Gerald Zwettler - Academia.edu (original) (raw)

Papers by Gerald Zwettler

2023 3rd International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME)

The precise orientation-estimation of humans relative to the pose of a monocular camera system is... more The precise orientation-estimation of humans relative to the pose of a monocular camera system is a challenging task due to the general aspects of camera calibration and the deformable nature of a human body in motion. Thus, novel approaches of Deep Learning for precise object pose-estimation in robotics are hard to adapt to human body analysis. In this work, a hybrid approach for the accurate estimation of a human body rotation relative to a camera system is presented, thereby significantly improving results derived from poseNet by applying analysis of optical flow in a frame to frame comparison. The human body in-place rotating in T-pose is thereby aligned in the center, applying object tracking methods to compensate for translations of the body movement. After 2D skeleton extraction, the optical flow is calculated for a region of interest (ROI) area aligned relative to the vertical skeleton joint representing the spine and compared frame by frame. To evaluate the eligibility of the clothing as a fundament for good feature, the local pixel homogeneity is taken into consideration to restrict the optical flow to heterogeneous regions with distinctive features like imprint patterns, buttons or buckles besides local illumination changes. Based on the mean optical flow with a coarse approximation of the axial body shape as ellipsis, an accuracy between 0.1° and 2.0° by a target rotation of 10° for orientation-estimation is achieved on a frame-toframe comparison evaluated and validated on both, Computer Generated Imagery (CGI) renderings and real-world videos of people wearing clothing of varying feature appropriateness.

Time activity curves are assessed from whole body szintigraphic time series for individual dose e... more Time activity curves are assessed from whole body szintigraphic time series for individual dose estimation to limit the applicable therapeutic dose, preventing critical organs from substantial damage during radiation therapy. Whole body scans are projective images, thus organ ROIs may overlap. After careful image registration rotated principal components analysis is applied to the time series, identifying image parts with similar dynamics. This method allows the separation of overlapping structures providing fuzzy regions, where fractions of the pixel counts are assigned to the respective accumulating morphology. The summed counts from theses fuzzy regions are modified with the physical decay constant of the considered therapeutic isotope, providing the correct time samples for further dose calculation. Mono-or bi-exponential regression yields the time activity curves for the respective morphologies, passed to a standard dose calculation program. The newly developed method allows the fuzzy separation of overlapping structures in projective planar imaging series, yielding more accurate dose calculation in individual radiation therapy.

Accurate and robust identification of the gyri and sulci of the human brain is a prerequisite of ... more Accurate and robust identification of the gyri and sulci of the human brain is a prerequisite of high importance for modelling the brain surface and thus to facilitate quantitative measurements and novel classification concepts. In this work we introduce a watershedinspired image processing strategy for topographical analysis of arbitrary surfaces in 3D. Thereby the object's topographical structure represented as depth profile is iteratively transformed into cyclic graph representations of both, the lowest and the highest characteristics of the particular shape. For graph analysis, the surface elements are partitioned according to their depth value. Neighbouring regions at different depth levels are iteratively merged. For region merging, the shape defining medial axes of the involved regions have to be connected by the optimum path with respect to a fitness function balancing shortness and minimal depth level changes of the solution.

Image guided surgery has established in modern surgery rooms, enabling high technology support fo... more Image guided surgery has established in modern surgery rooms, enabling high technology support for complicated surgical interventions. The ability to exactly position surgical tools, even if the target of surgery is subsurface, relying just on pre-acquired image data, causes the great success of surgical navigation. In cerebral surgery, image guidance has a long tradition, even in orthopedics; recently it also appears to abdominal surgery. A major prerequisite for accurate position navigation is the careful mutual registration of patient-, tracking-and imaging-domains. Only intuitive and precise handling of the registration procedure leads to satisfying results. An easy to use and accurate registration method, integrating the iterative closest point (ICP) algorithm was developed and implemented as showcase in a Matlab® based tracking environment. Image data from a diagnostic scan are preprocessed by anisotropic diffusion filtering and reformatted to cubic voxels. The point sets for registration are extracted from the image volume and acquired by a tracked pointing device. Rough reorientation of registration data is achieved by equalization of principal components. The ICP algorithm is applied to fully register both data sets. Accuracy of registration is quantified by distancemeasurements of the transformed tracking points from the surface and by measuring the summed distance of physical landmarks on the object's surface. The registration yields accurate overlay of the tracking and patient image domains, allowing exact navigation of surgical tools. The easy handling and accuracy of the developed registration method manifests the specific potential for clinical application.

