Search

Intracranial Hemorrhage is a potentially lethal condition whose manifestation is vastly diverse and shifts across clinical centers worldwide. Deep-learning-based solutions are starting to model complex relations between brain structures, but still struggle to generalize. While gathering more diverse data is the most natural approach, privacy regulations often limit the sharing of medical data. We propose the first application of Federated Scene Graph Generation. We show that our models can leverage the increased training data diversity. For Scene Graph Generation, they can recall up to 20% more clinically relevant relations across datasets compared to models trained on a single centralized dataset. Learning structured data representation in a federated setting can open the way to the development of new methods that can leverage this finer information to regularize across clients more effectively.

Whole-body CT is used for multi-trauma patients in the search of any and all injuries. Since an initial assessment needs to be rapid and the search for lesions is done for the whole body, very little time can be allocated for the inspection of a specific anatomy. In particular, intracranial hemorrhages are still missed, especially by clinical students. In this work, we present a Deep Learning approach for highlighting such lesions to improve the diagnostic accuracy. While most works on intracranial hemorrhages perform segmentation, detection only requires bounding boxes for the localization of the bleeding. In this paper, we propose a novel Voxel-Complete IoU (VC-IoU) loss that encourages the network to learn the 3D aspect ratios of bounding boxes and leads to more precise detections. We extensively experiment on brain bleeding detection using a publicly available dataset, and validate it on a private cohort, where we achieve 0.877 AR30, 0.728 AP30, and 0.653 AR30, 0.514 AP30 respectively. These results constitute a relative +5% improvement in Average Recall for both datasets compared to other loss functions. Finally, as there is little data currently publicly available for 3D object detection and as annotation resources are limited in the clinical setting, we evaluate the cost of different annotation methods, as well as the impact of imprecise bounding boxes in the training data on the detection performance.

Patients with Intracranial Hemorrhage (ICH) face a potentially life-threatening condition, and patient-centered individualized treatment remains challenging due to possible clinical complications. Deep-Learning-based methods can efficiently analyze the routinely acquired head CTs to support the clinical decision-making. The majority of early work focuses on the detection and segmentation of ICH, but do not model the complex relations between ICH and adjacent brain structures. In this work, we design a tailored object detection method for ICH, which we unite with segmentation-grounded Scene Graph Generation (SGG) methods to learn a holistic representation of the clinical cerebral scene. To the best of our knowledge, this is the first application of SGG for 3D voxel images. We evaluate our method on two head-CT datasets and demonstrate that our model can recall up to 74% of clinically relevant relations. This work lays the foundation towards SGG for 3D voxel data. The generated Scene Graphs can already provide insights for the clinician, but are also valuable for all downstream tasks as a compact and interpretable representation.

The integration of deep learning based systems in clinical practice is often impeded by challenges rooted in limited and heterogeneous medical datasets. In addition, prioritization of marginal performance improvements on a few, narrowly scoped benchmarks over clinical applicability has slowed down meaningful algorithmic progress. This trend often results in excessive fine-tuning of existing methods to achieve state-of-the-art performance on selected datasets rather than fostering clinically relevant innovations. In response, this work presents a comprehensive benchmark for the MedMNIST+ database to diversify the evaluation landscape and conduct a thorough analysis of common convolutional neural networks (CNNs) and Transformer-based architectures, for medical image classification. Our evaluation encompasses various medical datasets, training methodologies, and input resolutions, aiming to reassess the strengths and limitations of widely used model variants. Our findings suggest that computationally efficient training schemes and modern foundation models hold promise in bridging the gap between expensive end-to-end training and more resource-refined approaches. Additionally, contrary to prevailing assumptions, we observe that higher resolutions may not consistently improve performance beyond a certain threshold, advocating for the use of lower resolutions, particularly in prototyping stages, to expedite processing. Notably, our analysis reaffirms the competitiveness of convolutional models compared to ViT-based architectures emphasizing the importance of comprehending the intrinsic capabilities of different model architectures. Moreover, we hope that our standardized evaluation framework will help enhance transparency, reproducibility, and comparability on the MedMNIST+ dataset collection as well as future research within the field. Code is available at https://github.com/sdoerrich97 .

Quantum chemical methods and time-resolved laser spectroscopy are employed to elucidate ultrafast charge separation processes in triphenylamine (TPA) derivatives upon photoexcitation. When changing the ambient solvent from generic ones to those capable of accepting electrons, such as chloroform, a vastly extended and multifaceted photochemistry is observed. Following the initial excitation, two concurrent charge transfer processes are identified. Firstly, when the TPA derivative and solvent molecules are correctly positioned, an electron transfer to the solvent molecule with immediate charge separation takes place. Consequently, this process gives rise to the formation of the corresponding radical cation of the TPA derivative. This highly reactive species can subsequently combine with other TPA derivative molecules to yield dimeric species. Secondly, when the molecular positioning upon photoexcitation is not optimal, relaxation back to the $\mathrm{S_1}$ state occurs. From this state, an electron transfer process leads to the formation of a charge transfer complex. In this complex, the negatively charged solvent molecule remains closely associated with the positively charged TPA derivative. Within 30 picoseconds, the charges within this complex recombine, yielding a triplet state. This transition to the triplet state is driven by a lower reaction barrier for charge separation compared to the formation of the singlet state., Comment: 14 pages, 7 figures, 1 table

