Unsupervised Brain MRI Tumor Segmentation with Deformation-Based Feature

Deformation-based features has been proven effective for enhancing brain tumor segmentation accuracy. In our previous work, a component for extracting features based on brain lateral ventricular (LaV) deformation has been proposed. By employing the extracted feature on classifiers of artificial neural networks (ANN) and support vector machines (SVM), we have demonstrated its effect for enhancing brain magnetic resonance (MR) image tumor segmentation accuracy with supervised segmentation methods. In this paper, we propose an unsupervised brain tumor segmentation system with the use of extracted brain LaV deformation feature. By modifying the LaV deformation feature component, deformation-based feature is combined with MR image features as input dataset for the unsupervised fuzzy c-means (FCM) to perform clustering. Experimental results shows the positive effect from the deformation-based feature on FCM-based unsupervised brain tumor segmentation accuracy.