Jens Kleesiek

Fabian Hörst, Moritz Rempe, Lukas Heine et al.

Nuclei detection and segmentation in hematoxylin and eosin-stained (H&E) tissue images are important clinical tasks and crucial for a wide range of applications. However, it is a challenging task due to nuclei variances in staining and size, overlapping boundaries, and nuclei clustering. While convolutional neural networks have been extensively used for this task, we explore the potential of Transformer-based networks in this domain. Therefore, we introduce a new method for automated instance segmentation of cell nuclei in digitized tissue samples using a deep learning architecture based on Vision Transformer called CellViT. CellViT is trained and evaluated on the PanNuke dataset, which is one of the most challenging nuclei instance segmentation datasets, consisting of nearly 200,000 annotated Nuclei into 5 clinically important classes in 19 tissue types. We demonstrate the superiority of large-scale in-domain and out-of-domain pre-trained Vision Transformers by leveraging the recently published Segment Anything Model and a ViT-encoder pre-trained on 104 million histological image patches - achieving state-of-the-art nuclei detection and instance segmentation performance on the PanNuke dataset with a mean panoptic quality of 0.50 and an F1-detection score of 0.83. The code is publicly available at https://github.com/TIO-IKIM/CellViT

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.

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

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

Lars Heiliger, Zdravko Marinov, Max Hasin et al.

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

Oliver Ester, Fabian Hörst, Constantin Seibold et al.

The segmentation of histopathological whole slide images into tumourous and non-tumourous types of tissue is a challenging task that requires the consideration of both local and global spatial contexts to classify tumourous regions precisely. The identification of subtypes of tumour tissue complicates the issue as the sharpness of separation decreases and the pathologist's reasoning is even more guided by spatial context. However, the identification of detailed types of tissue is crucial for providing personalized cancer therapies. Due to the high resolution of whole slide images, existing semantic segmentation methods, restricted to isolated image sections, are incapable of processing context information beyond. To take a step towards better context comprehension, we propose a patch neighbour attention mechanism to query the neighbouring tissue context from a patch embedding memory bank and infuse context embeddings into bottleneck hidden feature maps. Our memory attention framework (MAF) mimics a pathologist's annotation procedure -- zooming out and considering surrounding tissue context. The framework can be integrated into any encoder-decoder segmentation method. We evaluate the MAF on a public breast cancer and an internal kidney cancer data set using famous segmentation models (U-Net, DeeplabV3) and demonstrate the superiority over other context-integrating algorithms -- achieving a substantial improvement of up to $17\%$ on Dice score. The code is publicly available at: https://github.com/tio-ikim/valuing-vicinity

Jianning Li, David G. Ellis, Antonio Pepe et al.

Designing implants for large and complex cranial defects is a challenging task, even for professional designers. Current efforts on automating the design process focused mainly on convolutional neural networks (CNN), which have produced state-of-the-art results on reconstructing synthetic defects. However, existing CNN-based methods have been difficult to translate to clinical practice in cranioplasty, as their performance on complex and irregular cranial defects remains unsatisfactory. In this paper, a statistical shape model (SSM) built directly on the segmentation masks of the skulls is presented. We evaluate the SSM on several cranial implant design tasks, and the results show that, while the SSM performs suboptimally on synthetic defects compared to CNN-based approaches, it is capable of reconstructing large and complex defects with only minor manual corrections. The quality of the resulting implants is examined and assured by experienced neurosurgeons. In contrast, CNN-based approaches, even with massive data augmentation, fail or produce less-than-satisfactory implants for these cases. Codes are publicly available at https://github.com/Jianningli/ssm

Amin Dada, Aokun Chen, Cheng Peng et al.

Traditionally, large language models have been either trained on general web crawls or domain-specific data. However, recent successes of generative large language models, have shed light on the benefits of cross-domain datasets. To examine the significance of prioritizing data diversity over quality, we present a German dataset comprising texts from five domains, along with another dataset aimed at containing high-quality data. Through training a series of models ranging between 122M and 750M parameters on both datasets, we conduct a comprehensive benchmark on multiple downstream tasks. Our findings demonstrate that the models trained on the cross-domain dataset outperform those trained on quality data alone, leading to improvements up to $4.45\%$ over the previous state-of-the-art. The models are available at https://huggingface.co/ikim-uk-essen

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.

Jianning Li, Jana Fragemann, Seyed-Ahmad Ahmadi et al.

The reconstruction loss and the Kullback-Leibler divergence (KLD) loss in a variational autoencoder (VAE) often play antagonistic roles, and tuning the weight of the KLD loss in $β$-VAE to achieve a balance between the two losses is a tricky and dataset-specific task. As a result, current practices in VAE training often result in a trade-off between the reconstruction fidelity and the continuity$/$disentanglement of the latent space, if the weight $β$ is not carefully tuned. In this paper, we present intuitions and a careful analysis of the antagonistic mechanism of the two losses, and propose, based on the insights, a simple yet effective two-stage method for training a VAE. Specifically, the method aggregates a learned Gaussian posterior $z \sim q_θ (z|x)$ with a decoder decoupled from the KLD loss, which is trained to learn a new conditional distribution $p_φ (x|z)$ of the input data $x$. Experimentally, we show that the aggregated VAE maximally satisfies the Gaussian assumption about the latent space, while still achieves a reconstruction error comparable to when the latent space is only loosely regularized by $\mathcal{N}(\mathbf{0},I)$. The proposed approach does not require hyperparameter (i.e., the KLD weight $β$) tuning given a specific dataset as required in common VAE training practices. We evaluate the method using a medical dataset intended for 3D skull reconstruction and shape completion, and the results indicate promising generative capabilities of the VAE trained using the proposed method. Besides, through guided manipulation of the latent variables, we establish a connection between existing autoencoder (AE)-based approaches and generative approaches, such as VAE, for the shape completion problem. Codes and pre-trained weights are available at https://github.com/Jianningli/skullVAE

Christina Gsaxner, Jianning Li, Antonio Pepe et al.

