Table 2.

Test results for predicting hematoma expansion in each model.

	Sensitivity	Specificity	Accuracy	AUC
CNN Model 1	1.000 (0.768–1.000)	0.163 (0.094–0.255)	0.274 (0.191–369)	0.582 (0.544–0.614)
CNN Model 2	1.000 (0.768–1.000)	0.511 (0.404–0.617)	0.575 (0.476–0.671)	0.755 (0.704–0.807)
Multimodal CNN Model 1	0.857 (0.572–0.982)	0.717 (0.614–0.806)	0.736 (0.641–0.817)	0.787 (0.682–0.893)
Multimodal CNN Model 2	1.000 (0.768–1.000)	0.598 (0.490–0.699)	0.615 (0.552–0.741)	0.799 (0.749–0.849)
Non-CNN Model 1	0.571 (0.289–0.823)	0.761 (0.661–0.844)	0.736 (0.641–0.817)	0.666 (0.525–0.808)
Non-CNN Model 2	0.643 (0.351–0.872)	0.696 (0.591–0.787)	0.689 (0.591–0.775)	0.669 (0.531–0.808)
Non-CNN Model 3	0.857 (0.572–0.982)	0.609 (0.501–0.709)	0.642 (0.543–0.732)	0.733 (0.625–0.840)
Non-CNN Model 4	0.786 (0.493–0.953)	0.761 (0.661–0.844)	0.764 (0.672–0.841)	0.773 (0.653–0.893)

Data are presented as value (95% confidence interval).

AUC = area under the receiver operating characteristic curve.