Search

Vision-Language Models (VLMs) have great potential in medical tasks, like Visual Question Answering (VQA), where they could act as interactive assistants for both patients and clinicians. Yet their robustness to distribution shifts on unseen data remains a critical concern for safe deployment. Evaluating such robustness requires a controlled experimental setup that allows for systematic insights into the model's behavior. However, we demonstrate that current setups fail to offer sufficiently thorough evaluations, limiting their ability to accurately assess model robustness. To address this gap, our work introduces a novel framework, called SURE-VQA, centered around three key requirements to overcome the current pitfalls and systematically analyze the robustness of VLMs: 1) Since robustness on synthetic shifts does not necessarily translate to real-world shifts, robustness should be measured on real-world shifts that are inherent to the VQA data; 2) Traditional token-matching metrics often fail to capture underlying semantics, necessitating the use of large language models (LLMs) for more accurate semantic evaluation; 3) Model performance often lacks interpretability due to missing sanity baselines, thus meaningful baselines should be reported that allow assessing the multimodal impact on the VLM. To demonstrate the relevance of this framework, we conduct a study on the robustness of various fine-tuning methods across three medical datasets with four different types of distribution shifts. Our study reveals several important findings: 1) Sanity baselines that do not utilize image data can perform surprisingly well; 2) We confirm LoRA as the best-performing PEFT method; 3) No PEFT method consistently outperforms others in terms of robustness to shifts. Code is provided at https://github.com/IML-DKFZ/sure-vqa.

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

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.

Dense prediction tasks such as object detection and segmentation require high-quality labels at pixel level, which are costly to obtain. Recent advances in foundation models have enabled the generation of autolabels, which we find to be competitive but not yet sufficient to fully replace human annotations, especially for more complex datasets. Thus, we consider the challenge of selecting a representative subset of images for labeling from a large pool of unlabeled images under a constrained annotation budget. This task is further complicated by imbalanced class distributions, as rare classes are often underrepresented in selected subsets. We propose object-focused data selection (OFDS) which leverages object-level representations to ensure that the selected image subsets semantically cover the target classes, including rare ones. We validate OFDS on PASCAL VOC and Cityscapes for object detection and semantic segmentation tasks. Our experiments demonstrate that prior methods which employ image-level representations fail to consistently outperform random selection. In contrast, OFDS consistently achieves state-of-the-art performance with substantial improvements over all baselines in scenarios with imbalanced class distributions. Moreover, we demonstrate that pre-training with autolabels on the full datasets before fine-tuning on human-labeled subsets selected by OFDS further enhances the final performance.

The field of medical imaging AI is currently undergoing rapid transformations, with methodical research increasingly translated into clinical practice. Despite these successes, research suffers from knowledge silos, hindering collaboration and progress: Existing knowledge is scattered across publications and many details remain unpublished, while privacy regulations restrict data sharing. In the spirit of democratizing of AI, we propose a framework for secure knowledge transfer in the field of medical image analysis. The key to our approach is dataset "fingerprints", structured representations of feature distributions, that enable quantification of task similarity. We tested our approach across 71 distinct tasks and 12 medical imaging modalities by transferring neural architectures, pretraining, augmentation policies, and multi-task learning. According to comprehensive analyses, our method outperforms traditional methods for identifying relevant knowledge and facilitates collaborative model training. Our framework fosters the democratization of AI in medical imaging and could become a valuable tool for promoting faster scientific advancement.

Background and objective: Diabetes is one of the four leading causes of death worldwide, necessitating daily blood glucose monitoring. While sweat offers a promising non-invasive alternative for glucose monitoring, its application remains limited due to the low to moderate correlation between sweat and blood glucose concentrations, which has been obtained until now by assuming a linear relationship. This study proposes a novel model-based strategy to estimate blood glucose concentrations from sweat samples, setting the stage for non-invasive glucose monitoring through sweat-sensing technology. Methods: We first developed a pharmacokinetic glucose transport model that describes the glucose transport from blood to sweat. Secondly, we designed a novel optimization strategy leveraging the proposed model to solve the inverse problem and infer blood glucose levels from measured glucose concentrations in sweat. To this end, the pharmacokinetic model parameters with the highest sensitivity were also optimized so as to achieve a personalized estimation. Our strategy was tested on a dataset composed of 108 samples from healthy volunteers and diabetic patients. Results: Our glucose transport model improves over the state-of-the-art in estimating sweat glucose concentrations from blood levels (higher accuracy, p<0.001). Additionally, our optimization strategy effectively solved the inverse problem, yielding a Pearson correlation coefficient of 0.98 across all 108 data points, with an average root-mean-square-percent-error of 12%+/-8%. This significantly outperforms the best sweat-blood glucose correlation reported in the existing literature (0.75). Conclusion: Our innovative optimization strategy, also leveraging more accurate modeling, shows promising results, paving the way for non-invasive blood glucose monitoring and, possibly, improved diabetes management.

Vision-Language models like CLIP have been shown to be highly effective at linking visual perception and natural language understanding, enabling sophisticated image-text capabilities, including strong retrieval and zero-shot classification performance. Their widespread use, as well as the fact that CLIP models are trained on image-text pairs from the web, make them both a worthwhile and relatively easy target for backdoor attacks. As training foundational models, such as CLIP, from scratch is very expensive, this paper focuses on cleaning potentially poisoned models via fine-tuning. We first show that existing cleaning techniques are not effective against simple structured triggers used in Blended or BadNet backdoor attacks, exposing a critical vulnerability for potential real-world deployment of these models. Then, we introduce PAR, Perturb and Recover, a surprisingly simple yet effective mechanism to remove backdoors from CLIP models. Through extensive experiments across different encoders and types of backdoor attacks, we show that PAR achieves high backdoor removal rate while preserving good standard performance. Finally, we illustrate that our approach is effective even only with synthetic text-image pairs, i.e. without access to real training data. The code and models are available at https://github.com/nmndeep/PerturbAndRecover.

