Fig. 1: Comparison of AUC performance between CNN, visual transformer, and GlioMT. | npj Digital Medicine

Fig. 1: Comparison of AUC performance between CNN, visual transformer, and GlioMT.

From: Interpretable multimodal transformer for prediction of molecular subtypes and grades in adult-type diffuse gliomas

Fig. 1

Receiver operating characteristic (ROC) curves for IDH mutation status, 1p/19q codeletion status, and tumor grade prediction tasks on the (a) TCGA and (b) UCSF external validation sets. For tumor grade, which involves three classes (Grade 2, Grade 3, and Grade 4), ROC curves were generated using a one-vs-rest approach, followed by macro-averaging to produce the final curves. (c) AUC comparison for each task across both TCGA and UCSF validation sets. (d) AUC comparison for each grade (Grade 2, Grade 3, and Grade 4) using a one-vs-rest approach across the TCGA and UCSF validation sets. Error bar represents the 95% confidence interval (CI). To compare the statistical differences in AUCs, DeLong’s test was used for the binary classification tasks (IDH mutation and 1p/19q codeletion), and the bootstrapping method was employed for the multiclass classification task (tumor grade).

Back to article page