Table 2 Classification performance on the BIMCV hold-out set. The Weighted Ensemble achieves the highest sensitivity (Recall) for csPCa, while the Stacked model offers the best overall discrimination (AUC).

From: Clinically significant prostate cancer detection with deep learning in a multi-center magnetic resonance imaging study

Model Strategy

Class

Precision

Recall

F1-Score

Accuracy

Ensemble (Weighted logits)

Non-csPC

0.77

0.65

0.70

0.75

csPC

0.73

0.83

0.78

Stacked (Meta-learner)

Non-csPC

0.72

0.68

0.70

0.73

csPC

0.73

0.77

0.75

No-pretrained (Baseline)

Non-csPC

0.71

0.63

0.67

0.71

csPC

0.71

0.77

0.74

Back to article page