is Professor for AI for Image-Guided Diagnosis and Therapy at TU Munich.
His research bridges the gap between medicine and computer science towards data-driven, personalized medicine for diagnosis and therapy. His research focuses on developing innovative computational analysis methods to extract actionable biomarkers for clinical decision-making from heterogeneous, multi-modal medical data. Translating these advancements into clinical application is a core motivation for his work.
The fifth edition of the WHO classification of brain tumors increasingly emphasizes the role of extensive genetic testing in the diagnosis of gliomas. In this context, computational pathology foundation models (FMs) present a promising approach for inferring molecular entities directly from conventional, H&E-stained histological images, potentially reducing the need for genetic analysis. We conducted a robust investigation into the ability of five established FMs to generate effective embeddings for downstream glioma classification using three datasets (TCGA, n=839 samples; EBRAINS, n=786 samples; TUM, n=250 samples) and state-of-the-art augmentation techniques. Our results demonstrate that FM embeddings enable competitive glioma classification performance, even with limited training data, achieving one-vs-rest AUC0.93 on all three datasets. However, we observed substantial differences between FMs in their downstream performance, susceptibility to perturbations, and consistency across multiple datasets. Dataset diversity and content of central nervous tissue were associated with improved generalization, while model and dataset size were not. Common to all FMs was a propensity to capture dataset-specific features in their embeddings. We examined Macenko normalization and random convolutions as potential solutions to combat dataset-dominated embeddings and show that ensembling FM embeddings over multiple augmented views improves downstream classifier performance. In summary, our findings highlight both the promise and current limitations of computational pathology foundation models for glioma classification, emphasizing the critical roles of training data composition and downstream augmentation to achieve strong task performance.
Vision foundation models like DINOv2 demonstrate remarkable potential in medical imaging despite their origin in natural image domains. However, their design inherently works best for uni-modal image analysis, limiting their effectiveness for multi-modal imaging tasks that are common in many medical fields, such as neurology and oncology. While supervised models perform well in this setting, they fail to leverage unlabeled datasets and struggle with missing modalities, a frequent challenge in clinical settings. To bridge these gaps, we introduce MM-DINOv2, a novel and efficient framework that adapts the pre-trained vision foundation model DINOv2 for multi-modal medical imaging. Our approach incorporates multi-modal patch embeddings, enabling vision foundation models to effectively process multi-modal imaging data. To address missing modalities, we employ full-modality masking, which encourages the model to learn robust cross-modality relationships. Furthermore, we leverage semi-supervised learning to harness large unlabeled datasets, enhancing both the accuracy and reliability of medical predictions. Applied to glioma subtype classification from multi-sequence brain MRI, our method achieves a Matthews Correlation Coefficient (MCC) of 0.6 on an external test set, surpassing state-of-the-art supervised approaches by +11.1%. Our work establishes a scalable and robust solution for multi-modal medical imaging tasks, leveraging powerful vision foundation models pre-trained on natural images while addressing real-world clinical challenges such as missing data and limited annotations.
Magnetic resonance imaging (MRI) is an invaluable tool for clinical and research applications. Yet, variations in scanners and acquisition parameters cause inconsistencies in image contrast, hindering data comparability and reproducibility across datasets and clinical studies. Existing scanner harmonization methods, designed to address this challenge, face limitations, such as requiring traveling subjects or struggling to generalize to unseen domains. We propose a novel approach using a conditioned diffusion autoencoder with a contrastive loss and domain-agnostic contrast augmentation to harmonize MR images across scanners while preserving subject-specific anatomy. Our method enables brain MRI synthesis from a single reference image. It outperforms baseline techniques, achieving a +7% PSNR improvement on a traveling subjects dataset and +18% improvement on age regression in unseen. Our model provides robust, effective harmonization of brain MRIs to target scanners without requiring fine-tuning. This advancement promises to enhance comparability, reproducibility, and generalizability in multi-site and longitudinal clinical studies, ultimately contributing to improved healthcare outcomes.
