Table 2.

DL model performance on the CQ500 test dataset at two selected operating points

	Accuracy	Sensitivity	Specificity	Positive likelihood ratio	Negative likelihood ratio
High-sensitivity operating point
Intracranial haemorrhage	0.88 (0.84–0.90)	0.95 (0.91–0.97)	0.81 (0.76–0.86)	5.100	0.061
EDH	0.90 (0.87–0.93)	1.00 (0.54–1.00)	0.90 (0.87–0.93)	10.173	0.000
ICH	0.87 (0.83–0.90)	0.99 (0.96–1.00)	0.83 (0.78–0.86)	5.700	0.009
IVH	0.89 (0.85–0.91)	1.00 (0.91–1.00)	0.88 (0.84–0.91)	8.051	0.000
SAH	0.79 (0.75–0.82)	0.96 (0.90–0.99)	0.74 (0.69–0.78)	3.622	0.058
SDH	0.77 (0.73–0.81)	0.94 (0.87–0.98)	0.73 (0.68–0.77)	3.497	0.078
Balanced operating point
Intracranial haemorrhage	0.91 (0.88–0.94)	0.88 (0.83–0.92)	0.94 (0.91–0.97)	15.400	0.129
EDH	0.90 (0.87–0.93)	1.00 (0.54–1.00)	0.90 (0.87–0.93)	10.173	0.000
ICH	0.94 (0.91–0.96)	0.95 (0.90–0.98)	0.93 (0.90–0.96)	14.341	0.053
IVH	0.96 (0.94–0.98)	0.95 (0.83–0.99)	0.96 (0.94–0.98)	26.349	0.051
SAH	0.86 (0.82–0.89)	0.87 (0.80–0.93)	0.85 (0.81–0.89)	5.927	0.150
SDH	0.93 (0.90–0.95)	0.90 (0.81–0.95)	0.93 (0.90–0.95)	13.185	0.111

95% CIs are provided in parentheses for accuracy, sensitivity and specificity

The greater the positive likelihood ratio, the greater the effect on the post-test probability of disease given a positive test result (0–5: slight increase, 5–10: moderate increase, > 10: large increase)

The smaller the negative likelihood ratio, the greater the effect on the post-test probability of disease given a negative test result (0.5–1: slight decrease, 0.2–0.5: moderate decrease, 0–0.1: large decrease)

DL Deep learning, EDH Extradural haemorrhage, ICH Intracerebral haemorrhage, IVH Intraventricular haemorrhage, SAH Subarachnoid haemorrhage, SDH Subdural haemorrhage