Estimating actual SARS-CoV-2 infections from secondary data.

Eminent in pandemic management is accurate information on infection dynamics to plan for timely installation of control measures and vaccination campaigns. Despite huge efforts in diagnostic testing of individuals, the underestimation of the actual number of SARS-CoV-2 infections remains significant due to the large number of undocumented cases. In this paper we demonstrate and compare three methods to estimate the dynamics of true infections based on secondary data i.e., (a) test positivity, (b) infection fatality and (c) wastewater monitoring. The concept is tested with Austrian data on a national basis for the period of April 2020 to December 2022. Further, we use the results of prevalence studies from the same period to generate (upper and lower bounds of) credible intervals for true infections for four data points. Model parameters are subsequently estimated by applying Approximate Bayesian Computation-rejection sampling and Genetic Algorithms. The method is then validated for the case study Vienna. We find that all three methods yield fairly similar results for estimating the true number of infections, which supports the idea that all three datasets contain similar baseline information. None of them is considered superior, as their advantages and shortcomings depend on the specific case study at hand.

Synthetic Optical Coherence Tomography Angiographs for Detailed Retinal Vessel Segmentation Without Human Annotations.

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.

Towards liver segmentation in the wild via contrastive distillation.

PURPOSE: Automatic liver segmentation is a key component for performing computer-assisted hepatic procedures. The task is challenging due to the high variability in organ appearance, numerous imaging modalities, and limited availability of labels. Moreover, strong generalization performance is required in real-world scenarios. However, existing supervised methods cannot be applied to data not seen during training (i.e. in the wild) because they generalize poorly. METHODS: We propose to distill knowledge from a powerful model with our novel contrastive distillation scheme. We use a pre-trained large neural network to train our smaller model. A key novelty is to map neighboring slices close together in the latent representation, while mapping distant slices far away. Then, we use ground-truth labels to learn a U-Net style upsampling path and recover the segmentation map. RESULTS: The pipeline is proven to be robust enough to perform state-of-the-art inference on target unseen domains. We carried out an extensive experimental validation using six common abdominal datasets, covering multiple modalities, as well as 18 patient datasets from the Innsbruck University Hospital. A sub-second inference time and a data-efficient training pipeline make it possible to scale our method to real-world conditions. CONCLUSION: We propose a novel contrastive distillation scheme for automatic liver segmentation. A limited set of assumptions and superior performance to state-of-the-art techniques make our method a candidate for application to real-world scenarios.

Machine learning for quantile regression of biogas production rates in anaerobic digesters.

Anaerobic digestion is a well-established tool at wastewater treatment plants for processing raw sludge; it can also be used to generate renewable energy by harvesting biogas in anaerobic digesters. Operational parameters, such as temperature, are usually set by plant operators according to expert knowledge. To completely utilize the potential of operational management, in this study, we calibrated a novel Temporal Fusion Transformer based on six years of life-scale time series data together with categorical features such as public holidays. The model design allows for the interpretability of the output in contrast to traditional data-driven techniques, using multi-head attention. In addition to forecasting the median biogas production rates for the following seven days, our model also yields quantiles, making it less prone to strong fluctuations. We used three well-known statistical techniques as benchmarks. The mean absolute percentage error of our forecasting approach is below 8 %.

Automatic modelling of human musculoskeletal ligaments - Framework overview and model quality evaluation.

BACKGROUND: Accurate segmentation of connective soft tissues in medical images is very challenging, hampering the generation of geometric models for bio-mechanical computations. Alternatively, one could predict ligament insertion sites and then approximate the shapes, based on anatomical knowledge and morphological studies. OBJECTIVE: In this work, we describe an integrated framework for automatic modelling of human musculoskeletal ligaments. METHOD: We combine statistical shape modelling with geometric algorithms to automatically identify insertion sites, based on which geometric surface/volume meshes are created. As clinical use case, the framework has been applied to generate models of the forearm interosseous membrane. Ligament insertion sites in the statistical model were defined according to anatomical predictions following a published approach. RESULTS: For evaluation we compared the generated sites, as well as the ligament shapes, to data obtained from a cadaveric study, involving five forearms with 15 ligaments. Our framework permitted the creation of 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. Average mean square errors as well as Hausdorff distances of the meshes could increase by an order of magnitude, as compared to employing known insertion locations of the cadaveric study. Using those, an average mean square error of 0.59 mm and an average Hausdorff distance of less than 7 mm resulted, for all ligaments. CONCLUSIONS: The presented approach for automatic generation of ligament shapes from insertion points appears to be feasible but the detection of the insertion sites with a SSM is too inaccurate, thus making a patient-specific approach necessary.

