Radiology has traditionally relied upon qualitative assessment of tumor growth to guide therapeutic decision-making. The adoption of quantitative techniques such as WHO and RECIST guidelines in radiology has been slow due to multiple factors, including the workload and inter-operator variability associated with manual measurement, the complexity and pitfalls of automated techniques, and the many inherent sources of variability in medical imaging making longitudinal comparison difficult. We will develop, design and evaluate a semi- automated morphometrics system for quantification of tumors in the brain. The input to our system will be multiple images from magnetic resonance imaging (MRI) systems, particularly post-contrast T1 studies. The final product will be a modular extension to our FDA-cleared Prism View(tm) software package, compatible with MRI systems currently used for brain imaging. We will demonstrate feasibility in Phase I by developing means for geometrical registration and intensity standardization in imaging protocols, collecting data to document intra-patient and intra- and inter-reader variability, and evaluating our system in a study of a small number of subjects. In Phase II, we will further refine algorithms based upon Phase I findings and evaluate our system in a much larger cohort of subjects to more rigorously evaluate accuracy and precision in sizing of tumors. If successful, greater confidence will be achieved in following tumor changes over time, and this should translate into more rapid understanding of the efficacy of medical treatment or management of brain tumors. Our focus is on 2D and 3D tumor geometry, not on tumor composition. However, the existing product line (Prism) already facilitates combination and viewing of functional and physiologic information including MR spectroscopy and MR perfusion studies, and this information will be readily superposed upon the regions of interest delineated in anatomical imagery. ? ? ?
