While federated learning (FL) enables collaborative medical image segmentation without centralizing sensitive data, real-world deployment is frequently complicated by cross-site label imperfections such as contour disagreement, missing or additional structures, and confused labels. Federated noisy label learning (FNLL) aims to mitigate these effects, yet remains underused in practice as existing evidence is largely based on synthetic noise, simplified settings, and limited real-world noisy evaluation. We address this gap by introducing a benchmark suite that combines diverse real-world noisy datasets, deployment-relevant client-noise scenarios, and label-noise-targeted evaluation to support systematic FNLL assessment and informed method selection. The suite combines curated real-world noisy medical image segmentation datasets from diverse sources with a comprehensive federated segmentation framework including various client-noise scenarios and noise-targeted evaluation. To demonstrate its capabilities, we compare representative FNLL methods across approaches, including noise-aware aggregation, robust personalization, label correction, and sample selection. In-depth data analysis shows that real-world segmentation label noise occurs both in isolation and in combination of characterized noise types. The benchmark shows that FedSelect performs strongest within noise-targeted evaluation, underlines FedAvg as a competitive baseline, and provides a practically informed decision guide for FNLL method selection based on label-noise type, client-noise scenario, and technical-cost considerations. The presented suite provides a realistic and discriminative basis for FNLL evaluation in medical image segmentation and establishes a reusable foundation for fair benchmarking, dataset-specific label-noise characterization, and future method development under realistic federated settings. Code is available at https://github.com/MIC-DKFZ/FedSegNoiseBench .

Multi-center studies are crucial for advancing medical and radiological research. Data exploration, collaboration discovery, and study progress monitoring are essential for maximizing their potential. However, in practice these processes often rely on manual communication and shared tables, which quickly become outdated and hinder efficient coordination in large distributed studies. This highlights the need for dedicated monitoring solutions that provide transparent and up-to-date insights into study progress. We propose a lightweight, open-source monitoring architecture for multi-center studies based on the widely used Grafana-Prometheus stack. The framework collects aggregated monitoring metrics from distributed study sites and visualizes them through configurable dashboards. As a real-world deployment example, the framework is integrated into the medical imaging platform Kaapana and evaluated within a large multi-center research network. By deploying our solution within the Germany-wide RACOON consortium, we demonstrate its ability to enable privacy-preserving data exploration and study progress monitoring across all 38 German university clinics. The monitoring framework supports transparent coordination of distributed research activities and can facilitate more efficient management of large-scale multi-center studies. The source code and Kaapana integration are publicly available at https://github.com/MIC-DKFZ/study-monitoring-kaapana.

Accurate disease diagnosis depends on effective collaboration between medical specialties, yet departments often use distinct data systems and proprietary formats. This heterogeneity hinders joint analysis and integration of complementary diagnostic information. The use of separate viewers for each modality further restricts cross-specialty collaboration. Although multimodal integration, particularly between radiology and pathology, has demonstrated potential for identifying novel biomarkers, it still relies heavily on manual, time-consuming data pairing. This project introduces an interdisciplinary toolbox that can operate within the Kaapana framework or as a standalone tool to bridge radiology and pathology. By linking modalityspecific viewers and extending them with automated image registration and alignment, the platform enables efficient, scalable multimodal analysis. The integrated environment promotes reproducible workflows, accelerates crossdisciplinary research, and facilitates deeper insights into disease mechanisms and patient care.

Postoperative placement of patients into a regular ward, an intermediate-care unit (IMC), or an intensive care unit (ICU) is critical for balancing patient safety against resource constraints. Most existing models collapse this decision into a binary ICU versus non-ICU choice and lack a mechanism to tune risk thresholds to local staffing ratios or definitions of ICU-level events. We developed Neuro-TACTIC, a cost-sensitive machine learning framework that stratifies postoperative neurosurgical patients into three monitoring levels: regular ward, intermediate care unit, and intensive care unit. An XGBoost-based classifier was trained on 27 demographic, intraoperative, and imaging-derived features from a retrospective cohort of 1072 patients undergoing elective craniotomy. A tunable parameter ζ integrates resource-related and harm-related costs to adjust the balance between over- and under-triage. Generalization was assessed in an independent cohort. Across repeated cross-validation and bootstrap analyses, the framework demonstrated stable behavior across cost settings. At the operating point ζ = 0.975, performance was AUCμ = 0.67 ± 0.03 and weighted F1 = 0.49 ± 0.03 in the development cohort, and AUCμ = 0.60 ± 0.04 and weighted F1 = 0.44 ± 0.06 in the independent evaluation cohort (n = 81). Feature importance analyses identified operative duration, tumor volume, surgical position, body mass index, and patient age as the most influential predictors. This study demonstrates the feasibility of cost-sensitive, three-tier postoperative triage modeling in neurosurgical patients. Neuro-TACTIC is a methodological proof-of-concept; prospective validation and multicenter evaluation are required before clinical deployment.

