Fig. 3: Comparison between IGNN and TACS1-8 models in external validation. | Nature Communications

Fig. 3: Comparison between IGNN and TACS1-8 models in external validation.

From: Intratumor graph neural network recovers hidden prognostic value of multi-biomarker spatial heterogeneity

Fig. 3

a Kaplan-Meier survival analysis of patients with the treatment guideline-derived low/moderate- and high-risk, re-stratified into high (red line) and low risk group (blue line) based on two prognostic models. A two-sided log-rank test was performed to determine significance, showing HR values with confidence intervals and exact p values. b Corresponding boxplots (left panels) showing TACS and IGNN scores for patients with DFS less and more than 5 years (in boxplots, middle line represents the median value, the upper and lower boundaries of boxes indicate 25th and 75th percentile, the whiskers reflect 1.5 times of interquartile range, the upper and lower tails indicate the maxima and minima, and single points indicate the distribution of values. A two-sided unpaired t-test was used to compare risk subgroups and determine significance from exact p values), and percentage histogram (right panels) showing survival distributions for model-based risk subgroups indicative of undertreatment and overtreatment. c, d Similar information obtained from 445 patients with a small (<2 cm) tumor. HR hazard ratio, DFS disease-free survival, TACS (TACS1-8) tumor-associated collagen signatures, IGNN intratumor graph neural network. Source data are provided as a Source Data file.

Back to article page