Table 5.

Results of all experiments from the deep learning approach for predicting good reperfusion (post-eTICI ≥ 2b).

Methods	AUC	F1-Score	Sensitivity	Specificity	PPV	NPV
Clinical experiment
Feed forward	0.53 (0.50–0.55)	0.57 (0.54–0.61)	0.51 (0.48–0.54)	0.53 (0.52–0.55)	0.65 (0.59–0.71)	0.38 (0.32–0.43)
Combination experiment
ResNet10 from scratch	0.50 (0.50–0.52)	0.13 (0.00–0.56)	0.12 (0.00–0.51)	0.87 (0.46–1.00)	0.15 (0.00–0.62)	0.36 (0.31–0.40)
Combination experiment
ResNet10 transfer learning	0.61 (0.50–0.72)	0.63 (0.54–0.71)	0.57 (0.50–0.64)	0.57 (0.50–0.64)	0.69 (0.60–0.80)	0.43 (0.40–0.45)

Average over 5-fold cross-validation.