Nearest-neighbor classification for identification of aggressive versus nonaggressive low-grade astrocytic tumors by means of image cytometry-generated variables

Abstract

✓ The authors investigated whether cytometry-related variables generated by means of computer-assisted microscopic analysis of Feulgen-stained nuclei can contribute significant information toward the characterization of low-grade astrocytic tumor aggressiveness. This investigation was conducted using the nearest-neighbor rule (a traditional classification method used in pattern recognition) to analyze a series of 250 supratentorial astrocytic tumors from adult patients. This series included 39 low-grade astrocytomas and 211 high-grade astrocytic tumors (including 47 anaplastic astrocytomas and 164 glioblastomas multiforme [GBMs]). The results show that the 3-nearest-neighbors rule enabled a subgroup of “atypical” astrocytomas to be distinguished from the “typical” tumors. The atypical astrocytoma species exhibited a DNA content (DNA ploidy level) and morphonuclear characteristics that were statistically more similar to the characteristics of GBMs than to those exhibited by the typical astrocytomas. An analysis of survival data revealed that patients with atypical astrocytomas survived for a significantly shorter period (p < 0.001) than patients with typical lesions of this kind. In fact, patients with atypical astrocytomas had a survival period similar to that of patients with anaplastic astrocytomas, whereas patients with typical astrocytomas had a survival period significantly longer (p < 0.0001) than those associated with anaplastic astrocytomas and GBMs.