Bernhard Kainz

Lena Maier-Hein, Annika Reinke, Patrick Godau et al. · utoronto

Increasing evidence shows that flaws in machine learning (ML) algorithm validation are an underestimated global problem. Particularly in automatic biomedical image analysis, chosen performance metrics often do not reflect the domain interest, thus failing to adequately measure scientific progress and hindering translation of ML techniques into practice. To overcome this, our large international expert consortium created Metrics Reloaded, a comprehensive framework guiding researchers in the problem-aware selection of metrics. Following the convergence of ML methodology across application domains, Metrics Reloaded fosters the convergence of validation methodology. The framework was developed in a multi-stage Delphi process and is based on the novel concept of a problem fingerprint - a structured representation of the given problem that captures all aspects that are relevant for metric selection, from the domain interest to the properties of the target structure(s), data set and algorithm output. Based on the problem fingerprint, users are guided through the process of choosing and applying appropriate validation metrics while being made aware of potential pitfalls. Metrics Reloaded targets image analysis problems that can be interpreted as a classification task at image, object or pixel level, namely image-level classification, object detection, semantic segmentation, and instance segmentation tasks. To improve the user experience, we implemented the framework in the Metrics Reloaded online tool, which also provides a point of access to explore weaknesses, strengths and specific recommendations for the most common validation metrics. The broad applicability of our framework across domains is demonstrated by an instantiation for various biological and medical image analysis use cases.

Hadrien Reynaud, Mengyun Qiao, Mischa Dombrowski et al.

Image synthesis is expected to provide value for the translation of machine learning methods into clinical practice. Fundamental problems like model robustness, domain transfer, causal modelling, and operator training become approachable through synthetic data. Especially, heavily operator-dependant modalities like Ultrasound imaging require robust frameworks for image and video generation. So far, video generation has only been possible by providing input data that is as rich as the output data, e.g., image sequence plus conditioning in, video out. However, clinical documentation is usually scarce and only single images are reported and stored, thus retrospective patient-specific analysis or the generation of rich training data becomes impossible with current approaches. In this paper, we extend elucidated diffusion models for video modelling to generate plausible video sequences from single images and arbitrary conditioning with clinical parameters. We explore this idea within the context of echocardiograms by looking into the variation of the Left Ventricle Ejection Fraction, the most essential clinical metric gained from these examinations. We use the publicly available EchoNet-Dynamic dataset for all our experiments. Our image to sequence approach achieves an $R^2$ score of 93%, which is 38 points higher than recently proposed sequence to sequence generation methods. Code and models will be available at: https://github.com/HReynaud/EchoDiffusion.

Hadrien Reynaud, Athanasios Vlontzos, Mischa Dombrowski et al.

Causally-enabled machine learning frameworks could help clinicians to identify the best course of treatments by answering counterfactual questions. We explore this path for the case of echocardiograms by looking into the variation of the Left Ventricle Ejection Fraction, the most essential clinical metric gained from these examinations. We combine deep neural networks, twin causal networks and generative adversarial methods for the first time to build D'ARTAGNAN (Deep ARtificial Twin-Architecture GeNerAtive Networks), a novel causal generative model. We demonstrate the soundness of our approach on a synthetic dataset before applying it to cardiac ultrasound videos to answer the question: "What would this echocardiogram look like if the patient had a different ejection fraction?". To do so, we generate new ultrasound videos, retaining the video style and anatomy of the original patient, while modifying the Ejection Fraction conditioned on a given input. We achieve an SSIM score of 0.79 and an R2 score of 0.51 on the counterfactual videos. Code and models are available at: https://github.com/HReynaud/dartagnan.

Berke Doga Basaran, Weitong Zhang, Mengyun Qiao et al.

Data augmentation has become a de facto component of deep learning-based medical image segmentation methods. Most data augmentation techniques used in medical imaging focus on spatial and intensity transformations to improve the diversity of training images. They are often designed at the image level, augmenting the full image, and do not pay attention to specific abnormalities within the image. Here, we present LesionMix, a novel and simple lesion-aware data augmentation method. It performs augmentation at the lesion level, increasing the diversity of lesion shape, location, intensity and load distribution, and allowing both lesion populating and inpainting. Experiments on different modalities and different lesion datasets, including four brain MR lesion datasets and one liver CT lesion dataset, demonstrate that LesionMix achieves promising performance in lesion image segmentation, outperforming several recent Mix-based data augmentation methods. The code will be released at https://github.com/dogabasaran/lesionmix.

Matthew Baugh, Jeremy Tan, Johanna P. Müller et al.

There is a growing interest in single-class modelling and out-of-distribution detection as fully supervised machine learning models cannot reliably identify classes not included in their training. The long tail of infinitely many out-of-distribution classes in real-world scenarios, e.g., for screening, triage, and quality control, means that it is often necessary to train single-class models that represent an expected feature distribution, e.g., from only strictly healthy volunteer data. Conventional supervised machine learning would require the collection of datasets that contain enough samples of all possible diseases in every imaging modality, which is not realistic. Self-supervised learning methods with synthetic anomalies are currently amongst the most promising approaches, alongside generative auto-encoders that analyse the residual reconstruction error. However, all methods suffer from a lack of structured validation, which makes calibration for deployment difficult and dataset-dependant. Our method alleviates this by making use of multiple visually-distinct synthetic anomaly learning tasks for both training and validation. This enables more robust training and generalisation. With our approach we can readily outperform state-of-the-art methods, which we demonstrate on exemplars in brain MRI and chest X-rays. Code is available at https://github.com/matt-baugh/many-tasks-make-light-work .

Mischa Dombrowski, Hadrien Reynaud, Johanna P. Müller et al.

Recent advancements in diffusion models have significantly impacted the trajectory of generative machine learning research, with many adopting the strategy of fine-tuning pre-trained models using domain-specific text-to-image datasets. Notably, this method has been readily employed for medical applications, such as X-ray image synthesis, leveraging the plethora of associated radiology reports. Yet, a prevailing concern is the lack of assurance on whether these models genuinely comprehend their generated content. With the evolution of text-conditional image generation, these models have grown potent enough to facilitate object localization scrutiny. Our research underscores this advancement in the critical realm of medical imaging, emphasizing the crucial role of interpretability. We further unravel a consequential trade-off between image fidelity as gauged by conventional metrics and model interpretability in generative diffusion models. Specifically, the adoption of learnable text encoders when fine-tuning results in diminished interpretability. Our in-depth exploration uncovers the underlying factors responsible for this divergence. Consequently, we present a set of design principles for the development of truly interpretable generative models. Code is available at https://github.com/MischaD/chest-distillation.

Mischa Dombrowski, Hadrien Reynaud, Matthew Baugh et al.