The HoloLens (Microsoft Corp., Redmond, WA), a head-worn, optically see-through augmented reality display, is the main player in the recent boost in medical augmented reality research. In medical settings, the HoloLens enables the physician to obtain immediate insight into patient information, directly overlaid with their view of the clinical scenario, the medical student to gain a better understanding of complex anatomies or procedures, and even the patient to execute therapeutic tasks with improved, immersive guidance. In this systematic review, we provide a comprehensive overview of the usage of the first-generation HoloLens within the medical domain, from its release in March 2016, until the year of 2021, were attention is shifting towards it's successor, the HoloLens 2. We identified 171 relevant publications through a systematic search of the PubMed and Scopus databases. We analyze these publications in regard to their intended use case, technical methodology for registration and tracking, data sources, visualization as well as validation and evaluation. We find that, although the feasibility of using the HoloLens in various medical scenarios has been shown, increased efforts in the areas of precision, reliability, usability, workflow and perception are necessary to establish AR in clinical practice.

Maximilian Rokuss, Yannick Kirchhoff, Saikat Roy et al.

Accurate segmentation of Multiple Sclerosis (MS) lesions in longitudinal MRI scans is crucial for monitoring disease progression and treatment efficacy. Although changes across time are taken into account when assessing images in clinical practice, most existing deep learning methods treat scans from different timepoints separately. Among studies utilizing longitudinal images, a simple channel-wise concatenation is the primary albeit suboptimal method employed to integrate timepoints. We introduce a novel approach that explicitly incorporates temporal differences between baseline and follow-up scans through a unique architectural inductive bias called Difference Weighting Block. It merges features from two timepoints, emphasizing changes between scans. We achieve superior scores in lesion segmentation (Dice Score, Hausdorff distance) as well as lesion detection (lesion-level $F_1$ score) as compared to state-of-the-art longitudinal and single timepoint models across two datasets. Our code is made publicly available at www.github.com/MIC-DKFZ/Longitudinal-Difference-Weighting.

Jianning Li, André Ferreira, Behrus Puladi et al.

We present a deep learning-based approach for skull reconstruction for MONAI, which has been pre-trained on the MUG500+ skull dataset. The implementation follows the MONAI contribution guidelines, hence, it can be easily tried out and used, and extended by MONAI users. The primary goal of this paper lies in the investigation of open-sourcing codes and pre-trained deep learning models under the MONAI framework. Nowadays, open-sourcing software, especially (pre-trained) deep learning models, has become increasingly important. Over the years, medical image analysis experienced a tremendous transformation. Over a decade ago, algorithms had to be implemented and optimized with low-level programming languages, like C or C++, to run in a reasonable time on a desktop PC, which was not as powerful as today's computers. Nowadays, users have high-level scripting languages like Python, and frameworks like PyTorch and TensorFlow, along with a sea of public code repositories at hand. As a result, implementations that had thousands of lines of C or C++ code in the past, can now be scripted with a few lines and in addition executed in a fraction of the time. To put this even on a higher level, the Medical Open Network for Artificial Intelligence (MONAI) framework tailors medical imaging research to an even more convenient process, which can boost and push the whole field. The MONAI framework is a freely available, community-supported, open-source and PyTorch-based framework, that also enables to provide research contributions with pre-trained models to others. Codes and pre-trained weights for skull reconstruction are publicly available at: https://github.com/Project-MONAI/research-contributions/tree/master/SkullRec

Jianning Li, Antonio Pepe, Gijs Luijten et al.

In this paper, we introduce a completion framework to reconstruct the geometric shapes of various anatomies, including organs, vessels and muscles. Our work targets a scenario where one or multiple anatomies are missing in the imaging data due to surgical, pathological or traumatic factors, or simply because these anatomies are not covered by image acquisition. Automatic reconstruction of the missing anatomies benefits many applications, such as organ 3D bio-printing, whole-body segmentation, animation realism, paleoradiology and forensic imaging. We propose two paradigms based on a 3D denoising auto-encoder (DAE) to solve the anatomy reconstruction problem: (i) the DAE learns a many-to-one mapping between incomplete and complete instances; (ii) the DAE learns directly a one-to-one residual mapping between the incomplete instances and the target anatomies. We apply a loss aggregation scheme that enables the DAE to learn the many-to-one mapping more effectively and further enhances the learning of the residual mapping. On top of this, we extend the DAE to a multiclass completor by assigning a unique label to each anatomy involved. We evaluate our method using a CT dataset with whole-body segmentations. Results show that our method produces reasonable anatomy reconstructions given instances with different levels of incompleteness (i.e., one or multiple random anatomies are missing). Codes and pretrained models are publicly available at https://github.com/Jianningli/medshapenet-feedback/ tree/main/anatomy-completor

Linara Adilova, Amr Abourayya, Jianning Li et al.

