A SOM algorithm based procedure for MRI image processing under significant Rician noise

Abstract

In this paper, a self-organizing map (SOM) based procedure is proposed for MRI image processing. Such images are usually corrupted by Rician noise generated during the image formation processing due to the nature of MRI technique. Rician noise is non-additive, signal dependent, and highly nonlinear, significantly different from those commonly discussed in images. These features make it very difficult to separate the signal from noise. A SOM algorithm is carefully applied to accurate MRI image processing by taking the decent Rician noise feature into consideration, resulting in a novel procedure for denosing and segmentation. The procedure is intuitively developed and justified, and demonstrated using simulation examples on a typical knee cartilage MRI image.