Image quality in single photon emission computed tomography (SPECT) is substantially influenced b... more Image quality in single photon emission computed tomography (SPECT) is substantially influenced by scatter and a finite volume of response associated with single detector elements. These effects are not restricted to the image plane, implying a shift in the tomographic imaging paradigm from 2D to 3D. The application of a 3D reconstruction model suffers from huge numerical efforts, affording for high performance computing hardware. A novel accelerated 3D ML-EM type reconstruction algorithm is developed by the implementation of a dual projector back-projector pair. An accurate 3D model of data acquisition is developed considering scatter and exact scanner geometry in opposite to a simple pencil-beam back-projection operator. This dual concept of projection and backprojection substantially accelerates the reconstruction process. Speed-up factors achieved by the novel algorithm are measured for several matrix sizes and collimator types. Accuracy of the accelerated reconstruction algorithm is shown by reconstruction of data from a physical Jaszczak phantom and a clinical endocrine study. In both cases the accelerated 3D reconstruction method achieves better results. The novel algorithm has a great potential to scale fully 3D reconstruction down to desktop applications, especially with the new possibilities employing massive parallel graphics hardware. The presented work is a step towards establishing sophisticated 3D reconstruction in a clinical workflow.

Springer eBooks, 2012

Accurate diagnostic assessment of metabolic processes in nuclear medicine diagnostic imaging, eg ... more Accurate diagnostic assessment of metabolic processes in nuclear medicine diagnostic imaging, eg SPECT and PET, rely on specific localization of physiological activity. The major step in precise staging of neuro-degenerative diseases is robust, patient-specific ...

The inspection of products and the assessment of quality is connected with high costs and time ef... more The inspection of products and the assessment of quality is connected with high costs and time effort in many industrial domains. This also applies to the forestry industry. Utilizing state-of-the-art deep learning models allows the analysis automation of wooden piles in a vision-based manner. In this work a three-step approach is presented for the localization, segmentation and multi-facet classification of individual logs based on a client/server architecture allowing to determine the quality, volume and like this the value of a wooden pile based on a smartphone application. Using multiple YOLOv4 and U-NET models leads to a client-side log localization accuracy of 82.9% with low storage requirements of 23 MB and a server-side log detection accuracy of 94.1%, together with a log type classification accuracy of 95% and 96% according to the quality assessment of spruce logs. In addition, the trained segmentation model reaches an accuracy of 89%.

Whenever generic segmentation strategies utilizing multivariate feature classification are develo... more Whenever generic segmentation strategies utilizing multivariate feature classification are developed in the medical domain, testing and training solely on real world data is insufficient to validate all possible feature aspects. Instead, utilization of a simulator is highly recommended to test against different feature distributions with varying levels of correlation. In this work a simulator is presented, not only generating correlated feature values but also providing geometric representations of the regions to classify, as well as simulated intensity volumes. Geometric representations of predefined shape primitives are simulated utilizing randomized dilation for steering surface characteristics. For intensity volume generation, gradient magnitude level, intra region homogeneity and border layout can be parameterized. The adjustable inter dataset variability allows for generation of entire training and validation datasets for model-based segmentation approaches.