Flatness of the loss curve around a model at hand has been shown to empirically correlate with its generalization ability. Optimizing for flatness has been proposed as early as 1994 by Hochreiter and Schmidthuber, and was followed by more recent successful sharpness-aware optimization techniques. Their widespread adoption in practice, though, is dubious because of the lack of theoretically grounded connection between flatness and generalization, in particular in light of the reparameterization curse - certain reparameterizations of a neural network change most flatness measures but do not change generalization. Recent theoretical work suggests that a particular relative flatness measure can be connected to generalization and solves the reparameterization curse. In this paper, we derive a regularizer based on this relative flatness that is easy to compute, fast, efficient, and works with arbitrary loss functions. It requires computing the Hessian only of a single layer of the network, which makes it applicable to large neural networks, and with it avoids an expensive mapping of the loss surface in the vicinity of the model. In an extensive empirical evaluation we show that this relative flatness aware minimization (FAM) improves generalization in a multitude of applications and models, both in finetuning and standard training. We make the code available at github.

Constantin Seibold, Simon Reiß, M. Saquib Sarfraz et al.

When reading images, radiologists generate text reports describing the findings therein. Current state-of-the-art computer-aided diagnosis tools utilize a fixed set of predefined categories automatically extracted from these medical reports for training. This form of supervision limits the potential usage of models as they are unable to pick up on anomalies outside of their predefined set, thus, making it a necessity to retrain the classifier with additional data when faced with novel classes. In contrast, we investigate direct text supervision to break away from this closed set assumption. By doing so, we avoid noisy label extraction via text classifiers and incorporate more contextual information. We employ a contrastive global-local dual-encoder architecture to learn concepts directly from unstructured medical reports while maintaining its ability to perform free form classification. We investigate relevant properties of open set recognition for radiological data and propose a method to employ currently weakly annotated data into training. We evaluate our approach on the large-scale chest X-Ray datasets MIMIC-CXR, CheXpert, and ChestX-Ray14 for disease classification. We show that despite using unstructured medical report supervision, we perform on par with direct label supervision through a sophisticated inference setting.

Alexander Jaus, Constantin Seibold, Simon Reiß et al.

Pathological structures in medical images are typically deviations from the expected anatomy of a patient. While clinicians consider this interplay between anatomy and pathology, recent deep learning algorithms specialize in recognizing either one of the two, rarely considering the patient's body from such a joint perspective. In this paper, we develop a generalist segmentation model that combines anatomical and pathological information, aiming to enhance the segmentation accuracy of pathological features. Our Anatomy-Pathology Exchange (APEx) training utilizes a query-based segmentation transformer which decodes a joint feature space into query-representations for human anatomy and interleaves them via a mixing strategy into the pathology-decoder for anatomy-informed pathology predictions. In doing so, we are able to report the best results across the board on FDG-PET-CT and Chest X-Ray pathology segmentation tasks with a margin of up to 3.3% as compared to strong baseline methods. Code and models will be publicly available at github.com/alexanderjaus/APEx.

Alexander Jaus, Constantin Seibold, Kelsey Hermann et al.

In this study, we present a method for generating automated anatomy segmentation datasets using a sequential process that involves nnU-Net-based pseudo-labeling and anatomy-guided pseudo-label refinement. By combining various fragmented knowledge bases, we generate a dataset of whole-body CT scans with $142$ voxel-level labels for 533 volumes providing comprehensive anatomical coverage which experts have approved. Our proposed procedure does not rely on manual annotation during the label aggregation stage. We examine its plausibility and usefulness using three complementary checks: Human expert evaluation which approved the dataset, a Deep Learning usefulness benchmark on the BTCV dataset in which we achieve 85% dice score without using its training dataset, and medical validity checks. This evaluation procedure combines scalable automated checks with labor-intensive high-quality expert checks. Besides the dataset, we release our trained unified anatomical segmentation model capable of predicting $142$ anatomical structures on CT data.

Constantin Seibold, Simon Reiß, Saquib Sarfraz et al.

In clinical radiology reports, doctors capture important information about the patient's health status. They convey their observations from raw medical imaging data about the inner structures of a patient. As such, formulating reports requires medical experts to possess wide-ranging knowledge about anatomical regions with their normal, healthy appearance as well as the ability to recognize abnormalities. This explicit grasp on both the patient's anatomy and their appearance is missing in current medical image-processing systems as annotations are especially difficult to gather. This renders the models to be narrow experts e.g. for identifying specific diseases. In this work, we recover this missing link by adding human anatomy into the mix and enable the association of content in medical reports to their occurrence in associated imagery (medical phrase grounding). To exploit anatomical structures in this scenario, we present a sophisticated automatic pipeline to gather and integrate human bodily structures from computed tomography datasets, which we incorporate in our PAXRay: A Projected dataset for the segmentation of Anatomical structures in X-Ray data. Our evaluation shows that methods that take advantage of anatomical information benefit heavily in visually grounding radiologists' findings, as our anatomical segmentations allow for up to absolute 50% better grounding results on the OpenI dataset as compared to commonly used region proposals. The PAXRay dataset is available at https://constantinseibold.github.io/paxray/.

Zdravko Marinov, Simon Reiß, David Kersting et al.

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

Hamza Kalisch, Fabian Hörst, Ken Herrmann et al.

Lesion segmentation in PET/CT imaging is essential for precise tumor characterization, which supports personalized treatment planning and enhances diagnostic precision in oncology. However, accurate manual segmentation of lesions is time-consuming and prone to inter-observer variability. Given the rising demand and clinical use of PET/CT, automated segmentation methods, particularly deep-learning-based approaches, have become increasingly more relevant. The autoPET III Challenge focuses on advancing automated segmentation of tumor lesions in PET/CT images in a multitracer multicenter setting, addressing the clinical need for quantitative, robust, and generalizable solutions. Building on previous challenges, the third iteration of the autoPET challenge introduces a more diverse dataset featuring two different tracers (FDG and PSMA) from two clinical centers. To this extent, we developed a classifier that identifies the tracer of the given PET/CT based on the Maximum Intensity Projection of the PET scan. We trained two individual nnUNet-ensembles for each tracer where anatomical labels are included as a multi-label task to enhance the model's performance. Our final submission achieves cross-validation Dice scores of 76.90% and 61.33% for the publicly available FDG and PSMA datasets, respectively. The code is available at https://github.com/hakal104/autoPETIII/ .

