Zdravko Marinov

Jianning Li, Zongwei Zhou, Jiancheng Yang et al.

Prior to the deep learning era, shape was commonly used to describe the objects. Nowadays, state-of-the-art (SOTA) algorithms in medical imaging are predominantly diverging from computer vision, where voxel grids, meshes, point clouds, and implicit surface models are used. This is seen from numerous shape-related publications in premier vision conferences as well as the growing popularity of ShapeNet (about 51,300 models) and Princeton ModelNet (127,915 models). For the medical domain, we present a large collection of anatomical shapes (e.g., bones, organs, vessels) and 3D models of surgical instrument, called MedShapeNet, created to facilitate the translation of data-driven vision algorithms to medical applications and to adapt SOTA vision algorithms to medical problems. As a unique feature, we directly model the majority of shapes on the imaging data of real patients. As of today, MedShapeNet includes 23 dataset with more than 100,000 shapes that are paired with annotations (ground truth). Our data is freely accessible via a web interface and a Python application programming interface (API) and can be used for discriminative, reconstructive, and variational benchmarks as well as various applications in virtual, augmented, or mixed reality, and 3D printing. Exemplary, we present use cases in the fields of classification of brain tumors, facial and skull reconstructions, multi-class anatomy completion, education, and 3D printing. In future, we will extend the data and improve the interfaces. The project pages are: https://medshapenet.ikim.nrw/ and https://github.com/Jianningli/medshapenet-feedback

Verena Jasmin Hallitschke, Tobias Schlumberger, Philipp Kataliakos et al.

Recently, deep learning enabled the accurate segmentation of various diseases in medical imaging. These performances, however, typically demand large amounts of manual voxel annotations. This tedious process for volumetric data becomes more complex when not all required information is available in a single imaging domain as is the case for PET/CT data. We propose a multimodal interactive segmentation framework that mitigates these issues by combining anatomical and physiological cues from PET/CT data. Our framework utilizes the geodesic distance transform to represent the user annotations and we implement a novel ellipsoid-based user simulation scheme during training. We further propose two annotation interfaces and conduct a user study to estimate their usability. We evaluated our model on the in-domain validation dataset and an unseen PET/CT dataset. We make our code publicly available: https://github.com/verena-hallitschke/pet-ct-annotate.

Lars Heiliger, Zdravko Marinov, Max Hasin et al.

Tumor volume and changes in tumor characteristics over time are important biomarkers for cancer therapy. In this context, FDG-PET/CT scans are routinely used for staging and re-staging of cancer, as the radiolabeled fluorodeoxyglucose is taken up in regions of high metabolism. Unfortunately, these regions with high metabolism are not specific to tumors and can also represent physiological uptake by normal functioning organs, inflammation, or infection, making detailed and reliable tumor segmentation in these scans a demanding task. This gap in research is addressed by the AutoPET challenge, which provides a public data set with FDG-PET/CT scans from 900 patients to encourage further improvement in this field. Our contribution to this challenge is an ensemble of two state-of-the-art segmentation models, the nn-Unet and the Swin UNETR, augmented by a maximum intensity projection classifier that acts like a gating mechanism. If it predicts the existence of lesions, both segmentations are combined by a late fusion approach. Our solution achieves a Dice score of 72.12\% on patients diagnosed with lung cancer, melanoma, and lymphoma in our cross-validation. Code: https://github.com/heiligerl/autopet_submission

Zdravko Marinov, David Schneider, Alina Roitberg et al.

