Machine learning based CAD system

Classifying brain Magnetic Resonance Images (MRI) as abnormal or healthy can be considered as the key for the preclinical state of a patient. In recent years, various methods were developed in this field. In this paper, a novel MRI classifier based on new Downsized Kernel Principal Component Analysis (DKPCA) and Artificial Neural Network (ANN) is presented. The proposed algorithm, called the DKPCA-ANN classifies brain MRIs as pathological or normal. The proposed study contains three main steps; Data acquisition and preprocessing stage, feature extraction and dimensionality reduction stage and finally the classification stage. Initially, to extract the image features, the scheme applied the Discrete Wavelet Transform (DWT). After feature vector normalization the DKPCA is applied to reduce features. The resulted matrix is used by the ANN classifier the predict the results. Seven common brain diseases have been used (Alzheimer's disease, glioma, meningioma, Huntington's disease, Alzheimer's disease plus visual agnosia, sarcoma and Pick's disease) as pathological brains. Brain MRIs were collected from the ‘Harvard Medical School’. The findings show that our scheme is robust and effective in comparison with other recent researches.