A Novel Segmentation and Contouring Scheme to Assist Accurate Brain Lesion Classification

Abstract

Segmentation of highly infiltrating glioblastoma multiforme (GBM) in BrainMR Images has been highly challenging as the grey levels of tumor and peritumoral vasogenic edema are quite homogenous and hence identifying a suitable scheme for isolating GBM from the background has been troublesome. This paper proposes a novel segmentation and contouring scheme, using shape sensitive derivative strategy for segmentation and energy minimizing contours for enhancing the edges of the GBM, under investigation. The efficiency of the algorithm has been tested with the aid of extracted tumor features, the Shape Features -circularity, irregularity, Area, Perimeter, Shape Index, Intensity features – Mean, Variance, Standard Variance, Median Intensity, Skewness, and Kurtosis, Texture features –Contrast, Correlation, Entropy, Energy, Homogeneity, cluster shade, sum of square variance. It is obvious, though this algorithm consumes more computational time, it segments the edges effectively and preserve the shape facilitating accurate extraction or estimation of the features further and provides stable and reproducible results. All the classification Schemes, with combined DA and SVM with Higher Rank Features and LDA techniques exhibited appreciable improvement in terms of sensitivity, specificity, positive Predictive value and negative Predictive value, because of the accuracy of shape sensitive derivative segmentation algorithm and energy minimizing contour algorithm.