-
Synthetic optical coherence tomography angiographs for detailed retinal vessel segmentation without human annotations
Authors:
Linus Kreitner,
Johannes C. Paetzold,
Nikolaus Rauch,
Chen Chen,
Ahmed M. Hagag,
Alaa E. Fayed,
Sobha Sivaprasad,
Sebastian Rausch,
Julian Weichsel,
Bjoern H. Menze,
Matthias Harders,
Benjamin Knier,
Daniel Rueckert,
Martin J. Menten
Abstract:
Optical coherence tomography angiography (OCTA) is a non-invasive imaging modality that can acquire high-resolution volumes of the retinal vasculature and aid the diagnosis of ocular, neurological and cardiac diseases. Segmenting the visible blood vessels is a common first step when extracting quantitative biomarkers from these images. Classical segmentation algorithms based on thresholding are st…
▽ More
Optical coherence tomography angiography (OCTA) is a non-invasive imaging modality that can acquire high-resolution volumes of the retinal vasculature and aid the diagnosis of ocular, neurological and cardiac diseases. Segmenting the visible blood vessels is a common first step when extracting quantitative biomarkers from these images. Classical segmentation algorithms based on thresholding are strongly affected by image artifacts and limited signal-to-noise ratio. The use of modern, deep learning-based segmentation methods has been inhibited by a lack of large datasets with detailed annotations of the blood vessels. To address this issue, recent work has employed transfer learning, where a segmentation network is trained on synthetic OCTA images and is then applied to real data. However, the previously proposed simulations fail to faithfully model the retinal vasculature and do not provide effective domain adaptation. Because of this, current methods are unable to fully segment the retinal vasculature, in particular the smallest capillaries. In this work, we present a lightweight simulation of the retinal vascular network based on space colonization for faster and more realistic OCTA synthesis. We then introduce three contrast adaptation pipelines to decrease the domain gap between real and artificial images. We demonstrate the superior segmentation performance of our approach in extensive quantitative and qualitative experiments on three public datasets that compare our method to traditional computer vision algorithms and supervised training using human annotations. Finally, we make our entire pipeline publicly available, including the source code, pretrained models, and a large dataset of synthetic OCTA images.
△ Less
Submitted 15 December, 2023; v1 submitted 19 June, 2023;
originally announced June 2023.
-
Automatic Modelling of Human Musculoskeletal Ligaments -- Framework Overview and Model Quality Evaluation
Authors:
Noura Hamze,
Lukas Nocker,
Nikolaus Rauch,
Markus Walzthöni,
Fabio Carrillo,
Philipp Fürnstahl,
Matthias Harders
Abstract:
Accurate segmentation of connective soft tissues is still a challenging task, which hinders the generation of corresponding geometric models for biomechanical computations. Alternatively, one could predict ligament insertion sites and then approximate the shapes, based on anatomical knowledge and morphological studies. Here, we describe a corresponding integrated framework for the automatic modell…
▽ More
Accurate segmentation of connective soft tissues is still a challenging task, which hinders the generation of corresponding geometric models for biomechanical computations. Alternatively, one could predict ligament insertion sites and then approximate the shapes, based on anatomical knowledge and morphological studies. Here, we describe a corresponding integrated framework for the automatic modelling of human musculoskeletal ligaments. We combine statistical shape modelling with geometric algorithms to automatically identify insertion sites, based on which geometric surface and volume meshes are created. For demonstrating a clinical use case, the framework has been applied to generate models of the interosseous membrane in the forearm. For the adoption to the forearm anatomy, ligament insertion sites in the statistical model were defined according to anatomical predictions following an approach proposed in prior work. For evaluation we compared the generated sites, as well as the ligament shapes, to data obtained from a cadaveric study, involving five forearms with a total of 15 ligaments. Our framework permitted the creation of 3D models approximating ligaments' shapes with good fidelity. However, we found that the statistical model trained with the state-of-the-art prediction of the insertion sites was not always reliable. Using that model, average mean square errors as well as Hausdorff distances of the meshes increased by more than one order of magnitude, as compared to employing the known insertion locations of the cadaveric study. Using the latter an average mean square error of 0.59 mm and an average Hausdorff distance of less than 7 mm resulted, for the complete set of ligaments. In conclusion, the presented approach for generating ligament shapes from insertion points appears to be feasible but the detection of the insertion sites with a SSM is too inaccurate.
△ Less
Submitted 24 March, 2020;
originally announced March 2020.
-
Accelerating Surface Tension Calculation in SPH via Particle Classification & Monte Carlo Integration
Authors:
Fernando Zorrilla,
Johannes Sappl,
Wolfgang Rauch,
Matthias Harders
Abstract:
Surface tension has a strong influence on the shape of fluid interfaces. We propose a method to calculate the corresponding forces efficiently. In contrast to several previous approaches, we discriminate to this end between surface and non-surface SPH particles. Our method effectively smooths the fluid interface, minimizing its curvature. We make use of an approach inspired by Monte Carlo integrat…
▽ More
Surface tension has a strong influence on the shape of fluid interfaces. We propose a method to calculate the corresponding forces efficiently. In contrast to several previous approaches, we discriminate to this end between surface and non-surface SPH particles. Our method effectively smooths the fluid interface, minimizing its curvature. We make use of an approach inspired by Monte Carlo integration to estimate local normals as well as curvatures, based on which the force can be calculated. The technique is applicable, but not limited to 2D and 3D simulations, and can be coupled with any common SPH formulation. It outperforms prior approaches with regard to total computation time per time step, while being stable and avoiding artifacts.
△ Less
Submitted 4 December, 2019;
originally announced December 2019.
-
Deep Learning of Preconditioners for Conjugate Gradient Solvers in Urban Water Related Problems
Authors:
Johannes Sappl,
Laurent Seiler,
Matthias Harders,
Wolfgang Rauch
Abstract:
Solving systems of linear equations is a problem occuring frequently in water engineering applications. Usually the size of the problem is too large to be solved via direct factorization. One can resort to iterative approaches, in particular the conjugate gradients method if the matrix is symmetric positive definite. Preconditioners further enhance the rate of convergence but hitherto only handcra…
▽ More
Solving systems of linear equations is a problem occuring frequently in water engineering applications. Usually the size of the problem is too large to be solved via direct factorization. One can resort to iterative approaches, in particular the conjugate gradients method if the matrix is symmetric positive definite. Preconditioners further enhance the rate of convergence but hitherto only handcrafted ones requiring expert knowledge have been used. We propose an innovative approach employing Machine Learning, in particular a Convolutional Neural Network, to unassistedly design preconditioning matrices specifically for the problem at hand. Based on an in-depth case study in fluid simulation we are able to show that our learned preconditioner is able to improve the convergence rate even beyond well established methods like incomplete Cholesky factorization or Algebraic MultiGrid.
△ Less
Submitted 17 June, 2019;
originally announced June 2019.
