Texture quantification of medical images using a novel complex space-frequency transform

Objective To develop an efficient space-frequency transform for texture analysis and demonstrate its application on magnetic resonance (MR) images of multiple sclerosis (MS) patients. Materials and methods We applied our new transform to MR images of three enhancing lesions from two relapsing-remitting MS patients acquired serially over 9 months. Local spectra of the images were generated using our new technique, and spatial frequencies corresponding to MS lesion activity were extracted by applying a band-pass filter and inverting. We examined the changes in T2 intensity and low-frequency energy (LFE) over time within the lesion, surrounding tissue and a region of normal-appearing white matter (NAWM). Results We calculated complex local spectra of 428 x 428 images in approximately 1 min and achieved a spatial frequency resolution of 0.05 cm(-1). We observed an increase in LFE within the lesion and a drop in LFE in the hyperintense border of tissue surrounding the lesion. Conclusion We have developed an efficient, invertible transform that produces high-resolution local frequency spectra of an MR image in approximately 1 min. Negative LFE values in the boundaries of an active lesion may help discriminate between the core lesion undergoing demyelination and a border of inflammation.