Tracking tumor growth rates in patients with malignant gliomas: A test of two algorithms

BACKGROUND AND PURPOSE: Two 3D image analysis algorithms, nearest-neighbor tissue segmentation and surface modeling, were applied separately to serial MR images in patients with glioblastoma multiforme (GBM), Rates of volumet ric change were tracked for contrast-enhancing tumor tissue. Our purpose wa s to compare the two image analysis algorithms in their ability to track tu mor volume relative to a manually defined standard of reference. METHODS: Three-dimensional T2-weighted and contrast-enhanced T1-weighted sp oiled gradient-echo MR volumes were acquired in 10 patients with GBM. One o f two protocols was observed: 1) a nearest-neighbor algorithm, which used m anually determined or propagated tags and automatically segmented tissues i nto specific classes to determine tissue volume; or 2) a surface modeling a lgorithm, which used operator-defined contrast-enhancing boundaries to conv ert traced points into a parametric mesh model. Volumes were automatically calculated from the mesh models. Volumes determined by each algorithm were compared with the standard of reference, generated by manual segmentation o f contrast-enhancing tissue in each cross section of a scan. RESULTS: Nearest-neighbor algorithm enhancement volumes were highly correla ted with manually segmented volumes, as were growth rates, which were measu red in terms of halving and doubling times. Enhancement volumes generated b y the surface modeling algorithm were also highly correlated with the stand ard of reference, although growth rates were not. CONCLUSION: The nearest-neighbor tissue segmentation algorithm provides sig nificant power in quantifying tumor volume and in tracking growth rates of contrast-enhancing tissue in patients with GEM. The surface modeling algori thm is able to quantify tumor volume reliably as well.