Curating datasets for object segmentation is a difficult task. With the advent of large-scale pre-trained generative models, conditional image generation has been given a significant boost in result quality and ease of use. In this paper, we present a novel method that enables the generation of general foreground-background segmentation models from simple textual descriptions, without requiring segmentation labels. We leverage and explore pre-trained latent diffusion models, to automatically generate weak segmentation masks for concepts and objects. The masks are then used to fine-tune the diffusion model on an inpainting task, which enables fine-grained removal of the object, while at the same time providing a synthetic foreground and background dataset. We demonstrate that using this method beats previous methods in both discriminative and generative performance and closes the gap with fully supervised training while requiring no pixel-wise object labels. We show results on the task of segmenting four different objects (humans, dogs, cars, birds) and a use case scenario in medical image analysis. The code is available at https://github.com/MischaD/fobadiffusion.

Annika Reinke, Minu D. Tizabi, Michael Baumgartner et al.

Validation metrics are key for the reliable tracking of scientific progress and for bridging the current chasm between artificial intelligence (AI) research and its translation into practice. However, increasing evidence shows that particularly in image analysis, metrics are often chosen inadequately in relation to the underlying research problem. This could be attributed to a lack of accessibility of metric-related knowledge: While taking into account the individual strengths, weaknesses, and limitations of validation metrics is a critical prerequisite to making educated choices, the relevant knowledge is currently scattered and poorly accessible to individual researchers. Based on a multi-stage Delphi process conducted by a multidisciplinary expert consortium as well as extensive community feedback, the present work provides the first reliable and comprehensive common point of access to information on pitfalls related to validation metrics in image analysis. Focusing on biomedical image analysis but with the potential of transfer to other fields, the addressed pitfalls generalize across application domains and are categorized according to a newly created, domain-agnostic taxonomy. To facilitate comprehension, illustrations and specific examples accompany each pitfall. As a structured body of information accessible to researchers of all levels of expertise, this work enhances global comprehension of a key topic in image analysis validation.

Shiyu Li, Julian Kreimeier, Hannah Schieber et al.

The handling and assembly of instruments during surgery imposes high cognitive demands on scrub nurses, particularly when instruments are unfamiliar. We present a supporting guidance system for surgical instrumentation that combines multi-camera 6D pose estimation with augmented reality in-situ visualization on a head-mounted display without the requirement for additional markers. Pose estimation and consecutive camera calibration are achieved through known objects. The 6D pose estimation network is trained purely on synthetic data, aiming for better generalizability and real-world applicability. The AR guidance displays tooltip localization cues and step-wise assembly animations. Via gaze-based selection and a foot pedal, users can switch between assembly steps in intraoperative use. In a technical evaluation, our approach outperforms state-of-art 6D pose estimation. A user study with 29 scrub nurses was conducted in a surgical simulation of knee arthroplasty, comparing the system against a paper manual. AR guidance significantly reduced the perceived workload compared. Objectively, AR guidance reduced task completion time by 21.3\% (4.76 minutes). Specifically, scrub nurses less experienced with the instrument set benefited when using the system. Error frequencies were comparable between conditions. Qualitative feedback highlighted improved process clarity, reduced information overload, and perceived independence. To summarize, our marker-free multi-camera AR guidance approach for surgical instruments can, subjectively and objectively, improve intraoperative instrumentation performance, particularly for untrained scrub nurses.

Veronika A. Zimmer, Alberto Gomez, Emily Skelton et al.

Automatic segmentation of the placenta in fetal ultrasound (US) is challenging due to the (i) high diversity of placenta appearance, (ii) the restricted quality in US resulting in highly variable reference annotations, and (iii) the limited field-of-view of US prohibiting whole placenta assessment at late gestation. In this work, we address these three challenges with a multi-task learning approach that combines the classification of placental location (e.g., anterior, posterior) and semantic placenta segmentation in a single convolutional neural network. Through the classification task the model can learn from larger and more diverse datasets while improving the accuracy of the segmentation task in particular in limited training set conditions. With this approach we investigate the variability in annotations from multiple raters and show that our automatic segmentations (Dice of 0.86 for anterior and 0.83 for posterior placentas) achieve human-level performance as compared to intra- and inter-observer variability. Lastly, our approach can deliver whole placenta segmentation using a multi-view US acquisition pipeline consisting of three stages: multi-probe image acquisition, image fusion and image segmentation. This results in high quality segmentation of larger structures such as the placenta in US with reduced image artifacts which are beyond the field-of-view of single probes.

Athanasios Vlontzos, Daniel Rueckert, Bernhard Kainz

Medical image analysis is a vibrant research area that offers doctors and medical practitioners invaluable insight and the ability to accurately diagnose and monitor disease. Machine learning provides an additional boost for this area. However, machine learning for medical image analysis is particularly vulnerable to natural biases like domain shifts that affect algorithmic performance and robustness. In this paper we analyze machine learning for medical image analysis within the framework of Technology Readiness Levels and review how causal analysis methods can fill a gap when creating robust and adaptable medical image analysis algorithms. We review methods using causality in medical imaging AI/ML and find that causal analysis has the potential to mitigate critical problems for clinical translation but that uptake and clinical downstream research has been limited so far.

Liu Li, Qiang Ma, Matthew Sinclair et al.

Fetal Magnetic Resonance Imaging (MRI) is used in prenatal diagnosis and to assess early brain development. Accurate segmentation of the different brain tissues is a vital step in several brain analysis tasks, such as cortical surface reconstruction and tissue thickness measurements. Fetal MRI scans, however, are prone to motion artifacts that can affect the correctness of both manual and automatic segmentation techniques. In this paper, we propose a novel network structure that can simultaneously generate conditional atlases and predict brain tissue segmentation, called CAS-Net. The conditional atlases provide anatomical priors that can constrain the segmentation connectivity, despite the heterogeneity of intensity values caused by motion or partial volume effects. The proposed method is trained and evaluated on 253 subjects from the developing Human Connectome Project (dHCP). The results demonstrate that the proposed method can generate conditional age-specific atlas with sharp boundary and shape variance. It also segment multi-category brain tissues for fetal MRI with a high overall Dice similarity coefficient (DSC) of $85.2\%$ for the selected 9 tissue labels.

Cheng Ouyang, Shuo Wang, Chen Chen et al.

Probability calibration for deep models is highly desirable in safety-critical applications such as medical imaging. It makes output probabilities of deep networks interpretable, by aligning prediction probability with the actual accuracy in test data. In image segmentation, well-calibrated probabilities allow radiologists to identify regions where model-predicted segmentations are unreliable. These unreliable predictions often occur to out-of-domain (OOD) images that are caused by imaging artifacts or unseen imaging protocols. Unfortunately, most previous calibration methods for image segmentation perform sub-optimally on OOD images. To reduce the calibration error when confronted with OOD images, we propose a novel post-hoc calibration model. Our model leverages the pixel susceptibility against perturbations at the local level, and the shape prior information at the global level. The model is tested on cardiac MRI segmentation datasets that contain unseen imaging artifacts and images from an unseen imaging protocol. We demonstrate reduced calibration errors compared with the state-of-the-art calibration algorithm.

