Affiliations 

  • 1 Department of Computer Science and Engineering, Indian Institute of Information Technology, Design and Manufacturing, Kancheepuram 600127, India. Electronic address: drnayak@ieee.org
  • 2 Department of Computer Science and Engineering, National Institute of Technology Rourkela, 769008, India. Electronic address: bm@iiitdm.ac.in
  • 3 Department of Computer Science and Engineering, Indian Institute of Information Technology, Design and Manufacturing, Kancheepuram 600127, India. Electronic address: bmajhi@nitrkl.ac.in
  • 4 Department of Electronics and Computer Engineering, Ngee Ann Polytechnic, Singapore 599489, Singapore; Department of Biomedical Engineering, School of Science and Technology, SUSS University, Singapore 599491, Singapore; School of Medicine, Faculty of Health and Medical Sciences, Taylor's University, Malaysia. Electronic address: aru@np.edu.sg
Comput Med Imaging Graph, 2019 10;77:101656.
PMID: 31563069 DOI: 10.1016/j.compmedimag.2019.101656

Abstract

Binary classification of brain magnetic resonance (MR) images has made remarkable progress and many automated systems have been developed in the last decade. Multiclass classification of brain MR images is comparatively more challenging and has great clinical significance. Hence, it has recently become an active area of research in biomedical image processing. In this paper, an automated multiclass brain MR classification framework is proposed to categorize the MR images into five classes such as brain stroke, degenerative disease, infectious disease, brain tumor, and normal brain. A texture based feature descriptor is proposed using curvelet transform and Tsallis entropy to extract salient features from MR images. The potential of Tsallis entropy features is compared with Shannon entropy features. A kernel extension of random vector functional link network (KRVFL) is used to perform multiclass classification and improve the generalization performance at faster training speed. To validate the proposed method, two standard multiclass brain MR datasets (MD-1 and MD-2) are used. The proposed system obtained classification accuracies of 97.33% and 94.00% for MD-1 and MD-2 datasets respectively using 5-fold cross validation approach. The experimental results demonstrated the effectiveness of our system compared to the state-of-the-art schemes and hence, can be utilized as a supportive tool by physicians to verify their screening.

* Title and MeSH Headings from MEDLINE®/PubMed®, a database of the U.S. National Library of Medicine.