Convolutional neural network applied to preoperative venous-phase CT images predicts risk category in patients with gastric gastrointestinal stromal tumors

Table 3 Diagnostic performance of CNN in risk tri-rating of GSTs in both training and validation cohorts

	Best epoch Acc	Kappa coefficient	Micro average roc	Macro average roc	Roc curve of class 0	Roc curve of class 1	Roc curve of class 2
Training cohort (297)
CNN_layer3	0.7864	0.6512	0.88	0.87	0.90	0.79	0.91
CNN_layer9	0.7835	0.7114	0.90	0.88	0.91	0.82	0.91
CNN_layer15	0.7806	0.7171	0.90	0.88	0.91	0.82	0.91
Validation cohort (128)
CNN_layer3	0.7891	0.6398	0.88	0.86	0.89	0.82	0.86
CNN_layer9	0.7891	0.6902	0.89	0.87	0.90	0.83	0.86
CNN_layer15	0.7891	0.7211	0.89	0.86	0.90	0.83	0.85

ISSN: 1471-2407