Matthew Baugh, James Batten, Johanna P. Müller et al.

This technical report outlines our submission to the zero-shot track of the Visual Anomaly and Novelty Detection (VAND) 2023 Challenge. Building on the performance of the WINCLIP framework, we aim to enhance the system's localization capabilities by integrating zero-shot segmentation models. In addition, we perform foreground instance segmentation which enables the model to focus on the relevant parts of the image, thus allowing the models to better identify small or subtle deviations. Our pipeline requires no external data or information, allowing for it to be directly applied to new datasets. Our team (Variance Vigilance Vanguard) ranked third in the zero-shot track of the VAND challenge, and achieve an average F1-max score of 81.5/24.2 at a sample/pixel level on the VisA dataset.

Franciskus Xaverius Erick, Johanna Paula Müller, Bernhard Kainz

Vision-language models (VLMs) can couple visual perception with open-ended clinical reasoning, making them attractive for computational histopathology. However, fine-tuning billions of parameters on scarce, expert-annotated pathology data is prohibitive, while in-context learning (ICL), which conditions the VLM on demonstrative image-text pairs without parameter updates, suffers from high sensitivity to which examples are selected and how the query is phrased, producing unreliable diagnostics. Existing selection strategies rely on query-dependent nearest-neighbour retrieval that ignores global data structure, require costly parameter updates, or disregard the joint vision-text embedding geometry of VLMs. We propose GAUC, a training-free coreset selection method operating directly in the pre-trained multimodal embedding space. GAUC jointly optimises three objectives: (1) a Maximum Mean Discrepancy term enforcing distributional fidelity between coreset and full dataset, (2) an Effective Mutual Information Difference regulariser bounding performance degradation under prompt paraphrases by exploiting the VLM's joint vision-text alignment, and (3) a predictive-variance penalty suppressing overconfident, unstable outputs. On CRC-100K and MHIST across multiple open-source VLM architectures, GAUC consistently improves accuracy, calibration, and prompt robustness over recent ICL selection methods and dataset-distillation baselines, all without a single gradient update.

Luca Hagen, Johanna P. Müller, Weitong Zhang et al.

Small vision-language models (2-8B) are well-suited for clin- ical deployment due to privacy constraints, limited connectivity, and low-latency requirements favouring on-device or on-premise inference. However, their limited capacity exacerbates the generation of plausible but incorrect outputs. We extend game-theoretic decoding, previously restricted to text-only, closed-ended NLP tasks, to vision-language mod- els for open-ended Medical VQA. We introduce a semantically aware Wasserstein stopping criterion that replaces lexical order matching, en- abling convergence based on semantic consensus among near-synonymous candidate answers and avoiding unnecessary iterations caused by clini- cally equivalent ranking swaps. On VQA-RAD and PathVQA, we ob- tain consistent, statistically significant improvements over greedy and discriminative baselines. On VQA-RAD, we improve Qwen3-VL-2B by +3.5 percentage points (p < 0.01), surpassing the greedy 4B model, with similar trends at larger scales. On PathVQA, Gemma-3-4B with BDG matches MedGemma-4B under greedy decoding despite no domain- specific fine-tuning. At accuracy parity with classic BDG, the Wasser- stein criterion reduces average convergence iterations by approximately 20%, improving inference efficiency while preserving the game-theoretic equilibrium behaviour. Code is available at https://github.com/luca-hagen/ Wasserstein-BDG-medical-VQA.

Andreas Hinterreiter, Christina Humer, Bernhard Kainz et al.

ParaDime is a framework for parametric dimensionality reduction (DR). In parametric DR, neural networks are trained to embed high-dimensional data items in a low-dimensional space while minimizing an objective function. ParaDime builds on the idea that the objective functions of several modern DR techniques result from transformed inter-item relationships. It provides a common interface for specifying these relations and transformations and for defining how they are used within the losses that govern the training process. Through this interface, ParaDime unifies parametric versions of DR techniques such as metric MDS, t-SNE, and UMAP. It allows users to fully customize all aspects of the DR process. We show how this ease of customization makes ParaDime suitable for experimenting with interesting techniques such as hybrid classification/embedding models and supervised DR. This way, ParaDime opens up new possibilities for visualizing high-dimensional data.

Sarah Cechnicka, James Ball, Hadrien Reynaud et al.

Poor performance of quantitative analysis in histopathological Whole Slide Images (WSI) has been a significant obstacle in clinical practice. Annotating large-scale WSIs manually is a demanding and time-consuming task, unlikely to yield the expected results when used for fully supervised learning systems. Rarely observed disease patterns and large differences in object scales are difficult to model through conventional patient intake. Prior methods either fall back to direct disease classification, which only requires learning a few factors per image, or report on average image segmentation performance, which is highly biased towards majority observations. Geometric image augmentation is commonly used to improve robustness for average case predictions and to enrich limited datasets. So far no method provided sampling of a realistic posterior distribution to improve stability, e.g. for the segmentation of imbalanced objects within images. Therefore, we propose a new approach, based on diffusion models, which can enrich an imbalanced dataset with plausible examples from underrepresented groups by conditioning on segmentation maps. Our method can simply expand limited clinical datasets making them suitable to train machine learning pipelines, and provides an interpretable and human-controllable way of generating histopathology images that are indistinguishable from real ones to human experts. We validate our findings on two datasets, one from the public domain and one from a Kidney Transplant study.

Foivos Paraperas Papantoniou, Stathis Galanakis, Rolandos Alexandros Potamias et al.

This paper presents STARCaster, an identity-aware spatio-temporal video diffusion model that addresses both speech-driven portrait animation and free-viewpoint talking portrait synthesis, given an identity embedding or reference image, within a unified framework. Existing 2D speech-to-video diffusion models depend heavily on reference guidance, leading to limited motion diversity. At the same time, 3D-aware animation typically relies on inversion through pre-trained tri-plane generators, which often leads to imperfect reconstructions and identity drift. We rethink reference- and geometry-based paradigms in two ways. First, we deviate from strict reference conditioning at pre-training by introducing softer identity constraints. Second, we address 3D awareness implicitly within the 2D video domain by leveraging the inherent multi-view nature of video data. STARCaster adopts a compositional approach progressing from ID-aware motion modeling, to audio-visual synchronization via lip reading-based supervision, and finally to novel view animation through temporal-to-spatial adaptation. To overcome the scarcity of 4D audio-visual data, we propose a decoupled learning approach in which view consistency and temporal coherence are trained independently. A self-forcing training scheme enables the model to learn from longer temporal contexts than those generated at inference, mitigating the overly static animations common in existing autoregressive approaches. Comprehensive evaluations demonstrate that STARCaster generalizes effectively across tasks and identities, consistently surpassing prior approaches in different benchmarks.

