Detection and classification of brain tumor using hybrid feature extraction technique
Multimedia Tools and Applications
Accurate manual detection of brain tumor by a team of radiologists may be a long and tedious process, and further rely on their skills in the subject. Nowadays various medical imaging modalities are extensively used to minimize the above complexities and enable the patients to live a long and healthy life. This paper mainly focuses on the suspected patients of the brain tumor. A new method for feature extraction has been introduced and the framework for it has been briefed in the following steps. To begin with, a dual segmentation i.e. Fuzzy K-mean and Expectation-Maximization method has been performed consequently. The Ranklet Transformation+ along with Statistical Feature Analysis, named as hybrid feature extraction has been proposed. Further, two classification techniques have been presented by involving Auto-Encoder Neural Network in addition to Support Vector Machine classifier. Here, Auto-Encoder is trained by using extracted feature vectors, and the resultant is subsequently trained by Support vector machine. Further, testing is performed by Support vector machine classifier. The experiments have been performed by applying BraTS 2013 and BraTS 2015 MRI brain datasets. The performance has been evaluated using various statistical matrices such as Sensitivity, Accuracy, Specificity, F-measure and Matthews’s correlation coefficient and found that the suggested model attained the accuracy.
Open Access Status
This publication is not available as open access