Dual-Tree Complex Wavelet Transform and Twin Support Vector Machine for Pathological Brain Detection

Abstract

(Aim) Classification of brain images as pathological or healthy case is a key pre-clinical step for potential patients. Manual classification is irreproducible and unreliable. In this study, we aim to develop an automatic classification system of brain images in magnetic resonance imaging (MRI). (Method) Three datasets were downloaded from the Internet. Those images are of T2-weighted along axial plane with size of 256ˆ256. We utilized an s-level decomposition on the basis of dual-tree complex wavelet transform (DTCWT), in order to obtain 12s "variance and entropy (VE)" features from each subband. Afterwards, we used support vector machine (SVM) and its two variants: the generalized eigenvalue proximal SVM (GEPSVM) and the twin SVM (TSVM), as the classifiers. In all, we proposed three novel approaches: DTCWT + VE + SVM, DTCWT + VE + GEPSVM, and DTCWT + VE + TSVM. (Results) The results showed that our "DTCWT + VE + TSVM" obtained an average accuracy of 99.57%, which was not only better than the two other proposed methods, but also superior to 12 state-of-the-art approaches. In addition, parameter estimation showed the classification accuracy achieved the largest when the decomposition level s was assigned with a value of 1. Further, we used 100 slices from real subjects, and we found our proposed method was superior to human reports from neuroradiologists. (Conclusions) This proposed system is effective and feasible.

Machine Learning Methods for Intelligent Abnormal Brain Identification

“…Yang (2016) [19] described three classifiers and a comparative study on them is reported. At the first stage, applying DTCWT to transform wavelets in order to improve the weakness of DWT.…”

Haar Wavelet Transform Based Facial Emotion Recognition

“…The SVM can analyze and classify data into two categories [16][17][18][19][20][21]. The SVM can generate a pair of hyperplanes, which separate the data of two different classes.…”

Teeth Category Classification via Hu Moment Invariant and Extreme Learning Machine

“…Currently there are numerous feature extraction methods, for example, Fourier transform [13][14][15][16], wavelet analysis [17][18][19], and so on. Moment invariant was proposed by Hu in 1962, with invariant character for translation, rotation and scale, and was widely applied in pattern recognition.…”