Brain tumor classification using modified kernel based softplus extreme learning machine

An uncontrollable growth of abnormal cells in the brain may result in brain tumor. Two different categories of brain tumor are benign and malignant. The doctors need to provide an efficient treatment for tumor affected patients, usually, the treatment process for both the types of tumors are different, as these two types may show diverse properties. Therefore it is necessary to accurately segment and classify the two types of brain tumor from MRI so that the doctors can provide proper treatment to each patient. For such segmentation and classification, a practical approach is introduced in this method. The tumor classification from MRI undergoes 4 different phases they are pre-processing, segmentation, feature extraction, and classification. During pre-processing, the Laplacian of Gaussian (LoG) and Contrast Limited Adaptive Histogram Equalization (CLAHE) is applied. Then, the features from the segmented image is extracted using three different extraction techniques. But sometimes the extracted features may found in large dimension with relevant and irrelevant features. To reduce that, an optimization based feature selection process is included before tumor classification phase. A kernel based Softplus extreme learning machine (KSELM) is used for classification. Finally, the experimental analysis is carried out with BRATS 2014, 2015, 2018, and BRT (Brain tumor) dataset. The performance metrics like accuracy, specificity, PPV, FNR, FPR, DSC, JSI, and sensitivity are determined. Different existing brain tumor classification techniques are compared with this proposed KSELM technique.