Skip to main content

View full-text article in PMC

. 2022 Mar 16;10(3):e28880. doi: 10.2196/28880

Table 2.

The performance of each framework.

Frameworks	Sensitivity (%), mean (range)	Specificity (%), mean (range)	Accuracy (%), mean (range)	Time cost (s)
RetinaNet [25]	89 (67.02-98.43)	80 (56.34-94.27)	84.50 (69.57-93.97)	0.24
Liao et al [15]	88 (65.76-98.06)	89 (67.02-98.43)	88.50 (74.44-96.39)	2.21
Duan et al [16]	87 (64.53-97.66)	86 (63.31-97.24)	86.50 (71.97-95.22)	0.33
Bi-input+RetinaNet+C-LSTM^a	89 (67.02-98.43)	93 (72.30-99.56)	91 (77.63-97.72)	2.72
Human experts	90 (68.30-98.77)	90 (68.30-98.77)	90 (76.34-97.21)	N/A^b

^aC-LSTM: convolutional long short-term memory.

^bN/A: not applicable.