13 - Comparative analysis of deformable models based segmentation methods for brain tumor classification

Abstract

In this chapter, the development and performance evaluation of brain tumors segmentation algorithms in magnetic resonance (MR) images using deformable models has been proposed and implemented using bench mark database. To calculate the efficiency of the segmentation methods presented and classification algorithms, three significant measurement metrics are considered. They are accuracy, sensitivity, and specificity. These techniques are experimented using BraTS and SimBRATS databases. The first proposed segmentation and classification algorithm uses the gradient vector flow (GVF) active contour model and K-means clustering approach with naive Bayes algorithm as classifier. The second proposed method is GVF active contour model based Fuzzy C-Means (FCM) segmentation algorithm along with fuzzy naive Bayes (FNB) classifier. The experimental result shows that the efficiency of the second approach is improved. To enhance the efficiency and also to make accurate decisions in diagnosis, convolutional neural network based deep learning method is used in the proposed approach. Hence, the earlier implementation is upgraded by fuzzy GVF deformable model with nature inspired optimization algorithms. These specifications are used along with deep convolutional neural network to classify brain tumor with Dolphin-SCA (Sine Cosine Algorithm) training algorithm. The experimental results of the proposed approach attained the sensitivity, specificity, and accuracy of 0.951, 0.902, and 0.978 for the BRATS database and 0.937, 0.906, and 0.915 of SimBRATS database, respectively. The outcome of final method shows the effectiveness in terms segmentation and accurate classification of tumors as compared to the previous techniques.