André Ferreira, Jianning Li, Kelsey L. Pomykala et al.

With the massive proliferation of data-driven algorithms, such as deep learning-based approaches, the availability of high-quality data is of great interest. Volumetric data is very important in medicine, as it ranges from disease diagnoses to therapy monitoring. When the dataset is sufficient, models can be trained to help doctors with these tasks. Unfortunately, there are scenarios where large amounts of data is unavailable. For example, rare diseases and privacy issues can lead to restricted data availability. In non-medical fields, the high cost of obtaining enough high-quality data can also be a concern. A solution to these problems can be the generation of realistic synthetic data using Generative Adversarial Networks (GANs). The existence of these mechanisms is a good asset, especially in healthcare, as the data must be of good quality, realistic, and without privacy issues. Therefore, most of the publications on volumetric GANs are within the medical domain. In this review, we provide a summary of works that generate realistic volumetric synthetic data using GANs. We therefore outline GAN-based methods in these areas with common architectures, loss functions and evaluation metrics, including their advantages and disadvantages. We present a novel taxonomy, evaluations, challenges, and research opportunities to provide a holistic overview of the current state of volumetric GANs.

Felix J. Dorfner, Amin Dada, Felix Busch et al.

Large language models (LLMs) have shown potential in biomedical applications, leading to efforts to fine-tune them on domain-specific data. However, the effectiveness of this approach remains unclear. This study evaluates the performance of biomedically fine-tuned LLMs against their general-purpose counterparts on a variety of clinical tasks. We evaluated their performance on clinical case challenges from the New England Journal of Medicine (NEJM) and the Journal of the American Medical Association (JAMA) and on several clinical tasks (e.g., information extraction, document summarization, and clinical coding). Using benchmarks specifically chosen to be likely outside the fine-tuning datasets of biomedical models, we found that biomedical LLMs mostly perform inferior to their general-purpose counterparts, especially on tasks not focused on medical knowledge. While larger models showed similar performance on case tasks (e.g., OpenBioLLM-70B: 66.4% vs. Llama-3-70B-Instruct: 65% on JAMA cases), smaller biomedical models showed more pronounced underperformance (e.g., OpenBioLLM-8B: 30% vs. Llama-3-8B-Instruct: 64.3% on NEJM cases). Similar trends were observed across the CLUE (Clinical Language Understanding Evaluation) benchmark tasks, with general-purpose models often performing better on text generation, question answering, and coding tasks. Our results suggest that fine-tuning LLMs to biomedical data may not provide the expected benefits and may potentially lead to reduced performance, challenging prevailing assumptions about domain-specific adaptation of LLMs and highlighting the need for more rigorous evaluation frameworks in healthcare AI. Alternative approaches, such as retrieval-augmented generation, may be more effective in enhancing the biomedical capabilities of LLMs without compromising their general knowledge.

Christian Strack, Kelsey L. Pomykala, Heinz-Peter Schlemmer et al.

With the rise in importance of personalized medicine, we trained personalized neural networks to detect tumor progression in longitudinal datasets. The model was evaluated on two datasets with a total of 64 scans from 32 patients diagnosed with glioblastoma multiforme (GBM). Contrast-enhanced T1w sequences of brain magnetic resonance imaging (MRI) images were used in this study. For each patient, we trained their own neural network using just two images from different timepoints. Our approach uses a Wasserstein-GAN (generative adversarial network), an unsupervised network architecture, to map the differences between the two images. Using this map, the change in tumor volume can be evaluated. Due to the combination of data augmentation and the network architecture, co-registration of the two images is not needed. Furthermore, we do not rely on any additional training data, (manual) annotations or pre-training neural networks. The model received an AUC-score of 0.87 for tumor change. We also introduced a modified RANO criteria, for which an accuracy of 66% can be achieved. We show that using data from just one patient can be used to train deep neural networks to monitor tumor change.

Zdravko Marinov, Rainer Stiefelhagen, Jens Kleesiek

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

Tianyu Han, Sven Nebelung, Firas Khader et al.

Large language models (LLMs) have broad medical knowledge and can reason about medical information across many domains, holding promising potential for diverse medical applications in the near future. In this study, we demonstrate a concerning vulnerability of LLMs in medicine. Through targeted manipulation of just 1.1% of the model's weights, we can deliberately inject an incorrect biomedical fact. The erroneous information is then propagated in the model's output, whilst its performance on other biomedical tasks remains intact. We validate our findings in a set of 1,038 incorrect biomedical facts. This peculiar susceptibility raises serious security and trustworthiness concerns for the application of LLMs in healthcare settings. It accentuates the need for robust protective measures, thorough verification mechanisms, and stringent management of access to these models, ensuring their reliable and safe use in medical practice.

Matthias Hadlich, Zdravko Marinov, Moon Kim et al.

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

Jana Fragemann, Lynton Ardizzone, Jan Egger et al.

Deep neural networks are commonly used for medical purposes such as image generation, segmentation, or classification. Besides this, they are often criticized as black boxes as their decision process is often not human interpretable. Encouraging the latent representation of a generative model to be disentangled offers new perspectives of control and interpretability. Understanding the data generation process could help to create artificial medical data sets without violating patient privacy, synthesizing different data modalities, or discovering data generating characteristics. These characteristics might unravel novel relationships that can be related to genetic traits or patient outcomes. In this paper, we give a comprehensive overview of popular generative models, like Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Flow-based Models. Furthermore, we summarize the different notions of disentanglement, review approaches to disentangle latent space representations and metrics to evaluate the degree of disentanglement. After introducing the theoretical frameworks, we give an overview of recent medical applications and discuss the impact and importance of disentanglement approaches for medical applications.

