A non-probabilistic approach to the histogram method in detecting abnormalities in MRI images

This paper describes modifications to a model previously proposed (1993), which is used to detect abnormalities in magnetic resonance imaging (MRI) images. The symmetry property of the human brain is exploited as the two hemispheres of the brain are relatively symmetrical. This is realised in the difference operation where profiles of pixel intensities of two regions, occupying symmetrical positions within the image, are used. These two profiles are simultaneously fed to the neural network which then determines whether half of its inputs coming from the first region is similar to the rest of the inputs coming from the second region. The neural network is a single layer discriminator-based net using the principle of WISARD