The assessment of cerebrovascular disease benefits from the availability of different neuroimaging modalities, each providing different aspects to the vasculature and surrounding brain parenchyma. Segmented vasculature may improve cross-modal alignment but is difficult to annotate in sequences without vessel contrast. We propose a novel method to segment cerebral arteries in three non-angiographic sequences: T1w, T2w and PDw. Our method further predicts annotations for four anatomical regions, which can be used to mask specific parts of the vascular network or analyze the topology. For our experiments, we annotated arterial vessels and anatomical regions in 2,218 scans of the IXI dataset using a novel automatic method. We used the nnU-Net framework to train models in a 5-fold cross validation and to predict on a separate test-set. Our results suggest that vascular structures can be segmented and annotated in the examined MRI sequences with reasonable quality. The approach may potentially be used to study vascular diseases, when trained on pathological images. We share our ground-truth and models to encourage future experiments.

The treatment of burns is supported by accurately estimating measurements such as total body surface area burned (TBSA-b) and total body surface area (TBSA). Computing these values automatically can lead to faster decisions and reduce human error. Deep learning-based methods rely on big data to perform well in practical applications. This is especially true for burn medicine, where real data is scarce. In this paper, we present a pipeline for synthesizing diagnostic burn images and wound annotations from virtual 3D models. We demonstrate how to generate a heterogeneous dataset by combining such features as body shape, real skin and wound textures, background scenes, camera settings and illumination. The resulting images can be used for various deep-learning tasks such as wound detection, segmentation or classification. Solely with these synthetic images, we train models for burn wound segmentation and show that they learn distinctive features of burn wounds.

As the U-Net architecture continues to push the state of the art in medical image segmentation, and keeping track of the multitude of derived models becomes increasingly difficult, opposing trends have emerged that prefer a coherent framework of ancillary processing tasks over the use of highly optimized and sophisticated models. This trend has culminated in the framework nnU-Net, which adapts preprocessing, training and inference to the respective dataset and thus managed to lead the Medical Segmentation Decathlon challenge despite relying on comparably simple models. In this paper, we focus on one ancillary technique that is commonly used but poorly addressed in literature: patchwise training. Since computational costs tend to increase drastically when using naive strategies, we discuss the benefits of content-sensitive sampling for patchwise training of deep learning segmentation models. We adapt this strategy in a content-driven architecture for abdominal aorta and stent-graft segmentation, where we evaluate and compare it with traditional sampling strategies based on a real-world clinical dataset.