The differentiation between tumor recurrence and radiationinduced contrast enhancements in post-treatment glioblastoma patients remains a major clinical challenge. Existing approaches rely on clinically sparsely available diffusion MRI or do not consider radiation maps, which are gaining increasing interest in the tumor board for this differentiation. We introduce RICE-NET, a multimodal 3D deep learning model that integrates longitudinal MRI data with radiotherapy dose distributions for automated lesion classification, using conventional T1-weighted MRI data. Using a cohort of 92 patients, the model achieved a performance of 0.92 [F1] on an independent test set. During extensive ablation experiments, we quantified the contribution of each timepoint and modality and showed that reliable classification largely depends on the radiation map. Occlusion-based interpretability analyses further confirmed the model’s focus on clinically relevant regions. These findings highlight the potential of multimodal deep learning to enhance diagnostic accuracy and support clinical decision-making in neuro-oncology.

This retrospective feasibility study aimed to train and test an AI-based algorithm for automated detection of focal bone marrow lesions (FL) on MRI. 444 patients with monoclonal plasma cell disorders were included, focusing on FLs in the left pelvis. Using the nnDetection framework, the algorithm was trained on 334 patients with 494 FLs from center 1 and evaluated on an internal test set (36 patients, 89 FLs) and a multicentric external test set (74 patients, 262 FLs, centers 2–11). On the internal/external test sets, the algorithm achieved a mAP of 0.44/0.34, F1-score 0.54/0.44, sensitivity 0.49/0.34, and a PPV of 0.61/0.61. In two high-quality external subsets, performance approached that of the internal test set (mAP 0.45/0.41, F1- score 0.50/0.53, sensitivity 0.44/0.43, PPV 0.60/0.71). Automated and reference FL counts correlated significantly (internal r = 0.51, p = 0.001; external r = 0.59, p < 0.001). These results demonstrate the feasibility and multicentric robustness of automated FL detection and quantification from MRI. [1]

Accurate disease diagnosis depends on effective collaboration between medical specialties, yet departments often use distinct data systems and proprietary formats. This heterogeneity hinders joint analysis and integration of complementary diagnostic information. The use of separate viewers for each modality further restricts cross-specialty collaboration. Although multimodal integration, particularly between radiology and pathology, has demonstrated potential for identifying novel biomarkers, it still relies heavily on manual, time-consuming data pairing. This project introduces an interdisciplinary toolbox that can operate within the Kaapana framework or as a standalone tool to bridge radiology and pathology. By linking modality-specific viewers and extending them with automated image registration and alignment, the platform enables efficient, scalable multimodal analysis. The integrated environment promotes reproducible workflows, accelerates cross-disciplinary research, and facilitates deeper insights into disease mechanisms and patient care. The method will be integrated in the Kaapana toolkit and as standalone package https://github.com/MIC-DKFZ/combinedmodalityviewer.

Digital pathology offers the opportunity to transform clinical practice in histopathological image analysis, yet faces a significant hurdle: the substantial file sizes of pathological WSIs. Whereas current digital pathology solutions rely on lossy JPEG compression to address this issue, lossy compression can introduce color and texture disparities, potentially impacting clinical decision-making. Prior research addresses perceptual image quality and downstream performance independently of each other, we jointly evaluate compression schemes for perceptual and downstream task quality on four different datasets. In addition, we collect an initially uncompressed dataset for an unbiased perceptual evaluation of compression schemes. Our results show that deep learning models fine-tuned for perceptual quality outperform conventional compression schemes like JPEG-XL or WebP for further compression of WSI. We introduce a novel evaluation metric based on feature similarity between original files and compressed files that aligns with the downstream performance on the compressedWSI. Our study provides novel insights for the assessment of lossy compression schemes forWSI and encourages a unified evaluation of lossy compression schemes to accelerate the clinical uptake of digital pathology [1].