Domain shifts, such as appearance changes, are a key challenge in real-world applications of activity recognition models, which range from assistive robotics and smart homes to driver observation in intelligent vehicles. For example, while simulations are an excellent way of economical data collection, a Synthetic-to-Real domain shift leads to a > 60% drop in accuracy when recognizing activities of Daily Living (ADLs). We tackle this challenge and introduce an activity domain generation framework which creates novel ADL appearances (novel domains) from different existing activity modalities (source domains) inferred from video training data. Our framework computes human poses, heatmaps of body joints, and optical flow maps and uses them alongside the original RGB videos to learn the essence of source domains in order to generate completely new ADL domains. The model is optimized by maximizing the distance between the existing source appearances and the generated novel appearances while ensuring that the semantics of an activity is preserved through an additional classification loss. While source data multimodality is an important concept in this design, our setup does not rely on multi-sensor setups, (i.e., all source modalities are inferred from a single video only.) The newly created activity domains are then integrated in the training of the ADL classification networks, resulting in models far less susceptible to changes in data distributions. Extensive experiments on the Synthetic-to-Real benchmark Sims4Action demonstrate the potential of the domain generation paradigm for cross-domain ADL recognition, setting new state-of-the-art results. Our code is publicly available at https://github.com/Zrrr1997/syn2real_DG

Weitong Kong, Di Wen, Kunyu Peng et al.

Correcting errors in long-video understanding is disproportionately costly: existing multimodal pipelines produce opaque, end-to-end outputs that expose no intermediate state for inspection, forcing annotators to revisit raw video and reconstruct temporal logic from scratch. The core bottleneck is not generation quality alone, but the absence of a supervisory interface through which human effort can be proportional to the scope of each error. We present IMPACT-CYCLE, a supervisory multi-agent system that reformulates long-video understanding as iterative claim-level maintenance of a shared semantic memory -- a structured, versioned state encoding typed claims, a claim dependency graph, and a provenance log. Role-specialized agents operating under explicit authority contracts decompose verification into local object-relation correctness, cross-temporal consistency, and global semantic coherence, with corrections confined to structurally dependent claims. When automated evidence is insufficient, the system escalates to human arbitration as the supervisory authority with final override rights; dependency-closure re-verification then ensures correction cost remains proportional to error scope. Experiments on VidOR show substantially improved downstream reasoning (VQA: 0.71 to 0.79) and a 4.8x reduction in human arbitration cost, with workload significantly lower than manual annotation. Code will be released at https://github.com/MKong17/IMPACT_CYCLE.

Alina Roitberg, Kunyu Peng, Zdravko Marinov et al.

Visual recognition inside the vehicle cabin leads to safer driving and more intuitive human-vehicle interaction but such systems face substantial obstacles as they need to capture different granularities of driver behaviour while dealing with highly limited body visibility and changing illumination. Multimodal recognition mitigates a number of such issues: prediction outcomes of different sensors complement each other due to different modality-specific strengths and weaknesses. While several late fusion methods have been considered in previously published frameworks, they constantly feature different architecture backbones and building blocks making it very hard to isolate the role of the chosen late fusion strategy itself. This paper presents an empirical evaluation of different paradigms for decision-level late fusion in video-based driver observation. We compare seven different mechanisms for joining the results of single-modal classifiers which have been both popular, (e.g. score averaging) and not yet considered (e.g. rank-level fusion) in the context of driver observation evaluating them based on different criteria and benchmark settings. This is the first systematic study of strategies for fusing outcomes of multimodal predictors inside the vehicles, conducted with the goal to provide guidance for fusion scheme selection.

Zdravko Marinov, Simon Reiß, David Kersting et al.

Positron Emission Tomography (PET) and Computer Tomography (CT) are routinely used together to detect tumors. PET/CT segmentation models can automate tumor delineation, however, current multimodal models do not fully exploit the complementary information in each modality, as they either concatenate PET and CT data or fuse them at the decision level. To combat this, we propose Mirror U-Net, which replaces traditional fusion methods with multimodal fission by factorizing the multimodal representation into modality-specific branches and an auxiliary multimodal decoder. At these branches, Mirror U-Net assigns a task tailored to each modality to reinforce unimodal features while preserving multimodal features in the shared representation. In contrast to previous methods that use either fission or multi-task learning, Mirror U-Net combines both paradigms in a unified framework. We explore various task combinations and examine which parameters to share in the model. We evaluate Mirror U-Net on the AutoPET PET/CT and on the multimodal MSD BrainTumor datasets, demonstrating its effectiveness in multimodal segmentation and achieving state-of-the-art performance on both datasets. Our code will be made publicly available.

