An adaptive fuzzy K-nearest neighbor approach for MR brain tumor image classification using parameter free bat optimization algorithm

摘要

This paper presents an automatic diagnosis system for the tumor grade classification through magnetic resonance imaging (MRI). The diagnosis system involves a region of interest (ROI) delineation using intensity and edge magnitude based multilevel thresholding algorithm. Then the intensity and the texture attributes are extracted from the segregated ROI. Subsequently, a combined approach known as Fisher+ Parameter-Free BAT (PFreeBAT) optimization is employed to derive the optimal feature subset. Finally, a novel learning approach dubbed as PFree BAT enhanced fuzzy K-nearest neighbor (FKNN) is proposed by combining FKNN with PFree BAT for the classification of MR images into two categories: High and Low-Grade. In PFree BAT enhanced FKNN, the model parameters, i.e., neighborhood size k and the fuzzy strength parameter m are adaptively specified by the PFree BAT optimization approach. Integrating PFree BAT with FKNN enhances the classification capability of the FKNN. The diagnostic system is rigorously evaluated on four MR images datasets including images from BRATS 2012 database and the Harvard repository using classification performance metrics. The empirical results illustrate that the diagnostic system reached to ceiling level of accuracy on the test MR image dataset via 5-fold cross-validation mechanism. Additionally, the proposed PFree BAT enhanced FKNN is evaluated on the Parkinson dataset (PD) from the UCI repository having the pre-extracted feature space. The proposed PFree BAT enhanced FKNN reached to an average accuracy of 98% and 97.45%. with and without feature selection on PD dataset. Moreover, solely to contrast, the performance of the proposed PFree BAT enhanced FKNN with the existing FKNN variants the experimentations were also done on six other standard datasets from KEEL repository. The results indicate that the proposed learning strategy achieves the best value of accuracy in contrast to the existing FKNN variants.