Expert eye movements provide a rich, passive source of domain knowledge in radiology, offering a powerful cue for integrating diagnostic reasoning into computer-aided analysis. However, direct integration into CNN-based systems, which historically have dominated the medical image analysis domain, is challenging: gaze recordings are sequential, temporally dense yet spatially sparse, noisy, and variable across experts. As a consequence, most existing image-based models utilize reduced representations such as heatmaps. In contrast, gaze naturally aligns with transformer architectures, as both are sequential in nature and rely on attention to highlight relevant input regions. In this work, we introduce FixationFormer, a transformer-based architecture that represents expert gaze trajectories as sequences of tokens, thereby preserving their temporal and spatial structure. By modeling gaze sequences jointly with image features, our approach addresses sparsity and variability in gaze data while enabling a more direct and fine-grained integration of expert diagnostic cues through explicit cross-attention between the image and gaze token sequences. We evaluate our method on three publicly available benchmark chest X-ray datasets and demonstrate that it achieves state-of-the-art classification performance, highlighting the value of representing gaze as a sequence in transformer-based medical image analysis. Code is available at https://zivgitlab.uni-muenster.de/cvmls/fixation_former .

We introduce a straightforward yet effective method to empirically measure and regularize memorization in deep neural networks for classification tasks. Our approach augments each training sample with auxiliary random labels, which are then predicted by a random label prediction head (RLP-head). RLP-heads can be attached at arbitrary depths of a network, predicting random labels from the corresponding intermediate representation and thereby enabling analysis of how memorization capacity evolves across layers. By interpreting the RLP-head performance as an empirical estimate of Rademacher complexity, we obtain a direct measure of both sample-level memorization and model capacity. We leverage this random label accuracy metric to analyze generalization and overfitting in different models and datasets. Building on this approach, we further propose a novel regularization technique based on the output of the RLP-head, which demonstrably reduces memorization. Interestingly, our experiments reveal that reducing memorization can either improve or impair generalization, depending on the dataset and training setup. These findings challenge the traditional assumption that overfitting is equivalent to memorization and suggest new hypotheses to reconcile these seemingly contradictory results.

The rapid loss of biodiversity worldwide is unprecedented, with more species facing extinction now than at any other time in human history. Key factors contributing to this decline include habitat destruction, overexploitation, and climate change. There is an urgent need for innovative and effective conservation practices that leverage advanced technologies, such as autonomous drones, to monitor wildlife, manage human-wildlife conflicts, and protect endangered species. While drones have shown promise in conservation efforts, significant technological challenges remain, particularly in developing reliable, cost-effective solutions capable of operating in remote, unstructured, and open-ended environments. This paper explores the technological advancements necessary for deploying autonomous drones in nature conservation and presents the interdisciplinary scientific methodology of the WildDrone doctoral network as a basis for integrating research in drones, computer vision, and machine learning for ecological monitoring. We report preliminary results demonstrating the potential of these technologies to enhance biodiversity conservation efforts. Based on our preliminary findings, we expect that drones and computer vision will develop to further automate time consuming observational tasks in nature conservation, thus allowing human workers to ground conservation actions on evidence based on large and frequent data.

Monocular imaging of animals inherently reduces 3D structures to 2D projections. Detection algorithms lead to 2D bounding boxes that lack information about animal's orientation relative to the camera. To build 3D detection methods for RGB animal images, there is a lack of labeled datasets; such labeling processes require 3D input streams along with RGB data. We present a pipeline that utilises Skinned Multi Animal Linear models to estimate 3D bounding boxes and to project them as robust labels into 2D image space using a dedicated camera pose refinement algorithm. To assess which sides of the animal are captured, cuboid face visibility metrics are computed. These 3D bounding boxes and metrics form a crucial step toward developing and benchmarking future monocular 3D animal detection algorithms. We evaluate our method on the Animal3D dataset, demonstrating accurate performance across species and settings.