Calvin Tanama, Kunyu Peng, Zdravko Marinov et al.

Deep learning-based models are at the forefront of most driver observation benchmarks due to their remarkable accuracies but are also associated with high computational costs. This is challenging, as resources are often limited in real-world driving scenarios. This paper introduces a lightweight framework for resource-efficient driver activity recognition. The framework enhances 3D MobileNet, a neural architecture optimized for speed in video classification, by incorporating knowledge distillation and model quantization to balance model accuracy and computational efficiency. Knowledge distillation helps maintain accuracy while reducing the model size by leveraging soft labels from a larger teacher model (I3D), instead of relying solely on original ground truth data. Model quantization significantly lowers memory and computation demands by using lower precision integers for model weights and activations. Extensive testing on a public dataset for in-vehicle monitoring during autonomous driving demonstrates that this new framework achieves a threefold reduction in model size and a 1.4-fold improvement in inference time, compared to an already optimized architecture. The code for this study is available at https://github.com/calvintanama/qd-driver-activity-reco.

Zdravko Marinov, Rainer Stiefelhagen, Jens Kleesiek

Interactive segmentation reduces the annotation time of medical images and allows annotators to iteratively refine labels with corrective interactions, such as clicks. While existing interactive models transform clicks into user guidance signals, which are combined with images to form (image, guidance) pairs, the question of how to best represent the guidance has not been fully explored. To address this, we conduct a comparative study of existing guidance signals by training interactive models with different signals and parameter settings to identify crucial parameters for the model's design. Based on our findings, we design a guidance signal that retains the benefits of other signals while addressing their limitations. We propose an adaptive Gaussian heatmaps guidance signal that utilizes the geodesic distance transform to dynamically adapt the radius of each heatmap when encoding clicks. We conduct our study on the MSD Spleen and the AutoPET datasets to explore the segmentation of both anatomy (spleen) and pathology (tumor lesions). Our results show that choosing the guidance signal is crucial for interactive segmentation as we improve the performance by 14% Dice with our adaptive heatmaps on the challenging AutoPET dataset when compared to non-interactive models. This brings interactive models one step closer to deployment on clinical workflows. We will make our code publically available.

Zdravko Marinov, Alina Roitberg, David Schneider et al.

Modality selection is an important step when designing multimodal systems, especially in the case of cross-domain activity recognition as certain modalities are more robust to domain shift than others. However, selecting only the modalities which have a positive contribution requires a systematic approach. We tackle this problem by proposing an unsupervised modality selection method (ModSelect), which does not require any ground-truth labels. We determine the correlation between the predictions of multiple unimodal classifiers and the domain discrepancy between their embeddings. Then, we systematically compute modality selection thresholds, which select only modalities with a high correlation and low domain discrepancy. We show in our experiments that our method ModSelect chooses only modalities with positive contributions and consistently improves the performance on a Synthetic-to-Real domain adaptation benchmark, narrowing the domain gap.

Matthias Hadlich, Zdravko Marinov, Rainer Stiefelhagen

Tumor segmentation in medical imaging is crucial and relies on precise delineation. Fluorodeoxyglucose Positron-Emission Tomography (FDG-PET) is widely used in clinical practice to detect metabolically active tumors. However, FDG-PET scans may misinterpret irregular glucose consumption in healthy or benign tissues as cancer. Combining PET with Computed Tomography (CT) can enhance tumor segmentation by integrating metabolic and anatomic information. FDG-PET/CT scans are pivotal for cancer staging and reassessment, utilizing radiolabeled fluorodeoxyglucose to highlight metabolically active regions. Accurately distinguishing tumor-specific uptake from physiological uptake in normal tissues is a challenging aspect of precise tumor segmentation. The AutoPET challenge addresses this by providing a dataset of 1014 FDG-PET/CT studies, encouraging advancements in accurate tumor segmentation and analysis within the FDG-PET/CT domain. Code: https://github.com/matt3o/AutoPET2-Submission/

Matthias Hadlich, Zdravko Marinov, Moon Kim et al.