State of the art deep learning (DL) manifested in image processing as an accurate segmentation me... more State of the art deep learning (DL) manifested in image processing as an accurate segmentation method. Nevertheless, its black-box nature hardly allows user interference. In this paper, we present a generic Graph cut (GC) and Graph segmentation (GS) approach for user-guided interactive post-processing of segmentations resulting from DL. The GC fitness function incorporates both, the original image characteristics and DL segmentation results, combining them with weights optimized by evolution strategy optimization. To allow for accurate user-guided processing, the fore-and background seeds of the Graph cut are automatically selected from the DL segmentations, but implementing effective features for expert input for adaptions of position and topology. The seamless integration of DL with GC/GS leads to marginal trade-off in quality, namely Jaccard (JI) 1.3% for automated GC and JI 0.46% for GS only. Yet, in specific areas where a welltrained DL model may potentially fail, precise adaptions at a low demand for user-interaction become feasible and thus even outperforming the original DL results. The potential of GC/GS is shown running on groundtruth seeds thereby outperforming DL by 0.44% JI for the GC and even by 1.16% JI for the GS. Iterative sliceby-slice progression of the post-processed and improved results keeps the demand for user-interaction low.

An automated and generally applicable method for segmentation is still in focus of medical image ... more An automated and generally applicable method for segmentation is still in focus of medical image processing research. Since a few years artificial inteligence methods show promising results, especially with widely available scalable Deep Learning libraries. In this work, a five layer hybrid U-net is developed for slice-byslice segmentation of liver data sets. Training data is taken from the Medical Segmentation Decathlon database, providing 131 fully segmented volumes. A slice-oriented segmentation model is implemented utilizing deep learning algorithms with adaptions for variable parenchyma shape along the stacking direction and similarities between adjacent slices. Both are transformed for coronal and sagittal views. The implementation is on a GPU rack with TensorFlow and Keras. For a quantitative measure of segmentation accuracy, standardized volume and surface metrics are used. Results DSC=97.59, JI=95.29 and NSD=99.37 show proper segmentation comparable to 3D U-Nets and other state of the art. The development of a 2D-slice oriented segmentation is justified by short training time and less complexity and therefore massively reduced memory consumption. This work manifests the high potential of AI methods for general use in medical segmentation as fully-or semi-automated tool supervised by the expert user.

Springer eBooks, 2012

ABSTRACT Considering both, the ecological and economical aspects in building construction enginee... more ABSTRACT Considering both, the ecological and economical aspects in building construction engineering, is of high importance for a balanced and efficient building design. For high competitiveness on the markets, the need for novel efficiency metrics and automated optimization of the building plan arises. We introduce an efficiency measure for balancing the trade-off between construction cost and the heating demand as energy efficiency aspect. Thereby the building physics are precisely modeled and all possible variations of the particular material choice can be evaluated via simulation. By considering all possible variations of the particular plan, a large multi-dimensional search space can be defined, allowing search for the optimal design utilizing heuristic methods. Exploitation of the search space allows the quantitative assessment of plans with respect to the theoretical optimum and generally the qualitative comparison of different construction settings and different material choice for development of current best standard practice in building construction engineering.

In the light of increasing energy prices and declining fossil resources, energy efficient design ... more In the light of increasing energy prices and declining fossil resources, energy efficient design is an important aspect of building construction planning. Software application BauOptimizer supports the planner in calculating, monitoring and optimizing both energy demand and cost aspects from the very first planning phase until the final architecture improving economic and ecologic properties of the building design. Furthermore the number of required planning phases is reduced as normative limits are kept considered right from the beginning. Costs for building hull creation and expected energy costs for the next decades are linked together as efficiency measure and all planning variants applicable for a concrete construction area are automatically evaluated, covering different material at different thicknesses, the window ratio, roof modality, and many more aspects. Within this full construction variants coverage, the planner optimizes his architectural design and building physics aspect to approximate efficiency optimum with respect to economy and ecology.