Conventional T1-weighted (T1w) magnetic resonance imaging (MRI) is commonly used in multiple sclerosis (MS) morphometry and volumetry research. However, arbitrary intensity scales preclude interpretation of signal values across patients, sites, and time. This requires quantitative MRI techniques, which are not always available. This study assessed T1w image intensity scaling methods, relying on extracerebral reference regions, for quantitative analysis of brain MRI in MS. In total, 701 people with a diagnosis of radiologically isolated syndrome, clinically isolated syndrome, or MS were included. Four intensity scaling strategies were applied: 1) MRI signal modeling, 2) linear scaling with reference regions, 3) z-score standardization, and 4) none (only bias field correction). Methods were evaluated using variance analysis, R1 map comparison, and normal-appearing white matter (NAWM) intensity group comparison, using mean and coefficient of variation (CoV), between low (≤3) and high (>3) expanded disability status scale (EDSS) scores. Statistical analysis was conducted using Pearson’s r, two-sided Welch two-sample t-test, ANCOVA, and Cohen’s d.
Linear scaling with temporal fatty tissue achieved the most consistent variance reduction and strong correlation with R1 maps (r = 0.84). R1 values in NAWM were significantly lower in people with high compared to low EDSS scores (d = -0.351). Similarly, group differences in mean NAWM intensity of fat-scaled images were significant (d = -0.252). The largest group differences were found in NAWM CoV in bias field-corrected T1w images (d = 0.818). Linear scaling with fatty tissue most accurately reproduced the results obtained with R1 maps. Changes in MS NAWM appear to increase intensity variability detectable in conventional T1w images.
The advent of large-scale vision foundation models, pre-trained on diverse natural images, has marked a paradigm shift in computer vision. However, how the frontier vision foundation models’ efficacies transfer to specialized domains remains such as medical imaging remains an open question. This report investigates whether DINOv3, a state-of-the-art self-supervised vision transformer (ViT) that features strong capability in dense prediction tasks, can directly serve as a powerful, unified encoder for medical vision tasks without domain-specific pre-training. To answer this, we benchmark DINOv3 across common medical vision tasks, including 2D/3D classification and segmentation on a wide range of medical imaging modalities. We systematically analyze its scalability by varying model sizes and input image resolutions. Our findings reveal that DINOv3 shows impressive performance and establishes a formidable new baseline. Remarkably, it can even outperform medical-specific foundation models like BiomedCLIP and CT-Net on several tasks, despite being trained solely on natural images. However, we identify clear limitations: The model’s features degrade in scenarios requiring deep domain specialization, such as in Whole-Slide Pathological Images (WSIs), Electron Microscopy (EM), and Positron Emission Tomography (PET). Furthermore, we observe that DINOv3 does not consistently obey scaling law in the medical domain; performance does not reliably increase with larger models or finer feature resolutions, showing diverse scaling behaviors across tasks. Ultimately, our work establishes DINOv3 as a strong baseline, whose powerful visual features can serve as a robust prior for multiple complex medical tasks. This opens promising future directions, such as leveraging its features to enforce multiview consistency in 3D reconstruction.
Ensuring the safety and reliability of large language models (LLMs) in clinical practice is critical to prevent patient harm and promote trustworthy healthcare applications of AI. However, LLMs are advancing so rapidly that static safety benchmarks often become obsolete upon publication, yielding only an incomplete and sometimes misleading picture of model trustworthiness. We demonstrate that a Dynamic, Automatic, and Systematic (DAS) red-teaming framework that continuously stress-tests LLMs can reveal significant weaknesses of current LLMs across four safety-critical domains: robustness, privacy, bias/fairness, and hallucination. A suite of adversarial agents is applied to autonomously mutate test cases, identify/evolve unsafe-triggering strategies, and evaluate responses, uncovering vulnerabilities in real time without human intervention. Applying DAS to 15 proprietary and open-source LLMs revealed a stark contrast between static benchmark performance and vulnerability under adversarial pressure. Despite a median MedQA accuracy exceeding 80%, 94% of previously correct answers failed our dynamic robustness tests. We observed similarly high failure rates across other domains: privacy leaks were elicited in 86% of scenarios, cognitive-bias priming altered clinical recommendations in 81% of fairness tests, and we identified hallucination rates exceeding 66% in widely used models. Such profound residual risks are incompatible with routine clinical practice. By converting red-teaming from a static checklist into a dynamic stress-test audit, DAS red-teaming offers the surveillance that hospitals/regulators/technology vendors require as LLMs become embedded in patient chatbots, decision-support dashboards, and broader healthcare workflows. Our framework delivers an evolvable, scalable, and reliable safeguard for the next generation of medical AI.
