Figure 5
From: Deep learning-based survival prediction of oral cancer patients
Random survival forest model. 9 features were used to construct the model: T stage, N stage, histologic grade (HG), perineural invasion (PNI), extranodal extension (ENE), lymphovascular permeation (LVP), overall recurrence (OR), bone marrow invasion (BM), presence of tumor at resection margin (RM). (A) OOB error rates. (B) Estimated survival of testing set. (C) Variable importance plot. Higher VIMP indicates the variable contributes more to predictive accuracy. (D) Variable interaction plot. Lower values indicate higher interactivity, with the target variable marked in red. T stage and N stage show relatively higher interactions with other variables.