Deep learning has revolutionized the accurate segmentation of diseases in medical imaging. However, achieving such results requires training with numerous manual voxel annotations. This requirement presents a challenge for whole-body Positron Emission Tomography (PET) imaging, where lesions are scattered throughout the body. To tackle this problem, we introduce SW-FastEdit - an interactive segmentation framework that accelerates the labeling by utilizing only a few user clicks instead of voxelwise annotations. While prior interactive models crop or resize PET volumes due to memory constraints, we use the complete volume with our sliding window-based interactive scheme. Our model outperforms existing non-sliding window interactive models on the AutoPET dataset and generalizes to the previously unseen HECKTOR dataset. A user study revealed that annotators achieve high-quality predictions with only 10 click iterations and a low perceived NASA-TLX workload. Our framework is implemented using MONAI Label and is available: https://github.com/matt3o/AutoPET2-Submission/

Zdravko Marinov, Paul F. Jäger, Jan Egger et al.

Interactive segmentation is a crucial research area in medical image analysis aiming to boost the efficiency of costly annotations by incorporating human feedback. This feedback takes the form of clicks, scribbles, or masks and allows for iterative refinement of the model output so as to efficiently guide the system towards the desired behavior. In recent years, deep learning-based approaches have propelled results to a new level causing a rapid growth in the field with 121 methods proposed in the medical imaging domain alone. In this review, we provide a structured overview of this emerging field featuring a comprehensive taxonomy, a systematic review of existing methods, and an in-depth analysis of current practices. Based on these contributions, we discuss the challenges and opportunities in the field. For instance, we find that there is a severe lack of comparison across methods which needs to be tackled by standardized baselines and benchmarks.

Marawan Elbatel, Mohamed Ghonim, Jiaji Mao et al.

Automated segmentation of liver lesions on non-contrast computed tomography (NCCT) is clinically important but fundamentally challenging, particularly in low-resource settings across Africa and Asia where contrast agents are frequently unavailable. Progress has been limited by the absence of annotated NCCT benchmarks. Here we describe the TriALS challenge for automated liver lesion segmentation under contrast-limited conditions, supported by a multi-centre dataset of 150 cases with four-phase CT acquisitions (600 volumes) from Egyptian and Chinese institutions. Algorithms were evaluated on 70 cases from three institutions, including an independent external cohort. The top-performing method achieved a mean venous-phase Dice of 0.754, consistent with human-level performance, yet dropped to 0.57 on NCCT. On external validation, the leading method outperformed off-the-shelf models by up to 28% in Dice on NCCT. Algorithm performance was most strongly predicted by training data scale and pre-training strategy. A cross-year comparison exposed a persistent perceptual barrier on NCCT that scaling pre-training alone cannot overcome. Data, annotations, and code are available at https://github.com/xmed-lab/TriALS.

Jonas Muth, Zdravko Marinov, Simon Reiß

While much of the medical computer vision community has focused on advancing performance for specific tasks, the underlying relationships between tasks, i.e., how they relate, overlap, or differ on a representational level, remain largely unexplored. Our work explores these intrinsic relationships between medical vision tasks, specifically, we investigate 30 tasks, such as semantic tasks (e.g., segmentation and detection), image generative tasks (e.g., denoising, inpainting, or colorization), and image transformation tasks (e.g., geometric transformations). Our goal is to probe whether a data-driven representation space can capture an underlying structure of tasks across a variety of 39 datasets from wildly different medical imaging modalities, including computed tomography, magnetic resonance, electron microscopy, X-ray ultrasound and more. By revealing how tasks relate to one another, we aim to provide insights into their fundamental properties and interconnectedness. To this end, we introduce Task-Contrastive Learning (TaCo), a contrastive learning framework designed to embed tasks into a shared representation space. Through TaCo, we map these heterogeneous tasks from different modalities into a joint space and analyze their properties: identifying which tasks are distinctly represented, which blend together, and how iterative alterations to tasks are reflected in the embedding space. Our work provides a foundation for understanding the intrinsic structure of medical vision tasks, offering a deeper understanding of task similarities and their interconnected properties in embedding spaces.