Conversational AI tools for generating and discussing accurate radiology reports could transform radiology by enabling collaborative, human-in-the-loop diagnostic processes, saving time and enhancing report quality. While, to this end, Large Vision-Language Models hold promise, current methods lack clinical correctness or are single-task models without conversational abilities. We propose a novel architecture and dataset to address these limitations. First, we propose a secondary image branch, explicitly focusing on structured clinical findings, improving the clinical correctness score by 13.3%. Second, we propose a catastrophic forgetting mitigation strategy and instruct dataset with variable dialog-based tasks, to enable our model to handle a multitude of different queries. RaDialog marks a foundational step toward clinical dialog systems, outperforming existing medical LVLMs by 15.0% in clinical correctness in report generation, 23.4% in interactive report correction, and is preferred by radiologists in 84.0% of cases over a comparative method.
Pathology detection in medical imaging is crucial for radiologists, yet current approaches that train specialized models for each region of interest often lack efficiency and robustness. Furthermore, the scarcity of annotated medical data, particularly for diverse phenotypes, poses significant challenges in achieving generalizability. To address these challenges, we present a novel language-guided object detection pipeline for medical imaging that leverages curriculum learning strategies, chosen for their ability to progressively train models on increasingly complex samples, thereby improving generalization across pathologies, phenotypes, and modalities. We developed a unified pipeline to convert segmentation datasets into bounding box annotations, and applied two curriculum learning approaches - teacher curriculum and bounding box size curriculum - to train a Grounding DINO model. Our method was evaluated on different tumor types in MRI and CT scans and showed significant improvements in detection accuracy. The teacher and bounding box size curriculum learning approaches yielded a 4.9% AP and 5.2% AP increase over baseline, respectively. The results highlight the potential of curriculum learning to optimize medical image analysis and clinical workflow by providing a versatile and efficient detection algorithm.
Butterfly tumors are a distinct class of gliomas that span the corpus callosum, producing a characteristic butterfly-shaped appearance on MRI. The distinctive growth pattern of these tumors highlights how white matter fibers and structural connectivity influence brain tumor cell migration. To investigate this relation, we applied biophysical tumor growth models to a large patient cohort, systematically comparing models that incorporate fiber tract information with those that do not. Our results demonstrate that including fiber orientation data significantly improves model accuracy, particularly for a subset of butterfly tumors. These findings highlight the critical role of white matter architecture in tumor spread and suggest that integrating fiber tract information can enhance the precision of radiotherapy target volume delineation.
Brain tumor growth is unique to each glioma patient and extends beyond what is visible in imaging scans, infiltrating surrounding brain tissue. Understanding these hidden patient-specific progressions is essential for effective therapies. Current treatment plans for brain tumors, such as radiotherapy, typically involve delineating a uniform margin around the visible tumor on pre-treatment scans to target this invisible tumor growth. This ‘one size fits all’ approach is derived from population studies and often fails to account for the nuances of individual patient conditions. We present the Glioma Optimizing the Discrete Loss (GliODIL) framework, which infers the full spatial distribution of tumor cell concentration from available multi-modal imaging, leveraging a Fisher-Kolmogorov type physics model to describe tumor growth. This is achieved through the newly introduced method of Optimizing the Discrete Loss (ODIL), where both data and physics-based constraints are softly assimilated into the solution. Our test dataset comprises 152 glioblastoma patients with pre-treatment imaging and post-treatment follow-ups for tumor recurrence monitoring. By blending data-driven techniques with physics-based constraints, GliODIL enhances recurrence prediction in radiotherapy planning, challenging traditional uniform margins and strict adherence to the Fisher-Kolmogorov partial differential equation model, which is adapted for complex cases.
