Identification of Enhancing MS Lesions in MR Images using Non-Parametric Image Subtraction

MIUA 2002, Portsmouth, U.K.

Simple pixel-by-pixel image subtraction is widely used in image analysis to identify changes in image pairs. For example, multiple sclerosis (MS) produces lesions in the brain that can be detected by subtraction of MRI scans taken before and after the injection of GdDTPA contrast agent, which highlights the lesions. However, the result is returned in arbitrary units of pixel grey-level, with no statistically well-defined meaning. We describe a new, non-parametric subtraction measure, analogous to standard statistical tests, which allows regional data fusion and direct probabilistic interpretation of image differences. We demonstrate the technique using scans of MS lesions, but is expected to be applicable to a wide range of image formation processes.

pab_miua2002_paper.pdf (238kB)