This paper presents our approach to scaling the nnU-Net framework for multi-structure segmentation on Cone Beam Computed Tomography (CBCT) images, specifically in the scope of the ToothFairy2 Challenge. We leveraged the nnU-Net ResEnc L model, introducing key modifications to patch size, network topology, and data augmentation strategies to address the unique challenges of dental CBCT imaging. Our method achieved a mean Dice coefficient of 0.9253 and HD95 of 18.472 on the test set, securing a mean rank of 4.6 and with it the first place in the ToothFairy2 challenge. The source code is publicly available, encouraging further research and development in the field., Comment: Fabian Isensee and Yannick Kirchhoff contributed equally

In multi-state models based on high-dimensional data, effective modeling strategies are required to determine an optimal, ideally parsimonious model. In particular, linking covariate effects across transitions is needed to conduct joint variable selection. A useful technique to reduce model complexity is to address homogeneous covariate effects for distinct transitions. We integrate this approach to data-driven variable selection by extended regularization methods within multi-state model building. We propose the fused sparse-group lasso (FSGL) penalized Cox-type regression in the framework of multi-state models combining the penalization concepts of pairwise differences of covariate effects along with transition grouping. For optimization, we adapt the alternating direction method of multipliers (ADMM) algorithm to transition-specific hazards regression in the multi-state setting. In a simulation study and application to acute myeloid leukemia (AML) data, we evaluate the algorithm's ability to select a sparse model incorporating relevant transition-specific effects and similar cross-transition effects. We investigate settings in which the combined penalty is beneficial compared to global lasso regularization.

This paper presents ETA-IK, a novel Execution-Time-Aware Inverse Kinematics method tailored for dual-arm robotic systems. The primary goal is to optimize motion execution time by leveraging the redundancy of both arms, specifically in tasks where only the relative pose of the robots is constrained, such as dual-arm scanning of unknown objects. Unlike traditional inverse kinematics methods that use surrogate metrics such as joint configuration distance, our method incorporates direct motion execution time and implicit collisions into the optimization process, thereby finding target joints that allow subsequent trajectory generation to get more efficient and collision-free motion. A neural network based execution time approximator is employed to predict time-efficient joint configurations while accounting for potential collisions. Through experimental evaluation on a system composed of a UR5 and a KUKA iiwa robot, we demonstrate significant reductions in execution time. The proposed method outperforms conventional approaches, showing improved motion efficiency without sacrificing positioning accuracy. These results highlight the potential of ETA-IK to improve the performance of dual-arm systems in applications, where efficiency and safety are paramount.

Current interactive segmentation approaches, inspired by the success of META's Segment Anything model, have achieved notable advancements, however, they come with substantial limitations that hinder their practical application in 3D radiological scenarios. These include unrealistic human interaction requirements, such as slice-by-slice operations for 2D models on 3D data, a lack of iterative interactive refinement, and insufficient evaluation experiments. These shortcomings prevent accurate assessment of model performance and lead to inconsistent outcomes across studies. The RadioActive benchmark overcomes these challenges by offering a comprehensive and reproducible evaluation of interactive segmentation methods in realistic, clinically relevant scenarios. It includes diverse datasets, target structures, and interactive segmentation methods, and provides a flexible, extendable codebase that allows seamless integration of new models and prompting strategies. We also introduce advanced prompting techniques to enable 2D models on 3D data by reducing the needed number of interaction steps, enabling a fair comparison. We show that surprisingly the performance of slice-wise prompted approaches can match native 3D methods, despite the domain gap. Our findings challenge the current literature and highlight that models not specifically trained on medical data can outperform the current specialized medical methods. By open-sourcing RadioActive, we invite the research community to integrate their models and prompting techniques, ensuring continuous and transparent evaluation of interactive segmentation models in 3D medical imaging., Comment: Undergoing Peer-Review

How can we test AI performance? This question seems trivial, but it isn't. Standard benchmarks often have problems such as in-distribution and small-size test sets, oversimplified metrics, unfair comparisons, and short-term outcome pressure. As a consequence, good performance on standard benchmarks does not guarantee success in real-world scenarios. To address these problems, we present Touchstone, a large-scale collaborative segmentation benchmark of 9 types of abdominal organs. This benchmark is based on 5,195 training CT scans from 76 hospitals around the world and 5,903 testing CT scans from 11 additional hospitals. This diverse test set enhances the statistical significance of benchmark results and rigorously evaluates AI algorithms across various out-of-distribution scenarios. We invited 14 inventors of 19 AI algorithms to train their algorithms, while our team, as a third party, independently evaluated these algorithms on three test sets. In addition, we also evaluated pre-existing AI frameworks--which, differing from algorithms, are more flexible and can support different algorithms--including MONAI from NVIDIA, nnU-Net from DKFZ, and numerous other open-source frameworks. We are committed to expanding this benchmark to encourage more innovation of AI algorithms for the medical domain., Comment: Accepted to NeurIPS-2024

Resonant anomaly detection methods have great potential for enhancing the sensitivity of traditional bump hunt searches. A key component of these methods is a high quality background template used to produce an anomaly score. Using the LHC Olympics R&D dataset, we demonstrate that this background template can also be repurposed to directly estimate the background expectation in a simple cut and count setup. In contrast to a traditional bump hunt, no fit to the invariant mass distribution is needed, thereby avoiding the potential problem of background sculpting. Furthermore, direct background estimation allows working with large background rejection rates, where resonant anomaly detection methods typically show their greatest improvement in significance., Comment: 26 pages, 9 figures, 2 tables