Voxel-based morphometry (VBM), a popular approach in neuroimaging research, uses magnetic resonance imaging data to assess variations in the local density of brain tissue and to examine its associations with biological and psychometric variables. Here we present deepmriprep, a preprocessing pipeline designed to leverage neural networks to perform all the necessary preprocessing steps for the VBM analysis of T1-weighted magnetic resonance imaging. Utilizing the graphics processing unit, deepmriprep is 37 times faster than CAT12, the leading VBM preprocessing toolbox. The proposed method matches CAT12 in accuracy for tissue segmentation and image registration across more than 100 datasets and shows strong correlations in the VBM results. Tissue segmentation maps from deepmriprep have more than 95% agreement with ground-truth maps, and its nonlinear registration predicts smooth deformation fields comparable to CAT12. The high computational speed of deepmriprep enables rapid preprocessing of large datasets and opens the door to real-time applications.

[18F]FDG-PET/CT is a cornerstone imaging modality for tumor staging and treatment response assessment across many cancer types, yet expert reader shortages necessitate more efficient diagnostic aids. While standalone AI models for automatic lesion segmentation exist, clinical translation remains hindered by concerns about interpretability, explainability, reliability, and workflow integration. We present GazeXPErT, a 4D eye-tracking dataset capturing expert search patterns during tumor detection and measurement on 346 FDG-PET/CT scans. Each study was read by a trainee and a board-certified nuclear medicine or radiology specialist using an eye-tracking-enabled annotation platform that simulates routine clinical reads. From 3,948 minutes of raw 60Hz eye-tracking data, 9,030 unique gaze-to-lesion trajectories were extracted, synchronized with PET/CT image slices, and rendered in COCO-style format for multiple machine learning applications. Baseline validation experiments demonstrate that a 3D nnUNet tumor segmentation model achieved superior performance when incorporating expert gaze patterns versus without (DICE score 0.6819 versus 0.6008), and that vision transformers trained on sequential gaze and PET/CT images can improve dynamic lesion localization (74.95

We introduce WildBox, a dataset and benchmark for monocular 3D detection of wildlife from drone video, comprising 237,505 3D bounding box annotations across seven African savanna species grouped into six benchmark classes. Annotations follow a KITTI/Omni3D-compatible format in a per-segment scale-normalised camera frame, with instance identities maintained across each segment. We evaluate two open-vocabulary monocular 3D architectures, OVMono3D-LIFT and DetAny3D, under zero-shot, ground-truth 2D box prompt, and supervised fine-tuning protocols. Open-vocabulary 2D foundation models provide usable zero-shot wildlife localisation (50.55 AP@50), but zero-shot 3D detection collapses to 0.00 AP across both architectures and every 2D-input condition tested, including ground-truth 2D box prompts, thus isolating the failure to the 3D stage. Fine-tuning on WildBox recovers performance to 8.68 +/- 0.47 AP-BEV@0.50 and 13.17 +/- 0.69 AP3D macro. Depth contributes 84

Recent progress in self- and weakly supervised occupancy estimation has largely relied on 2D projection or rendering-based supervision, which suffers from geometric inconsistencies and severe depth bleeding.We thus introduce ShelfOcc, a vision-only method that overcomes these limitations without relying on LiDAR.ShelfOcc brings supervision into native 3D space by generating metrically consistent semantic voxel labels from video, enabling true 3D supervision without any additional sensors or manual 3D annotations.While recent vision-based 3D geometry foundation models provide a promising source of prior knowledge, they do not work out of the box as a prediction due to sparse or noisy and inconsistent geometry, especially in dynamic driving scenes.Our method introduces a dedicated framework that mitigates these issues by filtering and accumulating static geometry consistently across frames, handling dynamic content and propagating semantic information into a stable voxel representation.This data-centric shift in supervision for weakly/shelf-supervised occupancy estimation allows the use of essentially any SOTA occupancy model architecture without relying on LiDAR data.We argue that such high-quality supervision is essential for robust occupancy learning and constitutes an important complementary avenue to architectural innovation.On the Occ3D-nuScenes benchmark, ShelfOcc substantially outperforms all previous weakly/shelf-supervised methods (up to a 34\% relative improvement), establishing a new data-driven direction for LiDAR-free 3D scene understanding.