Haoshen Zhang, Di Wen, Kunyu Peng et al.

We present IMPACT-HOI, a mixed-initiative framework for annotating egocentric procedural video by constructing structured event graphs for Human-Object Interactions (HOI), motivated by the need for high-quality structured supervision for learning robot manipulation from human demonstration. IMPACT-HOI frames this task as the incremental resolution of a partially specified, onset-anchored event state. A trust-calibrated controller selects among direct queries, human-confirmed suggestions, and conservative completions based on empirical annotator behavior and evidence quality. A risk-bounded execution protocol, utilizing atomic rollback, ensures that human-confirmed decisions are preserved against conflicting automated updates. A user study with 9 participants shows a 13.5% reduction in manual annotation actions, a 46.67% event match rate, and zero confirmed-field violations under the studied protocol. The code will be made publicly available at https://github.com/541741106/IMPACT_HOI.

Qian Yin, Di Wen, Kunyu Peng et al.

Dense temporal annotation of procedural activity videos is vital for action understanding and embodied intelligence but remains labor-intensive due to reactive tools. Each correction is treated as an isolated edit, limiting reuse of information on annotator uncertainty and model reliability. We introduce IMPACT-Scribe, a correction-driven framework for dense labeling that uses each correction to improve future human-machine collaboration. IMPACT-Scribe combines uncertainty-aware boundary scribble supervision, local proposal modeling, cost-aware query planning, structured propagation, and correction-driven adaptation. Experiments and a human study show that this closed-loop design improves labeling quality per effort, enhances boundary accuracy, and fosters better human-machine interaction over time. The code will be made publicly available at https://github.com/BanzQians/IMPACT_AS.

Jun Ma, Feifei Li, Sumin Kim et al.

Promptable segmentation foundation models have emerged as a transformative approach to addressing the diverse needs in medical images, but most existing models require expensive computing, posing a big barrier to their adoption in clinical practice. In this work, we organized the first international competition dedicated to promptable medical image segmentation, featuring a large-scale dataset spanning nine common imaging modalities from over 20 different institutions. The top teams developed lightweight segmentation foundation models and implemented an efficient inference pipeline that substantially reduced computational requirements while maintaining state-of-the-art segmentation accuracy. Moreover, the post-challenge phase advanced the algorithms through the design of performance booster and reproducibility tasks, resulting in improved algorithms and validated reproducibility of the winning solution. Furthermore, the best-performing algorithms have been incorporated into the open-source software with a user-friendly interface to facilitate clinical adoption. The data and code are publicly available to foster the further development of medical image segmentation foundation models and pave the way for impactful real-world applications.

Zdravko Marinov, Stanka Vasileva, Qing Wang et al.

Drones have become a common tool, which is utilized in many tasks such as aerial photography, surveillance, and delivery. However, operating a drone requires more and more interaction with the user. A natural and safe method for Human-Drone Interaction (HDI) is using gestures. In this paper, we introduce an HDI framework building upon skeleton-based pose estimation. Our framework provides the functionality to control the movement of the drone with simple arm gestures and to follow the user while keeping a safe distance. We also propose a monocular distance estimation method, which is entirely based on image features and does not require any additional depth sensors. To perform comprehensive experiments and quantitative analysis, we create a customized testing dataset. The experiments indicate that our HDI framework can achieve an average of 93.5\% accuracy in the recognition of 11 common gestures. The code is available at: https://github.com/Zrrr1997/Pose2Drone

Alexander Jaus, Constantin Seibold, Simon Reiß et al.

