3D convolutional neural networks applied to CT angiography in the detection of acute ischemic stroke

European Radiology Experimental

Table 2 Total number of infarct positive brain regions in 15 test subjects in group A and 15 non-infarcted test subjects in group B compared with radiologist evaluation

CNN features	Group A							Group B	Group A + group B
CNN features	TN	TP	FN	FP	DSC	Sensitivity	Specificity	FP	DSC
CTA	78	99	6	117	0.62	0.94	0.40	92	0.48
CTA + hemispheric comparison	158	98	7	37	0.82	0.93	0.81	46	0.69
CTA + hemispheric comparison + non-contrast CT	159	98	7	36	0.82	0.93	0.82	34	0.72

The number of true negative (TN), true positive (TP), false negative (FN), and false positive (FP) regions were used to calculate sensitivity specificity and Dice similarity coefficient (DSC). Convolution neural networks CNNs were evaluated for computed tomography angiography (CTA) alone, for CTA plus hemispheric comparison, and for CTA plus hemispheric comparison plus non-contrast CT. For the regions (M1–6, I, L, C, IC on both hemispheres), see text