Kai Geißler’s research while affiliated with Fraunhofer Institute for Digital Medicine and other places

Recent studies indicate that malignant breast lesions can be predicted from structural changes in prior exams of preventive breast MRI examinations. Due to non-rigid deformation between studies, spatial correspondences between structures in two consecutive studies are lost. Thus, deformable image registration can contribute to predicting individual cancer risks. This study evaluates a registration approach based on a novel breast mask segmentation and non-linear image registration based on data from 5 different sites. The landmark error (mean ± standard deviation [1st quartile, 3rd quartile]), annotated by three radiologists, is 2.9 ± 2.8 [1.3, 3.2] mm when leaving out two outlier cases from the evaluation for which the registration failed completely. We assess the inter-observer variabilities of keypoint errors and find an error of 3.6 ± 4.7 [1.6, 4.0] mm, 4.4 ± 4.9 [1.8, 4.8] mm, and 3.8 ± 4.0 [1.7, 4.1] mm when comparing each radiologist to the mean keypoints of the other two radiologists. Our study shows that the current state of the art in registration is well suited to recover spatial correspondences of structures in cancerous and non-cancerous cases, despite the high level of difficulty of this task.

In medical image segmentation with deep learning, large amounts of annotated data are needed to train precise models. Such annotations are timeconsuming and costly to create, since medical experts need to ensure their quality. Active learning techniques may reduce the expert effort. In this work, we compare different sample selection strategies for training a model for breast segmentation in MR images using 3D U-Nets: We evaluate two uncertainty-based approaches that compute the voxel-wise entropy or epistemic uncertainty based on a Bayesian neural network approximated via Monte Carlo dropout and compare them against a random selection as baseline. We find that both uncertainty-based approaches improve over the baseline in the earlier iterations, but lead to a similar performance in the long run. In early iterations they are 2-4 active learning iterations ahead of the "random selection" strategy, which corresponds to one or several days of saved annotation time.We also assess how well the different uncertainty measures correlate with the segmentation quality and find that epistemic uncertainty is a better surrogate measure than the commonly used plain entropy.