Brain Tumor Classification from MRI Images Based on Cumulant Features

Abstract

Automatic classification of brain magnetic resonance (MR) images using artificial intelligence methods has an important role in medical diagnosis. Cumulants are powerful and efficient high-order statistics. In this paper, we propose an efficient cumulant-based method to determine the lesion of brain. In the proposed method, at first one-level two-dimensional discrete wavelet transform of brain MR image is calculated and then cumulants of each sub-band are calculated to obtain primary feature vector. Next, discriminative features from the primary features are extracted using linear discriminant analysis. Support vector machine classifier is utilized to determine whether the MR image is normal or suffers from one of seven different diseases. Moreover, in another classification problem, the MR image is classified as normal or abnormal. The results demonstrate that cumulants are very effective in classification of different brain MR images. The proposed algorithm achieves high correct classification rate in both classification problems. Comparison with the recently proposed algorithms demonstrates that the proposed method has better performance and lower computational complexity.