Most recent research of deep neural networks in the field of computer vision has focused on improving performances of point predictions by developing network architectures or learning algorithms. Reliable uncertainty quantification accompanied by point estimation can lead to a more informed decision, and the quality of prediction can be improved. In this paper, we invoke a Bayesian neural network and propose a natural way of quantifying uncertainties in classification problems by decomposing the moment-based predictive uncertainty into two parts: aleatoric and epistemic uncertainty. The proposed method takes into account the discrete nature of the outcome, yielding the correct interpretation of each uncertainty. We demonstrate that the proposed uncertainty quantification method provides additional insights into the point prediction using two Ischemic Stroke Lesion Segmentation Challenge datasets and the Digital Retinal Images for Vessel Extraction dataset.
- Aleatoric and epistemic uncertainty
- Bayesian neural network
- Ischemic stroke lesion segmentation
- Retinal blood vessel segmentation
- Uncertainty quantification