We present Connected-Component~(CC)-Metrics, a novel semantic segmentation evaluation protocol, targeted to align existing semantic segmentation metrics to a multi-instance detection scenario in which each connected component matters. We motivate this setup in the common medical scenario of semantic metastases segmentation in a full-body PET/CT. We show how existing semantic segmentation metrics suffer from a bias towards larger connected components contradicting the clinical assessment of scans in which tumor size and clinical relevance are uncorrelated. To rebalance existing segmentation metrics, we propose to evaluate them on a per-component basis thus giving each tumor the same weight irrespective of its size. To match predictions to ground-truth segments, we employ a proximity-based matching criterion, evaluating common metrics locally at the component of interest. Using this approach, we break free of biases introduced by large metastasis for overlap-based metrics such as Dice or Surface Dice. CC-Metrics also improves distance-based metrics such as Hausdorff Distances which are uninformative for small changes that do not influence the maximum or 95th percentile, and avoids pitfalls introduced by directly combining counting-based metrics with overlap-based metrics as it is done in Panoptic Quality.

Simon Reiß, Zdravko Marinov, Alexander Jaus et al.

In this paper, we explore the potential of visual in-context learning to enable a single model to handle multiple tasks and adapt to new tasks during test time without re-training. Unlike previous approaches, our focus is on training in-context learners to adapt to sequences of tasks, rather than individual tasks. Our goal is to solve complex tasks that involve multiple intermediate steps using a single model, allowing users to define entire vision pipelines flexibly at test time. To achieve this, we first examine the properties and limitations of visual in-context learning architectures, with a particular focus on the role of codebooks. We then introduce a novel method for training in-context learners using a synthetic compositional task generation engine. This engine bootstraps task sequences from arbitrary segmentation datasets, enabling the training of visual in-context learners for compositional tasks. Additionally, we investigate different masking-based training objectives to gather insights into how to train models better for solving complex, compositional tasks. Our exploration not only provides important insights especially for multi-modal medical task sequences but also highlights challenges that need to be addressed.

Alexander Jaus, Zdravko Marinov, Constantin Seibold et al.

Improving label quality in medical image segmentation is costly, but its benefits remain unclear. We systematically evaluate its impact using multiple pseudo-labeled versions of CT datasets, generated by models like nnU-Net, TotalSegmentator, and MedSAM. Our results show that while higher-quality labels improve in-domain performance, gains remain unclear if below a small threshold. For pre-training, label quality has minimal impact, suggesting that models rather transfer general concepts than detailed annotations. These findings provide guidance on when improving label quality is worth the effort.

Lena Heinemann, Alexander Jaus, Zdravko Marinov et al.

Within this work, we introduce LIMIS: The first purely language-based interactive medical image segmentation model. We achieve this by adapting Grounded SAM to the medical domain and designing a language-based model interaction strategy that allows radiologists to incorporate their knowledge into the segmentation process. LIMIS produces high-quality initial segmentation masks by leveraging medical foundation models and allows users to adapt segmentation masks using only language, opening up interactive segmentation to scenarios where physicians require using their hands for other tasks. We evaluate LIMIS on three publicly available medical datasets in terms of performance and usability with experts from the medical domain confirming its high-quality segmentation masks and its interactive usability.

Zdravko Marinov, Moon Kim, Jens Kleesiek et al.

Interactive segmentation plays a crucial role in accelerating the annotation, particularly in domains requiring specialized expertise such as nuclear medicine. For example, annotating lesions in whole-body Positron Emission Tomography (PET) images can require over an hour per volume. While previous works evaluate interactive segmentation models through either real user studies or simulated annotators, both approaches present challenges. Real user studies are expensive and often limited in scale, while simulated annotators, also known as robot users, tend to overestimate model performance due to their idealized nature. To address these limitations, we introduce four evaluation metrics that quantify the user shift between real and simulated annotators. In an initial user study involving four annotators, we assess existing robot users using our proposed metrics and find that robot users significantly deviate in performance and annotation behavior compared to real annotators. Based on these findings, we propose a more realistic robot user that reduces the user shift by incorporating human factors such as click variation and inter-annotator disagreement. We validate our robot user in a second user study, involving four other annotators, and show it consistently reduces the simulated-to-real user shift compared to traditional robot users. By employing our robot user, we can conduct more large-scale and cost-efficient evaluations of interactive segmentation models, while preserving the fidelity of real user studies. Our implementation is based on MONAI Label and will be made publicly available.