Histological heterogeneity of human glioblastomas investigated with an unsupervised neural network (SOM).

摘要: Summary. The histological variability of Glioblastomas (GB) precludes the modern assimilation theses tumors into a single tumor group. As an alternative to statistical evaluation, we investigated 1489 human GB in order discover whether they could be correctly classified using Self-Organizing Maps (SOM). In all 50 features, as well age and sex patients, were examined. Four clusters with significance 52 (maximal 60) found. Cluster C1 contained 37.47% 41.09% polymorphic glioblastomas (PG). C2 included 35.06% 44.96% giant cell (GCG). C3 16.45% significant component astroblasts, glioblasts oligodendroglia. C4 11.01% GB, 87.80% gliosarcomas (GS) 36.72% GCG. Placing series windows their maps side by allows immediate recognition dependencies on variables determination necessary build specific clusters. SOM allow realistic classification, comparable actual classification WHO. addition, found new, small subclusters which may have clinical significance. With one can learn discriminate, discard delete data, select or genetic that are meaningful, consequently influence result patient management.