Curvelet processing of MRI for local image enhancement

Magnetic resonance imaging provides very good contrast between different soft tissues; however, in some cases, this technique is not so suitable to image calcified structures like bones. The quality of images is often degraded by blur edges or noises, which makes it difficult to accurately identify bone structures. In this paper, we proposed a new curvelet preprocessing method for local image enhancement to especially improve the quality of spinal MRI. Our objective is to both sharpen boundaries and smoothen the intensity variation of the vertebra. In the first phase, we extract features through curvelet coefficients and the gradient of the original image, then we utilize fuzzy cluster method to classify the whole image scope into the 'edge' region and the 'nonedge' region. In the second phase, we locally sharpen or smoothen the image by adaptive adjustment of curvelet coefficients and Gaussian smoothing method in different subregions. To evaluate the effect of the preprocessing method, we examine the gradient of the image and its segmentation results as the assessments. The experiment results show that the feature extraction method is effective for classification and the vertebra performs higher contrast on boundaries and less noises after the enhancement, which indeed helps increase the accuracy of further segmentation.