RATIONALE AND OBJECTIVES:To establish an advanced automated bone marrow (BM) segmentation model on whole-body (WB-)MRI in monoclonal plasma cell disorders (MPCD), and to demonstrate its robust performance on multicenter datasets with severe myeloma-related pathologies. MATERIALS AND METHODS:The study cohort comprised multi-vendor, multi-protocol imaging data acquired with varying field strength across 8 different centers. In total, 210 WB-MRIs of 207 MPCD patients were included. An nnU-Net algorithm was established for segmenting the individual bone marrow spaces (BMS) of the spine, pelvis, humeri and femora (advanced segmentation model). For this task, 186 T1-weighted (T1w) WB-MRIs from center 1 were used in the training set. Test sets included 12 T1w WB-MRIs from center 2 (I) and 9 T1w WB-MRIs from centers 3-8 (II). Example cases were included to showcase segmentation performance on T1w WB-MRIs with extensive tumor load. The segmentation accuracy of the advanced segmentation model was compared to a prior established basic segmentation model by calculating Dice scores and using the Wilcoxon signed-rank test. RESULTS:The mean Dice score on the individual BMS was 0.89±0.13 (test set I) and 0.88±0.11 (test set II), significantly higher than the Dice scores of a prior basic model (p<0.05). Dice scores for the BMS of the individual bones ranged from 0.77 to 0.96 (test set I), and 0.81 to 0.95 (test set II). BM altered by myeloma-relevant pathologies, artifacts or low imaging quality was precisely segmented. CONCLUSION:The advanced model performed reliable, automated segmentations, even on heterogeneously acquired multicenter WB-MRIs with severe pathologies.

OBJECTIVE:Federated Learning (FL) enables collaborative model training while keeping data locally. Currently, most FL studies in radiology are conducted in simulated environments due to numerous hurdles impeding its translation into practice. The few existing real-world FL initiatives rarely communicate specific measures taken to overcome these hurdles. To bridge this significant knowledge gap, we propose a comprehensive guide for real-world FL in radiology. Minding efforts to implement real-world FL, there is a lack of comprehensive assessments comparing FL to less complex alternatives in challenging real-world settings, which we address through extensive benchmarking. MATERIALS AND METHODS:We developed our own FL infrastructure within the German Radiological Cooperative Network (RACOON) and demonstrated its functionality by training FL models on lung pathology segmentation tasks across six university hospitals. Insights gained while establishing our FL initiative and running the extensive benchmark experiments were compiled and categorized into the guide. RESULTS:The proposed guide outlines essential steps, identified hurdles, and implemented solutions for establishing successful FL initiatives conducting real-world experiments. Our experimental results prove the practical relevance of our guide and show that FL outperforms less complex alternatives in all evaluation scenarios. DISCUSSION AND CONCLUSION:Our findings justify the efforts required to translate FL into real-world applications by demonstrating advantageous performance over alternative approaches. Additionally, they emphasize the importance of strategic organization, robust management of distributed data and infrastructure in real-world settings. With the proposed guide, we are aiming to aid future FL researchers in circumventing pitfalls and accelerating translation of FL into radiological applications.

Although advances in brain surgery techniques have led to fewer postoperative complications requiring intensive care unit (ICU) monitoring, the routine transfer of patients to the ICU remains the clinical standard, despite its high cost. Predictive Gradient Boosted Trees based on clinical data have attempted to optimize ICU admission by identifying key risk factors pre-operatively; however, these approaches overlook valuable imaging data that could enhance prediction accuracy. In this work, we show that multimodal approaches that combine clinical data with imaging data outperform the current clinical data only baseline from 0.29 [F1] to 0.30 [F1], when only pre-operative clinical data is used and from 0.37 [F1] to 0.41 [F1], for pre- and post-operative data. This study demonstrates that effective ICU admission prediction benefits from multimodal data fusion, especially in contexts of severe class imbalance.