Design and Characterization of an Actuated Drill Mockup for Orthopedic Surgical Training.

Haptic feedback in virtual reality-based trainers for surgical bone drilling is mostly provided via impedance-controlled haptic devices. Due to this, the displayable maximum stiffness is limited. In addition, vibration feedback is often only of reduced fidelity. To overcome these shortcomings, we have developed a hand-held, actuated admittance-controlled drill mockup, comprising enhanced kinesthetic and tactile feedback. This article reports on design and characterization of the device, and highlights its use for training. Kinesthetic feedback is provided through haptic augmentation, employing a ball-screw mechanism acting on a retractable drill-bit. Feedback computation relies on admittance control, thus allowing for stable display of very high resistance forces, and thus material stiffness, which cannot be achieved with standard impedance-control approaches. For the tactile mechanism, a modified linear vibration actuator is directly attached to the mockup handle, improving signal transmission. Tactile feedback computation is based on an extension of a previously proposed power spectral density control method. Frequency-specific gains are adjusted in real-time, compensating for differences between desired and measured vibrations. The performance of the device is characterized in several experiments, including comparisons to drilling with a real drill into artificial bone samples. In addition, several user studies have been carried out. We illustrate the capability of the mockup to render bone samples with different material layer stiffness and thickness. Moreover, we show that the mockup system allows for the same training effect as when rehearsing with a real drill.

Wrinkles, Folds, Creases, Buckles: Small-Scale Surface Deformations as Periodic Functions on 3D Meshes.

We propose a method for adding small-scale details to surfaces of 3D geometries in the context of interactive deformation computation of elastic objects. This is relevant in real-time applications, for instance, in surgical simulation or interactive animation. The key idea is the procedural generation of surface details via a weighted sum of periodic functions, applied as an on-surface displacement field. We first calculate local deformation strains of a low-resolution 3D input mesh, which are then employed to estimate amplitudes, orientations, and positions of high-resolution details. The shapes and spatial frequencies of the periodic details are obtained from mechanical parameters, assuming the physical model of a film-substrate aggregate. Finally, our approach creates the highly-detailed output mesh fully on the GPU. The performance is independent of the spatial frequency of the inserted details as well as, within certain limits, of the resolution of the output mesh. We can reproduce numerous commonly observed, characteristic surface deformation patterns, such as wrinkles or buckles, allowing for the representation of a wide variety of simulated materials and interaction processes. We highlight the performance of our method with several examples.

Exploring Feature-Based Learning for Data-Driven Haptic Rendering.

In this work we extend ideas of machine learning to the domain of data-driven haptic rendering. The proposed approach facilitates the processing of high-dimensional haptic interaction signals, which so far proved too difficult for existing data-driven methods. The key idea is to construct a compact feature space in the frequency domain which allows for efficient data reduction via a feature selection process. First, in a recording stage, extensive force and displacement datasets are acquired in automated measurements on deformable sample objects. These data are then transformed into a dimensionally reduced, compact frequency space representation. Next, feature-based learning is carried out in this feature space to significantly reduce the size of the original dataset. Based on this, time-domain haptic models capable of real-time performance are finally generated to encode the forces arising from bimanual object interactions. The presented processing chain is generally applicable and extendable to more complex interactions with even higher-dimensional data. The resulting haptic models are directly usable for data-driven haptic rendering. We illustrate the improved performance in comparison with previously existing data-processing approaches.

Identification of Vibrotactile Patterns Encoding Obstacle Distance Information.

Delivering distance information of nearby obstacles from sensors embedded in a white cane-in addition to the intrinsic mechanical feedback from the cane-can aid the visually impaired in ambulating independently. Haptics is a common modality for conveying such information to cane users, typically in the form of vibrotactile signals. In this context, we investigated the effect of tactile rendering methods, tactile feedback configurations and directions of tactile flow on the identification of obstacle distance. Three tactile rendering methods with temporal variation only, spatio-temporal variation and spatial/temporal/intensity variation were investigated for two vibration feedback configurations. Results showed a significant interaction between tactile rendering method and feedback configuration. Spatio-temporal variation generally resulted in high correct identification rates for both feedback configurations. In the case of the four-finger vibration, tactile rendering with spatial/temporal/intensity variation also resulted in high distance identification rate. Further, participants expressed their preference for the four-finger vibration over the single-finger vibration in a survey. Both preferred rendering methods with spatio-temporal variation and spatial/temporal/intensity variation for the four-finger vibration could convey obstacle distance information with low workload. Overall, the presented findings provide valuable insights and guidance for the design of haptic displays for electronic travel aids for the visually impaired.

