Comparison of methods for response analysis of central nervous system neoplasms

The primary objective of this study is to determine whether the conventionally utilized 'gold standard' methodology of bidimensional product is reproducible for the purpose of ascertaining response to treatment in CNS neoplasms. Secondary objectives include testing of volumetric methods for response analysis and examining the variability due to tumor pathology, observer, and other factors. Twenty-five patients with brain metastases (evaluated by serial CT) and 16 patients with glioblastoma multiforme (evaluated by serial MRI) were the subjects of this study. The images were reviewed independently by 5 observers and analyzed subjectively, bidimensionally, tridimensionally, and volumetrically, using an NIH-developed software tool, NIH Image. Volume proved to be the most precise measurement technique. Interobserver reproducibility for volume measurements was superior to that of all other methods tested, with the range of correlation coefficients being: bidimensional product = .88-.98, area = .94-.99, tridimensional product = .75-.99, and volume = .96-1.00. Misclassification in tumor response rates was lowest for volume measurements (mean = .23, range = .20-.27) and highest for bidimensional product (mean = .39, range = .34-.44). Volume derived from CT or MRI is a more precise measure of tumor size, yielding consistently lower misclassification rates than traditional bidimensional product or cross-sectional area. This superiority holds irrespective of pathology, tumor size, or observer.