples are used to construct the logistic regression model and to validate its performance. Finally, important criteria such as sensitivity, specificity, accuracy and receiver operating characteristic curve (ROC) analysis for this method are evaluated.
Our results show that the logistic regression model is able to classify correctly 28 out of 30 cases presented in the validation sample. The output of this method showed a high positive predictive value of 94%.
We have established a logistic discriminator approach which is able to predict the probability of stenosis in the carotid artery using features extracted from ultrasonic measurements on ultrasound imaging .