Evaluation of a virtual-reality-based simulator using passive haptic feedback for knee arthroscopy.

PURPOSE: The aim of this work is to determine face validity and construct validity of a new virtual-reality-based simulator for diagnostic and therapeutic knee arthroscopy. METHODS: The study tests a novel arthroscopic simulator based on passive haptics. Sixty-eight participants were grouped into novices, intermediates, and experts. All participants completed two exercises. In order to establish face validity, all participants filled out a questionnaire concerning different aspects of simulator realism, training capacity, and different statements using a seven-point Likert scale (range 1-7). Construct validity was tested by comparing various simulator metric values between novices and experts. RESULTS: Face validity could be established: overall realism was rated with a mean value of 5.5 points. Global training capacity scored a mean value of 5.9. Participants considered the simulator as useful for procedural training of diagnostic and therapeutic arthroscopy. In the foreign body removal exercise, experts were overall significantly faster in the whole procedure (6 min 24 s vs. 8 min 24 s, p < 0.001), took less time to complete the diagnostic tour (2 min 49 s vs. 3 min 32 s, p = 0.027), and had a shorter camera path length (186 vs. 246 cm, p = 0.006). CONCLUSION: The simulator achieved high scores in terms of realism. It was regarded as a useful training tool, which is also capable of differentiating between varying levels of arthroscopic experience. Nevertheless, further improvements of the simulator especially in the field of therapeutic arthroscopy are desirable. In general, the findings support that virtual-reality-based simulation using passive haptics has the potential to complement conventional training of knee arthroscopy skills. LEVEL OF EVIDENCE: II.

Haptic tumor augmentation: exploring multi-point interaction.

We currently explore the application of haptic augmentation in the context of palpation training systems. The key idea is to modify real touch sensations with computed haptic feedback. In earlier work, we have introduced an algorithmic framework for determining appropriate augmentation forces during interaction at one contact point. In this paper, we present an extension of the approach to deal with manipulations at more than one contact location. At the heart of our method is the data-driven estimation of Hunt-Crossley model parameters in a pre-computation step. Feeding the parameters into a contact dynamics model allows us to approximate the feedback behavior of various physical tissue mock-ups. Further, we combine the parameter estimation with the tracking of the position of a stiffer inclusion in the mock-up. These data are employed to create a model of movement due to external forces. The combination of these models then allows us to represent and render the mutual effects at multiple contact points. Several experiments have been carried out on a setup with two haptic devices. Comparisons of recorded with simulated interaction data demonstrate the performance and potential of our method.

Data-Driven Simulation of Detailed Surface Deformations for Surgery Training Simulators.

Data-driven methods have received increasing attention in recent years in order to meet real-time requirements in computationally intensive tasks. In our current work we examine the application of such approaches in soft-tissue simulation. The core idea is to split deformations into a coarse approximation and a differential part that contains the details. We employ the data-driven stamping approach to enrich a fast simulation surface with details that have been extracted from a set of example deformations obtained in offline computations. In this paper we detail our technique, and suggest further extensions over our previous work. First, we propose an improved method for correlating the current coarse approximation to the examples in the database. The new correlation metric combines Euclidean distances with cosine similarity. It allows for better example discrimination, resulting in a well-conditioned linear system. This also enables us to use a non-negative least squares solver that leads to a better regression and guarantees positive stamp blending weights. Second, we suggest a frequency-space stamp compression scheme that saves memory and, in most instances, is faster, since many operations can be done in the compressed space. Third, cutting is included by employing a physically-inspired influence map that allows for proper handling of material discontinuities that were not present in the original examples. We thoroughly evaluate our method and demonstrate its practical application in a surgical simulator prototype.

Adaptive space warping to enhance passive haptics in an arthroscopy surgical simulator.

Passive haptics, also known as tactile augmentation, denotes the use of a physical counterpart to a virtual environment to provide tactile feedback. Employing passive haptics can result in more realistic touch sensations than those from active force feedback, especially for rigid contacts. However, changes in the virtual environment would necessitate modifications of the physical counterparts. In recent work space warping has been proposed as one solution to overcome this limitation. In this technique virtual space is distorted such that a variety of virtual models can be mapped onto one single physical object. In this paper, we propose as an extension adaptive space warping; we show how this technique can be employed in a mixed-reality surgical training simulator in order to map different virtual patients onto one physical anatomical model. We developed methods to warp different organ geometries onto one physical mock-up, to handle different mechanical behaviors of the virtual patients, and to allow interactive modifications of the virtual structures, while the physical counterparts remain unchanged. Various practical examples underline the wide applicability of our approach. To the best of our knowledge this is the first practical usage of such a technique in the specific context of interactive medical training.

