Evidential Deep Learning with conditional entropy for glaucoma detection
Organizations: Gryphon Center for Artificial Intelligence and Theoretical Sciences, Singapore · Electrical and Computer Engineering Department, University of California San Diego, CA, USA
Abstract
We propose a novel conditional entropy function as a basis for quantifying predictive uncertainty in Evidential Deep Learning (EDL) for glaucoma detection from retinal fundus images. Using the EyePACS-AIROGS-light-V2 and RIM-ONE-DL datasets for validation, we found that our proposed EDL model regularized by conditional entropy function produced uncertainty estimates with substantially lower calibration errors than Deep Ensembles, MC Dropout, post-hoc calibration and the conventional EDL framework based on Dempster uncertainty. To further probe our models' interpretability, we analyzed Grad-CAM heatmaps using a focus index representing spatial concentration. Compared with the other uncertainty quantification methods, EDL models exhibited greater separation between the index distributions of correctly and incorrectly classified fundus images. In particular, false negatives were associated with localized Grad-CAM activations that were displaced from the optic disc, consistent with its established clinical importance in glaucoma assessment. Overall, our findings suggest that our proposed EDL framework for glaucoma detection provides not only enhanced uncertainty calibration but also admits a more physiologically consistent interpretability.
Figures & tables
| EyePACS dataset-trained model performance | ||||||||
| Models | Acc | AUC | ECE q | ECE w | OOD | |||
| (Baseline) MNetV3 | 0.90 | 0.96 | 0.054 | 0.063 | 0.08 | 0.038 | 0.050 | 0.82 |
| Temp. Scaling | 0.90 | 0.96 | 0.047 | 0.050 | 0.08 | 0.050 | 0.038 | 0.82 |
| MC Dropout | 0.90 | 0.96 | 0.038 | 0.044 | 0.08 | 0.103 | 0.043 | 0.79 |
| Deep Ensemble | 0.91 | 0.97 | 0.025 | 0.030 | 0.07 | 0.105 | 0.138 | 0.81 |
| EDL (w/o reg.) | 0.90 | 0.96 | 0.029 | 0.033 | 0.08 | 0.42 | 0.49 | 0.73 |
Evidential Deep Learning
Evidential deep learning estimates predictive uncertainty from neural-network evidence distributions to support calibrated, selective, and robust predictions.
See all 63 papers