Springer eBooks, 2013

The vendor specific interfaces and heterogeneous hardware of VR-devices is a major drawback for p... more The vendor specific interfaces and heterogeneous hardware of VR-devices is a major drawback for planning and realizing a VR-environment, necessitating an intermediate layer between the hardware and software interfaces, the hardware abstraction layer (HAL). In this work we present the implementation of a device server for generic and simple integration of devices like tracking tools, force sensors, micro controllers or 3D interaction devices. Utilizing this device server, harmonization of the message formats, data representations and transmission protocols is achieved. Furthermore, device-specific communication threads allow for precise timing capitalizing multi-core architecture of the host server. For application development, device-specific API code is automatically generated as a specific network proxy, thus allowing independence from programming language and platform. The presented device server is currently used for modern teaching aspects in the academic domain, as well in a research project developing a surgical training environment for kyphoplasty and vertebroplasty in the medical domain.

Three-dimensional segmentation of medical image data is crucial in modern diagnostics and still s... more Three-dimensional segmentation of medical image data is crucial in modern diagnostics and still subject of intensive research efforts. Most fully automated methods, e.g. the segmentation of the hippocampus, are highly specific for certain morphological regions and very sensitive to variations in input data, thus robustness is not sufficient to achieve sufficient accuracy to serve in differential diagnosis. In this work a processing pipeline for robust segmentation is presented. The flexibility of this novel generic segmentation method is based on entirely parameter-free pre-segmentation. Therefore a hybrid modification of the watershed algorithm is developed, employing both gradient and intensity metrics for the identification of connected regions depending on similar properties. In a further optimization step the vast number of small regions is condensed to anatomically meaningful structures by feature based classification. The core of the classification process is a topographical model of the segmented body region, representing a sufficient number of features from geometry and the texture domain. The model may learn from manual segmentation by experts or from its own results. The novel method is demonstrated for the human brain, based on the reference data set from brainweb. Results show high accuracy and the method proves to be robust. The method is easily extensible to other body regions and the novel concept shows high potential to introduce generic segmentation in the three-dimensional domain into a clinical work-flow. 104 Zwettler G. and Backfrieder W.. Generic 3D Segmentation in Medicine based on a Self-learning Topological Model.

2023 3rd International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME)

The precise orientation-estimation of humans relative to the pose of a monocular camera system is... more The precise orientation-estimation of humans relative to the pose of a monocular camera system is a challenging task due to the general aspects of camera calibration and the deformable nature of a human body in motion. Thus, novel approaches of Deep Learning for precise object pose-estimation in robotics are hard to adapt to human body analysis. In this work, a hybrid approach for the accurate estimation of a human body rotation relative to a camera system is presented, thereby significantly improving results derived from poseNet by applying analysis of optical flow in a frame to frame comparison. The human body in-place rotating in T-pose is thereby aligned in the center, applying object tracking methods to compensate for translations of the body movement. After 2D skeleton extraction, the optical flow is calculated for a region of interest (ROI) area aligned relative to the vertical skeleton joint representing the spine and compared frame by frame. To evaluate the eligibility of the clothing as a fundament for good feature, the local pixel homogeneity is taken into consideration to restrict the optical flow to heterogeneous regions with distinctive features like imprint patterns, buttons or buckles besides local illumination changes. Based on the mean optical flow with a coarse approximation of the axial body shape as ellipsis, an accuracy between 0.1° and 2.0° by a target rotation of 10° for orientation-estimation is achieved on a frame-toframe comparison evaluated and validated on both, Computer Generated Imagery (CGI) renderings and real-world videos of people wearing clothing of varying feature appropriateness.

Time activity curves are assessed from whole body szintigraphic time series for individual dose e... more Time activity curves are assessed from whole body szintigraphic time series for individual dose estimation to limit the applicable therapeutic dose, preventing critical organs from substantial damage during radiation therapy. Whole body scans are projective images, thus organ ROIs may overlap. After careful image registration rotated principal components analysis is applied to the time series, identifying image parts with similar dynamics. This method allows the separation of overlapping structures providing fuzzy regions, where fractions of the pixel counts are assigned to the respective accumulating morphology. The summed counts from theses fuzzy regions are modified with the physical decay constant of the considered therapeutic isotope, providing the correct time samples for further dose calculation. Mono-or bi-exponential regression yields the time activity curves for the respective morphologies, passed to a standard dose calculation program. The newly developed method allows the fuzzy separation of overlapping structures in projective planar imaging series, yielding more accurate dose calculation in individual radiation therapy.

