Deep learning (DL) has revolutionized the field of computer vision and image processing. In medical imaging, algorithmic solutions based on DL have been shown to achieve high performance on tasks that previously required medical experts. However, DL-based solutions for disease detection have been proposed without methods to quantify and control their uncertainty in a decision. In contrast, a physician knows whether she is uncertain about a case and will consult more experienced colleagues if needed. Here we evaluate drop-out based Bayesian uncertainty measures for DL in diagnosing diabetic retinopathy (DR) from fundus images and show that it captures uncertainty better than straightforward alternatives. Furthermore, we show that uncertainty informed decision referral can improve diagnostic performance. Experiments across different networks, tasks and datasets show robust generalization. Depending on network capacity and task/dataset difficulty, we surpass 85% sensitivity and 80% specificity as recommended by the NHS when referring 0−20% of the most uncertain decisions for further inspection. We analyse causes of uncertainty by relating intuitions from 2D visualizations to the high-dimensional image space. While uncertainty is sensitive to clinically relevant cases, sensitivity to unfamiliar data samples is task dependent, but can be rendered more robust.

Deep learning-based systems can achieve a diagnostic performance comparable to physicians in a variety of medical use cases including the diagnosis of diabetic retinopathy. To be useful in clinical practise, it is necessary to have well calibrated measures of the uncertainty with which these systems report their decisions. However, deep neural networks (DNNs) are being often overconfident in their predictions, and are not amenable to a straightforward probabilistic treatment. Here, we describe an intuitive framework based on test-time data augmentation for quantifying the diagnostic uncertainty of a state-of-the-art DNN for diagnosing diabetic retinopathy. We show that the derived measure of uncertainty is well-calibrated and that experienced physicians likewise find cases with uncertain diagnosis difficult to evaluate. This paves the way for an integrated treatment of uncertainty in DNN-based diagnostic systems.

Deep learning-based systems can achieve a diagnostic performance comparable to physicians in a variety of medical use cases including the diagnosis of diabetic retinopathy. To be useful in clinical practise, it is necessary to have well calibrated measures of the uncertainty with which these systems report their decisions. However, deep neural networks (DNNs) are being often overconfident in their predictions, and are not amenable to a straightforward probabilistic treatment. Here, we describe an intuitive framework based on test-time data augmentation for quantifying the diagnostic uncertainty of a state-of-the-art DNN for diagnosing diabetic retinopathy. We show that the derived measure of uncertainty is well-calibrated and that experienced physicians likewise find cases with uncertain diagnosis difficult to evaluate. This paves the way for an integrated treatment of uncertainty in DNN-based diagnostic systems.is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. certified by peer review)

No abstract

No abstract