Johanna P. Müller, Matthew Baugh, Jeremy Tan et al.

Universal anomaly detection still remains a challenging problem in machine learning and medical image analysis. It is possible to learn an expected distribution from a single class of normative samples, e.g., through epistemic uncertainty estimates, auto-encoding models, or from synthetic anomalies in a self-supervised way. The performance of self-supervised anomaly detection approaches is still inferior compared to methods that use examples from known unknown classes to shape the decision boundary. However, outlier exposure methods often do not identify unknown unknowns. Here we discuss an improved self-supervised single-class training strategy that supports the approximation of probabilistic inference with loosen feature locality constraints. We show that up-scaling of gradients with histogram-equalised images is beneficial for recently proposed self-supervision tasks. Our method is integrated into several out-of-distribution (OOD) detection models and we show evidence that our method outperforms the state-of-the-art on various benchmark datasets.

Mischa Dombrowski, Bernhard Kainz

Recent advances in score-based generative models have led to a huge spike in the development of downstream applications using generative models ranging from data augmentation over image and video generation to anomaly detection. Despite publicly available trained models, their potential to be used for privacy preserving data sharing has not been fully explored yet. Training diffusion models on private data and disseminating the models and weights rather than the raw dataset paves the way for innovative large-scale data-sharing strategies, particularly in healthcare, where safeguarding patients' personal health information is paramount. However, publishing such models without individual consent of, e.g., the patients from whom the data was acquired, necessitates guarantees that identifiable training samples will never be reproduced, thus protecting personal health data and satisfying the requirements of policymakers and regulatory bodies. This paper introduces a method for estimating the upper bound of the probability of reproducing identifiable training images during the sampling process. This is achieved by designing an adversarial approach that searches for anatomic fingerprints, such as medical devices or dermal art, which could potentially be employed to re-identify training images. Our method harnesses the learned score-based model to estimate the probability of the entire subspace of the score function that may be utilized for one-to-one reproduction of training samples. To validate our estimates, we generate anomalies containing a fingerprint and investigate whether generated samples from trained generative models can be uniquely mapped to the original training samples. Overall our results show that privacy-breaching images are reproduced at sampling time if the models were trained without care.

Jiayi Wang, Hadrien Reynaud, Ibrahim Ethem Hamamci et al.

Large-scale, volumetric medical imaging datasets typically aggregate scans from different vendors and devices, resulting in highly variable resolution, slice thicknesses, and numbers of slices per study. Consequently, training representation models usually requires cropping or interpolating along the z-axis to obtain fixed-size blocks, which inevitably causes information loss. We propose a new training approach to overcome this limitation. Instead of absolute position embeddings, we interpret volumes as sequences of 3D chunks and adopt Rotary Position Embeddings, allowing us to treat the z-axis as an unconstrained temporal dimensions. Building on this idea, we introduce a new vision-language model: SigVLP. In SigVLP, we implement Rotary Position Embedding as the positional encoding method, which is applied directly within the attention operation, generating input-conditioned sine and cosine weights on the fly. This design ensures consistent alignment between query and key projections and adapts to any input sizes. To allow for variable input size during training, we sample Computed Tomography volumes in chunks and pair them with localized organ-wise textual observations. Compared to using entire reports for conditioning, chunkwise alignment provides finer-grained supervision, enabling the model to establish stronger correlations between the text and volume representations, thereby improving the precision of text-to-volume alignment. Our models are trained with the Muon optimizer and evaluated on a diverse set of downstream tasks, including zero-shot abnormality and organ classification, segmentation, and retrieval tasks.

Matthew Baugh, Jeremy Tan, Athanasios Vlontzos et al.

The wide variety of in-distribution and out-of-distribution data in medical imaging makes universal anomaly detection a challenging task. Recently a number of self-supervised methods have been developed that train end-to-end models on healthy data augmented with synthetic anomalies. However, it is difficult to compare these methods as it is not clear whether gains in performance are from the task itself or the training pipeline around it. It is also difficult to assess whether a task generalises well for universal anomaly detection, as they are often only tested on a limited range of anomalies. To assist with this we have developed nnOOD, a framework that adapts nnU-Net to allow for comparison of self-supervised anomaly localisation methods. By isolating the synthetic, self-supervised task from the rest of the training process we perform a more faithful comparison of the tasks, whilst also making the workflow for evaluating over a given dataset quick and easy. Using this we have implemented the current state-of-the-art tasks and evaluated them on a challenging X-ray dataset.

Zhe Li, Hadrien Reynaud, Johanna P Müller et al.

Ultrasound echocardiography is essential for the non-invasive, real-time assessment of cardiac function, but the scarcity of labelled data, driven by privacy restrictions and the complexity of expert annotation, remains a major obstacle for deep learning methods. We propose the Motion Conditioned Diffusion Model (MCDM), a label-free latent diffusion framework that synthesises realistic echocardiography videos conditioned on self-supervised motion features. To extract these features, we design the Motion and Appearance Feature Extractor (MAFE), which disentangles motion and appearance representations from videos. Feature learning is further enhanced by two auxiliary objectives: a re-identification loss guided by pseudo appearance features and an optical flow loss guided by pseudo flow fields. Evaluated on the EchoNet-Dynamic dataset, MCDM achieves competitive video generation performance, producing temporally coherent and clinically realistic sequences without reliance on manual labels. These results demonstrate the potential of self-supervised conditioning for scalable echocardiography synthesis. Our code is available at https://github.com/ZheLi2020/LabelfreeMCDM.

Sudarshan Sreeram, Bernhard Kainz

Leveraging ML advancements to augment healthcare systems can improve patient outcomes. Yet, uninformed engineering decisions in early-stage research inadvertently hinder the feasibility of such solutions for high-throughput, on-device inference, particularly in settings involving legacy hardware and multi-modal gigapixel images. Through a preliminary case study concerning segmentation in cardiology, we highlight the excess operational complexity in a suboptimally configured ML model from prior work and demonstrate that it can be sculpted away using pruning to meet deployment criteria. Our results show a compression rate of 1148x with minimal loss in quality (~4%) and, at higher rates, achieve faster inference on a CPU than the GPU baseline, stressing the need to consider task complexity and architectural details when using off-the-shelf models. With this, we consider avenues for future research in streamlining workflows for clinical researchers to develop models quicker and better suited for real-world use.