Accurate and robust identification of the gyri and sulci of the human brain is a prerequisite of ... more Accurate and robust identification of the gyri and sulci of the human brain is a prerequisite of high importance for modelling the brain surface and thus to facilitate quantitative measurements and novel classification concepts. In this work we introduce a watershedinspired image processing strategy for topographical analysis of arbitrary surfaces in 3D. Thereby the object's topographical structure represented as depth profile is iteratively transformed into cyclic graph representations of both, the lowest and the highest characteristics of the particular shape. For graph analysis, the surface elements are partitioned according to their depth value. Neighbouring regions at different depth levels are iteratively merged. For region merging, the shape defining medial axes of the involved regions have to be connected by the optimum path with respect to a fitness function balancing shortness and minimal depth level changes of the solution.

Image guided surgery has established in modern surgery rooms, enabling high technology support fo... more Image guided surgery has established in modern surgery rooms, enabling high technology support for complicated surgical interventions. The ability to exactly position surgical tools, even if the target of surgery is subsurface, relying just on pre-acquired image data, causes the great success of surgical navigation. In cerebral surgery, image guidance has a long tradition, even in orthopedics; recently it also appears to abdominal surgery. A major prerequisite for accurate position navigation is the careful mutual registration of patient-, tracking-and imaging-domains. Only intuitive and precise handling of the registration procedure leads to satisfying results. An easy to use and accurate registration method, integrating the iterative closest point (ICP) algorithm was developed and implemented as showcase in a Matlab® based tracking environment. Image data from a diagnostic scan are preprocessed by anisotropic diffusion filtering and reformatted to cubic voxels. The point sets for registration are extracted from the image volume and acquired by a tracked pointing device. Rough reorientation of registration data is achieved by equalization of principal components. The ICP algorithm is applied to fully register both data sets. Accuracy of registration is quantified by distancemeasurements of the transformed tracking points from the surface and by measuring the summed distance of physical landmarks on the object's surface. The registration yields accurate overlay of the tracking and patient image domains, allowing exact navigation of surgical tools. The easy handling and accuracy of the developed registration method manifests the specific potential for clinical application.

Image quality in single photon emission computed tomography (SPECT) is substantially influenced b... more Image quality in single photon emission computed tomography (SPECT) is substantially influenced by scatter and a finite volume of response associated with single detector elements. These effects are not restricted to the image plane, implying a shift in the tomographic imaging paradigm from 2D to 3D. The application of a 3D reconstruction model suffers from huge numerical efforts, affording for high performance computing hardware. A novel accelerated 3D ML-EM type reconstruction algorithm is developed by the implementation of a dual projector back-projector pair. An accurate 3D model of data acquisition is developed considering scatter and exact scanner geometry in opposite to a simple pencil-beam back-projection operator. This dual concept of projection and backprojection substantially accelerates the reconstruction process. Speed-up factors achieved by the novel algorithm are measured for several matrix sizes and collimator types. Accuracy of the accelerated reconstruction algorithm is shown by reconstruction of data from a physical Jaszczak phantom and a clinical endocrine study. In both cases the accelerated 3D reconstruction method achieves better results. The novel algorithm has a great potential to scale fully 3D reconstruction down to desktop applications, especially with the new possibilities employing massive parallel graphics hardware. The presented work is a step towards establishing sophisticated 3D reconstruction in a clinical workflow.

Springer eBooks, 2012

Accurate diagnostic assessment of metabolic processes in nuclear medicine diagnostic imaging, eg ... more Accurate diagnostic assessment of metabolic processes in nuclear medicine diagnostic imaging, eg SPECT and PET, rely on specific localization of physiological activity. The major step in precise staging of neuro-degenerative diseases is robust, patient-specific ...

