IDENTIFICATION OF HIGH VERSUS LOWER RISK CLINICAL SUBGROUPS IN A GROUP OF ADULT PATIENTS WITH SUPRATENTORIAL ANAPLASTIC ASTROCYTOMAS

The present work investigates whether computer-assisted techniques can contribute any significant information to the characterization of ast rocytic tumor aggressiveness. Two complementary computer-assisted meth ods were used. The first method made use of the digital image analysis of Feulgen-stained nuclei, making it possible to compute 15 morphonuc lear and 8 nuclear DNA content-related (ploidy level) parameters. The second method enabled the most discriminatory parameters to be determi ned. This second method is the Decision Tree technique, which forms pa rt of the Supervised Learning Algorithms. These two techniques were ap plied to a series of 250 supratentorial astrocytic tumors of the adult . This series included 39 low-grade (astrocytomas, AST) and 211 high-g rade (47 anaplastic astrocytomas, ANA, and 164 glioblastomas, GBM) ast rocytic tumors. The results show that some AST, ANA and GBM did not fi t within simple logical rules. These ''complex'' cases were labeled NC -AST, NC-ANA and NC-GBM because they were ''non-classical'' (NC) with respect to their cytological features. An analysis of survival data re vealed that the patients with NC-GBM had the same survival period as p atients with GBM. In sharp contrast, patients with ANA survived signif icantly longer than patients with NC-ANA. In fact, the patients with A NA had the same survival period as patients who died from AST, while t he patients with NC-ANA had a survival period similar to those with GB M. All these data show that the computer-assisted techniques used in t his study can actually provide the pathologist with significant inform ation on the characterization of astrocytic tumor aggressiveness.