Graph neural networks (GNNs) effectively model chemical data by processing molecular graphs directly, eliminating the need for fixed-length fingerprints; however, their lack of intrinsic interpretability poses a key challenge. To address this limitation, we introduce an interpretability method for GNNs grounded in Myerson values from cooperative game theory. Myerson values extend the concept of Shapley values by incorporating the structural constraints of a graph, allowing only connected nodes to form coalitions. In this formulation, a GNN is treated as a coalition game where the nodes of the input graph represent the players, and the Myerson value quantifies each node's contribution to the overall model prediction. The resulting node-wise attributions can be visualized directly on the molecular graph as interpretable heatmaps. Since the exact computation of Myerson values scales exponentially with graph size, we propose a scalable Monte-Carlo approximation that maintains accuracy while significantly reducing computational cost. Our method provides a faithful and interpretable decomposition of GNN predictions, operating directly on graph features without the local approximations. For evaluation, we benchmark our interpretability method using the GraphXAI framework, which integrates several established explainers and chemical datasets. To further demonstrate its effectiveness in cheminformatics tasks, we evaluate our algorithm on two proof-of-concept datasets (log P and molecular weight) and a real-world kinase inhibitor dataset. Our results demonstrate the potential of Myerson-based explanations as an effective approach for interpreting graph-based models in cheminformatics and related domains.

Simulation-based medical education is essential for improving patient safety. In virtual reality (VR)–based simulation, immersion is primarily generated through visual and auditory cues, while other sensory modalities are typically absent. This sensory limitation may reduce the emergence of authentic safety-relevant behaviors. Olfaction plays an important role in clinical reasoning, risk perception, and self-protective behavior and is closely linked to memory and emotion. Although olfactory cues have been shown to influence hand hygiene behavior in real or simulated-real environments, their targeted integration into fully immersive VR-based medical simulation has not been systematically examined. This study aimed to investigate whether adding a real olfactory cue (disinfectant scent) to a fully virtual clinical simulation increases patient safety–relevant behavior, specifically hand hygiene compliance (hand disinfection and glove usage). In a randomized controlled study at the University of Münster (winter term 2025/26), 89 medical students participated in a VR-based clinical simulation. Study rooms were pre-assigned to either an olfactory intervention or a control condition, and participants selected their room without knowledge of the assigned condition. Hand hygiene and glove use were automatically tracked as outcomes. Odds ratios were calculated to assess the effect of the intervention on these behaviors. The olfactory intervention nearly tripled the odds of hand disinfection (OR = 2.81, 95% CI 1.09–7.75, P = 0.037), while no significant difference was observed for glove use (OR = 1.62, P = 0.278). The integration of a real olfactory cue into a fully immersive VR medical simulation significantly increased hand disinfection behavior, particularly after patient contact, but did not affect glove use. These findings suggest that olfactory augmentation can selectively reinforce safety-relevant behaviors in digital training environments. Incorporating real-world sensory cues into VR may represent a simple yet effective design strategy to enhance behavioral authenticity and patient safety outcomes in simulation-based medical education. not applicable

Sociable weaver nests function as complex ecological structures offering thermoregulatory microhabitats and sustaining diverse species; however, datasets used in prior studies lack fine-grained 3D structural detail. Producing usable and accurate 3D weaver nest data is challenging due to their irregular geometry and integration with complex host vegetation. We bridge this gap with an open-access, 1.4 TB multimodal drone dataset of 104 nest-bearing trees, comprising 27,945 RGB images, 111,780 multispectral images, approximately 781 million 3D points, and expert-annotated semantic segmentation labels. We benchmark semantic segmentation using KPConv, RandLA-Net, and Point Transformer V3, with PT-v3 achieving an mIoU of 86.35

