Showing papers by "Ivo Wolf published in 2017"
01 Nov 2017
TL;DR: In contrast to most existing table detection and structure recognition methods, which are applicable only to PDFs, DeepDeSRT processes document images, which makes it equally suitable for born-digital PDFs as well as even harder problems, e.g. scanned documents.
Abstract: This paper presents a novel end-to-end system for table understanding in document images called DeepDeSRT In particular, the contribution of DeepDeSRT is two-fold First, it presents a deep learning-based solution for table detection in document images Secondly, it proposes a novel deep learning-based approach for table structure recognition, ie identifying rows, columns, and cell positions in the detected tables In contrast to existing rule-based methods, which rely on heuristics or additional PDF metadata (like, for example, print instructions, character bounding boxes, or line segments), the presented system is data-driven and does not need any heuristics or metadata to detect as well as to recognize tabular structures in document images Furthermore, in contrast to most existing table detection and structure recognition methods, which are applicable only to PDFs, DeepDeSRT processes document images, which makes it equally suitable for born-digital PDFs (as they can automatically be converted into images) as well as even harder problems, eg scanned documents To gauge the performance of DeepDeSRT, the system is evaluated on the publicly available ICDAR 2013 table competition dataset containing 67 documents with 238 pages overall Evaluation results reveal that DeepDeSRT outperforms state-of-the-art methods for table detection and structure recognition and achieves F1-measures of 9677% and 9144% for table detection and structure recognition, respectively Additionally, DeepDeSRT is evaluated on a closed dataset from a real use case of a major European aviation company comprising documents which are highly unlike those in ICDAR 2013 Tested on a randomly selected sample from this dataset, DeepDeSRT achieves high detection accuracy for tables which demonstrates the sound generalization capabilities of our system
276 citations
10 Sep 2017
TL;DR: This paper uses an ensemble of UNet inspired architectures for segmentation of cardiac structures such as the left and right ventricular cavity (LVC, RVC) and the left ventricular myocardium (LVM) on each time instance of the cardiac cycle to address named limitations of cardiac magnetic resonance imaging.
Abstract: Cardiac magnetic resonance imaging improves on diagnosis of cardiovascular diseases by providing images at high spatiotemporal resolution. Manual evaluation of these time-series, however, is expensive and prone to biased and non-reproducible outcomes. In this paper, we present a method that addresses named limitations by integrating segmentation and disease classification into a fully automatic processing pipeline. We use an ensemble of UNet inspired architectures for segmentation of cardiac structures such as the left and right ventricular cavity (LVC, RVC) and the left ventricular myocardium (LVM) on each time instance of the cardiac cycle. For the classification task, information is extracted from the segmented time-series in form of comprehensive features handcrafted to reflect diagnostic clinical procedures. Based on these features we train an ensemble of heavily regularized multilayer perceptrons (MLP) and a random forest classifier to predict the pathologic target class. We evaluated our method on the ACDC dataset (4 pathology groups, 1 healthy group) and achieve dice scores of 0.945 (LVC), 0.908 (RVC) and 0.905 (LVM) in a cross-validation over the training set (100 cases) and 0.950 (LVC), 0.923 (RVC) and 0.911 (LVM) on the test set (50 cases). We report a classification accuracy of \(94 \%\) on a training set cross-validation and \(92\%\) on the test set. Our results underpin the potential of machine learning methods for accurate, fast and reproducible segmentation and computer-assisted diagnosis (CAD).
198 citations
TL;DR: In this article, an ensemble of UNet-inspired architectures was used for segmentation of cardiac structures such as the left and right ventricular cavity (LVC, RVC) and the left ventricular myocardium (LVM) on each time instance of the cardiac cycle.
