Cascade and Fusion of Multitask Convolutional Neural Networks for Detection of Thyroid Nodules in Contrast-Enhanced CT

Table 3

Performance comparison of different methods (percent).

Method	Accuracy	Recall	Precision	Specificity	F1 score	AUC	CI
CNN-F	95.73	87.14	98.10	99.30	92.30	98.49 (0.0029)	[97.91 99.06]
CNN-1	94.98	88.66	93.91	97.61	91.21	97.83 (0.0033)	[97.18 98.49]
CNN-2	93.14	84.10	91.87	96.90	87.81	95.63 (0.0052)	[94.60 96.66]
ResNet50	93.28	82.20	94.17	97.89	87.78	97.41 (0.003)	[97.26 98.28]
InceptionV3	93.78	87.29	91.15	96.48	89.18	97.31 (0.0054)	[96.85 98.46]
VGG-16	92.79	83.90	90.83	96.48	87.22	96.82 (0.0046)	[96.33 97.31]
Peng et al. [11]	88.80	82.10	91.70	93.30	—	95.30	—
Liu et al. [12]	86.73	91.30	82.35	82.96	—	91.05	—

Number format of AUC: mean (standard deviation).