Moritz Rempe, Florian Mentzel, Kelsey L. Pomykala et al.

Objectives: Present a novel deep learning-based skull stripping algorithm for magnetic resonance imaging (MRI) that works directly in the information rich k-space. Materials and Methods: Using two datasets from different institutions with a total of 36,900 MRI slices, we trained a deep learning-based model to work directly with the complex raw k-space data. Skull stripping performed by HD-BET (Brain Extraction Tool) in the image domain were used as the ground truth. Results: Both datasets were very similar to the ground truth (DICE scores of 92\%-98\% and Hausdorff distances of under 5.5 mm). Results on slices above the eye-region reach DICE scores of up to 99\%, while the accuracy drops in regions around the eyes and below, with partially blurred output. The output of k-strip often smoothed edges at the demarcation to the skull. Binary masks are created with an appropriate threshold. Conclusion: With this proof-of-concept study, we were able to show the feasibility of working in the k-space frequency domain, preserving phase information, with consistent results. Future research should be dedicated to discovering additional ways the k-space can be used for innovative image analysis and further workflows.

Jan Egger, Christina Gsaxner, Xiaojun Chen et al.

At the Worldwide Developers Conference (WWDC) in June 2023, Apple introduced the Vision Pro. The Vision Pro is a Mixed Reality (MR) headset, more specifically it is a Virtual Reality (VR) device with an additional Video See-Through (VST) capability. The VST capability turns the Vision Pro also into an Augmented Reality (AR) device. The AR feature is enabled by streaming the real world via cameras to the (VR) screens in front of the user's eyes. This is of course not unique and similar to other devices, like the Varjo XR-3. Nevertheless, the Vision Pro has some interesting features, like an inside-out screen that can show the headset wearers' eyes to "outsiders" or a button on the top, called "Digital Crown", that allows you to seamlessly blend digital content with your physical space by turning it. In addition, it is untethered, except for the cable to the battery, which makes the headset more agile, compared to the Varjo XR-3. This could actually come closer to the "Ultimate Display", which Ivan Sutherland had already sketched in 1965. Not available to the public yet, like the Ultimate Display, we want to take a look into the crystal ball in this perspective to see if it can overcome some clinical challenges that - especially - AR still faces in the medical domain, but also go beyond and discuss if the Vision Pro could support clinicians in essential tasks to spend more time with their patients.

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

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

Lukas Heine, Fabian Hörst, Jana Fragemann et al.

In industries such as healthcare, finance, and manufacturing, analysis of unstructured textual data presents significant challenges for analysis and decision making. Uncovering patterns within large-scale corpora and understanding their semantic impact is critical, but depends on domain experts or resource-intensive manual reviews. In response, we introduce Spacewalker in this system demonstration paper, an interactive tool designed to analyze, explore, and annotate data across multiple modalities. It allows users to extract data representations, visualize them in low-dimensional spaces and traverse large datasets either exploratory or by querying regions of interest. We evaluated Spacewalker through extensive experiments and annotation studies, assessing its efficacy in improving data integrity verification and annotation. We show that Spacewalker reduces time and effort compared to traditional methods. The code of this work is open-source and can be found at: https://github.com/code-lukas/Spacewalker

Frederic Jonske, Moon Kim, Enrico Nasca et al.

It is an open secret that ImageNet is treated as the panacea of pretraining. Particularly in medical machine learning, models not trained from scratch are often finetuned based on ImageNet-pretrained models. We posit that pretraining on data from the domain of the downstream task should almost always be preferred instead. We leverage RadNet-12M, a dataset containing more than 12 million computed tomography (CT) image slices, to explore the efficacy of self-supervised pretraining on medical and natural images. Our experiments cover intra- and cross-domain transfer scenarios, varying data scales, finetuning vs. linear evaluation, and feature space analysis. We observe that intra-domain transfer compares favorably to cross-domain transfer, achieving comparable or improved performance (0.44% - 2.07% performance increase using RadNet pretraining, depending on the experiment) and demonstrate the existence of a domain boundary-related generalization gap and domain-specific learned features.

Amr Abourayya, Jens Kleesiek, Kanishka Rao et al.

In many critical applications, sensitive data is inherently distributed and cannot be centralized due to privacy concerns. A wide range of federated learning approaches have been proposed to train models locally at each client without sharing their sensitive data, typically by exchanging model parameters, or probabilistic predictions (soft labels) on a public dataset or a combination of both. However, these methods still disclose private information and restrict local models to those that can be trained using gradient-based methods. We propose a federated co-training (FedCT) approach that improves privacy by sharing only definitive (hard) labels on a public unlabeled dataset. Clients use a consensus of these shared labels as pseudo-labels for local training. This federated co-training approach empirically enhances privacy without compromising model quality. In addition, it allows the use of local models that are not suitable for parameter aggregation in traditional federated learning, such as gradient-boosted decision trees, rule ensembles, and random forests. Furthermore, we observe that FedCT performs effectively in federated fine-tuning of large language models, where its pseudo-labeling mechanism is particularly beneficial. Empirical evaluations and theoretical analyses suggest its applicability across a range of federated learning scenarios.

Jonas Nasimzada, Jens Kleesiek, Ken Herrmann et al.

