Predicting microvascular invasion in hepatocellular carcinoma: a deep learning model validated across hospitals

Cancer Imaging

Table 5 Accuracy, sensitivity, specificity and AUC scores of MVI predictive models for training and validation sets

Model	Patient number (Train: Valid)		Training set				Validation set
Model	Patient number (Train: Valid)	Accuracy	Sensitivity	Specificity	AUC	Accuracy	Sensitivity	Specificity	AUC
SVM with LASSO in Ma et al. (AP)	157 (117:47)	0.72	0.62	0.77	0.70	0.68	0.61	0.72	0.68
SVM with LASSO in Ma et al. (AP + PVP + DP)	157 (117:47)	0.78	0.74	0.81	0.85	0.6	0.41	0.7	0.62
SVM with LASSO in Ma et al. (AP + PVP + DP + CF)	157 (117:47)	0.84	0.76	0.88	0.88	0.66	0.5	0.76	0.68
XGB in Jiang et al. (Radiological + radiomics + CF)	405	–	–	–	0.97	0.85	0.82	0.89	0.9
3D-CNN in Jiang et al. (AP + PVP + DP)	405	–	–	–	0.98	0.85	0.93	0.76	0.91
ResNet-18 in this study (AP)	309 (216:93)	0.95	0.91	0.97	0.98	0.68	0.96	0.56	0.82
ResNet-18 in this study (AP + CF)	309 (216:93)	0.97	0.94	0.98	0.97	0.72	0.96	0.62	0.85

Scores that were not reported in the study were represented by ‘-’. SVM support vector machine; LASSO least absolute shrinkage and selection operator; 3D-CNN three-dimensional convolutional neural network; XGB extreme gradient boosting; AP arterial phase; PVP portal venous phase; DP delay phase; CF clinical factors; AUC area under the curve.

ISSN: 1470-7330