Skip to main content

Table 4 Comparison of diagnostic performance between the DL model and radiologist evaluation in the patient subgroups

From: Deep learning for differentiation of osteolytic osteosarcoma and giant cell tumor around the knee joint on radiographs: a multicenter study

Age

Model

Accuracy 95% CI)

p value

Age (< 21 years, n = 27)

DL

100 [27/27]

 

Expert committee -A

88.9 (71.9–96.2) [24/27]

0.25

Expert committee -B

96.3 (81.7–99.3) [26/27]

0.99

Expert committee

96.3 (81.7–99.3) [26/27]

0.99

Age (21–30 years, n = 30)

DL

96.7 (83.3–99.4) [29/30]

 

Expert committee -A

60.0 (42.3–75.4) [18/30]

0.003

Expert committee -B

83.3 (66.4–92.7) [25/30]

0.22

Expert committee -C

86.8 (70.3–94.7) [26/30]

0.38

Age (30–38 years, n = 31)

DL

90.3 (75.1–96.7) [28/31]

 

Expert committee -A

74.2 (56.8–86.3) [23/31]

0.13

Expert committee -B

83.9 (67.4–92.9) [26/31]

0.63

Expert committee -C

83.9 (67.4–92.9) [26/31]

0.63

Age (≥ 38 years, n = 28)

DL

85.7 (68.5–94.3) [24/28]

 

Expert committee -A

85.7 (68.5–94.3) [24/28]

0.99

Expert committee -B

92.9 (77.4–98.0) [26/28]

0.63

Expert committee -C

92.9 (77.4–98.0) [26/28]

0.63

  1. p value indicates significant differences in accuracy between expert committee and DL