Recognizing pain in video is crucial for improving patient-computer interaction systems, yet traditional data collection in this domain raises significant ethical and logistical challenges. This study introduces a novel approach that leverages synthetic data to enhance video-based pain recognition models, providing an ethical and scalable alternative. We present a pipeline that synthesizes realistic 3D facial models by capturing nuanced facial movements from a small participant pool, and mapping these onto diverse synthetic avatars. This process generates 8,600 synthetic faces, accurately reflecting genuine pain expressions from varied angles and perspectives. Utilizing advanced facial capture techniques, and leveraging public datasets like CelebV-HQ and FFHQ-UV for demographic diversity, our new synthetic dataset significantly enhances model training while ensuring privacy by anonymizing identities through facial replacements. Experimental results demonstrate that models trained on combinations of synthetic data paired with a small amount of real participants achieve superior performance in pain recognition, effectively bridging the gap between synthetic simulations and real-world applications. Our approach addresses data scarcity and ethical concerns, offering a new solution for pain detection and opening new avenues for research in privacy-preserving dataset generation. All resources are publicly available to encourage further innovation in this field.

Fabian Hörst, Moritz Rempe, Helmut Becker et al.

Digital Pathology is a cornerstone in the diagnosis and treatment of diseases. A key task in this field is the identification and segmentation of cells in hematoxylin and eosin-stained images. Existing methods for cell segmentation often require extensive annotated datasets for training and are limited to a predefined cell classification scheme. To overcome these limitations, we propose $\text{CellViT}^{\scriptscriptstyle ++}$, a framework for generalized cell segmentation in digital pathology. $\text{CellViT}^{\scriptscriptstyle ++}$ utilizes Vision Transformers with foundation models as encoders to compute deep cell features and segmentation masks simultaneously. To adapt to unseen cell types, we rely on a computationally efficient approach. It requires minimal data for training and leads to a drastically reduced carbon footprint. We demonstrate excellent performance on seven different datasets, covering a broad spectrum of cell types, organs, and clinical settings. The framework achieves remarkable zero-shot segmentation and data-efficient cell-type classification. Furthermore, we show that $\text{CellViT}^{\scriptscriptstyle ++}$ can leverage immunofluorescence stainings to generate training datasets without the need for pathologist annotations. The automated dataset generation approach surpasses the performance of networks trained on manually labeled data, demonstrating its effectiveness in creating high-quality training datasets without expert annotations. To advance digital pathology, $\text{CellViT}^{\scriptscriptstyle ++}$ is available as an open-source framework featuring a user-friendly, web-based interface for visualization and annotation. The code is available under https://github.com/TIO-IKIM/CellViT-plus-plus.

Amin Dada, Marie Bauer, Amanda Butler Contreras et al.

Large Language Models (LLMs) are expected to significantly contribute to patient care, diagnostics, and administrative processes. Emerging biomedical LLMs aim to address healthcare-specific challenges, including privacy demands and computational constraints. Assessing the models' suitability for this sensitive application area is of the utmost importance. However, biomedical training has not been systematically evaluated on medical tasks. This study investigates the effect of biomedical training in the context of six practical medical tasks evaluating $25$ models. In contrast to previous evaluations, our results reveal a performance decline in nine out of twelve biomedical models after fine-tuning, particularly on tasks involving hallucinations, ICD10 coding, and instruction adherence. General-domain models like Meta-Llama-3.1-70B-Instruct outperformed their biomedical counterparts, indicating a trade-off between domain-specific fine-tuning and general medical task performance. We open-source all evaluation scripts and datasets at https://github.com/TIO-IKIM/CLUE to support further research in this critical area.

Moritz Rempe, Lukas Heine, Constantin Seibold et al.

Medical data employed in research frequently comprises sensitive patient health information (PHI), which is subject to rigorous legal frameworks such as the General Data Protection Regulation (GDPR) or the Health Insurance Portability and Accountability Act (HIPAA). Consequently, these types of data must be pseudonymized prior to utilisation, which presents a significant challenge for many researchers. Given the vast array of medical data, it is necessary to employ a variety of de-identification techniques. To facilitate the anonymization process for medical imaging data, we have developed an open-source tool that can be used to de-identify DICOM magnetic resonance images, computer tomography images, whole slide images and magnetic resonance twix raw data. Furthermore, the implementation of a neural network enables the removal of text within the images. The proposed tool automates an elaborate anonymization pipeline for multiple types of inputs, reducing the need for additional tools used for de-identification of imaging data. We make our code publicly available at https://github.com/code-lukas/medical_image_deidentification.

André Ferreira, Tiago Jesus, Behrus Puladi et al.

This paper presents the winning solution of task 1 and the third-placed solution of task 3 of the BraTS challenge. The use of automated tools in clinical practice has increased due to the development of more and more sophisticated and reliable algorithms. However, achieving clinical standards and developing tools for real-life scenarios is a major challenge. To this end, BraTS has organised tasks to find the most advanced solutions for specific purposes. In this paper, we propose the use of synthetic data to train state-of-the-art frameworks in order to improve the segmentation of adult gliomas in a post-treatment scenario, and the segmentation of meningioma for radiotherapy planning. Our results suggest that the use of synthetic data leads to more robust algorithms, although the synthetic data generation pipeline is not directly suited to the meningioma task. In task 1, we achieved a DSC of 0.7900, 0.8076, 0.7760, 0.8926, 0.7874, 0.8938 and a HD95 of 35.63, 30.35, 44.58, 16.87, 38.19, 17.95 for ET, NETC, RC, SNFH, TC and WT, respectively and, in task 3, we achieved a DSC of 0.801 and HD95 of 38.26, in the testing phase. The code for these tasks is available at https://github.com/ShadowTwin41/BraTS_2023_2024_solutions.

