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

(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 256 × 256. 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.