Self-Supervised Learning (SSL) presents an exciting opportunity to unlock the potential of vast, untapped clinical datasets, for various downstream applications that suffer from the scarcity of labeled data. While SSL has revolutionized fields like natural language processing and computer vision, its adoption in 3D medical image computing has been limited by three key pitfalls: Small pre-training dataset sizes, architectures inadequate for 3D medical image analysis, and insufficient evaluation practices. In this paper, we address these issues by i) leveraging a large-scale dataset of 39k 3D brain MRI volumes and ii) using a Residual Encoder U-Net architecture within the state-of-the-art nnU-Net framework. iii) A robust development framework, incorporating 5 development and 8 testing brain MRI segmentation datasets, allowed performance-driven design decisions to optimize the simple concept of Masked Auto Encoders (MAEs) for 3D CNNs. The resulting model not only surpasses previous SSL methods but also outperforms the strong nnU-Net baseline by an average of approximately 3 Dice points setting a new state-of-the-art. Our code and models are made available here., Comment: Arxiv Preprint. Revised and under review

What representation do deep neural networks learn? How similar are images to each other for neural networks? Despite the overwhelming success of deep learning methods key questions about their internal workings still remain largely unanswered, due to their internal high dimensionality and complexity. To address this, one approach is to measure the similarity of activation responses to various inputs. Representational Similarity Matrices (RSMs) distill this similarity into scalar values for each input pair. These matrices encapsulate the entire similarity structure of a system, indicating which input leads to similar responses. While the similarity between images is ambiguous, we argue that the spatial location of semantic objects does neither influence human perception nor deep learning classifiers. Thus this should be reflected in the definition of similarity between image responses for computer vision systems. Revisiting the established similarity calculations for RSMs we expose their sensitivity to spatial alignment. In this paper, we propose to solve this through semantic RSMs, which are invariant to spatial permutation. We measure semantic similarity between input responses by formulating it as a set-matching problem. Further, we quantify the superiority of semantic RSMs over spatio-semantic RSMs through image retrieval and by comparing the similarity between representations to the similarity between predicted class probabilities., Comment: Accepted at NeurIPS2024

With the goal of efficiently computing collision-free robot motion trajectories in dynamically changing environments, we present results of a novel method for Heuristics Informed Robot Online Path Planning (HIRO). Dividing robot environments into static and dynamic elements, we use the static part for initializing a deterministic roadmap, which provides a lower bound of the final path cost as informed heuristics for fast path-finding. These heuristics guide a search tree to explore the roadmap during runtime. The search tree examines the edges using a fuzzy collision checking concerning the dynamic environment. Finally, the heuristics tree exploits knowledge fed back from the fuzzy collision checking module and updates the lower bound for the path cost. As we demonstrate in real-world experiments, the closed-loop formed by these three components significantly accelerates the planning procedure. An additional backtracking step ensures the feasibility of the resulting paths. Experiments in simulation and the real world show that HIRO can find collision-free paths considerably faster than baseline methods with and without prior knowledge of the environment.

Path planning in a changing environment is a challenging task in robotics, as moving objects impose time-dependent constraints. Recent planning methods primarily focus on the spatial aspects, lacking the capability to directly incorporate time constraints. In this paper, we propose a method that leverages a generative model to decompose a complex planning problem into small manageable ones by incrementally generating subgoals given the current planning context. Then, we take into account the temporal information and use learned time estimators based on different statistic distributions to examine and select the generated subgoal candidates. Experiments show that planning from the current robot state to the selected subgoal can satisfy the given time-dependent constraints while being goal-oriented.

A plethora of jailbreaking attacks have been proposed to obtain harmful responses from safety-tuned LLMs. In their original settings, these methods all largely succeed in coercing the target output, but their attacks vary substantially in fluency and computational effort. In this work, we propose a unified threat model for the principled comparison of these methods. Our threat model combines constraints in perplexity, measuring how far a jailbreak deviates from natural text, and computational budget, in total FLOPs. For the former, we build an N-gram model on 1T tokens, which, in contrast to model-based perplexity, allows for an LLM-agnostic and inherently interpretable evaluation. We adapt popular attacks to this new, realistic threat model, with which we, for the first time, benchmark these attacks on equal footing. After a rigorous comparison, we not only find attack success rates against safety-tuned modern models to be lower than previously presented but also find that attacks based on discrete optimization significantly outperform recent LLM-based attacks. Being inherently interpretable, our threat model allows for a comprehensive analysis and comparison of jailbreak attacks. We find that effective attacks exploit and abuse infrequent N-grams, either selecting N-grams absent from real-world text or rare ones, e.g. specific to code datasets.

Novel optical imaging techniques, such as hyperspectral imaging (HSI) combined with machine learning-based (ML) analysis, have the potential to revolutionize clinical surgical imaging. However, these novel modalities face a shortage of large-scale, representative clinical data for training ML algorithms, while preclinical animal data is abundantly available through standardized experiments and allows for controlled induction of pathological tissue states, which is not ethically possible in patients. To leverage this situation, we propose a novel concept called "xeno-learning", a cross-species knowledge transfer paradigm inspired by xeno-transplantation, where organs from a donor species are transplanted into a recipient species. Using a total of 11,268 HSI images from humans as well as porcine and rat models, we show that although spectral signatures of organs differ across species, shared pathophysiological mechanisms manifest as comparable relative spectral changes across species. Such changes learnt in one species can thus be transferred to a new species via a novel "physiology-based data augmentation" method, enabling the large-scale secondary use of preclinical animal data for humans. The resulting ethical, monetary, and performance benefits of the proposed knowledge transfer paradigm promise a high impact of the methodology on future developments in the field., Comment: Jan Sellner and Alexander Studier-Fischer contributed equally to this work

