Abdominal aortic aneurysms (AAA) may remain clinically silent until they enlarge and patients present with a potentially lethal rupture. This necessitates early detection and elective treatment. The goal of this study was to develop an easy-to-train algorithm which is capable of automated AAA screening in CT scans and can be applied to an intra-hospital environment. Three deep convolutional neural networks (ResNet, VGG-16 and AlexNet) were adapted for 3D classification and applied to a dataset consisting of 187 heterogenous CT scans. The 3D ResNet outperformed both other networks. Across the five folds of the first training dataset it achieved an accuracy of 0.856 and an area under the curve (AUC) of 0.926. Subsequently, the algorithms performance was verified on a second data set containing 106 scans, where it ran fully automated and resulted in an accuracy of 0.953 and an AUC of 0.971. A layer-wise relevance propagation (LRP) made the decision process interpretable and showed that the network correctly focused on the aortic lumen. In conclusion, the deep learning-based screening proved to be robust and showed high performance even on a heterogeneous multi-center data set. Integration into hospital workflow and its effect on aneurysm management would be an exciting topic of future research.

Objective: Three-dimensional (3D) blood vessel structure information is important for diagnosis and treatment in various clinical scenarios. We present a fully automatic method for the extraction and differentiation of the arterial and venous vessel trees from abdominal contrast enhanced computed tomography (CE-CT) volumes using convolutional neural networks (CNNs). Methods: We used a novel ratio-based sampling method to train 2D and 3D versions of the U-Net, the V-Net and the DeepVesselNet. Networks were trained with a combination of the Dice and cross entropy loss. Performance was evaluated on 20 IR-CAD subjects. Best performing networks were combined into an ensemble. We investigated seven different weight-ing schemes. Trained networks were additionally applied to 26 BTCV cases to validate the generalizability. Results: Based on our experiments, the optimal configuration is an equally weighted ensemble of 2D and 3D U-and V-Nets. Our method achieved Dice similarity coefficients of 0.758 ± 0.050 (veins) and 0.838 ± 0.074 (arteries) on the IRCAD data set. Application to the BTCV data set showed a high transfer ability. Conclusion: Abdominal vascular structures can be segmented more accurately using ensembles than individual CNNs. 2D and 3D networks have complementary strengths and weaknesses. Our ensemble of 2D and 3D U-Nets and V-Nets in combination with ratio-based sampling achieves a high agreement with manual annotations for both artery and vein segmentation. Our results surpass other state-of-the-art methods. Significance: Our segmen-tation pipeline can provide valuable information for the planning of living donor organ transplantations.

Magnetic resonance image formation is not trivial and remains a difficult subject for teaching. Therefore, we saw an urgent need to facilitate teaching by developing a practical and easily accessible MR image generator. Due to the increasing interest in X-nuclei MRI, sodium image generation is also offered. The tool is implemented as a web application that is compatible with all standard desktop browsers and is open source. The user interface focuses on the parameters needed for the creation and display of the resulting images. Available MR sequences range from the standard Spin Echo and Inversion Recovery over steady-state to conventional sodium and more advanced single and triple quantum sequences. Additionally, the user interface has parameters to alter the resolution, the noise, and the k-space sampling. Our software is free to use and specifically suited for teaching purposes.

Purpose: Development of an algorithm to self-calibrate arbitrary CBCT trajectories which can be used to reduce metal artifacts. By using feature detection and matching we want to reduce the amount of parameters for the BFGS optimization and thus reduce the runtime. Methods: Each projection is 2D-3D registered on a prior image with AKAZE feature detection and brute force matching. Translational misalignment is calculated directly from the misalignment of feature positions, rotations are aligned using a minimization algorithm that fits a quartic function and determines the minimum of this function. Evaluation: We did three experiments to compare how well the algorithm can handle noise on the different degrees of freedom. Our algorithms are compared to Broyden–Fletcher–Goldfarb–Shanno (BFGS) minimizer with Normalized Gradient Information (NGI) objective function, and BFGS with distance between features objective function using SSIM, nRMSE, and the Dice coefficient of segmented metal object. Results: Our algorithm (Feature ORiented Calibration for Arbitrary Scan Trajectories with Enhanced Reliability (FORCASTER)) performs on par with the state-of-the-art algorithms (BFGS with NGI objective). nRMSE: FORCASTER = 0.3390, BFGS+NGI = 0.3441; SSIM: FORCASTER = 0.83, BFGS + NGI = 0.79; Dice: FORCASTER = 0.86, BFGS + NGI = 0.87. Conclusion: The proposed algorithm can determine the parameters of the projection orientations for arbitrary trajectories with calibration quality comparable to state-of-the-art algorithms, but faster and with higher tolerance to errors in the initially guessed parameters.

Background: For the reconstruction of Cone-Beam CT volumes, the exact position of each projection is needed; however, in some situations, this information is missing. Purpose: The development of a self-calibration algorithm for arbitrary CBCT trajectories that does not need initial positions. Methods: Projections are simulated in a spherical grid around the center of rotation. Through using feature detection and matching, an acquired projection is compared to each simulated image in this grid. The position with the most matched features was used as a starting point for a fine calibration with a state-of-the-art algorithm. Evaluation: This approach is compared with the calibration of nearly correct starting positions when using FORCASTER and CMA-ES minimization with a normalized gradient information (NGI) objective function. The comparison metrics were the normalized root mean squared error, structural similarity index, and the dice coefficient, which were evaluated on the segmentation of a metal object. Results: The parameter estimation for a regular Cone-Beam CT with a 496 projection took 1:26 h with the following metric values: NRMSE = 0.0669; SSIM = 0.992; NGI = 0.75; and Dice = 0.96. FORCASTER with parameter estimation took 3:28 h with the following metrics: NRMSE = 0.0190; SSIM = 0.999; NGI = 0.92; and Dice = 0.99. CMA-ES with parameter estimation took 5:39 h with the following metrics: NRMSE = 0.0037; SSIM = 1.0; NGI = 0.98; and Dice = 1.0. Conclusions: The proposed algorithm can determine the parameters of the projection orientations for arbitrary trajectories with enough accuracy to reconstruct a 3D volume with low errors.