Table 1 Comparing the AUC scores (with a 95% confidence interval) of the multimodal classifiers to single modality classifiers and existing clinical scores.
From: Multimodal fusion models for pulmonary embolism mortality prediction
Model | AUC | Accuracy | Specificity | Sensitivity | PPV | NPV |
|---|---|---|---|---|---|---|
Single modalities | ||||||
 EHR | 0.87 [0.78–0.95] | 0.81 | 0.79 | 0.9 | 0.41 | 0.98 |
 CTPA-SANet | 0.82 [0.74–0.91] | 0.79 | 0.79 | 0.8 | 0.38 | 0.96 |
 CTPA-Swin UNETR | 0.73 [0.6–0.85] | 0.56 | 0.52 | 0.8 | 0.21 | 0.94 |
Clinical predictive rules | ||||||
 PESI | 0.8 [0.7–0.87] | 0.61 | 0.56 | 0.9 | 0.25 | 0.97 |
 sPESI | 0.77 [0.63–0.87] | 0.48 | 0.42 | 0.9 | 0.20 | 0.96 |
Fusion models | ||||||
 Late-average (Fig. 7a) | 0.85 [0.75–0.93] | 0.68 | 0.65 | 0.9 | 0.29 | 0.98 |
 Late-TabNet (Fig. 7b) | 0.92 [0.87–0.98] | 0.89 | 0.89 | 0.9 | 0.56 | 0.98 |
 Late-XGBoost (Fig. 7c) | 0.88 [0.8–0.95] | 0.76 | 0.74 | 0.9 | 0.36 | 0.98 |
 Early (Fig. 7d) | 0.9 [0.83–0.96] | 0.81 | 0.79 | 0.9 | 0.41 | 0.98 |
 Intermediate + SANet(Fig. 7e) | \({\textbf {0.96 [0.93-1.0]}}\) | \({\textbf {0.93}}\) | \({\textbf {0.94}}\) | \({\textbf {0.9}}\) | \({\textbf {0.69}}\) | \({\textbf {0.98}}\) |
 Intermediate + Swin UNETR (Fig. 7f) | 0.95 [0.9–0.98] | 0.87 | 0.87 | 0.9 | 0.53 | 0.98 |
 MBT (Fig. 7g) | 0.9 [0.85–0.96] | 0.82 | 0.81 | 0.9 | 0.43 | 0.98 |