Abstract: Cardiac magnetic resonance imaging improves on diagnosis of cardiovascular diseases by providing images at high spatiotemporal resolution. Manual evaluation of these time-series, however, is expensive and prone to biased and non-reproducible outcomes. In this paper, we present a method that addresses named limitations by integrating segmentation and disease classification into a fully automatic processing pipeline. We use an ensemble of UNet inspired architectures for segmentation of cardiac structures such as the left and right ventricular cavity (LVC, RVC) and the left ventricular myocardium (LVM) on each time instance of the cardiac cycle. For the classification task, information is extracted from the segmented time-series in form of comprehensive features handcrafted to reflect diagnostic clinical procedures. Based on these features we train an ensemble of heavily regularized multilayer perceptrons (MLP) and a random forest classifier to predict the pathologic target class. We evaluated our method on the ACDC dataset (4 pathology groups, 1 healthy group) and achieve dice scores of 0.945 (LVC), 0.908 (RVC) and 0.905 (LVM) in a cross-validation over the training set (100 cases) and 0.950 (LVC), 0.923 (RVC) and 0.911 (LVM) on the test set (50 cases). We report a classification accuracy of 94% on a training set cross-validation and 92% on the test set. Our results underpin the potential of machine learning methods for accurate, fast and reproducible segmentation and computer-assisted diagnosis (CAD).
101 citations
Posted Content•
TL;DR: It is proposed the establishment of a new ontology for data and algorithms in surgical data science, which can be used to provide common data sets for the community, encouraging sharing of knowledge and comparison of algorithms on common data.
Abstract: Every year approximately 234 million major surgeries are performed, leading to plentiful, highly diverse data. This is accompanied by a matching number of novel algorithms for the surgical domain. To garner all benefits of surgical data science it is necessary to have an unambiguous, shared understanding of algorithms and data. This includes inputs and outputs of algorithms and thus their function, but also the semantic content, i.e. meaning of data such as patient parameters. We therefore propose the establishment of a new ontology for data and algorithms in surgical data science. Such an ontology can be used to provide common data sets for the community, encouraging sharing of knowledge and comparison of algorithms on common data. We hold that this is a necessary foundation towards new methods for applications such as semantic-based content retrieval and similarity measures and that it is overall vital for the future of surgical data science.
5 citations
TL;DR: This novel comprehensive computer-based analysis reveals that the surveyed sizing methods led to the selection of significantly different annuloplasty rings and therefore underscore the ambiguity of routinely applied annuoplasty sizing strategies.
Abstract: Ring sizing for mitral valve annuloplasty is conventionally done intraoperatively using specific ‘sizer’ instruments, which are placed onto the valve tissue. This approach is barely reproducible since different sizing strategies have been established among surgeons. The goal of this study is to virtually apply different sizing methods on the basis of pre-repair echocardiography to find out basic differences between sizing strategies. In three-dimensional echocardiographs of 43 patients, the mitral annulus and the contour of the anterior mitral leaflet were segmented using MITK Mitralyzer software. Similarly, three-dimensional virtual models of Carpentier-Edwards Physio II annuloplasty rings and their corresponding sizers were interactively generated from computer tomography images. For each patient, the matching annuloplasty ring was selected repeatedly according to popular sizing strategies, such as the height of anterior mitral leaflet, the intercommissural distance and the surface area of anterior mitral leaflet. The areas of the selected rings were considered as the neo-surface area of the mitral annulus after implantation. The sizing of the mitral valve according to the height of anterior mitral leaflet (mean ring size = 29.9 ± 3.90), intercommissural distance (mean ring size = 37.5 ± 1.92) or surface area of anterior mitral leaflet (mean ring size = 32.7 ± 3.3) led to significantly different measurements (p ≤ 0.01). In contrary to intercommissural distance, height and surface area of the anterior mitral leaflet exhibited significant variations between the patients (p ≤ 0.01). The sizing according to the height of anterior mitral leaflet led to the maximal reduction of the mitral annulus surface area followed by the sizing according to the surface area of anterior mitral leaflet and finally by the intercommissural distance. This novel comprehensive computer-based analysis reveals that the surveyed sizing methods led to the selection of significantly different annuloplasty rings and therefore underscore the ambiguity of routinely applied annuloplasty sizing strategies.
4 citations
01 Jan 2017
TL;DR: This work focuses on the intelligent setup of the biomechanical model and the flexible interfaces of the HPC-based implementation of the resulting MVR simulation, thereby aiming at a cognition-guided, patient-specific integration into systems for surgery assistance.