Clara Lebbos, Jen Barcroft, Jeremy Tan et al.

Ovarian cancer is the most lethal gynaecological malignancy. The disease is most commonly asymptomatic at its early stages and its diagnosis relies on expert evaluation of transvaginal ultrasound images. Ultrasound is the first-line imaging modality for characterising adnexal masses, it requires significant expertise and its analysis is subjective and labour-intensive, therefore open to error. Hence, automating processes to facilitate and standardise the evaluation of scans is desired in clinical practice. Using supervised learning, we have demonstrated that segmentation of adnexal masses is possible, however, prevalence and label imbalance restricts the performance on under-represented classes. To mitigate this we apply a novel pathology-specific data synthesiser. We create synthetic medical images with their corresponding ground truth segmentations by using Poisson image editing to integrate less common masses into other samples. Our approach achieves the best performance across all classes, including an improvement of up to 8% when compared with nnU-Net baseline approaches.

Sergio Naval Marimont, Vasilis Siomos, Matthew Baugh et al.

Unsupervised Anomaly Detection (UAD) methods aim to identify anomalies in test samples comparing them with a normative distribution learned from a dataset known to be anomaly-free. Approaches based on generative models offer interpretability by generating anomaly-free versions of test images, but are typically unable to identify subtle anomalies. Alternatively, approaches using feature modelling or self-supervised methods, such as the ones relying on synthetically generated anomalies, do not provide out-of-the-box interpretability. In this work, we present a novel method that combines the strengths of both strategies: a generative cold-diffusion pipeline (i.e., a diffusion-like pipeline which uses corruptions not based on noise) that is trained with the objective of turning synthetically-corrupted images back to their normal, original appearance. To support our pipeline we introduce a novel synthetic anomaly generation procedure, called DAG, and a novel anomaly score which ensembles restorations conditioned with different degrees of abnormality. Our method surpasses the prior state-of-the art for unsupervised anomaly detection in three different Brain MRI datasets.

Sergio Naval Marimont, Matthew Baugh, Vasilis Siomos et al.

Unsupervised Anomaly Detection (UAD) techniques aim to identify and localize anomalies without relying on annotations, only leveraging a model trained on a dataset known to be free of anomalies. Diffusion models learn to modify inputs $x$ to increase the probability of it belonging to a desired distribution, i.e., they model the score function $\nabla_x \log p(x)$. Such a score function is potentially relevant for UAD, since $\nabla_x \log p(x)$ is itself a pixel-wise anomaly score. However, diffusion models are trained to invert a corruption process based on Gaussian noise and the learned score function is unlikely to generalize to medical anomalies. This work addresses the problem of how to learn a score function relevant for UAD and proposes DISYRE: Diffusion-Inspired SYnthetic REstoration. We retain the diffusion-like pipeline but replace the Gaussian noise corruption with a gradual, synthetic anomaly corruption so the learned score function generalizes to medical, naturally occurring anomalies. We evaluate DISYRE on three common Brain MRI UAD benchmarks and substantially outperform other methods in two out of the three tasks.

Mischa Dombrowski, Felix Nützel, Bernhard Kainz

Recent advances in generative image modeling have achieved visual realism sufficient to deceive human experts, yet their potential for privacy preserving data sharing remains insufficiently understood. A central obstacle is the absence of reliable memorization detection mechanisms, limited quantitative evaluation, and poor generalization of existing privacy auditing methods across domains. To address this, we propose to view memorization detection as a unified problem at the intersection of re-identification and copy detection, whose complementary goals cover both identity consistency and augmentation-robust duplication, and introduce Latent Contrastive Memorization Network (LCMem), a cross-domain model evaluated jointly on both tasks. LCMem achieves this through a two-stage training strategy that first learns identity consistency before incorporating augmentation-robust copy detection. Across six benchmark datasets, LCMem achieves improvements of up to 16 percentage points on re-identification and 30 percentage points on copy detection, enabling substantially more reliable memorization detection at scale. Our results show that existing privacy filters provide limited performance and robustness, highlighting the need for stronger protection mechanisms. We show that LCMem sets a new standard for cross-domain privacy auditing, offering reliable and scalable memorization detection. Code and model is publicly available at https://github.com/MischaD/LCMem.

Luca Schmidtke, Benjamin Hou, Athanasios Vlontzos et al.

Inferring 3D human pose from 2D images is a challenging and long-standing problem in the field of computer vision with many applications including motion capture, virtual reality, surveillance or gait analysis for sports and medicine. We present preliminary results for a method to estimate 3D pose from 2D video containing a single person and a static background without the need for any manual landmark annotations. We achieve this by formulating a simple yet effective self-supervision task: our model is required to reconstruct a random frame of a video given a frame from another timepoint and a rendered image of a transformed human shape template. Crucially for optimisation, our ray casting based rendering pipeline is fully differentiable, enabling end to end training solely based on the reconstruction task.

Ibrahim Ethem Hamamci, Sezgin Er, Chenyu Wang et al.

Advancements in medical imaging AI, particularly in 3D imaging, have been limited due to the scarcity of comprehensive datasets. We introduce CT-RATE, a public dataset that pairs 3D medical images with corresponding textual reports. CT-RATE comprises 25,692 non-contrast 3D chest CT scans from 21,304 unique patients. Each scan is accompanied by its corresponding radiology report. Leveraging CT-RATE, we develop CT-CLIP, a CT-focused contrastive language-image pretraining framework designed for broad applications without the need for task-specific training. We demonstrate how CT-CLIP can be used in multi-abnormality detection and case retrieval, and outperforms state-of-the-art fully supervised models across all key metrics. By combining CT-CLIP's vision encoder with a pretrained large language model, we create CT-CHAT, a vision-language foundational chat model for 3D chest CT volumes. Finetuned on over 2.7 million question-answer pairs derived from the CT-RATE dataset, CT-CHAT underscores the necessity for specialized methods in 3D medical imaging. Collectively, the open-source release of CT-RATE, CT-CLIP, and CT-CHAT not only addresses critical challenges in 3D medical imaging but also lays the groundwork for future innovations in medical AI and improved patient care.

Bernhard Kainz, Johanna P Mueller, Matthew Baugh et al.

