A virtual biopsy study of microsatellite instability in gastric cancer based on deep learning radiomics

Insights into Imaging

Table 3 Comparison of the prediction performance of three models for MSI status

Models	AUC (95%CI)	Sensitivity (95%CI)	Specificity (95%CI)
Clinical model
Training (n = 167)	0.725 (0.611–0.840)	0.683 (0.478–0.870)	0.775 (0.404–0.851)
Testing (n = 56)	0.738 (0.586–0.889)	0.553 (0.300–0.812)	0.812 (0.454–0.960)
Radiomics model
Training (n = 167)	0.856 (0.792–0.919)	0.957 (0.782–1.000)	0.681 (0.518–0.794)
Testing (n = 56)	0.753 (0.606–0.901)	0.625 (0.312–0.875)	0.750 (0.325–0.925)
Hybrid model
Training (n = 167)	0.883 (0.822–0.945)	0.870 (0.565–1.000)	0.723 (0.539–0.823)
Testing (n = 56)	0.802 (0.666–0.937)	0.688 (0.375–0.938)	0.775 (0.370–0.950)