André Ferreira, Gijs Luijten, Behrus Puladi et al.

This paper presents the second-placed solution for task 8 and the participation solution for task 7 of BraTS 2024. The adoption of automated brain analysis algorithms to support clinical practice is increasing. However, many of these algorithms struggle with the presence of brain lesions or the absence of certain MRI modalities. The alterations in the brain's morphology leads to high variability and thus poor performance of predictive models that were trained only on healthy brains. The lack of information that is usually provided by some of the missing MRI modalities also reduces the reliability of the prediction models trained with all modalities. In order to improve the performance of these models, we propose the use of conditional 3D wavelet diffusion models. The wavelet transform enabled full-resolution image training and prediction on a GPU with 48 GB VRAM, without patching or downsampling, preserving all information for prediction. The code for these tasks is available at https://github.com/ShadowTwin41/BraTS_2023_2024_solutions.

Gijs Luijten, Roberto Maria Scardigno, Lisle Faray de Paiva et al.

Ultrasound (US) is widely accessible and radiation-free but has a steep learning curve due to its dynamic nature and non-standard imaging planes. Additionally, the constant need to shift focus between the US screen and the patient poses a challenge. To address these issues, we integrate deep learning (DL)-based semantic segmentation for real-time (RT) automated kidney volumetric measurements, which are essential for clinical assessment but are traditionally time-consuming and prone to fatigue. This automation allows clinicians to concentrate on image interpretation rather than manual measurements. Complementing DL, augmented reality (AR) enhances the usability of US by projecting the display directly into the clinician's field of view, improving ergonomics and reducing the cognitive load associated with screen-to-patient transitions. Two AR-DL-assisted US pipelines on HoloLens-2 are proposed: one streams directly via the application programming interface for a wireless setup, while the other supports any US device with video output for broader accessibility. We evaluate RT feasibility and accuracy using the Open Kidney Dataset and open-source segmentation models (nnU-Net, Segmenter, YOLO with MedSAM and LiteMedSAM). Our open-source GitHub pipeline includes model implementations, measurement algorithms, and a Wi-Fi-based streaming solution, enhancing US training and diagnostics, especially in point-of-care settings.

Michael Lackner, Behrus Puladi, Jens Kleesiek et al.

In traumatic medical emergencies, the patients heavily depend on cranioplasty - the craft of neurocranial repair using cranial implants. Despite the improvements made in recent years, the design of a patient-specific implant (PSI) is among the most complex, expensive, and least automated tasks in cranioplasty. Further research in this area is needed. Therefore, we created a prototype application with a graphical user interface (UI) specifically tailored for semi-automatic implant generation, where the users only need to perform high-level actions. A general outline of the proposed implant generation process involves setting an area of interest, aligning the templates, and then creating the implant in voxel space. Furthermore, we show that the alignment can be improved significantly, by only considering clipped geometry in the vicinity of the defect border. The software prototype will be open-sourced at https://github.com/3Descape/Cranial_Implant_Design

Michael Götz, Christian Weber, Franciszek Binczyk et al.

We propose a new method that employs transfer learning techniques to effectively correct sampling selection errors introduced by sparse annotations during supervised learning for automated tumor segmentation. The practicality of current learning-based automated tissue classification approaches is severely impeded by their dependency on manually segmented training databases that need to be recreated for each scenario of application, site, or acquisition setup. The comprehensive annotation of reference datasets can be highly labor-intensive, complex, and error-prone. The proposed method derives high-quality classifiers for the different tissue classes from sparse and unambiguous annotations and employs domain adaptation techniques for effectively correcting sampling selection errors introduced by the sparse sampling. The new approach is validated on labeled, multi-modal MR images of 19 patients with malignant gliomas and by comparative analysis on the BraTS 2013 challenge data sets. Compared to training on fully labeled data, we reduced the time for labeling and training by a factor greater than 70 and 180 respectively without sacrificing accuracy. This dramatically eases the establishment and constant extension of large annotated databases in various scenarios and imaging setups and thus represents an important step towards practical applicability of learning-based approaches in tissue classification.

Yannick Kirchhoff, Maximilian R. Rokuss, Saikat Roy et al.

Accurately segmenting thin tubular structures, such as vessels, nerves, roads or concrete cracks, is a crucial task in computer vision. Standard deep learning-based segmentation loss functions, such as Dice or Cross-Entropy, focus on volumetric overlap, often at the expense of preserving structural connectivity or topology. This can lead to segmentation errors that adversely affect downstream tasks, including flow calculation, navigation, and structural inspection. Although current topology-focused losses mark an improvement, they introduce significant computational and memory overheads. This is particularly relevant for 3D data, rendering these losses infeasible for larger volumes as well as increasingly important multi-class segmentation problems. To mitigate this, we propose a novel Skeleton Recall Loss, which effectively addresses these challenges by circumventing intensive GPU-based calculations with inexpensive CPU operations. It demonstrates overall superior performance to current state-of-the-art approaches on five public datasets for topology-preserving segmentation, while substantially reducing computational overheads by more than 90%. In doing so, we introduce the first multi-class capable loss function for thin structure segmentation, excelling in both efficiency and efficacy for topology-preservation.

Markus R. Bujotzek, Ünal Akünal, Stefan Denner et al.