Given a digraph $D=(V,A)$ with vertex-set $V=\{1,\ldots,n\}$ and arc-set $A$, we denote by $Q(D)$ the set of all real $n\times n$ matrices $B=[b_{u,w}]$ with $b_{u,u}\not=0$ for all $u\in V$, $b_{u,w} \not= 0$ if $u\not=w$ and there is an arc from $u$ to $w$, and $b_{u,w}=0$ if $u\not=w$ and there is no arc from $u$ to $w$. We say that a matrix $B\in Q(D)$ has the Asymmetric Strong Arnold Property (ASAP) if $X\circ B = 0$, $X^T B = 0$, and $B X^T = 0$ implies $X=0$. We define the stable maximum nullity, $\overrightarrow{\nu}(D)$, of a digraph $D$ as the largest nullity of any matrix $A\in Q(D)$ that has the ASAP\@. We show that a digraph $D$ has $\overrightarrow{\nu}(D)\leq 1$ if and only $D$ and $\overleftarrow{D}$ a partial $1$-DAGs.

CubeSats offer a cost-effective platform for various space missions, but their limited fuel capacity and susceptibility to environmental disturbances pose significant challenges for precise orbital maneuvering. This paper presents a novel control strategy that integrates a J2-optimized sequence with an LSTM-based low-level control layer to address these issues. The J2-optimized sequence leverages the Earth's oblateness to minimize fuel consumption during orbital corrections, while the LSTM network provides real-time adjustments to compensate for external disturbances and unmodeled dynamics. The LSTM network was trained on a dataset generated from simulated orbital scenarios, including factors such as atmospheric drag, solar radiation pressure, and gravitational perturbations. The proposed system was evaluated through numerical simulations, demonstrating significant improvements in maneuver accuracy and robustness compared to traditional methods. The results show that the combined system efficiently reduces miss distances, even under conditions of high uncertainty. This hybrid approach offers a powerful and adaptive solution for CubeSat missions, balancing fuel efficiency with precise orbital control., Comment: Pre-print of the paper submitted to 75th International Astronautical Congress (IAC), Milan, Italy, 14-18 October 2024

Grading Objective Structured Clinical Examinations (OSCEs) is a time-consuming and expensive process, traditionally requiring extensive manual effort from human experts. In this study, we explore the potential of Large Language Models (LLMs) to assess skills related to medical student communication. We analyzed 2,027 video-recorded OSCE examinations from the University of Texas Southwestern Medical Center (UTSW), spanning four years (2019-2022), and several different medical cases or "stations." Specifically, our focus was on evaluating students' ability to summarize patients' medical history: we targeted the rubric item 'did the student summarize the patients' medical history?' from the communication skills rubric. After transcribing speech audio captured by OSCE videos using Whisper-v3, we studied the performance of various LLM-based approaches for grading students on this summarization task based on their examination transcripts. Using various frontier-level open-source and proprietary LLMs, we evaluated different techniques such as zero-shot chain-of-thought prompting, retrieval augmented generation, and multi-model ensemble methods. Our results show that frontier LLM models like GPT-4 achieved remarkable alignment with human graders, demonstrating a Cohen's kappa agreement of 0.88 and indicating strong potential for LLM-based OSCE grading to augment the current grading process. Open-source models also showed promising results, suggesting potential for widespread, cost-effective deployment. Further, we present a failure analysis identifying conditions where LLM grading may be less reliable in this context and recommend best practices for deploying LLMs in medical education settings.

Radiologists play a crucial role by translating medical images into medical reports. However, the field faces staffing shortages and increasing workloads. While automated approaches using vision-language models (VLMs) show promise as assistants, they require exceptionally high accuracy. Most current VLMs in radiology rely solely on supervised fine-tuning (SFT). Meanwhile, in the general domain, additional preference fine-tuning has become standard practice. The challenge in radiology lies in the prohibitive cost of obtaining radiologist feedback. We propose a scalable automated preference alignment technique for VLMs in radiology, focusing on chest X-ray (CXR) report generation. Our method leverages publicly available datasets with an LLM-as-a-Judge mechanism, eliminating the need for additional expert radiologist feedback. We evaluate and benchmark five direct alignment algorithms (DAAs). Our results show up to a 57.4% improvement in average GREEN scores, a LLM-based metric for evaluating CXR reports, and a 9.2% increase in an average across six metrics (domain specific and general), compared to the SFT baseline. We study reward overoptimization via length exploitation, with reports lengthening by up to 3.2x. To assess a potential alignment tax, we benchmark on six additional diverse tasks, finding no significant degradations. A reader study involving four board-certified radiologists indicates win rates of up to 0.62 over the SFT baseline, while significantly penalizing verbosity. Our analysis provides actionable insights for the development of VLMs in high-stakes fields like radiology.

Selective Classification, wherein models can reject low-confidence predictions, promises reliable translation of machine-learning based classification systems to real-world scenarios such as clinical diagnostics. While current evaluation of these systems typically assumes fixed working points based on pre-defined rejection thresholds, methodological progress requires benchmarking the general performance of systems akin to the $\mathrm{AUROC}$ in standard classification. In this work, we define 5 requirements for multi-threshold metrics in selective classification regarding task alignment, interpretability, and flexibility, and show how current approaches fail to meet them. We propose the Area under the Generalized Risk Coverage curve ($\mathrm{AUGRC}$), which meets all requirements and can be directly interpreted as the average risk of undetected failures. We empirically demonstrate the relevance of $\mathrm{AUGRC}$ on a comprehensive benchmark spanning 6 data sets and 13 confidence scoring functions. We find that the proposed metric substantially changes metric rankings on 5 out of the 6 data sets.

