Deep learning has been widely accepted as a promising solution for medical image segmentation, given a sufficiently large representative dataset of images with corresponding annotations. With ever increasing amounts of annotated medical datasets, it is infeasible to train a learning method always with all data from scratch. This is also doomed to hit computational limits, e.g., memory or runtime feasible for training. Incremental learning can be a potential solution, where new information (images or anatomy) is introduced iteratively. Nevertheless, for the preservation of the collective information, it is essential to keep some "important" (i.e., representative) images and annotations from the past, while adding new information. In this paper, we introduce a framework for applying incremental learning for segmentation and propose novel methods for selecting representative data therein. We comparatively evaluate our methods in different scenarios using MR images and validate the increased learning capacity with using our methods.

Segmentation is essential for medical image analysis tasks such as intervention planning, therapy guidance, diagnosis, treatment decisions. Deep learning is becoming increasingly prominent for segmentation, where the lack of annotations, however, often becomes the main limitation. Due to privacy concerns and ethical considerations, most medical datasets are created, curated, and allow access only locally. Furthermore, current deep learning methods are often suboptimal in translating anatomical knowledge between different medical imaging modalities. Active learning can be used to select an informed set of image samples to request for manual annotation, in order to best utilize the limited annotation time of clinical experts for optimal outcomes, which we focus on in this work. Our contributions herein are two fold: (1) we enforce domain-representativeness of selected samples using a proposed penalization scheme to maximize information at the network abstraction layer, and (2) we propose a Borda-count based sample querying scheme for selecting samples for segmentation. Comparative experiments with baseline approaches show that the samples queried with our proposed method, where both above contributions are combined, result in significantly improved segmentation performance for this active learning task.

BackgroundOpening-wedge osteotomies of the distal radius, performed with three-dimensional printed patient-specific instruments, are a promising technique for accurate correction of malunions. Nevertheless, reports of residual malalignments and discrepancies in the plate and screw position from the planned fixation exist. Consequently, we developed a patient-specific ramp-guide technique, combining navigation of plate positioning, osteotomy cutting, and reduction. The aim of this study is to compare the accuracy of navigation of three-dimensional planned opening-wedge osteotomies, using a ramp-guide, over state-of-the-art guide techniques relying solely on pre-drilled holes.MethodsA retrospective analysis was carried out on opening-wedge osteotomies of the distal radius, performed between May 2016 and April 2017, with patient-specific instruments. Eight patients were identified in which a ramp-guide for the distal plate fixation was used. We compared the reduction accuracy with a control group of seven patients, where the reduction was performed with pre-drilled screw holes placed with the patient-specific instruments. The navigation accuracy was assessed by comparing the preoperative plans with the postoperative segmented, computed tomography scans. The accuracy was expressed using a 3D angle and in measurements of all six degrees of freedom (3 translations, 3 rotations), with respect to an anatomical coordinate system.ResultsThe duration of the surgery of the ramp-guide group was significantly shorter compared to the control group. Significantly less rotational and translational residual malalignment error was observed in the open-wedged osteotomies, where patient-specific instruments with ramp-guides were used. On average, a residual rotational malalignment error of 2.0° (± 2.2°) and a translational malalignment error of 0.6 mm (± 0.2 mm) was observed in the ramp-guide group, as compared to the 4.2° (± 15.0°) and 1.0 mm (± 0.4 mm) error in the control group. The used plate was not significantly positioned more accurately, but significantly fewer screws (15.6%) were misaligned in the distal fragment compared to the control group (51.9%).ConclusionThe use of the presented ramp-guide technique in opening-wedge osteotomies is improving reduction accuracy, screw position, and surgical duration, compared to the existing patient-specific instrument based navigation methods.

A registration framework based on the bone surface extraction from 3D freehand US and a subsequent fast, automatic surface alignment robust to single-sided view and large false-positive rates from US was shown to achieve registration accuracy feasible for practical orthopedic scenarios and a qualitative outcome indicating good visual image alignment.