1,721,241 research outputs found
Visual Analytics for Digital Radiotherapy: Towards a Comprehensible Pipeline
No full textProstate cancer is one of the most frequently occurring types of cancer in males. It is often treated with radiation therapy,which aims at irradiating tumors with a high dose, while sparing the surrounding healthy tissues. In the course of the years,radiotherapy technology has undergone great advancements. However, tumors are not only different from each other, theyare also highly heterogeneous within, consisting of regions with distinct tissue characteristics, which should be treated withdifferent radiation doses. Tailoring radiotherapy planning to the specific needs and intra-tumor tissue characteristics of eachpatient is expected to lead to more effective treatment strategies. Currently, clinical research is moving towards this direction,but an understanding of the specific tumor characteristics of each patient, and the integration of all available knowledge into apersonalizable radiotherapy planning pipeline are still required. The present work describes solutions from the field of VisualAnalytics, which aim at incorporating the information from the distinct steps of the personalizable radiotherapy planningpipeline, along with eventual sources of uncertainty, into comprehensible visualizations. All proposed solutions are meantto increase the - up to now, limited - understanding and exploratory capabilities of clinical researchers. These approachescontribute towards the interactive exploration, visual analysis and understanding of the involved data and processes at differentsteps of the radiotherapy planning pipeline, creating a fertile ground for future research in radiotherapy planning
Strategies for alleviating overdraw effects in information visualization
No full textWe are surrounded by devices that capture data. From personal devices like mobile phones and smartwatches to large industrial machines with thousands of sensors. This data can be of great value in different aspects, from optimizing personal fitness training to predictive maintenance of industrial machines. While statistical analysis allows insights from this data, such analysis usually requires some hypotheses of present patterns. However, visualization allows for an exploratory ap- proach to finding possible patterns. Designing effective visualizations is a complex process, especially as the amount and complexity of data increases.
One of the most important visual artifacts that can hinder the ability to read visualizations is overdrawing between visual elements. While this might not be a big problem for small datasets, as the size of data increases, this issue gets more and more prominent. Aggregation techniques like box plots or histograms can solve this issue but do not allow to see individual data points. In this thesis, I present a series of publications that address the problem of overdraw for a wide range of types of data and various visualization techniques.
In the first paper, we minimize the overdraw appearing in a Jitter Plot by applying Lloyd relaxation. Another publication presents a perception-based measurement for overdraw in scatterplots and derives a recommendation of shapes that appear to suffer less from overdraw, mitigating the adverse effects. Next, I present a series of encodings for missing values in parallel coordinates, a visualization technique usually involving strong overdraw. Using these encodings, we show that, while suffering from overdraw, the choice of viable encodings is critical to mitigating the negative effects. Finally, I investigate two glyph-based encodings in two-dimensional embeddings to visualize high-dimensional data in their ability to enable pattern identification. Here, we use complex glyphs in scenarios where the glyphs suffer from strong visual clutter, including overdraw.
From the publications presented in this thesis, we show that while minimization of overdraw is an essential factor when designing a visualization, it is not always possible. In these scenarios, where over- draw can not be avoided, the choice of viable visual encodings can significantly help mitigate the adverse effects
Proceedings of SIGRAD 2013, Visual Computing, June 13-14, 2013, Norrköping, Sweden
No full textSIGRAD 2013 We are happy to announce the 12th SIGRAD Conference Proceedings. SIGRAD 2013 will be held in Norrköping, Sweden, on June 13 and 14, 2013. SIGRAD 2013 focuses on visual computing, and solicits the submission of original research papers that advance the state-ofthe-art of one of the subareas of visual computing, ranging from computer graphics and visualization to human-computer-interaction. SIGRAD 2013 is the premier Nordic forum for computer graphics and visualization advances for academia, and industry. This annual event brings together researchers and practitioners with interest in techniques, tools, and technology from various fields such as computer graphics,visualization, visual analytics, or human-computer interaction. Each paper in this conference proceedings was peerreviewed by at least three reviewers from the international program committee consisting of 26 experts listed below. Based on this set of reviews, the conference co-chairs accepted 9 papers in total and compiled the final program. Jonas Unger and Timo Ropinsk
Projection-driven medical visualization
No full textProjection techniques are one of the fundamental methods to transform data and make it available to the viewer when it comes to medical visualization. In this context, the type of data involved is spatial, such as CT volumes or ultrasound scans. Modern imaging modalities capture a lot of information, however, not all of it is always of great importance. Medical visualization applications often hide dispensable data and accentuate important details to focus on the individual task at hand. Combining the right information with an effective visual encoding is a difficult challenge. Nevertheless, the goal is to enable an expert to form a well-founded decision, which allows the best possible treatment of a patient.
