Classification of post contrast T1 weighted MRI brain images using support vector machine

In the field of clinical neuroscience, Magnetic Resonance Imaging (MRI) is amongst the most rapidly progressing diagnostic tools. However, the diagnostic accuracy is limited by the competency of the operating personnel to distinguish between physiological and pathological images. A computational system trained with various unhealthy and healthy brain images can help detect abnormalities in the 2D brain scans to aid in diagnosis. The pre-classification steps required are feature extraction and feature reduction. A Gaussian membership function was used for the former and principal component analysis for the latter. After testing the algorithm with various grid sizes, an optimum of 2×2 was chosen for the study. For classification, support vector machine with a quadratic kernel was used. An equal number of healthy and unhealthy post-contrast T1 weighted (radiological) images were used for training and test purposes (n=384). Accuracy as high as 87.5% was reported with this method. Therefore, this methodology can be applied effectively to classify MRI brain images.