Medical imaging is spearheading the AI transformation of healthcare. Performance reporting is key to determine which methods should be translated into clinical practice. Frequently, broad conclusions are simply derived from mean performance values. In this paper, we argue that this common practice is often a misleading simplification as it ignores performance variability. Our contribution is threefold. (1) Analyzing all MICCAI segmentation papers (n = 221) published in 2023, we first observe that more than 50% of papers do not assess performance variability at all. Moreover, only one (0.5%) paper reported confidence intervals (CIs) for model performance. (2) To address the reporting bottleneck, we show that the unreported standard deviation (SD) in segmentation papers can be approximated by a second-order polynomial function of the mean Dice similarity coefficient (DSC). Based on external validation data from 56 previous MICCAI challenges, we demonstrate that this approximation can accurately reconstruct the CI of a method using information provided in publications. (3) Finally, we reconstructed 95% CIs around the mean DSC of MICCAI 2023 segmentation papers. The median CI width was 0.03 which is three times larger than the median performance gap between the first and second ranked method. For more than 60% of papers, the mean performance of the second-ranked method was within the CI of the first-ranked method. We conclude that current publications typically do not provide sufficient evidence to support which models could potentially be translated into clinical practice., Comment: Paper accepted at MICCAI 2024 conference

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., Comment: Accepted at MICCAI 2024 LDTM

Virtual Reality technology, when integrated with Surgical Digital Twins (SDTs), offers significant potential in medical training and surgical planning. We present SurgTwinVR, a VR application that immerses users within an SDT and enables them to navigate a high-fidelity virtual replica of the surgical environment. SurgTwinVR is the first VR application to utilize a dynamic 3D environment that is a clone of a real surgery, encompassing the entire surgical scene, including the surgeon, anatomy, and instruments. Our system utilizes a SDT with important improvements for real-time rendering and features to showcase the potential benefits of such an application in surgical education., Comment: Accepted demo at ISMAR 2024. Project page: https://jonashein.github.io/surgerydigitization/

Segmentation of the fetal and maternal structures, particularly intrapartum ultrasound imaging as advocated by the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) for monitoring labor progression, is a crucial first step for quantitative diagnosis and clinical decision-making. This requires specialized analysis by obstetrics professionals, in a task that i) is highly time- and cost-consuming and ii) often yields inconsistent results. The utility of automatic segmentation algorithms for biometry has been proven, though existing results remain suboptimal. To push forward advancements in this area, the Grand Challenge on Pubic Symphysis-Fetal Head Segmentation (PSFHS) was held alongside the 26th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2023). This challenge aimed to enhance the development of automatic segmentation algorithms at an international scale, providing the largest dataset to date with 5,101 intrapartum ultrasound images collected from two ultrasound machines across three hospitals from two institutions. The scientific community's enthusiastic participation led to the selection of the top 8 out of 179 entries from 193 registrants in the initial phase to proceed to the competition's second stage. These algorithms have elevated the state-of-the-art in automatic PSFHS from intrapartum ultrasound images. A thorough analysis of the results pinpointed ongoing challenges in the field and outlined recommendations for future work. The top solutions and the complete dataset remain publicly available, fostering further advancements in automatic segmentation and biometry for intrapartum ultrasound imaging.

Flagellated bacteria are hydrodynamically attracted to rigid walls, yet past work shows a 'hovering' state where they swim stably at a finite height above surfaces. We use numerics and theory to reveal the physical origin of hovering. Simulations first show that hovering requires an elongated cell body and results from a tilt away from the wall. Theoretical models then identify two essential asymmetries: the response of width-asymmetric cells to active flows created by length-asymmetric cells. A minimal model reconciles near and far-field hydrodynamics, capturing all key features of hovering., Comment: 6 pages, 3 figures. To appear in Physical Review Research (2024)

The third autoPET challenge introduced a new data-centric task this year, shifting the focus from model development to improving metastatic lesion segmentation on PET/CT images through data quality and handling strategies. In response, we developed targeted methods to enhance segmentation performance tailored to the characteristics of PET/CT imaging. Our approach encompasses two key elements. First, to address potential alignment errors between CT and PET modalities as well as the prevalence of punctate lesions, we modified the baseline data augmentation scheme and extended it with misalignment augmentation. This adaptation aims to improve segmentation accuracy, particularly for tiny metastatic lesions. Second, to tackle the variability in image dimensions significantly affecting the prediction time, we implemented a dynamic ensembling and test-time augmentation (TTA) strategy. This method optimizes the use of ensembling and TTA within a 5-minute prediction time limit, effectively leveraging the generalization potential for both small and large images. Both of our solutions are designed to be robust across different tracers and institutional settings, offering a general, yet imaging-specific approach to the multi-tracer and multi-institutional challenges of the competition. We made the challenge repository with our modifications publicly available at \url{https://github.com/MIC-DKFZ/miccai2024_autopet3_datacentric}., Comment: Contribution to the data-centric task of the autoPET III Challenge 2024