Body mass index (BMI) does not account for substantial inter-individual differences in regional fat and muscle compartments, which are relevant for the prevalence of cardiometabolic and cancer conditions. We applied a validated deep learning pipeline for automated segmentation of whole-body MRI scans in 45,851 adults from the UK Biobank and German National Cohort, enabling harmonized quantification of visceral (VAT), gluteofemoral (GFAT), and abdominal subcutaneous adipose tissue (ASAT), liver fat fraction (LFF), and trunk muscle volume. Associations with clinical conditions were evaluated using compartment measures adjusted for age, sex, height, and BMI. Our analysis demonstrates that regional adiposity and muscle volume show distinct associations with cardiometabolic and cancer prevalence, and that substantial disease heterogeneity exists within BMI strata. The analytic framework and reference data presented here will support future risk stratification efforts and facilitate the integration of automated MRI phenotyping into large-scale population and clinical research.
Perceptualizing tool interactions with deformable structures in surgical procedures remains challenging, as unimodal visualization techniques often fail to capture the complexity of these interactions due
to constraints such as occlusion and limited depth perception. This paper presents a novel approach to augment tool navigation in mixed reality environments by providing auditory representations of tool-tissue dynamics, particularly for interactions with soft tissue. BioSonix, a physics-informed design framework, utilizes tissue displacements in 3D space to compute excitation forces for a sound model encoding tissue properties such as stiffness and density. Biomechanical simulations were employed to model particle displacements resulting from tool-tissue interactions, establishing a robust foundation for the method. An optimization approach was used to define configurations for capturing diverse interaction scenarios with varying tool trajectories. Experiments were conducted to validate the accuracy of the sound-displacement mappings. Additionally, two user studies were performed: the first involved two clinical professionals (a neuroradiologist and a cardiologist), who confirmed the method’s impact and achieved high task accuracy; the second included 22 biomedical experts, who demonstrated high discrimination accuracy in tissue differentiation and targeting tasks. The results revealed a strong correlation between tool-tissue dynamics and their corresponding auditory profiles, highlighting the potential of these sound representations to en-
hance the intuitive understanding of complex interactions.
In this paper, we propose a methodology for extracting molecular tumor biomarkers from hyperspectral imaging (HSI), an emerging technology for intraoperative tissue assessment. To achieve this, we employ spectral unmixing, allowing to decompose the spectral signals recorded by the HSI camera into their constituent molecular components. Traditional unmixing approaches are based on physical models that establish a relationship between tissue molecules and the recorded spectra. However, these methods commonly assume a linear relationship between the spectra and molecular content, which does not capture the whole complexity of light-matter interaction. To address this limitation, we introduce a novel unmixing procedure that allows to take into account non-linear optical effects while preserving the computational benefits of linear spectral unmixing. We validate our methodology on an in-vivo brain tissue HSI dataset and demonstrate that the extracted molecular information leads to superior classification performance.
Normative representation learning focuses on understanding the typical anatomical distributions from large datasets of medical scans from healthy individuals. Generative Artificial Intelligence (AI) leverages this attribute to synthesize images that accurately reflect these normative patterns. This capability enables the AI allowing them to effectively detect and correct anomalies in new, unseen pathological data without the need for expert labeling. Traditional anomaly detection methods often evaluate the anomaly detection performance, overlooking the crucial role of normative learning. In our analysis, we introduce novel metrics, specifically designed to evaluate this facet in AI models. We apply these metrics across various generative AI frameworks, including advanced diffusion models, and rigorously test them against complex and diverse brain pathologies. In addition, we conduct a large multi-reader study to compare these metrics to experts’ evaluations. Our analysis demonstrates that models proficient in normative learning exhibit exceptional versatility, adeptly detecting a wide range of unseen medical conditions.