Robust interactive collision handling between tools and thin volumetric objects.

Treating the interactions of soft tissue with rigid user-guided tools is a difficult problem. This is particularly true if the soft tissue has a slender shape, i.e., resembling a thin shell, and if the underlying numerical time-integration scheme employs large time steps. In this case, large mutual displacements of both the tool and the soft tissue occur frequently, resulting in deep interpenetrations or breakthroughs. As a consequence, the computation of spatially and temporally coherent contact spaces turns out to be very challenging. In this paper, an approach is proposed that is tailored to these kinds of interactions. To solve this problem, a novel spatially reduced representation of the soft tissue geometry is employed where the dominant dimensions of the object are approximated by a 2D triangle surface, while the third dimension is given in terms of nodal radii. To construct a feasible, nonpenetrating configuration, a novel manifold projection scheme is presented where the colliding triangles are rasterized into a distance field in order to robustly estimate the contact spaces, even for large intersections. The method produces physically plausible results, albeit it is purely geometric, and the material parameters are neglected at the collision response stage. Various examples, including an interactive prototype arthroscopy simulator, underline the wide applicability of the approach.

Maintaining large time steps in explicit finite element simulations using shape matching.

We present a novel hybrid method to allow large time steps in explicit integrations for the simulation of deformable objects. In explicit integration schemes, the time step is typically limited by the size and the shape of the discretization elements as well as by the material parameters. We propose a two-step strategy to enable large time steps for meshes with elements potentially destabilizing the integration. First, the necessary time step for a stable computation is identified per element using modal analysis. This allows determining which elements have to be handled specially given a desired simulation time step. The identified critical elements are treated by a geometric deformation model, while the remaining ones are simulated with a standard deformation model (in our case, a corotational linear Finite Element Method). In order to achieve a valid deformation behavior, we propose a strategy to determine appropriate parameters for the geometric model. Our hybrid method allows taking much larger time steps than using an explicit Finite Element Method alone. The total computational costs per second are significantly lowered. The proposed scheme is especially useful for simulations requiring interactive mesh updates, such as for instance cutting in surgical simulations.

Computer assisted reconstruction of complex proximal humerus fractures for preoperative planning.

Operative treatment of displaced fractures of the proximal humerus is among the most difficult problems in orthopedic shoulder surgery. An accurate preoperative assessment of fragment displacement is crucial for a successful joint restoration. We present a computer assisted approach to precisely quantify these displacements. The bone is virtually reconstructed by multi-fragment alignment. In case of largely displaced pieces, a reconstruction template based on the contralateral humerus is incorporated in the algorithm to determine the optimal assembly. Cadaver experiments were carried out to evaluate our approach. All cases could be successfully reconstructed with little user interaction, and only requiring a few minutes of processing time. On average, the reassembled bone geometries resulted in a translational displacement error of 1.3±0.4 mm and a rotational error of 3.4±2.2°, respectively.

Filtering of high modal frequencies for stable real-time explicit integration of deformable objects using the Finite Element Method.

The behavior, performance, and run-time of mechanical simulations in interactive virtual surgery depend heavily on the type of numerical differential equation solver used to integrate in time the dynamic equations obtained from simulation methods, such as the Finite Element Method. Explicit solvers are fast but only conditionally stable. The condition number of the stiffness matrix limits the highest possible time step. This limit is related to the geometrical properties of the underlying mesh, such as element shape and size. In fact, it can be governed by a small set of ill-shaped elements. For many applications this issue can be solved a priori by a careful meshing. However, when meshes are cut during interactive surgery simulation, it is difficult and computationally expensive to control the quality of the resulting elements. As an alternative, we propose to modify the elemental stiffness matrices directly in order to ensure stability. In this context, we first investigate the behavior of the eigenmodes of the elemental stiffness matrix in a Finite Element Method. We then propose a simple filter to reduce high model frequencies and thus allow larger time steps, while maintaining the general mechanical behavior.

An interactive surgical planning tool for acetabular fractures: initial results.

