Linear filter design for CNR enhancement of MR images with multiple interfering features

Maximizing the minimum absolute contrast-to-noise ratios (CNRs) between a desired feature and multiple interfering features, by linear combination of images in an MRI scene sequence, is attractive for medical image analysis and interpretation. Previously, we published a general formulation of the problem and a novel solution utilizing the simple and numerically stable method of Gram-Schmidt orthogonalization. Our approach derives explicit solutions for the general case of multiple interfering features in the scene. Here, we review the theoretical basis for the filter and present its applications to MRI scene sequences of the human brain with cerebral infarction, tumor, or multiple sclerosis. For these applications, a 20% to 100% improvement for the smallest absolute CNR is obtained.