Brain Tumor Classification: Feature Fusion

Brain tumor detection is a challenging task in medical image analysis. The manual process performs through domain specialists is a more time-consuming task. Numerous works are represented for brain tumor detection and discrimination but still, there is a need for a fast and efficient technique. In this article, a technique is presented to distinguish between benign and malignant tumor. Our method integrates the enhancement of the lesion, features extraction, selection, and classification methods. In-lesion enhancement phase, the N4ITK3 method is applied to normalize the input image. A serial fusion based technique is used for feature extraction and fusion by using HOG (shape), SFTA and LBP (texture) based features. The fused features are selected through the Boltzmann entropy technique. The fused feature vector is supplied to the multiple classifiers to compare the better prediction rate. The three publically available BRATS databases are utilized for tumor detection. The method is trained separately on benchmark databases. The performance outcomes demonstrate that the mean dice similarity coefficient (DSC) is 0.99 for tumor estimation. The performance outcome illustrates that the proposed model effectively classifies the abnormal and normal brain region. The comparison performed with previous techniques proves the effectiveness of the presented technique.