Application of multiresolution analysis for automated detection of brain abnormality using MR images: A comparative study.

Neurological disorders are abnormalities related to the human nervous system, and comprise electrical, biochemical, or structural changes in the spinal cord, brain, or central nervous system that induce various symptoms. These symptoms may be in the form of muscle weakness, paralysis, and poor coordination, among other factors. Early diagnosis of these changes is important for treatment, to limit disease progression. Magnetic resonance (MR) imaging is a widely used modality for diagnosing brain abnormality. Expert reading and interpretation of MR images is time-consuming, tedious, and subject to interobserver variability. Hence, various automated computer aided diagnosis (CAD) tools have been developed to detect brain abnormalities from MR imaging. Multiresolution analysis involves the transformation of images to capture obscure signatures. In this paper, we compare the performance of three different multi-resolution analysis techniques – the discrete wavelet transform, curvelet transform and shearlet transform – for detecting brain abnormality. Further, textural features extracted from the transformed image are optimally selected using particle swarm optimization (PSO), and classified using a support vector machine (SVM). The proposed method is applied on 83 control images, as well as 529 abnormal images from patients with cerebrovascular, neoplastic, degenerative and inflammatory diseases. For quantitative analysis, a cross validation scheme is implemented to improve system generality. Among the three techniques, the shearlet transform achieves a highest classification accuracy of 97.38% using only fifteen optimally selected features. The proposed system requires testing on a large data set prior to implementation as a standalone system to assist neurologists and radiologists in the early detection of brain abnormality.