12 results
Now showing1 - 10 of 12
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Dose robustness of deep learning models for anatomic segmentation of computed tomography imagesPurpose: The trend towards lower radiation doses and advances in computed tomography (CT) reconstruction may impair the operation of pretrained segmentation models, giving rise to the problem of estimating the dose robustness of existing segmentation models. Previous studies addressing the issue suffer either from a lack of registered low- and full-dose CT images or from simplified simulations. Approach: We employed raw data from full-dose acquisitions to simulate low-dose CT scans, avoiding the need to rescan a patient. The accuracy of the simulation is validated using a real CT scan of a phantom. We consider down to 20% reduction of radiation dose, for which we measure deviations of several pretrained segmentation models from the full-dose prediction. In addition, compatibility with existing denoising methods is considered. Results: The results reveal the surprising robustness of the TotalSegmentator approach, showing minimal differences at the pixel level even without denoising. Less robust models show good compatibility with the denoising methods, which help to improve robustness in almost all cases. With denoising based on a convolutional neural network (CNN), the median Dice between low- and full-dose data does not fall below 0.9 (12 for the Hausdorff distance) for all but one model. We observe volatile results for labels with effective radii less than 19 mm and improved results for contrasted CT acquisitions. Conclusion: The proposed approach facilitates clinically relevant analysis of dose robustness for human organ segmentation models. The results outline the robustness properties of a diverse set of models. Further studies are needed to identify the robustness of approaches for lesion segmentation and to rank the factors contributing to dose robustness.Publicationtype: Journal Article8 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Learning CT Segmentation from Label Masks Only(2024-04-27); ; ; ; Training segmentation models for CT scans in the absence of input data is a challenging problem. Methods based on generative adversarial networks translate images from other modalities but still require additional data and training. Synthesizing images directly from segmentation masks using heuristics can overcome this limitation. However, capabilities for model generalization remain underexplored for these methods. In this study, we generate synthetic data for liver segmentation using organ labels and prior CT knowledge. Ground truth labels serve as a source of information about global structures and are filled with artificial textures in various settings. Segmentation models trained on synthetic data demonstrate sufficient generalization to real CT data, highlighting a perspective of a simple yet powerful approach to data bootstrapping.Publicationtype: Conference Paper26 48 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , On TotalSegmentator’s performance on low-dose CT imagesDeep neural networks have emerged as the preferred method for semantic segmentation of CT images in recent years. However, understanding their limitations and generalization properties remains an active area of research and a relevant topic for clinical applications. One crucial factor among many is the X-ray radiation dose, which is always kept as low as reasonably possible during CT acquisition. Therefore, potential dose reductions may pose a challenge for existing segmentation models. In this paper, we investigate robustness of the recently proposed TotalSegmentator model for anatomical segmentation with respect to dose reduction. TotalSegmentator combines a large CT dataset and the well-established nnU-Net framework to train deep learning models, resulting in state-of-the-art performance for anatomical segmentation. Our method relies on accurate low-dose simulations derived from acquired full-dose projections. For a set of registered low- and full-dose CT images, we measure the Dice score between the corresponding segmentations. Our results reveal a high level of robustness in the segmentation outcomes. Comprehensive quantitative comparisons demonstrate that at a 20% dose level, the Dice score declines by at most 3%. Visual comparisons reveal only minor differences at the boundaries of the segmented organs. These findings may have a large potential for dose reduction when CT data are acquired predominantly for segmentation purposes, such as for the planning of interventional or surgical procedures.Publicationtype: Conference Paper38 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Deep learning for calcium segmentation in intravascular ultrasound images(De Gruyter, 2021-08-01); ; ; ; Knowing the shape of vascular calcifications is crucial for appropriate planning and conductance of percutaneous coronary interventions. The clinical workflow can therefore benefit from automatic segmentation of calcified plaques in intravascular ultrasound (IVUS) images. To solve segmentation problems with convolutional neural networks (CNNs), large datasets are usually required. However, datasets are often rather small in the medical domain. Hence, developing and investigating methods for increasing CNN performance on small datasets can help on the way towards clinically relevant results. We compared two state-of-the-art CNN architectures for segmentation, U-Net and DeepLabV3, and investigated how incorporating auxiliary image data with vessel wall and lumen annotations improves the calcium segmentation performance by using these either for pretraining or multi-task training. DeepLabV3 outperforms U-Net with up to 6.3 % by means of the Dice coefficient and 36.5 % by means of the average Hausdorff distance. Using auxiliary data improves the segmentation performance in both cases, whereas the multi-task approach outperforms the pre-training approach. The improvements of the multi-task approach in contrast to not using auxiliary data at all is 5.7 % for the Dice coefficient and 42.9 % for the average Hausdorff distance. Automatic segmentation of calcified plaques in IVUS images is a demanding task due to their relatively small size compared to the image dimensions and due to visual ambiguities with other image structures. We showed that this problem can generally be tackled by CNNs. Furthermore, we were able to improve the performance by a multi-task learning approach with auxiliary segmentation data.Publicationtype: Journal Article121 342 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Deep learning for guidewire detection in intravascular ultrasound images(De Gruyter, 2021-08-01); ; ; ; Algorithms for automated analysis of intravascular ultrasound (IVUS) images can be disturbed by guidewires, which are often encountered when treating bifurcations in percutaneous coronary interventions. Detecting guidewires in advance can therefore help avoiding potential errors. This task is not trivial, since guidewires appear rather small compared to other relevant objects in IVUS images. We employed CNNs with additional multi-task learning