BMC Cancer

Table 2 Predictive performance of CNN in both training and validation cohort

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

	CNN_layer3			CNN_layer9			CNN_layer15
Training cohort(297)	0 (154)	1 (84)	2 (59)	0 (154)	1 (84)	2 (59)	0 (154)	1 (84)	2 (59)
Predicted number	148	34	8	146	50	35	137	56	34
AUC	0.90	0.79	0.91	0.91	0.82	0.91	0.91	0.82	0.91
Sensitivity	0.961	0.4047	0.5253	0.948	0.5952	0.5953	0.8896	0.6667	0.5762
Specificity	0.6293	0.9014	0.9579	0.7552	0.9061	0.9537	0.8041	0.8450	0.9015
True positive	0.9610	0.4047	0.5254	0.9480	0.5952	0.5953	0.8896	0.6667	0.5762
False negative	0.0390	0.6953	0.4746	0.0520	0.4048	0.4047	0.1104	0.3333	0.4238
Validation cohort(128)	0 (67)	1 (35)	2 (26)	0 (67)	1 (35)	2 (26)	0 (67)	1 (35)	2 (26)
Predicted number	66	24	11	63	25	13	61	28	12
AUC	0.89	0.82	0.86	0.90	0.83	0.86	0.90	0.83	0.85
Sensitivity	0.985	0.6857	0.423	0.9402	0.7142	0.50	0.9104	0.8	0.4615
Specificity	0.6885	0.9139	1.0	0.7704	0.8817	0.9803	0.8360	0.8279	0.99
True positive	0.9850	0.6857	0.4230	0.9402	0.7142	0.5000	0.9104	0.800	0.4615
False negative	0.0150	0.3143	0.5770	0.0598	0.2858	0.5000	0.0896	0.2000	0.5385

Note.—Except where indicated, data in parentheses are numbers of tumors

Back to article page

ISSN: 1471-2407

Contact us

Submission enquiries: bmccancer@biomedcentral.com
General enquiries: ORSupport@springernature.com