Glioblastoma, a highly aggressive brain tumor, poses major challenges due to its poor prognosis and high morbidity rates. Partial differential equation-based models offer promising potential to enhance therapeutic outcomes by simulating patient-specific tumor behavior for improved radiotherapy planning. However, model calibration remains a bottleneck due to the high computational demands of optimization methods like Monte Carlo sampling and evolutionary algorithms. To address this, we recently introduced an approach leveraging a neural forward solver with gradient-based optimization to significantly reduce calibration time. This approach requires a highly accurate and fully differentiable forward model. We investigate multiple architectures, including (i) an enhanced TumorSurrogate, (ii) a modified nnU-Net, and (iii) a 3D Vision Transformer (ViT). The optimized TumorSurrogate achieved the best overall results, excelling in both tumor outline matching and voxel-level prediction of tumor cell concentration. It halved the MSE relative to the baseline model and achieved the highest Dice score across all tumor cell concentration thresholds. Our study demonstrates significant enhancement in forward solver performance and outlines important future research directions.
Predicting future brain states is crucial for understanding healthy aging and neurodegenerative diseases. Longitudinal brain MRI registration, a cornerstone for such analyses, has long been limited by its inability to forecast future developments, reliance on extensive, dense longitudinal data, and the need to balance registration accuracy with temporal smoothness. In this work, we present emph{TimeFlow}, a novel framework for longitudinal brain MRI registration that overcomes all these challenges. Leveraging a U-Net architecture with temporal conditioning inspired by diffusion models, TimeFlow enables accurate longitudinal registration and facilitates prospective analyses through future image prediction. Unlike traditional methods that depend on explicit smoothness regularizers and dense sequential data, TimeFlow achieves temporal consistency and continuity without these constraints. Experimental results highlight its superior performance in both future timepoint prediction and registration accuracy compared to state-of-the-art methods. Additionally, TimeFlow supports novel biological brain aging analyses, effectively differentiating neurodegenerative conditions from healthy aging. It eliminates the need for segmentation, thereby avoiding the challenges of non-trivial annotation and inconsistent segmentation errors. TimeFlow paves the way for accurate, data-efficient, and annotation-free prospective analyses of brain aging and chronic diseases.
Biophysical modeling of brain tumors has emerged as a promising strategy for personalizing radiotherapy planning by estimating the otherwise hidden distribution of tumor cells within the brain. However, many existing state-of-the-art methods are computationally intensive, limiting their widespread translation into clinical practice. In this work, we propose an efficient and direct method that utilizes soft physical constraints to estimate the tumor cell concentration from preoperative MRI of brain tumor patients. Our approach optimizes a 3D tumor concentration field by simultaneously minimizing the difference between the observed MRI and a physically informed loss function. Compared to existing state-of-the-art techniques, our method significantly improves predicting tumor recurrence on two public datasets with a total of 192 patients while maintaining a clinically viable runtime of under one minute - a substantial reduction from the 30 minutes required by the current best approach. Furthermore, we showcase the generalizability of our framework by incorporating additional imaging information and physical constraints, highlighting its potential to translate to various medical diffusion phenomena with imperfect data.
This research explores the integration of language models and unsupervised anomaly detection in medical imaging, addressing two key questions: (1) Can language models enhance the interpretability of anomaly detection maps? and (2) Can anomaly maps improve the generalizability of language models in open-set anomaly detection tasks? To investigate these questions, we introduce a new dataset for multi-image visual question-answering on brain magnetic resonance images encompassing multiple conditions. We propose KQ-Former (Knowledge Querying Transformer), which is designed to optimally align visual and textual information in limited-sample contexts. Our model achieves a 60.81% accuracy on closed questions, covering disease classification and severity across 15 different classes. For open questions, KQ-Former demonstrates a 70% improvement over the baseline with a BLEU-4 score of 0.41, and achieves the highest entailment ratios (up to 71.9%) and lowest contradiction ratios (down to 10.0%) among various natural language inference models. Furthermore, integrating anomaly maps results in an 18% accuracy increase in detecting open-set anomalies, thereby enhancing the language model’s generalizability to previously unseen medical conditions.
