Model uncertainty estimates for deep learning mammographic density prediction using ordinal and classification approaches

PurposeMammographic density is associated with the risk of developing breast cancer and can be predicted using deep learning methods. Model uncertainty estimates are not produced by standard regression approaches but would be valuable for clinical and research purposes. Our objective is to produce deep learning models with in-built uncertainty estimates without degrading predictive performance. ApproachWe analyse data from over 150,000 mammogram images with associated continuous density scores from expert readers in the Predicting Risk Of Cancer At Screening (PROCAS) study. We re-designate the continuous density scores to 100 density classes then train classification and ordinal deep learning models. Distributions and distribution-free methods are applied to extract predictions and uncertainties. A deep learning regression model is trained on the continuous density scores to act as a direct comparison. ResultsThe root mean squared error (RMSE) between expert assigned density labels and predictions of the standard regression model are 8.42 (8.34-8.51) while the RMSE for the classification and ordinal classification are 8.37 (8.28-8.46) and 8.44 (8.35-8.53) respectively. The average uncertainties produced by the models are higher when the density scores from pairs of expert readers density scores differ more, are higher when different mammogram views of the same views are more variable and when two separately trained models show higher variation. ConclusionsUsing either a classification or ordinal approach we can produce model uncertainty estimates without loss of predictive performance.