Fig. 3

Survival prediction results. a-k Kaplan-Meier plots for TIP and TMEC features that result in patient stratification into two groups: with high and low values of the feature. Only features with statistically significant differences in patient survival are shown, as measured using the log rank test and the Benjamini-Hochberg procedure (p-values and critical values c in the top right corner, significance confirmed if p < 0.05 or p_r < c, where p_r is a p-value ranked in ascending order). For the latter, we set the False Discovery Rate at 0.1 and included all TIP and TMEC features. The cutoff value ρ (lower left corner) indicates the selected threshold yielding patient strata with high and low values of the feature. The results correlate with previous studies of the relationship between these features and patient survival. l c-index scores for Cox models from survival prediction experiments performed with different feature sets. The best results were obtained for models with such feature sets that included TIP and TMEC features. m Hazard ratios for the best model that utilized the TIP features. The prevalence of the necrosis tissue class in the whole slide has a statistically significant negative effect on survival. n Hazard ratios for the best model that utilized the TMEC features. The presence of the necrosis tissue class and the vessel tissue class in the TME has a statistically significant negative effect on survival