EVALUATION OF NO-N-UNIFORMITY CORRECTIONS FOR TUMOR RESPONSE MEASUREMENTS
Robert P. Velthuizen, Laurence P. Clarke, Hongbo Lin, Bruce B. Downs Blvd · 1997
Two popular non-uniformity techniques were evaluated for the effect on tumor response measurements (change in tumor volume over time): a phantom correction method and homomorphic filtering. No improvement in tumor segmentation was achieved, and the tumor response measurement was less accurate when using a correction. INTRODUCTION. Segmentation of magnetic resonance images (MRI) is important for quantification of normal tissues and pathology for areas such as Alzheimer's disease, multiple sclerosis and clinical management of brain tumor patients (l). In many cases, longitudinal studies are required to measure the effect of treatments, or to estimate changes over time as diagnostic indicators. The MRI sensor has various measurable imperfections, such as geometric distortions due to main field variations, radio-frequency inhomogeneities causing shading, and imperfect slice profiles ( 1,2). Moreover, several artifacts are influenced by the object being imaged, such as Fourier artifacts (Gibb's phenomenon), susceptibility artifacts and partial volume effects. In the literature on MRI segmentation, various correction methods have been proposed (3-91. Often individual artifacts are corrected with the assumption that other artifacts are negligible. For example, some attribute the smearing in feature space to partial volume effects only, while others attribute the same streaks in the feature space to image non-uniformity (Wl. Image non-unifomity corrections can be separated in two broad classes: methods based on the images themselves, and methods based on some external correction matrix. The latter includes phantom based correction methods (5) and calculated RF response characteristics (6). The methods that use the images themselves for the correction all assume that the non- uniformity is separable from the true signal, and that the non- uniformity is slow varying over the image (7-91. A more extensive review of non-uniformity correction methods can be found in (l). In this paper, we will evaluate two non-uniformity techniques for the measurement of change in tumor volumes as the patients are treated. The first is phantom based (5), the second is image based (7). The question of interest is: does a non-uniformity correction change the response assessment using serial MRIs?