VoxelMorph, proposed in 2018, utilizes Convolutional Neural Networks (CNNs) to address medical image registration problems. In 2021 TransMorph advanced this approach by replacing CNNs with Attention mechanisms, claiming enhanced performance. More recently, the rise of Mamba with selective state space models has led to MambaMorph, which substituted Attention with Mamba blocks, asserting superior registration. These developments prompt a critical question: does chasing the latest computational trends with “more advanced” computational blocks genuinely enhance registration accuracy, or is it merely hype? Furthermore, the role of classic high-level registration-specific designs, such as coarse-to-fine pyramid mechanism, correlation calculation, and iterative optimization, warrants scrutiny, particularly in differentiating their influence from the aforementioned low-level computational blocks. In this study, we critically examine these questions through a rigorous evaluation in brain MRI registration. We employed modularized components for each block and ensured unbiased comparisons across all methods and designs to disentangle their effects on performance. Our findings indicate that adopting “advanced” computational elements fails to significantly improve registration accuracy. Instead, well-established registration-specific designs offer fair improvements, enhancing results by a marginal 1.5% over the baseline. Our findings emphasize the importance of rigorous, unbiased evaluation and contribution disentanglement of all low- and high-level registration components, rather than simply following the computer vision trends with “more advanced” computational blocks. We advocate for simpler yet effective solutions and novel evaluation metrics that go beyond conventional registration accuracy, warranting further research across various organs and modalities.
The rapid development of artificial intelligence (AI) has gained importance, with many tools already entering our daily lives. The medical field of radiation oncology is also subject to this development, with AI entering all steps of the patient journey. In this review article, we summarize contemporary AI techniques and explore the clinical applications of AI-based automated segmentation models in radiotherapy planning, focusing on delineation of organs at risk (OARs), the gross tumor volume (GTV), and the clinical target volume (CTV). Emphasizing the need for precise and individualized plans, we review various commercial and freeware segmentation tools and also state-of-the-art approaches. Through our own findings and based on the literature, we demonstrate improved efficiency and consistency as well as time savings in different clinical scenarios. Despite challenges in clinical implementation such as domain shifts, the potential benefits for personalized treatment planning are substantial. The integration of mathematical tumor growth models and AI-based tumor detection further enhances the possibilities for refining target volumes. As advancements continue, the prospect of one-stop-shop segmentation and radiotherapy planning represents an exciting frontier in radiotherapy, potentially enabling fast treatment with enhanced precision and individualization.
Image synthesis is increasingly being adopted in medical image processing, for example for data augmentation or inter-modality image translation. In these critical applications, the generated images must fulfill a high standard of biological correctness. A particular requirement for these images is global consistency, i.e an image being overall coherent and structured so that all parts of the image fit together in a realistic and meaningful way. Yet, established image quality metrics do not explicitly quantify this property of synthetic images. In this work, we introduce two metrics that can measure the global consistency of synthetic images on a per-image basis. To measure the global consistency, we presume that a realistic image exhibits consistent properties, e.g., a person’s body fat in a whole-body MRI, throughout the depicted object or scene. Hence, we quantify global consistency by predicting and comparing explicit attributes of images on patches using supervised trained neural networks. Next, we adapt this strategy to an unlabeled setting by measuring the similarity of implicit image features predicted by a self-supervised trained network. Our results demonstrate that predicting explicit attributes of synthetic images on patches can distinguish globally consistent from inconsistent images. Implicit representations of images are less sensitive to assess global consistency but are still serviceable when labeled data is unavailable. Compared to established metrics, such as the FID, our method can explicitly measure global consistency on a per-image basis, enabling a dedicated analysis of the biological plausibility of single synthetic images.
Do black-box neural network models learn clinically relevant features for fracture diagnosis? The answer not only establishes reliability, quenches scientific curiosity, but also leads to explainable and verbose findings that can assist the radiologists in the final and increase trust. This work identifies the concepts networks use for vertebral fracture diagnosis in CT images. This is achieved by associating concepts to neurons highly correlated with a specific diagnosis in the dataset. The concepts are either associated with neurons by radiologists pre-hoc or are visualized during a specific prediction and left for the user’s interpretation. We evaluate which concepts lead to correct diagnosis and which concepts lead to false positives. The proposed frameworks and analysis pave the way for reliable and explainable vertebral fracture diagnosis.
Ashkan Khakzar
Dr.
©all images: LMU | TUM
2024-12-27 - Last modified: 2024-12-27