Graph neural networks (GNNs) are a natural choice to represent chemical data, due to their inherent ability to handle arbitrary input topologies. They avoid the need to convert molecules into molecular fingerprints with a fixed vector length. However, like most deep learning models, GNNs are not interpretable and common explainability methods fail because of the variable input size. We introduce a novel method to interpret the predictions of GNNs based on Myerson values from cooperative game theory. Myerson values are closely related to Shapley values , which have been adapted to explain a wide variety of machine learning model predictions. Applying these approaches to GNNs have, however, proven to be challenging because of their varying graph size. Our approach treats a GNN as a coalition game and the nodes of an input layer graph as players. The Myerson value of a node then determines the contribution to the prediction of the model, with only connected nodes contributing to coalitions. All Myerson values add up to the predicted value of the model allowing for a simple and intuitive interpretation of the prediction. Because calculating Myerson values becomes computationally infeasible for large graphs, we have also implemented a scalable approximation technique using Monte Carlo sampling. We developed the technique for applications in cheminformatics and drug discovery, but it can also be used in any application that uses GNNs. The effectiveness of our approach is validated through successful applications to two proof-of-concept datasets (logP and molecular weight) as well as a real-world dataset featuring kinase inhibitors, highlighting its broad applicability and promise in explaining graph-based cheminformatic models.

Personalized medicine requires physicians to adapt clinical communication and decision-making to patients’ individual motives, emotions, and interpersonal behaviors. However, training these skills remains challenging, as established simulation-based formats—particularly actor-based simulations—are resource-intensive and difficult to scale. Consequently, there is a need for scalable and flexible training approaches that allow repeated practice across diverse patient personalities and clinical contexts. The PerTRAIN (Personalization Training in Medicine) project addresses this need by leveraging large language models (LLMs) to simulate virtual patients with dynamically adapting personality expression at scale. Grounded in Contemporary Integrative Interpersonal Theory (CIIT), patient behavior is modelled along the dimensions of agency and communion and updated in response to the medical trainee’s behavior, enabling systematic variation and dynamic adaptation of interpersonal behavior within medical scenarios. This allows trainees to learn how patient personality shapes communication, and to practice adaptive, patient-centered communication strategies and appropriate clinical decision-making. For the personalization trainings, we developed clinical cases in which personality expression substantially influences behavior in doctor–patient interactions, that represent common encounters in primary care, and that align with national guidelines for patient-centered care. An initial chat-based implementation enables structured interactions, dynamic personality adaptation, and iterative refinement. The system is designed as a complementary tool to existing simulation formats, offering a scalable, low-threshold environment for repeated practice and reflection. A first application in curricular teaching is scheduled for 2026. Future extensions of the framework are discussed, such as large-scale empirical validation, modelling long-term interpersonal trajectories, and the extension of interactions to multimodal formats.

Penile measurement is clinically relevant across male reproductive and urogenital health, including conditions such as micropenis, congenital and endocrine disorders, and sexual or urinary dysfunction. However, quantitative assessment of penile size has relied mainly on external length or circumference measurements, which are difficult to standardize, sensitive to measurement conditions, and unable to capture the internal portion of the penis. MRI enables volumetric assessment of the whole penis in vivo, but automated segmentation has not previously been established at population scale. Automated whole-organ volumetry would enable high-throughput phenotyping for multi-omics and clinical studies of male reproductive disease. Here, we present a deep learning framework for whole-penis segmentation in multi-channel DIXON MRI. Using a newly curated expert-annotated training dataset (n = 145 subjects; 13,050 annotated slices) and a double-annotated independent test benchmark (n = 24 subjects; 2,160 double-annotated slices), we optimized a 3D nnU-Net architecture. The model achieved a 5-fold cross-validation Dice score of 0.90 and performed at observer-level accuracy on the independent test set (Dice: 0.92; Hausdorff distance: 3.58). We deployed the model in 34,412 UK Biobank participants, enabling automated quantification of total penile tissue, including both external and internal components. Longitudinal evaluation in 2,282 men demonstrated high inter-session reproducibility (r = 0.87). This framework establishes a reproducible and population-scalable method for MRI-based assessment of penile anatomy and provides an open technical resource for future studies in urological imaging and male reproductive health. The trained model weights will be publicly released.

