Information measures-based intensity standardization of MRI

Scan-to-scan intensity variation, even with the same imaging modality, affects a number of intensity-based image processing methods such as feature map based segmentation and non-rigid registration techniques that minimize sum of squared differences (SSD). Current intensity standardization techniques based on either percentile alignment or polynomial mapping suffer from a number of limitations. We present a novel intensity standardization techniques that exploits information measures obtained from the images. A probability similarity measure obtained by using polynomial mapping with Kullback-Leibler (KL) divergence is used for intensity standardization of pair-wise magnetic resonance (MR) images. For standardization of group-wise MR images, polynomial mapping with minimum entropy as a group probability similarity measure is used for attaining standardization in a group to attain common feature without bias. Our method is more flexible, particularly in mapping high intensity regions, such as lesions, since it does not set any hard limit. The mappings were realized through optimization of cost functions with Powell's search. The performance of the proposed method is demonstrated for non-rigid registration and feature map-based image segmentation of MR brain images.