Automated lesion segmentation in PET/CT scans is crucial for improving clinical workflows and advancing cancer diagnostics. However, the task is challenging due to physiological variability, different tracers used in PET imaging, and diverse imaging protocols across medical centers. To address this, the autoPET series was created to challenge researchers to develop algorithms that generalize across diverse PET/CT environments. This paper presents our solution for the autoPET III challenge, targeting multitracer, multicenter generalization using the nnU-Net framework with the ResEncL architecture. Key techniques include misalignment data augmentation and multi-modal pretraining across CT, MR, and PET datasets to provide an initial anatomical understanding. We incorporate organ supervision as a multitask approach, enabling the model to distinguish between physiological uptake and tracer-specific patterns, which is particularly beneficial in cases where no lesions are present. Compared to the default nnU-Net, which achieved a Dice score of 57.61, or the larger ResEncL (65.31) our model significantly improved performance with a Dice score of 68.40, alongside a reduction in false positive (FPvol: 7.82) and false negative (FNvol: 10.35) volumes. These results underscore the effectiveness of combining advanced network design, augmentation, pretraining, and multitask learning for PET/CT lesion segmentation. After evaluation on the test set, our approach was awarded the first place in the model-centric category (Team LesionTracer). Code is publicly available at https://github.com/MIC-DKFZ/autopet-3-submission., Comment: Winning method of the autoPET III challenge (model-centric) - Team LesionTracer

Panel data arises when transitions between different states are interval-censored in multi-state data. The analysis of such data using non-parametric multi-state models was not possible until recently, but is very desirable as it allows for more flexibility than its parametric counterparts. The single available result to date has some unique drawbacks. We propose a non-parametric estimator of the transition intensities for panel data using an Expectation Maximisation algorithm. The method allows for a mix of interval-censored and right-censored (exactly observed) transitions. A condition to check for the convergence of the algorithm to the non-parametric maximum likelihood estimator is given. A simulation study comparing the proposed estimator to a consistent estimator is performed, and shown to yield near identical estimates at smaller computational cost. A data set on the emergence of teeth in children is analysed. Code to perform the analyses is publicly available., Comment: 32 pages, 17 figures. Software in the form of an R package for the proposed method can be found on GitHub: https://github.com/d-gomon/icmstate

Hyperspectral imaging (HSI) is emerging as a promising novel imaging modality with various potential surgical applications. Currently available cameras, however, suffer from poor integration into the clinical workflow because they require the lights to be switched off, or the camera to be manually recalibrated as soon as lighting conditions change. Given this critical bottleneck, the contribution of this paper is threefold: (1) We demonstrate that dynamically changing lighting conditions in the operating room dramatically affect the performance of HSI applications, namely physiological parameter estimation, and surgical scene segmentation. (2) We propose a novel learning-based approach to automatically recalibrating hyperspectral images during surgery and show that it is sufficiently accurate to replace the tedious process of white reference-based recalibration. (3) Based on a total of 742 HSI cubes from a phantom, porcine models, and rats we show that our recalibration method not only outperforms previously proposed methods, but also generalizes across species, lighting conditions, and image processing tasks. Due to its simple workflow integration as well as high accuracy, speed, and generalization capabilities, our method could evolve as a central component in clinical surgical HSI., Comment: Oral at MICCAI 2024

In the presence of unmeasured heterogeneity, the hazard ratio for exposure has a complex causal interpretation. To address this, accelerated failure time (AFT) models, which assess the effect on the survival time ratio scale, are often suggested as a better alternative. AFT models also allow for straightforward confounder adjustment. In this work, we formalize the causal interpretation of the acceleration factor in AFT models using structural causal models and data under independent censoring. We prove that the acceleration factor is a valid causal effect measure, even in the presence of frailty and treatment effect heterogeneity. Through simulations, we show that the acceleration factor better captures the causal effect than the hazard ratio when both AFT and proportional hazards models apply. Additionally, we extend the interpretation to systems with time-dependent acceleration factors, revealing the challenge of distinguishing between a time-varying homogeneous effect and unmeasured heterogeneity. While the causal interpretation of acceleration factors is promising, we caution practitioners about potential challenges in estimating these factors in the presence of effect heterogeneity.

In realistic medical settings, the data are often inherently long-tailed, with most samples concentrated in a few classes and a long tail of rare classes, usually containing just a few samples. This distribution presents a significant challenge because rare conditions are critical to detect and difficult to classify due to limited data. In this paper, rather than attempting to classify rare classes, we aim to detect these as out-of-distribution data reliably. We leverage low-rank adaption (LoRA) and diffusion guidance to generate targeted synthetic data for the detection problem. We significantly improve the OOD detection performance on a challenging histopathological task with only ten samples per tail class without losing classification accuracy on the head classes.

Medical image classification in radiology faces significant challenges, particularly in generalizing to unseen pathologies. In contrast, CLIP offers a promising solution by leveraging multimodal learning to improve zero-shot classification performance. However, in the medical domain, lesions can be small and might not be well represented in the embedding space. Therefore, in this paper, we explore the potential of visual prompt engineering to enhance the capabilities of Vision Language Models (VLMs) in radiology. Leveraging BiomedCLIP, trained on extensive biomedical image-text pairs, we investigate the impact of embedding visual markers directly within radiological images to guide the model's attention to critical regions. Our evaluation on the JSRT dataset, focusing on lung nodule malignancy classification, demonstrates that incorporating visual prompts $\unicode{x2013}$ such as arrows, circles, and contours $\unicode{x2013}$ significantly improves classification metrics including AUROC, AUPRC, F1 score, and accuracy. Moreover, the study provides attention maps, showcasing enhanced model interpretability and focus on clinically relevant areas. These findings underscore the efficacy of visual prompt engineering as a straightforward yet powerful approach to advance VLM performance in medical image analysis., Comment: Accepted at ECCV 2024 Workshop on Emergent Visual Abilities and Limits of Foundation Models

