[Morphometrical and statistical detection of cancer cells].

ABSTRACT

Morphometrical and multivariate statistical methods were applied to detection of carcinoma cells in cytologic studies. A total of 70 clumps of epithelial cells were selected from endometrial samples (10 adenocarcinomas, 4 hyperplasias and 56 normal controls), and their cytologic characters were reduced to a combination of five quantitative parameters (nuclear size, degree of anisokaryosis, nuclear from index, homogeneity of nuclear chromatin texture and regularity of cellular arrangement). The 5-variate cluster analysis demonstrated that the 70 clumps could be classified into three definite groups; A(17), B(41), and C(12). Group C, characterized by cells of large nuclear size, marked anisokaryosis, heterogeneous chromatin texture and irregular cellular arrangement, was derived from 10 samples of adenocarcinoma and 2 hyperplasia, while Groups A and B were not derived from any malignant samples. In conclusion, morphometrical-statistical classification can be of great aid in improving the cytodiagnostic validity and reproducibility.