Abstract Plant–pollinator interactions are fundamental to the functioning of natural ecosystems and the enhancement of agricultural productivity. An effective support of pollinator communities necessitates a comprehensive understanding of the mechanisms underlying pollinator attraction and nutritional preferences. Pollinators depend on olfactory, visual, gustatory and tactile signals to locate and select floral resources. Investigating olfactory and gustatory preferences in flying insects, however, presents notable challenges due to the lack of standardised experimental frameworks, the labour‐intensive nature of current methodologies and the limitations inherent in existing experimental systems, particularly their inadequacy in simultaneously delivering and measuring olfactory and gustatory cues. These constraints hinder the detailed understanding of how insects integrate multiple sensory modalities during foraging. In this study, we introduce a binary‐choice flight arena and assay specifically designed for the investigation of scent and taste preferences in flying insects. To facilitate robust and unbiased behavioural data analysis, we developed a computer‐vision‐based tool capable of quantifying insect preferences from video recordings. The system's effectiveness was validated through a series of experimental trials, utilising definitive and quantifiable metrics, including (i) the number of visits at each feeding tube, (ii) the total number of insects interacting, (iii) the cumulative time spent by insects at each stimulus source per hour and (iv) the volume of reward solution consumed. Our findings demonstrate that the experimental framework, detection algorithm and arena design collectively facilitate balanced visitation rates under controlled conditions, yielding consistent and reliable quantifications. Our results also indicate that the setup is highly suitable for and will facilitate exploring the influence of olfactory and gustatory compounds on insect visitation patterns and feeding behaviours.

Animal ecology data collection using drones represents a substantial investment of time, expertise, and financial resources. Yet most existing datasets serve only a single research community, limiting interdisciplinary reuse. We propose a unified drone dataset standard, FAIR^2 Drones, that bridges ecology, robotics, and computer vision by building on existing FAIR and AI-ready data frameworks while adding essential platform metadata and annotation specifications. Our standard enables datasets to simultaneously support ecological analysis, robotics algorithm development, and computer vision benchmarking. We provide open-source validation tools, reference implementations, and multimodal extensions linking drone imagery with complementary sensors such as camera traps, GPS, and acoustics. By standardizing metadata across disciplines, this framework maximizes the scientific return on investment for costly field deployments and accelerates cross-domain collaboration in environmental monitoring.

Sampling error yields exclusively reactive, non-lesional brain parenchyma in a significant proportion of intracranial biopsies, leaving the underlying disease undiagnosed. We benchmark four pathology foundation models (UNI2-h, Virchow2, Prov-GigaPath, H-optimus-0) as frozen patch encoders within a shared attention-based multiple-instance learning framework using 245 whole-slide images from 186 patients with confirmed downstream diagnoses. We first show that coarse disease-category prediction can be reproduced largely from slide size alone. After restricting classification to three finer diagnostic distinctions within common tissue categories, this confound no longer explains performance, yet disease labels remain predictable above chance under permutation testing (p ≤ 10^-4 throughout). Surprisingly, performance is statistically indistinguishable across all foundation-model encoders, suggesting that recovering these weak morphological signatures is not limited by current patch representations. Signed instance-contribution maps and expert review further test whether predictive evidence localizes to reactive parenchyma rather than sampling-induced bias like blood introduced during tissue sampling. These results position acquisition-shortcut auditing via a provenance-only baseline as a necessary control in computational-pathology benchmarks, and show, once that confound is removed, that weakly supervised models still recover disease signal from tissue conventionally regarded as non-diagnostic.

