Additional file 1 of Artificial intelligence for detecting acute heart failure on chest CT: prospective clinical proof-of-concept validation

Abstract

Additional file 1: Table S1 Objective parameters supporting respiratory imbalance. Fig. S1 The performance of the artificial intelligence algorithm in comparison to the secondary outcomes: research radiologist 1 (dark green) and research radiologist 2 (green). The two research radiologists independently identified radiologic signs of AHF in 60 patients (25%) and 64 patients (27%), respectively. Fig. S2 Model calibration analysis. (a) Calibration plot in the independent prospective cohort showing suboptimal agreement of predicted probabilities and observed outcomes (Brier score 0.16). Vertical lines indicate the 95% binomial CIs for the observed event rates. (b) Calibration curves obtained by randomly splitting the FACTUAL data (n = 232) into two, calibrating the model on one part (n = 116) and predicting on the other part (n = 116). Each curve corresponds to one random split. The AUROC is provided in the legend. Recalibration improves agreement between predicted probabilities and observed outcomes, with substantial variability across splits reflecting the limited calibration sample size (n = 116). Fig. S3—Aggregated feature importance plots across the whole cohort (a), and stratified by true positives (b), false positives (c), and false negatives (d). Table S2—Patient characteristics for false positive cases and false negative cases. Appendix 1—Overview of the primary and secondary outcomes

Ähnliche Arbeiten