Table 2 Video-level performance of coronary angiography clinical reports and DeepCoro against expert annotated videos present in Dataset B
From: Evaluation of stenoses using AI video models applied to coronary angiography
Task | Metric | Coronary artery | |||||
|---|---|---|---|---|---|---|---|
LCA | RCA | RCA + LCA | |||||
Clinical reports | DeepCoro | Clinical reports | DeepCoro | Clinical reports | DeepCoro | ||
Number of videos | 475 | 490 | 965 | ||||
Number of severe stenoses, ≥70% (n (%)) | 105 (22%) | 92 (19%) | 197 (20%) | ||||
Number of non-severe stenoses, <70% (n (%)) | 370 (78%) | 398 (81%) | 768 (80%) | ||||
Number of healthy vessels, 0% stenoses (n (%)) | 335 (71%) | 322 (66%) | 657 (68%) | ||||
Classification | AUROC | 0.7430 (0.7136–0.7735) | 0.8781 (0.8608–0.8965) | 0.7649 (0.7359–0.7959) | 0.8599 (0.8386–0.8867) | 0.7533 (0.7328–0.7744) | 0.8699 (0.8564–0.8853) |
AUPRC | 0.4908 (0.4400–0.5369) | 0.7066 (0.6659–0.7494) | 0.4684 (0.4168–0.5213) | 0.7040 (0.6592–0.7483) | 0.4737 (0.4382–0.5086) | 0.7042 (0.6717–0.7339) | |
Sensitivity (%) | 46.63 (41.57–51.76) | 79.91 (76.54–83.95) | 47.88 (42.31–52.78) | 77.20 (72.86–81.94) | 47.30 (43.95–50.64) | 78.74 (76.05–81.70) | |
Specificity (%) | 91.87 (90.57–93.27) | 77.05 (75.00–79.12) | 93.73 (92.63–94.94) | 77.40 (75.39–79.50) | 92.81 (91.95–93.77) | 77.20 (75.78–78.60) | |
Regression | MAE (%) | 22.28 (21.18–23.39) | 19.76 (19.02–20.53) | 19.79 (18.74–20.84) | 18.45 (17.75–19.15) | 21.00 (20.20–21.76) | 19.09 (18.55–19.58) |
r | 0.4661 (0.4221–0.5084) | 0.6653 (0.6406–0.6916) | 0.5332 (0.4929–0.5757) | 0.6914 (0.6624–0.7232) | 0.5000 (0.4702–0.5302) | 0.6792 (0.6598–0.7004) | |
