Brain tumor segmentation and classification via adaptive CLFAHE with hybrid classification

This article exploits a new brain tumor classification model that includes five steps like (a) denoising, (b) skull stripping, (c) segmentation, (d) feature extraction and (e) classification. Initially, the image is subjected under the denoising process, where the noise removal procedure is carried out by employing the entropy-based trilateral filter. Then, the denoised image is applied to the skull stripping process via Otsu thresholding and morphology segmentation. Subsequently, the next step is the segmentation, where the image is segmented by deploying the adaptive CLFAHE (contrast limited fuzzy adaptive histogram equalization) technique. Once the segmentation is completed, gray-level co-occurrence matrix (GLCM) based features are extracted. Finally, the extracted features are processed under hybrid classification model to attain enhanced classification rate. Here, hybrid classification hybrids two classifiers namely deep belief network (DBN) and Bayesian regularization classifier. The vital contribution of this research work exists in the optimal selection of hidden neurons in the DBN. Along with this, the membership function (bounding limits) of fuzzy logic is optimally selected. For this, a new lion exploration based whale optimization (LE-WO) algorithm is proposed in this article that hybrids the concept of (lion algorithm) LA and (whale optimization algorithm) WOA. Finally, the performance of proposed LE-WO is compared over the other methods in terms of accuracy, sensitivity, specificity, precision, negative predictive value (NPV), F1 _ score and Matthews correlation coefficient (MCC), False positive rate (FPR), False negative rate (FNR) and false discovery rate (FDR) and proves the betterments of proposed work. From the outcomes, the accuracy measure of proposed model at 60th population size is 1.98%, 1.81%, 1.32%, 3.46% and 0.75% better than PSO, FF, GWO, WOA and LA, respectively. Similarly, in 80th population size, the performance of the implemented model is 4.47%, 5.04%, 3.96%, 6.29% and 1.37% superior to PSO, FF, GWO, WOA and LA, respectively. Thus, the betterment of the adopted scheme is validated in an effective manner.