Objective: Federated Learning (FL) enables collaborative model training while keeping data locally. Currently, most FL studies in radiology are conducted in simulated environments due to numerous hurdles impeding its translation into practice. The few existing real-world FL initiatives rarely communicate specific measures taken to overcome these hurdles, leaving behind a significant knowledge gap. Minding efforts to implement real-world FL, there is a notable lack of comprehensive assessment comparing FL to less complex alternatives. Materials & Methods: We extensively reviewed FL literature, categorizing insights along with our findings according to their nature and phase while establishing a FL initiative, summarized to a comprehensive guide. We developed our own FL infrastructure within the German Radiological Cooperative Network (RACOON) and demonstrated its functionality by training FL models on lung pathology segmentation tasks across six university hospitals. We extensively evaluated FL against less complex alternatives in three distinct evaluation scenarios. Results: The proposed guide outlines essential steps, identified hurdles, and proposed solutions for establishing successful FL initiatives conducting real-world experiments. Our experimental results show that FL outperforms less complex alternatives in all evaluation scenarios, justifying the effort required to translate FL into real-world applications. Discussion & Conclusion: Our proposed guide aims to aid future FL researchers in circumventing pitfalls and accelerating translation of FL into radiological applications. Our results underscore the value of efforts needed to translate FL into real-world applications by demonstrating advantageous performance over alternatives, and emphasize the importance of strategic organization, robust management of distributed data and infrastructure in real-world settings.

Amin Dada, Osman Alperen Koras, Marie Bauer et al.

While increasing patients' access to medical documents improves medical care, this benefit is limited by varying health literacy levels and complex medical terminology. Large language models (LLMs) offer solutions by simplifying medical information. However, evaluating LLMs for safe and patient-friendly text generation is difficult due to the lack of standardized evaluation resources. To fill this gap, we developed MeDiSumQA. MeDiSumQA is a dataset created from MIMIC-IV discharge summaries through an automated pipeline combining LLM-based question-answer generation with manual quality checks. We use this dataset to evaluate various LLMs on patient-oriented question-answering. Our findings reveal that general-purpose LLMs frequently surpass biomedical-adapted models, while automated metrics correlate with human judgment. By releasing MeDiSumQA on PhysioNet, we aim to advance the development of LLMs to enhance patient understanding and ultimately improve care outcomes.

Alexander Jaus, Constantin Seibold, Simon Reiß et al.

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

Ahmad Idrissi-Yaghir, Amin Dada, Henning Schäfer et al.

Recent advances in natural language processing (NLP) can be largely attributed to the advent of pre-trained language models such as BERT and RoBERTa. While these models demonstrate remarkable performance on general datasets, they can struggle in specialized domains such as medicine, where unique domain-specific terminologies, domain-specific abbreviations, and varying document structures are common. This paper explores strategies for adapting these models to domain-specific requirements, primarily through continuous pre-training on domain-specific data. We pre-trained several German medical language models on 2.4B tokens derived from translated public English medical data and 3B tokens of German clinical data. The resulting models were evaluated on various German downstream tasks, including named entity recognition (NER), multi-label classification, and extractive question answering. Our results suggest that models augmented by clinical and translation-based pre-training typically outperform general domain models in medical contexts. We conclude that continuous pre-training has demonstrated the ability to match or even exceed the performance of clinical models trained from scratch. Furthermore, pre-training on clinical data or leveraging translated texts have proven to be reliable methods for domain adaptation in medical NLP tasks.

Ting Han, Linara Adilova, Henning Petzka et al.

Neural collapse, i.e., the emergence of highly symmetric, class-wise clustered representations, is frequently observed in deep networks and is often assumed to reflect or enable generalization. In parallel, flatness of the loss landscape has been theoretically and empirically linked to generalization. Yet, the causal role of either phenomenon remains unclear: Are they prerequisites for generalization, or merely by-products of training dynamics? We disentangle these questions using grokking, a training regime in which memorization precedes generalization, allowing us to temporally separate generalization from training dynamics and we find that while both neural collapse and relative flatness emerge near the onset of generalization, only flatness consistently predicts it. Models encouraged to collapse or prevented from collapsing generalize equally well, whereas models regularized away from flat solutions exhibit delayed generalization, resembling grokking, even in architectures and datasets where it does not typically occur. Furthermore, we show theoretically that neural collapse leads to relative flatness under classical assumptions, explaining their empirical co-occurrence. Our results support the view that relative flatness is a potentially necessary and more fundamental property for generalization, and demonstrate how grokking can serve as a powerful probe for isolating its geometric underpinnings.

Maximilian Fischer, Peter Neher, Peter Schüffler et al.

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

Nikoo Moradi, André Ferreira, Behrus Puladi et al.

Radiation therapy (RT) is essential in treating head and neck cancer (HNC), with magnetic resonance imaging(MRI)-guided RT offering superior soft tissue contrast and functional imaging. However, manual tumor segmentation is time-consuming and complex, and therfore remains a challenge. In this study, we present our solution as team TUMOR to the HNTS-MRG24 MICCAI Challenge which is focused on automated segmentation of primary gross tumor volumes (GTVp) and metastatic lymph node gross tumor volume (GTVn) in pre-RT and mid-RT MRI images. We utilized the HNTS-MRG2024 dataset, which consists of 150 MRI scans from patients diagnosed with HNC, including original and registered pre-RT and mid-RT T2-weighted images with corresponding segmentation masks for GTVp and GTVn. We employed two state-of-the-art models in deep learning, nnUNet and MedNeXt. For Task 1, we pretrained models on pre-RT registered and mid-RT images, followed by fine-tuning on original pre-RT images. For Task 2, we combined registered pre-RT images, registered pre-RT segmentation masks, and mid-RT data as a multi-channel input for training. Our solution for Task 1 achieved 1st place in the final test phase with an aggregated Dice Similarity Coefficient of 0.8254, and our solution for Task 2 ranked 8th with a score of 0.7005. The proposed solution is publicly available at Github Repository.