Robust semantic segmentation of intraoperative image data holds promise for enabling automatic surgical scene understanding and autonomous robotic surgery. While model development and validation are primarily conducted on idealistic scenes, geometric domain shifts, such as occlusions of the situs, are common in real-world open surgeries. To close this gap, we (1) present the first analysis of state-of-the-art (SOA) semantic segmentation models when faced with geometric out-of-distribution (OOD) data, and (2) propose an augmentation technique called "Organ Transplantation", to enhance generalizability. Our comprehensive validation on six different OOD datasets, comprising 600 RGB and hyperspectral imaging (HSI) cubes from 33 pigs, each annotated with 19 classes, reveals a large performance drop in SOA organ segmentation models on geometric OOD data. This performance decline is observed not only in conventional RGB data (with a dice similarity coefficient (DSC) drop of 46 %) but also in HSI data (with a DSC drop of 45 %), despite the richer spectral information content. The performance decline increases with the spatial granularity of the input data. Our augmentation technique improves SOA model performance by up to 67 % for RGB data and 90 % for HSI data, achieving performance at the level of in-distribution performance on real OOD test data. Given the simplicity and effectiveness of our augmentation method, it is a valuable tool for addressing geometric domain shifts in surgical scene segmentation, regardless of the underlying model. Our code and pre-trained models are publicly available at https://github.com/IMSY-DKFZ/htc., Comment: Silvia Seidlitz and Jan Sellner contributed equally

Competing risks models can involve more than one time scale. A relevant example is the study of mortality after a cancer diagnosis, where time since diagnosis but also age may jointly determine the hazards of death due to different causes. Multiple time scales have rarely been explored in the context of competing events. Here, we propose a model in which the cause-specific hazards vary smoothly over two times scales. It is estimated by two-dimensional $P$-splines, exploiting the equivalence between hazard smoothing and Poisson regression. The data are arranged on a grid so that we can make use of generalized linear array models for efficient computations. The R-package TwoTimeScales implements the model. As a motivating example we analyse mortality after diagnosis of breast cancer and we distinguish between death due to breast cancer and all other causes of death. The time scales are age and time since diagnosis. We use data from the Surveillance, Epidemiology and End Results (SEER) program. In the SEER data, age at diagnosis is provided with a last open-ended category, leading to coarsely grouped data. We use the two-dimensional penalised composite link model to ungroup the data before applying the competing risks model with two time scales., Comment: 36 pages, 9 figures, 2 tables

Accurate assessment of lymph node size in 3D CT scans is crucial for cancer staging, therapeutic management, and monitoring treatment response. Existing state-of-the-art segmentation frameworks in medical imaging often rely on fully annotated datasets. However, for lymph node segmentation, these datasets are typically small due to the extensive time and expertise required to annotate the numerous lymph nodes in 3D CT scans. Weakly-supervised learning, which leverages incomplete or noisy annotations, has recently gained interest in the medical imaging community as a potential solution. Despite the variety of weakly-supervised techniques proposed, most have been validated only on private datasets or small publicly available datasets. To address this limitation, the Mediastinal Lymph Node Quantification (LNQ) challenge was organized in conjunction with the 26th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2023). This challenge aimed to advance weakly-supervised segmentation methods by providing a new, partially annotated dataset and a robust evaluation framework. A total of 16 teams from 5 countries submitted predictions to the validation leaderboard, and 6 teams from 3 countries participated in the evaluation phase. The results highlighted both the potential and the current limitations of weakly-supervised approaches. On one hand, weakly-supervised approaches obtained relatively good performance with a median Dice score of $61.0\%$. On the other hand, top-ranked teams, with a median Dice score exceeding $70\%$, boosted their performance by leveraging smaller but fully annotated datasets to combine weak supervision and full supervision. This highlights both the promise of weakly-supervised methods and the ongoing need for high-quality, fully annotated data to achieve higher segmentation performance., Comment: Submitted to MELBA

With sepsis remaining a leading cause of mortality, early identification of septic patients and those at high risk of death is a challenge of high socioeconomic importance. The driving hypothesis of this study was that hyperspectral imaging (HSI) could provide novel biomarkers for sepsis diagnosis and treatment management due to its potential to monitor microcirculatory alterations. We conducted a comprehensive study involving HSI data of the palm and fingers from more than 480 patients on the day of their intensive care unit (ICU) admission. The findings demonstrate that HSI measurements can predict sepsis with an area under the receiver operating characteristic curve (AUROC) of 0.80 (95 % confidence interval (CI) [0.76; 0.84]) and mortality with an AUROC of 0.72 (95 % CI [0.65; 0.79]). The predictive performance improves substantially when additional clinical data is incorporated, leading to an AUROC of up to 0.94 (95 % CI [0.92; 0.96]) for sepsis and 0.84 (95 % CI [0.78; 0.89]) for mortality. We conclude that HSI presents novel imaging biomarkers for the rapid, non-invasive prediction of sepsis and mortality, suggesting its potential as an important modality for guiding diagnosis and treatment., Comment: Markus A. Weigand, Lena Maier-Hein and Maximilian Dietrich contributed equally

Measuring the similarity of different representations of neural architectures is a fundamental task and an open research challenge for the machine learning community. This paper presents the first comprehensive benchmark for evaluating representational similarity measures based on well-defined groundings of similarity. The representational similarity (ReSi) benchmark consists of (i) six carefully designed tests for similarity measures, (ii) 23 similarity measures, (iii) eleven neural network architectures, and (iv) six datasets, spanning over the graph, language, and vision domains. The benchmark opens up several important avenues of research on representational similarity that enable novel explorations and applications of neural architectures. We demonstrate the utility of the ReSi benchmark by conducting experiments on various neural network architectures, real world datasets and similarity measures. All components of the benchmark are publicly available and thereby facilitate systematic reproduction and production of research results. The benchmark is extensible, future research can build on and further expand it. We believe that the ReSi benchmark can serve as a sound platform catalyzing future research that aims to systematically evaluate existing and explore novel ways of comparing representations of neural architectures., Comment: Feedback welcome! Code and data at https://github.com/mklabunde/resi