The inspection of products and the assessment of quality is connected with high costs and time ef... more The inspection of products and the assessment of quality is connected with high costs and time effort in many industrial domains. This also applies to the forestry industry. Utilizing state-of-the-art deep learning models allows the analysis automation of wooden piles in a vision-based manner. In this work a three-step approach is presented for the localization, segmentation and multi-facet classification of individual logs based on a client/server architecture allowing to determine the quality, volume and like this the value of a wooden pile based on a smartphone application. Using multiple YOLOv4 and U-NET models leads to a client-side log localization accuracy of 82.9% with low storage requirements of 23 MB and a server-side log detection accuracy of 94.1%, together with a log type classification accuracy of 95% and 96% according to the quality assessment of spruce logs. In addition, the trained segmentation model reaches an accuracy of 89%.

Whenever generic segmentation strategies utilizing multivariate feature classification are develo... more Whenever generic segmentation strategies utilizing multivariate feature classification are developed in the medical domain, testing and training solely on real world data is insufficient to validate all possible feature aspects. Instead, utilization of a simulator is highly recommended to test against different feature distributions with varying levels of correlation. In this work a simulator is presented, not only generating correlated feature values but also providing geometric representations of the regions to classify, as well as simulated intensity volumes. Geometric representations of predefined shape primitives are simulated utilizing randomized dilation for steering surface characteristics. For intensity volume generation, gradient magnitude level, intra region homogeneity and border layout can be parameterized. The adjustable inter dataset variability allows for generation of entire training and validation datasets for model-based segmentation approaches.

State of the art deep learning (DL) manifested in image processing as an accurate segmentation me... more State of the art deep learning (DL) manifested in image processing as an accurate segmentation method. Nevertheless, its black-box nature hardly allows user interference. In this paper, we present a generic Graph cut (GC) and Graph segmentation (GS) approach for user-guided interactive post-processing of segmentations resulting from DL. The GC fitness function incorporates both, the original image characteristics and DL segmentation results, combining them with weights optimized by evolution strategy optimization. To allow for accurate user-guided processing, the fore-and background seeds of the Graph cut are automatically selected from the DL segmentations, but implementing effective features for expert input for adaptions of position and topology. The seamless integration of DL with GC/GS leads to marginal trade-off in quality, namely Jaccard (JI) 1.3% for automated GC and JI 0.46% for GS only. Yet, in specific areas where a welltrained DL model may potentially fail, precise adaptions at a low demand for user-interaction become feasible and thus even outperforming the original DL results. The potential of GC/GS is shown running on groundtruth seeds thereby outperforming DL by 0.44% JI for the GC and even by 1.16% JI for the GS. Iterative sliceby-slice progression of the post-processed and improved results keeps the demand for user-interaction low.

An automated and generally applicable method for segmentation is still in focus of medical image ... more An automated and generally applicable method for segmentation is still in focus of medical image processing research. Since a few years artificial inteligence methods show promising results, especially with widely available scalable Deep Learning libraries. In this work, a five layer hybrid U-net is developed for slice-byslice segmentation of liver data sets. Training data is taken from the Medical Segmentation Decathlon database, providing 131 fully segmented volumes. A slice-oriented segmentation model is implemented utilizing deep learning algorithms with adaptions for variable parenchyma shape along the stacking direction and similarities between adjacent slices. Both are transformed for coronal and sagittal views. The implementation is on a GPU rack with TensorFlow and Keras. For a quantitative measure of segmentation accuracy, standardized volume and surface metrics are used. Results DSC=97.59, JI=95.29 and NSD=99.37 show proper segmentation comparable to 3D U-Nets and other state of the art. The development of a 2D-slice oriented segmentation is justified by short training time and less complexity and therefore massively reduced memory consumption. This work manifests the high potential of AI methods for general use in medical segmentation as fully-or semi-automated tool supervised by the expert user.

