Identification of early apoptosis in feulgen-stained cultured cells in situ by computerized image analysis.

A method is presented in which Feulgen-stained apoptotic cells are identified among cultured cells in situ using computerized image analysis of their nuclei. Images of either control (untreated) human lymphoblastoid cell nuclei or of similar cells treated with 3mM methyl methanesulfonate for 1 h, which induces apoptosis in 100% of the cells, were converted to a standard size (80 x 80 pixels) and Fourier transforms and boundary images based on 21 proportional mean gray-level thresholds obtained. The perimeter and the mean fractal dimension of the latter and 39 selected coefficients of the former were then obtained from which data were chosen heuristically and subjected to multivariate linear discrimination analysis. Eighty-two percent of a teaching set of 50 nuclei and 69% of a test set of 29 nuclei were correctly identified as apoptotic versus nonapoptotic by the computer, compared with 57% identified correctly by a panel of 7 pathologists. This study shows the feasibility of using this type of analysis to directly identify apoptotic cells in culture, and probably also in tissues, by direct observation using computerized imaging technology.