Zero-shot anomaly localisation via vision-language models (VLMs) offers a compelling approach for rare pathology detection, yet its performance is fundamentally limited by the absence of healthy anatomical context. We reformulate zero-shot localisation as a comparative inference problem in which anomalies are identified through structured comparison against reference distributions of normal anatomy. We introduce WALDO, a training-free framework grounded in optimal transport theory that enables comparative reasoning through: (i) entropy-weighted Sliced Wasserstein distances for anatomically-aware reference selection from DINOv2 patch distributions, (ii) Goldilocks zone sampling exploiting the non-monotonic relationship between reference similarity and localisation accuracy, and (iii) self-consistency aggregation via weighted non-maximum suppression. We theoretically analyse the Goldilocks effect through distributional divergence, and show that references with moderate similarity minimize a bias-variance trade-off in comparative visual reasoning. On the NOVA brain MRI benchmark, WALDO with Qwen2.5-VL-72B achieves $43.5_{\pm1.6}\%$ mAP@30 (95\% CI: [40.4, 46.7]), representing a 19\% relative improvement over zero-shot baselines. Cross-model evaluation shows consistent gains: GPT-4o achieves $32.0_{\pm6.5}\%$ and Qwen3-VL-32B achieves $32.0_{\pm6.6}\%$ mAP@30. Paired McNemar tests confirm statistical significance ($p<0.01$). Source code is available at https://github.com/bkainz/WALDO_MICCAI26_demo .

Glejdis Shkëmbi, Johanna P. Müller, Zhe Li et al.

Breast cancer is a major concern for women's health globally, with axillary lymph node (ALN) metastasis identification being critical for prognosis evaluation and treatment guidance. This paper presents a deep learning (DL) classification pipeline for quantifying clinical information from digital core-needle biopsy (CNB) images, with one step less than existing methods. A publicly available dataset of 1058 patients was used to evaluate the performance of different baseline state-of-the-art (SOTA) DL models in classifying ALN metastatic status based on CNB images. An extensive ablation study of various data augmentation techniques was also conducted. Finally, the manual tumor segmentation and annotation step performed by the pathologists was assessed.

Zhi-Song Liu, Markus Büttner, Matthew Scarborough et al.

Learning the fine-scale details of a coastal ocean simulation from a coarse representation is a challenging task. For real-world applications, high-resolution simulations are necessary to advance understanding of many coastal processes, specifically, to predict flooding resulting from tsunamis and storm surges. We propose a Downscaling Neural Network for Coastal Simulation (DNNCS) for spatiotemporal enhancement to learn the high-resolution numerical solution. Given images of coastal simulations produced on low-resolution computational meshes using low polynomial order discontinuous Galerkin discretizations and a coarse temporal resolution, the proposed DNNCS learns to produce high-resolution free surface elevation and velocity visualizations in both time and space. To model the dynamic changes over time and space, we propose grid-aware spatiotemporal attention to project the temporal features to the spatial domain for non-local feature matching. The coordinate information is also utilized via positional encoding. For the final reconstruction, we use the spatiotemporal bilinear operation to interpolate the missing frames and then expand the feature maps to the frequency domain for residual mapping. Besides data-driven losses, the proposed physics-informed loss guarantees gradient consistency and momentum changes, leading to a 24% reduction in root-mean-square error compared to the model trained with only data-driven losses. To train the proposed model, we propose a coastal simulation dataset and use it for model optimization and evaluation. Our method shows superior downscaling quality and fast computation compared to the state-of-the-art methods.

Sarah Cechnicka, James Ball, Matthew Baugh et al.

Diagnosing medical conditions from histopathology data requires a thorough analysis across the various resolutions of Whole Slide Images (WSI). However, existing generative methods fail to consistently represent the hierarchical structure of WSIs due to a focus on high-fidelity patches. To tackle this, we propose Ultra-Resolution Cascaded Diffusion Models (URCDMs) which are capable of synthesising entire histopathology images at high resolutions whilst authentically capturing the details of both the underlying anatomy and pathology at all magnification levels. We evaluate our method on three separate datasets, consisting of brain, breast and kidney tissue, and surpass existing state-of-the-art multi-resolution models. Furthermore, an expert evaluation study was conducted, demonstrating that URCDMs consistently generate outputs across various resolutions that trained evaluators cannot distinguish from real images. All code and additional examples can be found on GitHub.

Athanasios Vlontzos, Hadrien Reynaud, Bernhard Kainz

Curating a large scale medical imaging dataset for machine learning applications is both time consuming and expensive. Balancing the workload between model development, data collection and annotations is difficult for machine learning practitioners, especially under time constraints. Causal analysis is often used in medicine and economics to gain insights about the effects of actions and policies. In this paper we explore the effect of dataset interventions on the output of image classification models. Through a causal approach we investigate the effects of the quantity and type of data we need to incorporate in a dataset to achieve better performance for specific subtasks. The main goal of this paper is to highlight the potential of causal analysis as a tool for resource optimization for developing medical imaging ML applications. We explore this concept with a synthetic dataset and an exemplary use-case for Diabetic Retinopathy image analysis.

Liu Li, Hanchun Wang, Matthew Baugh et al.

Although existing medical image segmentation methods provide impressive pixel-wise accuracy, they often neglect topological correctness, making their segmentations unusable for many downstream tasks. One option is to retrain such models whilst including a topology-driven loss component. However, this is computationally expensive and often impractical. A better solution would be to have a versatile plug-and-play topology refinement method that is compatible with any domain-specific segmentation pipeline. Directly training a post-processing model to mitigate topological errors often fails as such models tend to be biased towards the topological errors of a target segmentation network. The diversity of these errors is confined to the information provided by a labelled training set, which is especially problematic for small datasets. Our method solves this problem by training a model-agnostic topology refinement network with synthetic segmentations that cover a wide variety of topological errors. Inspired by the Stone-Weierstrass theorem, we synthesize topology-perturbation masks with randomly sampled coefficients of orthogonal polynomial bases, which ensures a complete and unbiased representation. Practically, we verified the efficiency and effectiveness of our methods as being compatible with multiple families of polynomial bases, and show evidence that our universal plug-and-play topology refinement network outperforms both existing topology-driven learning-based and post-processing methods. We also show that combining our method with learning-based models provides an effortless add-on, which can further improve the performance of existing approaches.

Hadrien Reynaud, Matthew Baugh, Mischa Dombrowski et al.

We introduce the Joint Video-Image Diffusion model (JVID), a novel approach to generating high-quality and temporally coherent videos. We achieve this by integrating two diffusion models: a Latent Image Diffusion Model (LIDM) trained on images and a Latent Video Diffusion Model (LVDM) trained on video data. Our method combines these models in the reverse diffusion process, where the LIDM enhances image quality and the LVDM ensures temporal consistency. This unique combination allows us to effectively handle the complex spatio-temporal dynamics in video generation. Our results demonstrate quantitative and qualitative improvements in producing realistic and coherent videos.

Mathias Öttl, Frauke Wilm, Jana Steenpass et al.

