M.R. brain tumor classification employing ICA and kernel-based support vector machine
- PDF / 771,752 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 80 Downloads / 219 Views
ORIGINAL PAPER
M.R. brain tumor classification employing ICA and kernel-based support vector machine Rahul Singh1
· Aditya Goel1 · D. K. Raghuvanshi1
Received: 23 December 2019 / Revised: 14 June 2020 / Accepted: 25 August 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract A revolutionary and robust magnetic resonance (MR) brain tumor detection and segmentation approach has been presented in this work. We have put forward a robust technique to extract the tumor and classify the same as benign or malignant. The extraction of features from detected lesions is achieved through usage of wavelet transform. There is an immediate need to eliminate the redundant features from extracted subset because they degrade the performance of classification. Now, these reduced features are fed to kernel-based SVM (K-SVM). The kernel that is being utilized in this framework is Gaussian radial basis as it is quite efficient. Once the MR images are classified as benign or malignant, the next obvious step is to segment out the infected portion. We have relied upon region growing technique for the segmentation of the infected area. K-fold cross-validation approach has been adopted to optimize the performance K-SVM. We have compared the outcomes of our approach to that of various in-class conventional approaches. Experimental outcomes clearly suggest that our approach has performed efficiently and robustly for almost entire set of data and has performed way better when compared to existing in-class methodologies both qualitatively as well as quantitatively. Keywords Discrete wavelet transform (DWT) · Kernel support vector machine (K-SVM) · Independent component analysis (ICA) · K-fold cross-validation
1 Introduction The key objective of medical image classification is to classify and group the image pixels into various tissues classes on the basis of certain specified criteria. This objective is fulfilled by using trained pattern classifiers followed by requisite post-processing techniques such as morphological filtering. In tumor segmentation, the key goal is to extract out the tumor data from non-tumor data. Therefore, the multi-class classification can be named as bi-class classification because of treating both foresaid tumors as two different classes. Several techniques have already been implemented for the segmentation of brain tumor. Some of them are machine
B
Rahul Singh [email protected] Aditya Goel [email protected] D. K. Raghuvanshi [email protected]
1
Department of Electronics and Communication, Maulana Azad National Institute of Technology, Bhopal, M.P., India
learning, level-set method, fuzzy clustering and region growing. But, the most of these techniques depends manual intervention. Region growing technique needs seed point to be defined initially [1]. Region growing is immune to noise and it doesn’t depend on complex attribute tuning. In the recent past, new approaches have been put forward for automated selection of these seeds [2]. However, segmentation results will be
Data Loading...