Monocular RGB cameras mounted on drones are widely used for wildlife monitoring, yet most analytical pipelines remain confined to two-dimensional image space, leaving geometric information in video underexploited. We present WildLIFT, a computational framework that integrates three-dimensional scene geometry from monocular drone video with open-vocabulary 2D instance segmentation to enable species-agnostic 3D detection and tracking. Oriented 3D bounding box labels with semantic face information enable quantitative assessment of viewpoint coverage and inter-animal occlusion, producing structured metadata for downstream ecological analyses. We validate the framework on 2,581 manually curated frames comprising over 6,700 3D detections across four large mammal species. WildLIFT maintains high identity consistency in multi-animal scenes and substantially reduces manual 3D annotation effort through keyframe-based refinement. By transforming standard drone footage into structured 3D and viewpoint-aware representations, WildLIFT extends the analytical utility of aerial wildlife datasets for behavioural research and population monitoring.

We introduce a straightforward yet effective method to empirically study memorization in deep neural networks for classification tasks. Our approach augments each training sample with auxiliary random labels, which are then predicted by a random label prediction head (RLP-head). RLP-heads can be attached at arbitrary depths of a network, predicting random labels from the corresponding intermediate representation and thereby enabling analysis of how memorization capacity evolves across layers. By interpreting the RLP-head performance as an empirical estimate of Rademacher complexity, we obtain a direct measure of both sample-level memorization and model capacity. We leverage this random label accuracy metric to analyze generalization and overfitting in different models and datasets. Building on this approach, we further propose a novel regularization technique based on the output of the RLP-head, which demonstrably reduces memorization. Interestingly, our experiments reveal that reducing memorization can either improve or impair generalization, depending on the dataset and training setup. These findings challenge the traditional assumption that overfitting is equivalent to memorization and suggest new hypotheses to reconcile these seemingly contradictory results. The source code is available at https://github.com/MarlonBecker/RandomLabelHeads

The increasing adoption of virtual reality (VR) in medical education offers substantial opportunities for immersive, practice-oriented training that complements traditional teaching methods. In particular, VR enables repeated, risk-free exposure to complex clinical scenarios and supports the development of clinical reasoning, communication skills, and procedural competence. However, implementing VR-based courses remains challenging due to high development costs, technical complexity, and the need for close interdisciplinary collaboration. This tutorial presents key insights and best practices from the medical tr.AI.ning project, a 3-year interdisciplinary initiative funded by the German Federal Ministry of Education and Research. The project’s objective was to develop an artificial intelligence (AI)-supported, VR-based training platform that allows medical students to practice clinical decision-making in immersive, interactive scenarios. The paper is structured as a tutorial and offers recommendations for planning, developing, and integrating VR courses into medical curricula. Each recommendation is illustrated with concrete examples from our project, serving as a practical blueprint to guide educators and developers in applying these guidelines in their own contexts. Successful implementation of a VR project in medical education requires strategic planning and collaboration, starting with a thorough identification of curricular gaps that VR can address and a clear justification of its added educational value. An interdisciplinary consortium that combines expertise from medical didactics experts, computer science, and design is essential to ensure the development of high-quality, pedagogically sound simulations and intuitive user interfaces. Key factors for success include defining specific learning objectives aligned with competency-based frameworks; iterative development with continuous feedback from medical experts, educators, and students; and structured pilot testing with systematic collection of quantitative and qualitative data to assess usability, immersion, and learning outcomes. Early engagement and walkthroughs with end users help identify practical challenges and inform iterative improvements. A dedicated authoring tool within the project allows medical teachers to create and adapt VR scenarios without prior technical experience, supporting the scalability and sustainability of the approach. Effective project management frameworks facilitate collaboration, clear task allocation, and adaptive progress throughout development. Additionally, considerations for hardware selection, technical infrastructure, and sustainable dissemination strategies, including open-access publications, project websites, and professional networking, are crucial to ensure long-term viability and broad adoption across institutions. By combining a tutorial format with practical, step-by-step recommendations, this article provides a comprehensive guide for educators and developers on implementing immersive, AI-supported VR courses to enhance medical education. It highlights key lessons learned in interdisciplinary collaboration, iterative testing, systematic evaluation, and alignment with educational objectives, thereby facilitating the effective, evidence-based, and sustainable integration of VR into medical curricula across diverse institutions.