Table 4.

External validation of trained AI model

Prediction target	Patient-based performance (n/N)
Undifferentiated histology
Accuracy (%)	75.6 (183/242)
Sensitivity (%)	81.7 (67/82)
Specificity (%)	72.5 (116/160)
PPV (%)	60.4 (67/111)
NPV (%)	88.6 (116/131)
Submucosal invasion
Accuracy (%)	71.9 (174/242)
Sensitivity (%)	53.3 (41/77)
Specificity (%)	80.6 (133/169)
PPV (%)	56.2 (41/73)
NPV (%)	78.2 (133/169)
Lymphovascular invasion
Accuracy (%)	88.8 (215/242)
Sensitivity (%)	31.8 (7/22)
Specificity (%)	94.5 (208/220)
PPV (%)	36.8 (7/19)
NPV (%)	93.3 (208/223)
Lymph node metastasis
Accuracy (%)	87.0 (131/153)
Sensitivity (%)	10.0 (2/20)
Specificity (%)	98.5 (129/133)
PPV (%)	50.0 (2/4)
NPV (%)	86.6 (129/149)

AI artificial intelligence, PPV positive prediction value, NPV negative prediction value, n number of correct predictions, N number of patients