Table 1 Accuracy of individual predicted labels
From: Deep learning to convert unstructured CT pulmonary angiography reports into structured reports
Predicted label | Number of statements | Accuracy by strict criteria | Accuracy by modified criteria | Problematic statements |
|---|---|---|---|---|
Cardiovascular | 840 | 805/840 (95.8%) | 815/840 (97.0%) | 23/840 (2.7%) |
Lines/tubes | 118 | 111/118 (94.1%) | 113/118 (95.8%) | 2/118 (1.7%) |
Lungs and airways | 821 | 717/821 (87.3%) | 768/821 (93.5%) | 68/821 (8.3%) |
Mediastinum and lymph nodes | 447 | 402/447 (89.9%) | 444/447 (99.3%) | 48/447 (10.7%) |
Pleura | 371 | 307/371 (82.7%) | 369/371 (99.5%) | 62/371 (16.8%) |
Pulmonary arteries | 502 | 485/502 (96.6%) | 487/502 (97.0%) | 21/502 (4.2%) |
Soft tissues and bones | 583 | 553/583 (94.8%) | 556/583 (95.4%) | 16/583 (2.7%) |
Upper abdomen | 475 | 426/475 (89.7%) | 434/475 (91.4%) | 34/475 (7.2%) |
Total | 4,157 | 3,806/4,157 (91.6%) | 3,986/4,157 (95.9%) | 274/4,157 (6.6%) |