BACKGROUND: Acetabular fractures still are among the most challenging fractures to treat because of complex anatomy, involved surgical access to fracture sites and the relatively low incidence of these lesions. Proper evaluation and surgical planning is necessary to achieve anatomic reduction of the articular surface and stable fixation of the pelvic ring. The goal of this study was to test the feasibility of preoperative surgical planning in acetabular fractures using a new prototype planning tool based on an interactive virtual reality-style environment. METHODS: 7 patients (5 male and 2 female; median age 53 y (25 to 92 y)) with an acetabular fracture were prospectively included. Exclusion criterions were simple wall fractures, cases with anticipated surgical dislocation of the femoral head for joint debridement and accurate fracture reduction. According to the Letournel classification 4 cases had two column fractures, 2 cases had anterior column fractures and 1 case had a T-shaped fracture including a posterior wall fracture.The workflow included following steps: (1) Formation of a patient-specific bone model from preoperative computed tomography scans, (2) interactive virtual fracture reduction with visuo-haptic feedback, (3) virtual fracture fixation using common osteosynthesis implants and (4) measurement of implant position relative to landmarks. The surgeon manually contoured osteosynthesis plates preoperatively according to the virtually defined deformation. Screenshots including all measurements for the OR were available.The tool was validated comparing the preoperative planning and postoperative results by 3D-superimposition. RESULTS: Preoperative planning was feasible in all cases. In 6 of 7 cases superimposition of preoperative planning and postoperative follow-up CT showed a good to excellent correlation. In one case part of the procedure had to be changed due to impossibility of fracture reduction from an ilioinguinal approach. In 3 cases with osteopenic bone patient-specific prebent fixation plates were helpful in guiding fracture reduction. Additionally, anatomical landmark based measurements were helpful for intraoperative navigation. CONCLUSION: The presented prototype planning tool for pelvic surgery was successfully integrated in a clinical workflow to improve patient-specific preoperative planning, giving visual and haptic information about the injury and allowing a patient-specific adaptation of osteosynthesis implants to the virtually reduced pelvis.

Complex radius shaft malunion: osteotomy with computer-assisted planning.

We report about two cases with a combined axial and angular malunion of the radius shaft with functional loss of pro-supination. For the preoperative planning, a computer simulation was developed that allows the quantification of the malunion by comparing the 3-d surface model of the impaired bone with the contralateral anatomy. The proximal parts of the left and right radii are superimposed, while the different positions of the distal parts are used to quantify the malunion. This task is performed fully automatically which reduces the overall planning time. The osteotomies were performed according to the results of the computer-aided planning. The first case showed 1 year postoperatively an increase of pronation from 40° to 70° at expense of supination from 95° to 90°. The patient was practically pain-free and reported functional improvement. The second case showed 6 months postoperatively an improvement of supination from 15° to 40° and of pronation from 50° to 60°. The computer-assisted operation planning facilitated the quantification of combined axial and angular malunions which were difficult to detect on plain radiographs.

Establishing construct validity of a virtual-reality training simulator for hysteroscopy via a multimetric scoring system.

BACKGROUND: The aims of this study are to determine construct validity for the HystSim virtual-reality (VR) training simulator for hysteroscopy via a new multimetric scoring system (MMSS) and to explore learning curves for both novices and experienced surgeons. METHODS: Fifteen relevant metrics had been identified for diagnostic hysteroscopy by means of hierarchical task decomposition. They were grouped into four modules (visualization, ergonomics, safety, and fluid handling) and individually weighted, building the MMSS for this study. In a first step, 24 novice medical students and 12 experienced gynecologists went through a self-paced teaching tutorial, in which all participants received clearly stated goals and instructions on how to carry out hysteroscopic procedures properly for this study. All subjects performed five repeated trials on two different exercises on HystSim (exploration and diagnosis exercises). After each trial the results were presented to the participants in the form of an automated objective feedback report (AOFR). Construct validity for the MMSS and learning curves were investigated by comparing the performance between novices and experienced surgeons and in between the repeated trials. To study the effect of repeated practice, 23 of the novices returned 2 weeks later for a second training session. RESULTS: Comparing novices with the experienced group, the ergonomics and fluid handling modules resulted in construct validity, while the visualization module did not, and for the safety module the experienced group even scored significantly lower than novices in both exercises. The overall score showed only construct validity when the safety module was excluded. Concerning learning curves, all subjects improved significantly during the training on HystSim, with clear indication that the second training session was beneficial for novice surgeons. CONCLUSIONS: Construct validity for HystSim has been established for different modules of VR metrics on a new MMSS developed for diagnostic hysteroscopy. Careful refinement and further testing of metrics and scores is required before using them as assessment tools for operative skills.