Extended Data Fig. 3: Validation of the TEA-graph on the external NLST dataset. | Nature Biomedical Engineering

Extended Data Fig. 3: Validation of the TEA-graph on the external NLST dataset.

From: Derivation of prognostic contextual histopathological features from whole-slide images of tumours via graph deep learning

Extended Data Fig. 3: Validation of the TEA-graph on the external NLST dataset.

a, Kaplan-Meier survival analysis using the TEA-graph predicted-risk value (right) and the original stage (left). P-values were calculated through two-sided log-rank test (n = 445). b, Number of patches belong to low, mid, and high IG group for each risk group. IQR of box plot is between Q1 and Q3 and center line indicates median value. Maxima is Q3 + 1.5*IQR and minima is Q1 – 1.5*IQR (n = 378 for each risk group). c, Merged scatter plot between the risk value and the number of patches belonging to each IG group. d, Predicted risk heat map of NLST patients. Scale bar, 4 mm e, Risk-related contextual features predicted by the TEA-graph. Scale bar, 400 μm.

Back to article page