as well as different guidewire-specific regularizations to enable and improve guidewire detection. In this context, we developed a network block which generates heatmaps that highlight guidewires without the need of localization annotations. The guidewire detection results reach values of 0.931 in terms of the F1-score and 0.996 in terms of area under curve (AUC). Comparing thresholded guidewire heatmaps with ground truth segmentation masks leads to a Dice score of 23.1 % and an average Hausdorff distance of 1.45 mm. Guidewire detection has proven to be a task that CNNs can handle quite well. Employing multi-task learning and guidewire-specific regularizations further improve detection results and enable generation of heatmaps that indicate the position of guidewires without actual labels.Publicationtype: Journal Article101 295 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Attention via Scattering Transforms for Segmentation of Small Intravascular Ultrasound Data Sets(Microtome Publishing, 2021-07); ; ; ; Using intracoronary imaging modalities like intravascular ultrasound (IVUS) has a positive impact on the results of percutaneous coronary interventions. Efficient extraction of important vessel metrics like lumen diameter, vessel wall thickness or plaque burden via automatic segmentation of IVUS images can improve the clinical workflow. State-of-the-art segmentation results are usually achieved by data-driven methods like convolutional neural networks (CNNs). However, clinical data sets are often rather small leading to extraction of image features which are not very meaningful and thus decreasing performance. This is also the case for some applications which inherently allow for only small amounts of available data, e.g., detection of diseases with extremely small prevalence or online-adaptation of an existing algorithm to individual patients. In this work we investigate how integrating scattering transformations - as special forms of wavelet transformations - into CNNs could improve the extraction of meaningful features. To this end, we developed a novel network module which uses features of a scattering transform for an attention mechanism. We observed that this approach improves the results of calcium segmentation up to 8.2% (relatively) in terms of the Dice coefficient and 24.8% in terms of the modified Hausdorff distance. In the case of lumen and vessel wall segmentation, the improvements are up to 2.3% (relatively) in terms of the Dice coefficient and 30.8% in terms of the modified Hausdorff distance.Incorporating scattering transformations as a component of an attention block into CNNs improves the segmentation results on small IVUS segmentation data sets. In general, scattering transformations can help in situations where efficient feature extractors can not be learned via the training data. This makes our attention module an interesting candidate for applications like few-shot learning for patient adaptation or detection of rare diseases.Publicationtype: Conference Paper45 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Tailored methods for segmentation of intravascular ultrasound images via convolutional neural networks(SPIE, 2021); ; ; ; Publicationtype: Conference Paper129 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Learning metal artifact reduction in cardiac CT images with moving pacemakersMetal objects in the human heart such as implanted pacemakers frequently lead to heavy artifacts in reconstructed CT image volumes. Due to cardiac motion, common metal artifact reduction methods which assume a static object during CT acquisition are not applicable. We propose a fully automatic Dynamic Pacemaker Artifact Reduction (DyPAR+) pipeline which is built of three convolutional neural network (CNN) ensembles. In a first step, pacemaker metal shadows are segmented directly in the raw projection data by the SegmentationNets. Second, resulting metal shadow masks are passed to the InpaintingNets which replace metal-affected line integrals in the sinogram for subsequent reconstruction of a metal-free image volume. Third, the metal locations in a pre-selected motion state are predicted by the ReinsertionNets based on a stack of partial angle back-projections generated from the segmented metal shadow mask. We generate the data required for the supervised learning processes by introducing synthetic, moving pacemaker leads into 14 clinical cases without pacemakers. The SegmentationNets and the ReinsertionNets achieve average Dice coefficients of 94.16% ± 2.01% and 55.60% ± 4.79% during testing on clinical data with synthetic metal leads. With a mean absolute reconstruction error of 11.54 HU ± 2.49 HU in the image domain, the InpaintingNets outperform the hand-crafted approaches PatchMatch and inverse distance weighting. Application of the proposed DyPAR+ pipeline to nine clinical test cases with real pacemakers leads to significant reduction of metal artifacts and demonstrates the transferability to clinical practice. Especially the SegmentationNets and InpaintingNets generalize well to unseen acquisition modes and contrast protocols.Publicationtype: Journal Article116 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Motion estimation and correction in cardiac CT angiography images using convolutional neural networks(2019-09); ; ; ; Cardiac motion artifacts frequently reduce the interpretability of coronary computed tomography angiography (CCTA) images and potentially lead to misinterpretations or preclude the diagnosis of coronary artery disease (CAD). In this paper, a novel motion compensation approach dealing with Coronary Motion estimation by Patch Analysis in CT data (CoMPACT) is presented. First, the required data for supervised learning is generated by the Coronary Motion Forward Artifact model for CT data (CoMoFACT) which introduces simulated motion to 19 artifact-free clinical CT cases with step-and-shoot acquisition protocol. Second, convolutional neural networks (CNNs) are trained to estimate underlying 2D motion vectors from 2.5D image patches based on the coronary artifact appearance. In a phantom study with computer-simulated vessels, CNNs predict the motion direction and the motion magnitude with average test accuracies of 13.37°±1.21° and 0.77 ± 0.09 mm, respectively. On clinical data with simulated motion, average test accuracies of 34.85°±2.09° and 1.86 ± 0.11 mm are achieved, whereby the precision of the motion direction prediction increases with the motion magnitude. The trained CNNs are integrated into an iterative motion compensation pipeline which includes distance-weighted motion vector extrapolation. Alternating motion estimation and compensation in twelve clinical cases with real cardiac motion artifacts leads to significantly reduced artifact levels, especially in image data with severe artifacts. In four observer studies, mean artifact levels of 3.08 ± 0.24 without MC and 2.28 ± 0.29 with CoMPACT MC are rated in a five point Likert scale.Publicationtype: Journal Article149 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Dynamic pacemaker artifact removal (DyPAR) from CT data using CNNs(Microtome Publishing, 2019-07); ; ; 11