Abstract: Medical simulations play an increasingly important role in today’s clinical and surgical treatment processes. The scope of this work is the support of the surgical operation of a mitral valve reconstruction (MVR) by means of biomechanical simulations. Based on numerical simulation, the natural anatomical setting, the ring implantation and the valve closure are modelled and efficiently computed in order to provide surgeons during the operation with additional morphological and functional information. Our simulation is based on the Finite Element Method (FEM) and implemented using the open-source C++ FEM software HiFlow3. Integrating patient data and surgical expert knowledge, and making efficient use of High-Performance Computing (HPC) methods allows for obtaining valuable simulation results for surgery assistance in adequate times. In this work, we focus on the intelligent setup of the biomechanical model and the flexible interfaces of the HPC-based implementation of the resulting MVR simulation, thereby aiming at a cognition-guided, patient-specific integration into systems for surgery assistance.
4 citations
TL;DR: Implantation of different ring types in patients with different annuli shapes and dimensions did not lead to any significant change in the configuration of mitral annuli after the virtual implantation of the tested annuloplasty rings.
Abstract: Background Different types of mitral annuloplasty rings are commercially available The aim of this study was to investigate the effect of implantation of six types of annuloplasty rings on the geometry and dynamics of the mitral valve Methods Three-dimensional echocardiography images of 42 patients were acquired to visualize the mitral valve annulus Virtual representations of six commercially available annuloplasty rings were matched to anatomical mitral annuli of each patient according to anterolateral-posteromedial diameter The virtual displacement of each annuloplasty ring after the implantation was measured and compared with the other rings Results Patients with severe mitral regurgitation had significantly dilated annuli according to anterolateral-posteromedial diameter, anterior-posterior diameter and to annulus circumference Anterior and posterior heights of the mitral annuli and non-planarity angle showed no significant differences among different patients with different degree of mitral regurgitation The ratio of anterior-posterior to anterolateral-posteromedial diameter was almost identical in all groups with identical annular shapes The implantation of the Carpentier-Edwards Classic Annuloplasty Ring™ led to maximal displacement of mitral annulus, followed by the IM-Ring™, without a statistical significance In contrary, the implantation of a MyxoETlogix Ring™ was associated with minimal displacement of mitral annulus throughout the groups, but without statistical significance Conclusions The implantation of different ring types in patients with different annuli shapes and dimensions did not lead to any significant change in the configuration of mitral annuli after the virtual implantation of the tested annuloplasty rings
2 citations
01 Jan 2017
TL;DR: Zur Unterstutzung onkologischer Interventionen konnen durch die Registrierung praoperativer Bildaten zu intraoperativen Cone-Beam-Computertomographieaufnahmen (CBCT) zusatzliche Informationen uber die Anatomie and Morphologie des Patienten erhalten werden.
Abstract: Zur Unterstutzung onkologischer Interventionen konnen durch die Registrierung praoperativer Bildaten zu intraoperativen Cone-Beam-Computertomographieaufnahmen (CBCT) zusatzliche Informationen uber die Anatomie und Morphologie des Patienten erhalten werden. In der vorliegenden Arbeit wird eine neuartige Metrik fur die gradientenbasierte Bildregistrierung vorgestellt.
TL;DR: The results show that 3D HOG feature descriptors are suitable as image-similarity metric and that the novel metric can compete with established methods in terms of registration accuracy.
Abstract: Image registration of preprocedural contrast-enhanced CTs to intraprocedual cone-beam computed tomography (CBCT) can provide additional information for interventional liver oncology procedures such as transcatheter arterial chemoembolisation (TACE). In this paper, a novel similarity metric for gradient-based image registration is proposed. The metric relies on the patch-based computation of histograms of oriented gradients (HOG) building the basis for a feature descriptor. The metric was implemented in a framework for rigid 3D-3D-registration of pre-interventional CT with intra-interventional CBCT data obtained during the workflow of a TACE. To evaluate the performance of the new metric, the capture range was estimated based on the calculation of the mean target registration error and compared to the results obtained with a normalized cross correlation metric. The results show that 3D HOG feature descriptors are suitable as image-similarity metric and that the novel metric can compete with established methods in terms of registration accuracy