Deep learning-based image generation has seen significant advancements with diffusion models, notably improving the quality of generated images. Despite these developments, generating images with unseen characteristics beneficial for downstream tasks has received limited attention. To bridge this gap, we propose Style-Extracting Diffusion Models, featuring two conditioning mechanisms. Specifically, we utilize 1) a style conditioning mechanism which allows to inject style information of previously unseen images during image generation and 2) a content conditioning which can be targeted to a downstream task, e.g., layout for segmentation. We introduce a trainable style encoder to extract style information from images, and an aggregation block that merges style information from multiple style inputs. This architecture enables the generation of images with unseen styles in a zero-shot manner, by leveraging styles from unseen images, resulting in more diverse generations. In this work, we use the image layout as target condition and first show the capability of our method on a natural image dataset as a proof-of-concept. We further demonstrate its versatility in histopathology, where we combine prior knowledge about tissue composition and unannotated data to create diverse synthetic images with known layouts. This allows us to generate additional synthetic data to train a segmentation network in a semi-supervised fashion. We verify the added value of the generated images by showing improved segmentation results and lower performance variability between patients when synthetic images are included during segmentation training. Our code will be made publicly available at [LINK].

Mischa Dombrowski, Felix Nützel, Bernhard Kainz

Generative data augmentation with latent diffusion models is a promising strategy for addressing class imbalance in medical imaging, yet current approaches focus on perceptual fidelity and domain-specific autoencoder fine-tuning while neglecting a more fundamental bottleneck. We identify and formalize the learnability gap: large-scale pretrained autoencoders faithfully encode discriminative features for medical classification, as evidenced by near-lossless performance in reconstruction space, yet their latent representations are structured in ways that are difficult for classifiers to learn from. Across five autoencoder families and four medical benchmarks spanning chest radiography, dermatoscopy, computed tomography, and echocardiography, we show that this gap persists regardless of architecture, initialization strategy, or hyperparameter tuning, and that medical-domain fine-tuning of the autoencoder does not close it. To probe and partially narrow the gap, we develop noise-conditioned latent classifiers with FiLM layers and image-space distillation that offer 64x throughput and 120x memory gains over image-space models while serving as diagnostic tools for latent space quality. Our analysis provides a new framework for evaluating autoencoder latent spaces and identifies their structure, rather than their fidelity or domain specificity, as the primary obstacle to closing the performance gap between real and synthetic medical training data.

Mischa Dombrowski, Weitong Zhang, Sarah Cechnicka et al.

Generative methods now produce outputs nearly indistinguishable from real data but often fail to fully capture the data distribution. Unlike quality issues, diversity limitations in generative models are hard to detect visually, requiring specific metrics for assessment. In this paper, we draw attention to the current lack of diversity in generative models and the inability of common metrics to measure this. We achieve this by framing diversity as an image retrieval problem, where we measure how many real images can be retrieved using synthetic data as queries. This yields the Image Retrieval Score (IRS), an interpretable, hyperparameter-free metric that quantifies the diversity of a generative model's output. IRS requires only a subset of synthetic samples and provides a statistical measure of confidence. Our experiments indicate that current feature extractors commonly used in generative model assessment are inadequate for evaluating diversity effectively. Consequently, we perform an extensive search for the best feature extractors to assess diversity. Evaluation reveals that current diffusion models converge to limited subsets of the real distribution, with no current state-of-the-art models superpassing 77% of the diversity of the training data. To address this limitation, we introduce Diversity-Aware Diffusion Models (DiADM), a novel approach that improves diversity of unconditional diffusion models without loss of image quality. We do this by disentangling diversity from image quality by using a diversity aware module that uses pseudo-unconditional features as input. We provide a Python package offering unified feature extraction and metric computation to further facilitate the evaluation of generative models https://github.com/MischaD/beyondfid.

Felix Nützel, Mischa Dombrowski, Bernhard Kainz

Rare diseases dominate the diagnostic challenge in medical imaging yet are severely underrepresented in clinical datasets, causing classifiers to fail on exactly the conditions where reliable detection matters most. Generative augmentation can supply the missing tail-class coverage, but coarse disease labels aggregate diverse subtypes and acquisition settings into multi-modal conditionals that bias generators toward dominant submodes, while a shared Gaussian source forces rare subpopulations through disproportionately long transport paths. We propose an offline strategy that introduces informative priors at two levels: first, we partition each coarse label into coherent submodes via Gaussian mixture modeling in the generative model's latent space; second, we learn subclass-conditioned source distributions that re-center and re-scale the starting distribution per submode, shortening trajectories and reducing within-subclass dispersion. To prevent degenerate solutions we impose explicit geometric control, moderately concentrating normalized displacement directions around learnable prototypes while capping path-length outliers. On long-tailed chest X-ray (MIMIC-LT, NIH-LT) and CT slice (CT-RATE) benchmarks the proposed method consistently improves tail-class generation fidelity and diversity (FID, IRS) and is a promising augmentation strategy that reliably improves downstream balanced accuracy and macro-F1 over a non-augmented baseline across modalities.

Hadrien Reynaud, Alberto Gomez, Paul Leeson et al.

Advances in deep learning have significantly enhanced medical image analysis, yet the availability of large-scale medical datasets remains constrained by patient privacy concerns. We present EchoFlow, a novel framework designed to generate high-quality, privacy-preserving synthetic echocardiogram images and videos. EchoFlow comprises four key components: an adversarial variational autoencoder for defining an efficient latent representation of cardiac ultrasound images, a latent image flow matching model for generating accurate latent echocardiogram images, a latent re-identification model to ensure privacy by filtering images anatomically, and a latent video flow matching model for animating latent images into realistic echocardiogram videos conditioned on ejection fraction. We rigorously evaluate our synthetic datasets on the clinically relevant task of ejection fraction regression and demonstrate, for the first time, that downstream models trained exclusively on EchoFlow-generated synthetic datasets achieve performance parity with models trained on real datasets. We release our models and synthetic datasets, enabling broader, privacy-compliant research in medical ultrasound imaging at https://huggingface.co/spaces/HReynaud/EchoFlow.

Zhe Li, Hadrien Reynaud, Alberto Gomez et al.

Echocardiography playing a critical role in the diagnosis and monitoring of cardiovascular diseases as a non-invasive real-time assessment of cardiac structure and function. However, the growing scale of echocardiographic video data presents significant challenges in terms of storage, computation, and model training efficiency. Dataset distillation offers a promising solution by synthesizing a compact, informative subset of data that retains the key clinical features of the original dataset. In this work, we propose a novel approach for distilling a compact synthetic echocardiographic video dataset. Our method leverages motion feature extraction to capture temporal dynamics, followed by class-wise graph construction and representative sample selection using the Infomap algorithm. This enables us to select a diverse and informative subset of synthetic videos that preserves the essential characteristics of the original dataset. We evaluate our approach on the EchoNet-Dynamic datasets and achieve a test accuracy of \(69.38\%\) using only \(25\) synthetic videos. These results demonstrate the effectiveness and scalability of our method for medical video dataset distillation.