This dissertation examines projections combined with other forms of effective visualization methods to create novel medical visualization approaches. The here presented work comprises of a survey regarding flattening-based medical visualizations, among several individual visualization techniques for specific applications. Some of the addressed medical procedures in this dissertation have so far not gained a lot of attention. We have also conducted several studies to investigate if our developed approaches successfully support the users with their tasks. Additionally, we introduce a design space for our visualizations to characterize their fundamental composition.In reference to IEEE copyrighted material which is used with permission in this thesis, the IEEE does not endorse any of Ulm University’s products or services. Internal or personal use of this material is permitted. If interested in reprinting/republishing IEEE copyrighted material for advertising or promotional purposes or for creating new collective works for resale or redistribution, please go to http://www.ieee.org/ publications standards/publications/rights/rights_link.html to learn how to obtain a License from RightsLink
Going Beyond Counting First Authors in Author Co-citation Analysis
No full textThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Improving interactive rendering of volumetric effects in computer graphics and visualization
No full textVolumetric effects are some of the most challenging effects to render in the field of computer generated images. Depending on the data to generate images from, the challenge lies either in the time budget, or the memory resources available during rendering. Long rendering times often occur when a physically based approach is taken in presence of participating media. Since the goal of physically based rendering is to generate images that are as close to a photo as possible, the rendering process needs to take into account finest details. Rendering large volume data sets, on the other hand, is more severely affected by memory shortage, as these data sets are often present in case of data visualization of three or four dimensional medical or, more general, scientific data. Fortunately, in many cases not all details are important for the viewer and can therefore be omitted, while other details might need to be emphasized to attain an effective visualization.
This dissertation aims to address both of these challenges encountered when generating images capturing volumetric effects. It presents several novel approaches to physically based rendering that improve image quality in comparison to other state of the art techniques in that field, while keeping the impact on rendering times low. These approaches include a deep learning technique based on point cloud data, as well as several classic rasterization methods that have a special focus on the translucency effect.
The dissertation also includes an evaluation that compares the visual impact of different methods of memory reduction techniques for volume data
Proceedings of SIGRAD 2013, Visual Computing, June 13-14, 2013, Norrköping, Sweden
No full textSIGRAD 2013 We are happy to announce the 12th SIGRAD Conference Proceedings. SIGRAD 2013 will be held in Norrköping, Sweden, on June 13 and 14, 2013. SIGRAD 2013 focuses on visual computing, and solicits the submission of original research papers that advance the state-ofthe-art of one of the subareas of visual computing, ranging from computer graphics and visualization to human-computer-interaction. SIGRAD 2013 is the premier Nordic forum for computer graphics and visualization advances for academia, and industry. This annual event brings together researchers and practitioners with interest in techniques, tools, and technology from various fields such as computer graphics,visualization, visual analytics, or human-computer interaction. Each paper in this conference proceedings was peerreviewed by at least three reviewers from the international program committee consisting of 26 experts listed below. Based on this set of reviews, the conference co-chairs accepted 9 papers in total and compiled the final program. Jonas Unger and Timo Ropinsk
Visualization-based neural network introspection
No full textArtificial intelligence (AI) and the use of neural networks have become omnipresent in recent years. Neural networks model complex mathematical functions that can be based on billions, or even trillions, of parameters. At the same time, neural networks make decisions that can deeply impact our lives. Therefore, understanding, testing, and interpreting these networks and their decisions is an integral part of model development and deployment. While there exist introspection techniques that support such understanding, testing, and interpretation, their adoption for diagnosing systems and explaining decisions can be difficult. Current problems with the adoption of introspection techniques are that they are not easily implemented in one's framework, do not work well in combination to create more meaningful analyses, and are difficult to interpret.
Through the integration of existing and novel introspection techniques into visualization interfaces, extensive analysis, actionable insights, and effective diagnosis are made widely available. These visualization interfaces can be incorporated into existing development workflows and are designed to support bespoke analysis needs, which makes the interpretation of findings easier. In this thesis, we present published visualization interfaces in three different areas, namely quality assurance, communication, and AI education. These publications include a visualization approach for correcting mislabeled training data, an interface for automatic network figure generation to communicate network architectures, and an educational environment for recurrent neural networks (RNNs). Finally, to unify the diverse landscape of AI visualization interfaces, we also present a framework for composing, reusing, exploring, and sharing such interactive machine learning (ML) interfaces
Viewpoint guided learning for 3D scene understanding and representations
No full textDeep Learning has become the dominating technique in many domains, including computer vision. The visual perception, of both
humans and machines, is rooted in the 2D space, as visual sensors are
ultimately only able to capture a 2D representation from their specific
viewpoint. Even 3D sensors, such as indirect Time-of-Flight (iToF)
cameras, can only perceive information which is visible from their
respective viewpoint, and thus can only model the observed 3D space
as a 2D manifold, which is commonly referred to as a 2.5D image.
Consequently, the visual observation is highly dependent on both
the observed 3D geometry and the chosen viewpoint. This interdependency of the viewpoint and the observed 3D structure can be
exploited to improve the performance of neural networks for visual
tasks. To this end, this dissertation presents algorithmic optimizations
for neural network training, which allow for the use of viewpoint
information when addressing 3D problems and, vice versa, the use of
3D information when addressing viewpoint related problems.
First, a dynamic labeling strategy is presented, which enables 3D
point networks to identify informative viewpoints for 3D models.
Second, it is shown how the integration of information about the
view direction into 3D point convolutional networks can improve the
error correction rates for iToF cameras. Lastly, a training method for
compensating motion artifacts in iToF images, e.g., through changes
of the viewpoint, is derived, which allows for the supervision of 2D
networks via the reconstructed 2.5D depth image.In reference to IEEE copyrighted material which is used with permission in this thesis, the IEEE does not endorse any of Ulm University’s products or services. Internal or personal use of this material is permitted. If interested in reprinting/republishing IEEE copyrighted material for advertising or promotional purposes or for creating new collective works for resale or redistribution, please go to http://www.ieee.org/ publications standards/publications/rights/rights_link.html to learn how to obtain a License from RightsLink
- …