Introduction Cervical central spinal stenosis (CCSS) is a leading cause of adult spinal cord dysfunction. Magnetic resonance imaging (MRI) is the reference standard, but reporting is time-consuming and subject to inter-observer variability. Artificial intelligence (AI)—especially deep-learning—may enable automated, consistent assessment. Research Question Evaluation of performance metrics of AI models for diagnosing CCSS. Material and Methods Following PRISMA 2020, we searched PubMed, Cochrane, Embase, IEEE Xplore, and Web of Science (2015–July 2025) for studies training and evaluating AI models using MRI to diagnose or grade CCSS. We excluded studies limited to foraminal stenosis, non-MRI modalities, thoracic/lumbar levels, segmentation-only or image-enhancement tools without diagnostic output, and studies focused solely on non-stenotic cervical pathologies. Data were extracted on MRI protocol, model type, data splits and external validation, stenosis classification, and diagnostic performance. Results Ten studies (2019–2025) met inclusion criteria, predominantly single-centre and retrospective. Most models used T2-weighted axial and/or sagittal MRI; CNNs (e.g., ResNet-50, EfficientNet) and Transformer-based architectures were common. Sensitivities ranged roughly 0.67–1.00 and specificities 0.42–0.97 across models, with many reporting AUCs ≥0.90 and accuracies ≥0.85. Only one study reported true external test performance. Reporting of confidence intervals, processing time, and explainability (e.g., Grad-CAM) was inconsistent. Discussion and Conclusion Deep-learning shows promising diagnostic performance for automated CCSS assessment on MRI and could reduce variability and reporting time. However, generalisability remains uncertain due to small, retrospective, largely single-centre cohorts and scarce external validation. Standardized reporting (e.g., CLAIM) and prospective, multi-centre validation is needed before routine clinical deployment.

Digital pathology offers a groundbreaking opportunity to transform clinical practice in histopathological image analysis, yet faces a significant hurdle: the substantial file sizes of pathological whole slide images (WSIs). Whereas current digital pathology solutions rely on lossy JPEG compression to address this issue, lossy compression can introduce color and texture disparities, potentially impacting clinical decision-making. Whereas prior research addresses perceptual image quality and downstream performance independently of each other, we jointly evaluate compression schemes for perceptual and downstream task quality on four different datasets. In addition, we collect an initially uncompressed dataset for an unbiased perceptual evaluation of compression schemes. Our results show that deep learning models fine-tuned for perceptual quality outperform conventional compression schemes like JPEG-XL or WebP for further compression of WSI. However, they exhibit a significant bias towards the compression artifacts present in the training data and struggle to generalize across various compression schemes. We introduce a novel evaluation metric based on feature similarity between original files and compressed files that aligns very well with the actual downstream performance on the compressed WSI. Our metric allows for a general and standardized evaluation of lossy compression schemes and mitigates the requirement to independently assess different downstream tasks. Our study provides novel insights for the assessment of lossy compression schemes for WSI and encourages a unified evaluation of lossy compression schemes to accelerate the clinical uptake of digital pathology.

OBJECTIVES:Breast diffusion-weighted imaging (DWI) has shown potential as a standalone imaging technique for certain indications, eg, supplemental screening of women with dense breasts. This study evaluates an artificial intelligence (AI)-powered computer-aided diagnosis (CAD) system for clinical interpretation and workload reduction in breast DWI. MATERIALS AND METHODS:This retrospective IRB-approved study included: n = 824 examinations for model development (2017-2020) and n = 235 for evaluation (01/2021-06/2021). Readings were performed by three readers using either the AI-CAD or manual readings. BI-RADS-like (Breast Imaging Reporting and Data System) classification was based on DWI. Histopathology served as ground truth. The model was nnDetection-based, trained using 5-fold cross-validation and ensembling. Statistical significance was determined using McNemar's test. Inter-rater agreement was calculated using Cohen's kappa. Model performance was calculated using the area under the receiver operating curve (AUC). RESULTS:The AI-augmented approach significantly reduced BI-RADS-like 3 calls in breast DWI by 29% (P =.019) and increased interrater agreement (0.57 ± 0.10 vs 0.49 ± 0.11), while preserving diagnostic accuracy. Two of the three readers detected more malignant lesions (63/69 vs 59/69 and 64/69 vs 62/69) with the AI-CAD. The AI model achieved an AUC of 0.78 (95% CI: [0.72, 0.85]; P <.001), which increased for women at screening age to 0.82 (95% CI: [0.73, 0.90]; P <.001), indicating a potential for workload reduction of 20.9% at 96% sensitivity. DISCUSSION AND CONCLUSION:Breast DWI might benefit from AI support. In our study, AI showed potential for reduction of BI-RADS-like 3 calls and increase of inter-rater agreement. However, given the limited study size, further research is needed.