Skip to main content

Table 1 Performance of sub-models and CLA-HDM in the diagnosis of unexplained CLA

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

Models Cohorts AUC ACC (%) SENS (%) SPEC (%)
Sub-model 1 Training cohort (n = 395) 0.986 (0.977, 0.998) 96.8 (95.2, 98.4) 97.9 (96.7, 99.7) 94.4 (91.1, 98.0)
Internal testing cohort (n = 171) 0.932 (0.901, 0.966) 86.0 (81.9, 90.1) 89.5 (85.2, 94.5) 78.9 (70.2, 88.3)
External testing cohort 1 (n = 105) 0.963 (0.939, 0.993) 87.6 (82.9, 93.3) 83.3 (76.7, 90.9) 96.9 (93.9, 103.0)
External testing cohort 2 (n = 92) 0.896 (0.846, 0.963) 82.6 (76.1, 90.2) 81.8 (73.4, 90.9) 83.8 (74.0, 95.2)
Sub-model 2 Training cohort (n = 136) 0.935 (0.902, 0.976) 86.3 (81.5, 91.9) 84.6 (78.7, 91.1) 90.9 (81.8, 100.0)
Internal testing cohort (n = 57) 0.922 (0.866, 0.986) 84.2 (77.2, 91.2) 85.7 (76.8, 94.8) 80.0 (65.6, 97.5)
External testing cohort 1 (n = 33) 0.857 (0.758, 0.981) 75.8 (63.6, 87.9) 76.2 (61.9, 93.6) 75.0 (56.7, 96.2)
External testing cohort 2 (n = 37) 0.872 (0.771, 0.986) 78.4 (67.6, 89.2) 71.4 (54.9, 87.9) 87.5 (75.0, 102.8)
Sub-model 3 Training cohort (n = 259) 0.979 (0.96, 1.012) 93.2 (90.8, 96.0) 92.6 (89.9, 95.6) 96.9 (93.8, 102.6)
Internal testing cohort (n = 114) 0.852 (0.759, 0.968) 86.0 (80.7, 91.2) 87.9 (82.8, 93.4) 73.3 (55.0, 93.3)
External testing cohort 1 (n = 72) 0.847 (0.742, 0.969) 86.1 (79.2, 93.1) 88.7 (82.2, 95.8) 70.0 (40.0, 97.1)
External testing cohort 2 (n = 55) 0.827 (0.715, 0.964) 83.6 (76.4, 92.7) 87.2 (78.9, 95.6) 62.5 (35.0, 91.7)
CLA-HDM Training cohort (n = 395) 0.964 (0.951, 0.978) 94.1 (92.6, 95.6) 88.2 (85.3, 91.2) 96.1 (95.1, 97.1)
Internal testing cohort (n = 171) 0.873 (0.838, 0.908) 87.1 (84.2, 90.1) 74.3 (68.4, 80.1) 91.4 (89.5, 93.4)
External testing cohort 1 (n = 105) 0.837 (0.789, 0.889) 82.9 (78.6, 86.7) 65.7 (57.1, 73.3) 88.6 (85.7, 91.1)
External testing cohort 2 (n = 92) 0.840 (0.789, 0.898) 85.9 (82.1, 89.7) 71.7 (64.1, 79.3) 90.6 (88.0, 93.1)
  1. The data in brackets represent the 95% confidence intervals