A comparison of Covid-19 early detection between convolutional neural networks and radiologists

Insights into Imaging

Table 2 Results over the 836 test images for the different architectures and RandAugments parameters

Neural Net	N	M	AUC	Sensitivity (%)	Specificity (%)	TP	TN	FP	FN	Accuracy (%)
ResNet50	6	25	0.833	75	77	233	407	119	77	76.5
Densenet121	4	30	0.845	76	78	237	414	112	73	77.8
InceptionV3	4	30	0.830	80	73	250	382	144	60	75.6
InceptionResNetV2	4	30	0.822	74	76	229	400	126	81	75.2
Ensemble4Covid			0.856	78	81	242	425	101	68	79.8

The threshold over the scores used to obtain the sensitivity, specificity, true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN) shown in the table is the one that maximizes the Youden index