Springer eBooks, 2012

ABSTRACT Considering both, the ecological and economical aspects in building construction enginee... more ABSTRACT Considering both, the ecological and economical aspects in building construction engineering, is of high importance for a balanced and efficient building design. For high competitiveness on the markets, the need for novel efficiency metrics and automated optimization of the building plan arises. We introduce an efficiency measure for balancing the trade-off between construction cost and the heating demand as energy efficiency aspect. Thereby the building physics are precisely modeled and all possible variations of the particular material choice can be evaluated via simulation. By considering all possible variations of the particular plan, a large multi-dimensional search space can be defined, allowing search for the optimal design utilizing heuristic methods. Exploitation of the search space allows the quantitative assessment of plans with respect to the theoretical optimum and generally the qualitative comparison of different construction settings and different material choice for development of current best standard practice in building construction engineering.

In the light of increasing energy prices and declining fossil resources, energy efficient design ... more In the light of increasing energy prices and declining fossil resources, energy efficient design is an important aspect of building construction planning. Software application BauOptimizer supports the planner in calculating, monitoring and optimizing both energy demand and cost aspects from the very first planning phase until the final architecture improving economic and ecologic properties of the building design. Furthermore the number of required planning phases is reduced as normative limits are kept considered right from the beginning. Costs for building hull creation and expected energy costs for the next decades are linked together as efficiency measure and all planning variants applicable for a concrete construction area are automatically evaluated, covering different material at different thicknesses, the window ratio, roof modality, and many more aspects. Within this full construction variants coverage, the planner optimizes his architectural design and building physics aspect to approximate efficiency optimum with respect to economy and ecology.

Springer eBooks, 2013

The vendor specific interfaces and heterogeneous hardware of VR-devices is a major drawback for p... more The vendor specific interfaces and heterogeneous hardware of VR-devices is a major drawback for planning and realizing a VR-environment, necessitating an intermediate layer between the hardware and software interfaces, the hardware abstraction layer (HAL). In this work we present the implementation of a device server for generic and simple integration of devices like tracking tools, force sensors, micro controllers or 3D interaction devices. Utilizing this device server, harmonization of the message formats, data representations and transmission protocols is achieved. Furthermore, device-specific communication threads allow for precise timing capitalizing multi-core architecture of the host server. For application development, device-specific API code is automatically generated as a specific network proxy, thus allowing independence from programming language and platform. The presented device server is currently used for modern teaching aspects in the academic domain, as well in a research project developing a surgical training environment for kyphoplasty and vertebroplasty in the medical domain.

Three-dimensional segmentation of medical image data is crucial in modern diagnostics and still s... more Three-dimensional segmentation of medical image data is crucial in modern diagnostics and still subject of intensive research efforts. Most fully automated methods, e.g. the segmentation of the hippocampus, are highly specific for certain morphological regions and very sensitive to variations in input data, thus robustness is not sufficient to achieve sufficient accuracy to serve in differential diagnosis. In this work a processing pipeline for robust segmentation is presented. The flexibility of this novel generic segmentation method is based on entirely parameter-free pre-segmentation. Therefore a hybrid modification of the watershed algorithm is developed, employing both gradient and intensity metrics for the identification of connected regions depending on similar properties. In a further optimization step the vast number of small regions is condensed to anatomically meaningful structures by feature based classification. The core of the classification process is a topographical model of the segmented body region, representing a sufficient number of features from geometry and the texture domain. The model may learn from manual segmentation by experts or from its own results. The novel method is demonstrated for the human brain, based on the reference data set from brainweb. Results show high accuracy and the method proves to be robust. The method is easily extensible to other body regions and the novel concept shows high potential to introduce generic segmentation in the three-dimensional domain into a clinical work-flow. 104 Zwettler G. and Backfrieder W.. Generic 3D Segmentation in Medicine based on a Self-learning Topological Model.