Table 4.
Internal dataset: Stanford | External dataset: Intermountain | |
---|---|---|
Metric (AUROC) | ||
PENet—1 slice | 0.48 [0.45–0.51] | 0.51 [0.47–0.54] |
PENet—6 slices | 0.57 [0.53–0.60] | 0.58 [0.55–0.59] |
PENet—12 slices | 0.74 [0.70–0.77] | 0.69 [0.67–0.72] |
PENet—24 slices | 0.84 [0.82–0.87] | 0.85 [0.81–0.88] |
PENet—48 slices | 0.80 [0.77–0.83] | 0.83 [0.76–0.86] |
AUROC on the internal test set (Stanford) and external test set (Intermountain) with 95% confidence interval: smaller input slice number does not provide enough structural information to learn while too many input slices makes pulmonary embolism hard to detect.