Skip to main content

Table 2 Comparison of diagnostic performance between CLA-HDM and six radiologists, and between radiologists with and without AI assistance

From: Deep learning radiomics of dual-modality ultrasound images for hierarchical diagnosis of unexplained cervical lymphadenopathy

Radiologists Internal testing cohort (n = 171) External testing cohort 1 (n = 105) External testing cohort 2 (n = 92)
Without AI (%) With AI (%) Without AI (%) With AI (%) Without AI (%) With AI (%)
1 Accuracy 84.2 (81.3, 87.4) 86.8 (83.9, 89.8)↑# 82.9 (79.1, 86.7) 83.8 (80.0, 87.6)↑ 83.2 (78.8, 87.0) 82.6 (78.3, 86.4)
Sensitivity 68.4 (62.6, 74.9) 73.7 (67.8, 79.5)↑ 65.7 (58.1, 73.3) 67.6 (60.0, 75.2)↑ 66.3 (57.6, 73.9) 65.2 (56.5, 72.8)
Specificity 89.5 (87.5, 91.6) 91.2 (89.3, 93.2)↑ 88.6 (86.0, 91.1) 89.2 (86.7, 91.8)↑ 88.8 (85.9, 91.3) 88.4 (85.5, 90.9)
2 Accuracy 82.2 (79.5, 85.1) ** 85.7 (83.0, 88.3)↑## 80.5 (76.7, 84.8) 84.3 (80.5, 88.1)↑# 79.9 (76.1, 84.2) 82.6 (78.8, 87.0)↑
Sensitivity 64.3 (59.1, 70.2) * 71.4 (66.1, 76.6)↑## 60.9 (53.3, 69.5) 68.6 (61.0, 76.2)↑ 59.8 (52.2, 68.5) 65.2 (57.6, 73.9)↑
Specificity 88.1 (86.4, 90.1) 90.5 (88.7, 92.2)↑ 86.9 (84.4, 89.8) 89.5 (87.0, 92.1)↑ 86.6 (84.1, 89.5) 88.4 (85.9, 91.3)↑
3 Accuracy 81.0 (78.1, 83.9) ** 84.8 (82.2, 87.7)↑## 80.5 (76.2, 84.8) 80.5 (76.7, 84.3)↑ 79.4 (75.0, 83.7) 81.0 (76.6, 85.3)↑
Sensitivity 62.0 (56.1, 67.8) * 69.6 (64.3, 75.4)↑# 60.9 (52.4, 69.5) 61.0 (53.3, 68.6)↑ 58.7 (50.0, 67.4) 62.0 (53.3, 70.7)↑
Specificity 87.3 (85.4, 89.3) * 89.9 (88.1, 91.8)↑ 87.0 (84.1, 89.8) 87.0 (84.4, 89.5)↑ 86.2 (83.3, 89.1) 87.3 (84.4, 90.2)↑
4 Accuracy 81.0 (78.1, 84.2) ** 86.3 (83.6, 89.2)↑### 75.2 (71.4, 79.5) ** 81.0 (77.1, 84.8)↑## 77.7 (73.9, 82.2) 82.1 (78.3, 86.4)↑#
Sensitivity 62.0 (56.1, 68.4) * 72.5 (67.3, 78.4)↑### 50.5 (42.9, 59.1)* 61.9 (54.3, 69.5)↑# 55.4 (47.8, 64.1) * 64.1 (56.5, 72.8)↑
Specificity 87.3 (85.4, 89.5) * 90.8 (89.1, 92.8)↑ 83.5 (81.0, 86.4) 87.3 (84.8, 89.8)↑ 85.1 (82.6, 88.0) 88.0 (85.5, 90.9)↑
5 Accuracy 78.7 (75.4, 81.9) *** 79.8 (76.9, 83.0)↑ 77.1 (72.9, 81.0) * 82.4 (78.6, 86.7)↑# 76.6 (72.3, 81.0) 79.4 (75.5, 84.2)↑
Sensitivity 57.3 (50.9, 63.7) ** 59.7 (53.8, 66.1)↑ 54.3 (45.7, 61.9) 64.8 (57.1, 73.3)↑# 53.3 (44.6, 62.0) * 58.7 (51.1, 68.5)↑
Specificity 85.7 (83.6, 87.9) ** 86.6 (84.6, 88.7)↑ 84.8 (81.9, 87.3) 88.3 (85.7, 91.1)↑ 84.4 (81.5, 87.3) * 86.2 (83.7, 89.5)↑
6 Accuracy 76.9 (73.7, 80.1) *** 82.2 (78.9, 85.1)↑## 77.1 (73.3, 81.0) * 81.9 (78.1, 86.2)↑# 74.5 (70.1, 78.8) * 78.3 (73.9, 82.6)↑
Sensitivity 53.8 (47.4, 60.2) *** 64.3 (57.9, 70.2)↑## 54.3 (46.7, 61.9) 63.8 (56.2, 72.4)↑ 48.9 (40.2, 57.6) ** 56.5 (47.8, 65.2)↑
Specificity 84.6 (82.5, 86.7) ** 88.1 (85.9, 90.1)↑ 84.8 (82.2, 87.3) 87.9 (85.4, 90.8)↑ 83.0 (80.1, 85.9) ** 85.5 (82.6, 88.4)↑
  1. The data in brackets represent the 95% confidence intervals. * indicates a statistically significant difference between CLA-HDM and radiologist without AI assistance (*P < 0.05, **P < 0.01, and ***P < 0.001); # indicates a statistically significant difference between radiologist without and with CLA-HDM assistance (#P < 0.05, ##P < 0.01, and ###P < 0.001). The upward arrow (↑) represents indicators that improved owing to AI assistance