Intensity standardization of longitudinal images using 4D clustering

Longitudinal magnetic resonance (MR) images of the same subject often vary significantly in their overall contrast. Intensity standardization aims to minimize the inter-scan intensity variations by transforming the intensities into a standard gray scale, but true anatomical changes over time are often masked out. We propose an intensity standardization method based on four dimensional Fuzzy C-means (FCM) clustering over longitudinal images. Assuming that the images in the longitudinal series of the same subject have been spatially aligned, our method tries to find for each image a piecewise linear intensity transformation function that minimizes the 4D energy function. The performance of our method is evaluated through the volume measurements of the tissue segmentation. Results show that our method can minimize the scanner induced intensity variation among longitudinal images, while preserving intensity variations caused by anatomical changes.