Computer Simulations of Histologic Patterns in Melanoma Using a Cellular Automaton Provide Correlations with Prognosis

Computer simulations have been used frequently in the life sciences to investigate the mechanisms of morphologic pattern formation. The cellular automation programs SMN5 is designed to simulate tumor growth and to estimate biologic properties by comparing real tumor patterns with computer-simulated reference patterns. This method was applied to 195 cases of primary melanoma of the skin. S-100-stained sections were evaluated by image analysis and compared statistically to a references set of 4000 simulated patterns. Estimates of tumor cell proliferation, motility, cell loss, cohesion, stroma destruction, and intercellular signals (autocrine and paracrine factors affecting growth, motility, and cell loss) were calculated. Twelve of 18 estimated parameters correlated significantly with tumor progression, as indicated by vertical tumor thickness (linear regression analysis: p≤ 0.05), and 13 of 18 parameters carried prognostic significance (log rank test: p ≤ 0.05). Poor prognosis was associated particularly with a pronounced increase in the estimates of proliferation, tumor cell motility, and stromal degradation. Poor prognosis was also associated with a decrease in the estimates of cell loss, tumor cell cohesion, and paracrine growth factor dependence. In multivariate analysis using Cox's proportional hazard model, stromal degradation and motility showed prognostic information in addition to conventional prognostic parameters. The study shows that analytical comparison of real tumors with computer-simulated patterns of a cellular automaton facilitates a functional interpretation of tumor morphology, which carries prognostic significance in cutaneous melanoma.