New level set approach based on Parzen estimation for stroke segmentation in skull CT images

Abstract

Stroke is the second most common cause of death and one of the leading causes of disability in industrialized countries. Among the 17.5 million deaths caused by cardiovascular disease in 2012, approximately 6.7 million were caused by stroke. This study is focused on the hemorrhagic type of stroke, which accounts for 40% of all stroke deaths. This work proposes a new approach to segment the stroke from cranial CT images, in order to aid medical diagnosis. This approach proposes to automatically start the level set method within the stroke region and to use a nonparametric estimation approach based on the Parzen window to segment the stroke. The results obtained by the proposed approach are compared with the results of the level set algorithms using fuzzy C-means, and the level set based on the method of coherent propagation, fuzzy C-means, Watershed and Region Growth, which are commonly used in this field. The assessment is based on the validation of the segmentation from a radiologist. The experimental results showed that the proposed method presented a superior performance compared to the other commonly used methods, thus indicating that it is a promising tool for medical diagnosis. The results show that the proposed method has the highest mean of accuracy with 99.84% and lowest standard deviation of 0.08%, demonstrating that the proposed method is superior to the others in the literature. These results are confirmed by the high indexes of accuracy, sensitivity and specificity.