Ben Schaper, Maxime Di Folco, Bernhard Kainz et al.

Vision-Language Models show strong zero-shot performance for chest X-ray classification, but standard flat metrics fail to distinguish between clinically minor and severe errors. This work investigates how to quantify and mitigate abstraction errors by leveraging medical taxonomies. We benchmark several state-of-the-art VLMs using hierarchical metrics and introduce Catastrophic Abstraction Errors to capture cross-branch mistakes. Our results reveal substantial misalignment of VLMs with clinical taxonomies despite high flat performance. To address this, we propose risk-constrained thresholding and taxonomy-aware fine-tuning with radial embeddings, which reduce severe abstraction errors to below 2 per cent while maintaining competitive performance. These findings highlight the importance of hierarchical evaluation and representation-level alignment for safer and more clinically meaningful deployment of VLMs.

Hadrien Reynaud, Bernhard Kainz

This work presents an extensive hyperparameter search on Image Diffusion Models for Echocardiogram generation. The objective is to establish foundational benchmarks and provide guidelines within the realm of ultrasound image and video generation. This study builds over the latest advancements, including cutting-edge model architectures and training methodologies. We also examine the distribution shift between real and generated samples and consider potential solutions, crucial to train efficient models on generated data. We determine an Optimal FID score of $0.88$ for our research problem and achieve an FID of $2.60$. This work is aimed at contributing valuable insights and serving as a reference for further developments in the specialized field of ultrasound image and video generation.

Felix Nützel, Mischa Dombrowski, Bernhard Kainz

Recent advances in text-to-image diffusion models enable high-fidelity generation across diverse prompts. However, these models falter in long-tail settings, such as medical imaging, where rare pathologies comprise a small fraction of data. This results in mode collapse: tail-class outputs lack quality and diversity, undermining the goal of synthetic data augmentation for underrepresented conditions. We pinpoint gradient conflicts between frequent head and rare tail classes as the primary culprit, a factor unaddressed by existing sampling or conditioning methods that mainly steer inference without altering the learned distribution. To resolve this, we propose GRASP: Guided Residual Adapters with Sample-wise Partitioning. GRASP uses external priors to statically partition samples into clusters that minimize intra-group gradient clashes. It then fine-tunes pre-trained models by injecting cluster-specific residual adapters into transformer feedforward layers, bypassing learned gating for stability and efficiency. On the long-tail MIMIC-CXR-LT dataset, GRASP yields superior FID and diversity metrics, especially for rare classes, outperforming baselines like vanilla fine-tuning and Mixture of Experts variants. Downstream classification on NIH-CXR-LT improves considerably for tail labels. Generalization to ImageNet-LT confirms broad applicability. Our method is lightweight, scalable, and readily integrates with diffusion pipelines.

Ibrahim Ethem Hamamci, Sezgin Er, Suprosanna Shit et al.

Recent progress in vision-language modeling for 3D medical imaging has been fueled by large-scale computed tomography (CT) corpora with paired free-text reports, stronger architectures, and powerful pretrained models. This has enabled applications such as automated report generation and text-conditioned 3D image synthesis. Yet, current approaches struggle with high-resolution, long-sequence volumes: contrastive pretraining often yields vision encoders that are misaligned with clinical language, and slice-wise tokenization blurs fine anatomy, reducing diagnostic performance on downstream tasks. We introduce BTB3D (Better Tokens for Better 3D), a causal convolutional encoder-decoder that unifies 2D and 3D training and inference while producing compact, frequency-aware volumetric tokens. A three-stage training curriculum enables (i) local reconstruction, (ii) overlapping-window tiling, and (iii) long-context decoder refinement, during which the model learns from short slice excerpts yet generalizes to scans exceeding 300 slices without additional memory overhead. BTB3D sets a new state-of-the-art on two key tasks: it improves BLEU scores and increases clinical F1 by 40% over CT2Rep, CT-CHAT, and Merlin for report generation; and it reduces FID by 75% and halves FVD compared to GenerateCT and MedSyn for text-to-CT synthesis, producing anatomically consistent 512*512*241 volumes. These results confirm that precise three-dimensional tokenization, rather than larger language backbones alone, is essential for scalable vision-language modeling in 3D medical imaging. The codebase is available at: https://github.com/ibrahimethemhamamci/BTB3D

Sarah Cechnicka, Matthew Baugh, Weitong Zhang et al.

Recent advances in Diffusion Probabilistic Models (DPMs) have set new standards in high-quality image synthesis. Yet, controlled generation remains challenging, particularly in sensitive areas such as medical imaging. Medical images feature inherent structure such as consistent spatial arrangement, shape or texture, all of which are critical for diagnosis. However, existing DPMs operate in noisy latent spaces that lack semantic structure and strong priors, making it difficult to ensure meaningful control over generated content. To address this, we propose graph-based object-level representations for Graph-Conditioned-Diffusion. Our approach generates graph nodes corresponding to each major structure in the image, encapsulating their individual features and relationships. These graph representations are processed by a transformer module and integrated into a diffusion model via the text-conditioning mechanism, enabling fine-grained control over generation. We evaluate this approach using a real-world histopathology use case, demonstrating that our generated data can reliably substitute for annotated patient data in downstream segmentation tasks. The code is available here.

Felix Nützel, Mischa Dombrowski, Bernhard Kainz

Retrieval-augmented learning based on radiology reports has emerged as a promising direction to improve performance on long-tail medical imaging tasks, such as rare disease detection in chest X-rays. Most existing methods rely on comparing high-dimensional text embeddings from models like CLIP or CXR-BERT, which are often difficult to interpret, computationally expensive, and not well-aligned with the structured nature of medical knowledge. We propose a novel, ontology-driven alternative for comparing radiology report texts based on clinically grounded concepts from the Unified Medical Language System (UMLS). Our method extracts standardised medical entities from free-text reports using an enhanced pipeline built on RadGraph-XL and SapBERT. These entities are linked to UMLS concepts (CUIs), enabling a transparent, interpretable set-based representation of each report. We then define a task-adaptive similarity measure based on a modified and weighted version of the Tversky Index that accounts for synonymy, negation, and hierarchical relationships between medical entities. This allows efficient and semantically meaningful similarity comparisons between reports. We demonstrate that our approach outperforms state-of-the-art embedding-based retrieval methods in a radiograph classification task on MIMIC-CXR, particularly in long-tail settings. Additionally, we use our pipeline to generate ontology-backed disease labels for MIMIC-CXR, offering a valuable new resource for downstream learning tasks. Our work provides more explainable, reliable, and task-specific retrieval strategies in clinical AI systems, especially when interpretability and domain knowledge integration are essential. Our code is available at https://github.com/Felix-012/ontology-concept-distillation