Table 4 Performance of Bayesian Convolutional Neural Network (BCNN) models (AUC and Accuracy) with varying amounts of data retained.
From: Uncertainty-aware diabetic retinopathy detection using deep learning enhanced by Bayesian approaches
Method | 50% data retained | 70% data retained | 100% data retained | |||
|---|---|---|---|---|---|---|
AUC \(\uparrow\) | Accuracy \(\uparrow\) | AUC \(\uparrow\) | Accuracy \(\uparrow\) | AUC \(\uparrow\) | Accuracy \(\uparrow\) | |
MC Dropout | 90.4 ± 0.8 | 93.7 ± 0.3 | 94.8 ± 0.6 | 95.3 ± 0.3 | 96.2 ± 0.2 | 98.2 ± 0.1 |
Mean-field VI | 89.2 ± 1.0 | 91.3 ± 0.4 | 91.7 ± 0.8 | 93.4 ± 0.4 | 93.6 ± 0.8 | 95.8 ± 0.3 |
Deterministic | 85.2 ± 0.7 | 88.7 ± 0.6 | 90.4 ± 0.9 | 91.8 ± 0.8 | 92.1 ± 0.3 | 93.8 ± 0.2 |
