Capability and reliability of deep learning models to make density predictions on low dose mammograms

PurposeBreast density is associated with risk of developing cancer and can be automatically estimated, using deep learning models, from digital mammograms. Our aim is to estimate the capacity and reliability of such models to estimate density from low dose mammograms taken to enable risk estimates for younger women. MethodsWe trained deep learning models on standard and simulated low dose mammograms. The models were then tested on a mammography data-set with paired standard and low-dose image. The effect of different factors (including age, density and dose ratio) on the differences between predictions on standard and low dose are analysed. Methods to improve performance are assessed and factors that reduce model quality are demonstrated. ResultsWe showed that whilst many factors have no significant effect on the quality of low dose density prediction both density and breast area have an impact. For example correlation between density predictions on low and standard dose images of breasts with the largest breast area is 0.985 (0.949-0.995) while with the smallest is 0.882 (0.697-0.961). We also demonstrated that averaging across CC-MLO images and across repeatedly trained models can improve predictive performance. ConclusionLow dose mammography can be used to produce density and risk estimates that are comparable to standard dose images. Averaging across CC-MLO and across model predictions should improve this performance. Model quality is reduced when making predictions on denser and smaller breasts. Code is available at: https://github.com/stevensquires/