In many supply chains, the current efforts at digitalization have led to improved information exchanges between manufacturers and their customers. Specifically, demand forecasts are often provided by the customers and regularly updated as the related customer information improves. In this paper, we investigate the influence of forecast updates on the production planning method of Material Requirements Planning (MRP). A simulation study was carried out to assess how updates in information affect the setting of planning parameters in a rolling horizon MRP planned production system. An intuitive result is that information updates lead to disturbances in the production orders for the MRP standard, and, therefore, an extension for MRP to mitigate these effects is developed. A large numerical simulation experiment shows that the MRP safety stock exploitation heuristic, that has been developed, leads to significantly improved results as far as inventory and backorder costs are concerned. An interesting result is that the fixed-order-quantity lotsizing policy performs - in most instances - better than the fixed-order-period lotsizing policy, when periodic forecast updates occur. In addition, the simulation study shows that underestimating demand is marginally more costly than overestimating it, based on the comparative analysis of all instances. Furthermore, the results indicate that the MRP safety stock exploitation heuristic can mitigate the negative effects of biased forecasts.

One of the most important tools available to limit the spread and impact of infectious diseases is vaccination. It is therefore important to understand what factors determine people's vaccination decisions. To this end, previous behavioural research made use of, (i) controlled but often abstract or hypothetical studies (e.g., vignettes) or, (ii) realistic but typically less flexible studies that make it difficult to understand individual decision processes (e.g., clinical trials). Combining the best of these approaches, we propose integrating real-world Bluetooth contacts via smartphones in several rounds of a game scenario, as a novel methodology to study vaccination decisions and disease spread. In our 12-week proof-of-concept study conducted with $N$ = 494 students, we found that participants strongly responded to some of the information provided to them during or after each decision round, particularly those related to their individual health outcomes. In contrast, information related to others' decisions and outcomes (e.g., the number of vaccinated or infected individuals) appeared to be less important. We discuss the potential of this novel method and point to fruitful areas for future research.

During the operation of industrial robots, unusual events may endanger the safety of humans and the quality of production. When collecting data to detect such cases, it is not ensured that data from all potentially occurring errors is included as unforeseeable events may happen over time. Therefore, anomaly detection (AD) delivers a practical solution, using only normal data to learn to detect unusual events. We introduce a dataset that allows training and benchmarking of anomaly detection methods for robotic applications based on machine data which will be made publicly available to the research community. As a typical robot task the dataset includes a pick-and-place application which involves movement, actions of the end effector and interactions with the objects of the environment. Since several of the contained anomalies are not task-specific but general, evaluations on our dataset are transferable to other robotics applications as well. Additionally, we present MVT-Flow (multivariate time-series flow) as a new baseline method for anomaly detection: It relies on deep-learning-based density estimation with normalizing flows, tailored to the data domain by taking its structure into account for the architecture. Our evaluation shows that MVT-Flow outperforms baselines from previous work by a large margin of 6.2% in area under ROC., Comment: 14 pages, 14 figures, accepted to Transactions on Robotics

Disease propagation between countries strongly depends on their effective distance, a measure derived from the world air transportation network (WAN). It reduces the complex spreading patterns of a pandemic to a wave-like propagation from the outbreak country, i.e. a linear relationship to the arrival time of the unmitigated spread of a disease. However, in the early stage of an outbreak, what matters to countries' decision makers is knowledge about the relative risk of arrival of active cases, i.e. how likely it is that an active case that boarded at the outbreak location will arrive in their country. As accurate mechanistic models to estimate such risks are still lacking, we propose here the ``import risk'' model that defines an import probability by means of the effective-distance framework. The model assumes that airline passengers are distributed along the shortest path tree that starts at the outbreak's origin. In combination with a random walk, we account for all possible paths, thus inferring predominant connecting flights. Our model outperforms other mobility models, such as the radiation and gravity model with varying distance types, and it improves further if additional geographic information is included. The import risk model's precision increases for countries that are more connected within the WAN, and recovers a geographic distance-dependence that suggests a pull- rather than a push- dynamic of the distribution process., Comment: 11 pages, 7 figures

Two genomes over the same set of gene families form a canonical pair when each of them has exactly one gene from each family. Different distances of canonical genomes can be derived from a structure called breakpoint graph, which represents the relation between the two given genomes as a collection of cycles of even length and paths. Then, the breakpoint distance is equal to n - (c_2 + p_0/2), where n is the number of genes, c_2 is the number of cycles of length 2 and p_0 is the number of paths of length 0. Similarly, when the considered rearrangements are those modeled by the double-cut-and-join (DCJ) operation, the rearrangement distance is n - (c + p_e/2), where c is the total number of cycles and p_e is the total number of even paths. The distance formulation is a basic unit for several other combinatorial problems related to genome evolution and ancestral reconstruction, such as median or double distance. Interestingly, both median and double distance problems can be solved in polynomial time for the breakpoint distance, while they are NP-hard for the rearrangement distance. One way of exploring the complexity space between these two extremes is to consider the {\sigma}_k distance, defined to be n - [c_2 + c_4 + ... + c_k + (p_0 + p_2 + ... +p_k)/2], and increasingly investigate the complexities of median and double distance for the {\sigma}_4 distance, then the {\sigma}_6 distance, and so on. While for the median much effort was done in our and in other research groups but no progress was obtained even for the {\sigma}_4 distance, for solving the double distance under {\sigma}_4 and {\sigma}_6 distances we could devise linear time algorithms, which we present here., Comment: 24 pages, 26 figures