Automated image analysis of gliomas an objective and reproducible method for tumor grading

Publisher Website
Google Scholar
Abstract
Summary A system of automated microscopic picture analysis was used in an examination of 272 gliomas (70 glioblastomas, 91 astrocytomas, 56 pilocytic astrocytomas or spongioblastomas, and 55 oligodendrogliomas). The specimens were prepared as Feulgen sections, 4μm in thickness. Thirteen morphometricdensitometric parameters of tumor cell nuclei were tested together with two mitotic parameters. Objective and reproducible data on numerical nuclear density (KRNZ, AREA), nuclear size (KOFL, KFRL, P250), nuclear shape (FOFK, FOFR, P150), optical density (EXTU, EXTS, EXSR, EXTM, EXMR), and mitotic activity (MITZ, VHMK) of the gliomas were obtained from the morphometric-densitometric parameters. All gliomas but glioblastomas were subdivided by four tumor grades. The morphometric-densitometric and mitotic data recorded were statistically checked, depending on tumor grade (Student'st-test, Wilcoxon's test, α=0.05). Numerical nuclear density, deformation of nuclei, and mitotic activity were found to grow with significance along with increasing tumor grade up to glioblastoma. The relative standard deviation (SD) of nuclear size (KFRL), relative SD of shape factors (FOFR), and relative SD of extinction sums (EXSR) are high-accuracy parameters for the pathologist to describe variability of sizes, polymorphism, and polychromasia of nuclei. These parameters show a significant increase of values in parallel with rising tumor grades, with maximum values being recordable from cases of glioblastomas. In cases of astrocytomas, optical values of nuclei decrease along with rising tumor grade. The data thus obtained were used as reference values for objective, reproducibel automatic glioma grading. The classifier method, described in an earlier publication, proved to be more effective than the regression method.