The application of machine learning in medicine and healthcare has led to the creation of numerous diagnostic and prognostic models. However, despite their success, current approaches generally issue predictions using data from a single modality. This stands in stark contrast with clinician decision-making which employs diverse information from multiple sources. While several multimodal machine learning approaches exist, significant challenges in developing multimodal systems remain that are hindering clinical adoption. In this paper, we introduce a multimodal framework, AutoPrognosis-M, that enables the integration of structured clinical (tabular) data and medical imaging using automated machine learning. AutoPrognosis-M incorporates 17 imaging models, including convolutional neural networks and vision transformers, and three distinct multimodal fusion strategies. In an illustrative application using a multimodal skin lesion dataset, we highlight the importance of multimodal machine learning and the power of combining multiple fusion strategies using ensemble learning. We have open-sourced our framework as a tool for the community and hope it will accelerate the uptake of multimodal machine learning in healthcare and spur further innovation.

Inverse problems, such as accelerated MRI reconstruction, are ill-posed and an infinite amount of possible and plausible solutions exist. This may not only lead to uncertainty in the reconstructed image but also in downstream tasks such as semantic segmentation. This uncertainty, however, is mostly not analyzed in the literature, even though probabilistic reconstruction models are commonly used. These models can be prone to ignore plausible but unlikely solutions like rare pathologies. Building on MRI reconstruction approaches based on diffusion models, we add guidance to the diffusion process during inference, generating two meaningfully diverse reconstructions corresponding to an upper and lower bound segmentation. The reconstruction uncertainty can then be quantified by the difference between these bounds, which we coin the 'uncertainty boundary'. We analyzed the behavior of the upper and lower bound segmentations for a wide range of acceleration factors and found the uncertainty boundary to be both more reliable and more accurate compared to repeated sampling. Code is available at https://github.com/NikolasMorshuis/SGR, Comment: Accepted at DGM4MICCAI 2024

This paper does not describe a novel method. Instead, it studies an essential foundation for reliable benchmarking and ultimately real-world application of AI-based image analysis: generating high-quality reference annotations. Previous research has focused on crowdsourcing as a means of outsourcing annotations. However, little attention has so far been given to annotation companies, specifically regarding their internal quality assurance (QA) processes. Therefore, our aim is to evaluate the influence of QA employed by annotation companies on annotation quality and devise methodologies for maximizing data annotation efficacy. Based on a total of 57,648 instance segmented images obtained from a total of 924 annotators and 34 QA workers from four annotation companies and Amazon Mechanical Turk (MTurk), we derived the following insights: (1) Annotation companies perform better both in terms of quantity and quality compared to the widely used platform MTurk. (2) Annotation companies' internal QA only provides marginal improvements, if any. However, improving labeling instructions instead of investing in QA can substantially boost annotation performance. (3) The benefit of internal QA depends on specific image characteristics. Our work could enable researchers to derive substantially more value from a fixed annotation budget and change the way annotation companies conduct internal QA., Comment: Accepted at ECCV 2024, preprint, Computer Vision, Data Annotation

Coherent phonons, light-induced coherent lattice vibrations in solids, provide a powerful route to engineer structural and electronic degrees of freedom using light. In this manuscript, we formulate an ab initio theory of the displacive excitation of coherent phonons (DECP), the primary mechanism for light-induced structural control in semimetals. Our study - based on the ab initio simulations of the ultrafast electron and coherent-phonon dynamics in presence of electron-phonon interactions - establishes a predictive computational framework for describing the emergence of light-induced structural changes and the ensuing transient band-structure renormalization arising from the DECP mechanism. We validate this framework via a combined theoretical and experimental investigation of coherent phonons in the elemental semimetal antimony. Via a Fourier analysis of time- and angle-resolved photoemission spectroscopy (tr-ARPES) measurements, we retrieve information about transient spectral features and quasiparticle renormalization arising from the coherent A1g phonon as a function of momentum, energy, time, and fluence. The qualitative and quantitative agreement between experiment and theory corroborates the first-principles approach formulated in this study. Besides advancing the fundamental understanding of electron-phonon interactions mediated by coherent phonons, this study opens new opportunities for predictively engineering structural and electronic degrees of freedom in semimetals via the DECP mechanism.

The upcoming 6G mobile communication standard will offer a revolutionary new feature: Integrated sensing and communication (ISAC) reuses mobile communication signals to realize multi-static radar for various applications including localization. Consequently, applied ISAC propagation research necessitates to evolve from classical monostatic radar cross section (RCS) measurement of static targets on to bistatic radar reflectivity characterization of dynamic objects. Here, we introduce our "Bistatic Radar" (BIRA) and antenna measurement facility for bistatic spherical positioning with sub-millimeter accuracy on a diameter of up to 7 m and with almost continuous frequency coverage from 0.7 up to 260 GHz. Currently, BIRA is the only bistatic measurement facility capable of unrestricted ISAC research: In addition to vector network analysis, BIRA employs advanced wideband transceiver technology with an instantaneous bandwidth of up to 4 GHz. These transceivers grant BIRA the unique ability to characterize dynamic targets in both Doppler and range, while also significantly accelerating RCS measurements of static objects., Comment: 9 pages, 8 figures

This paper explores the use of model-based offline reinforcement learning with long model rollouts. While some literature criticizes this approach due to compounding errors, many practitioners have found success in real-world applications. The paper aims to demonstrate that long rollouts do not necessarily result in exponentially growing errors and can actually produce better Q-value estimates than model-free methods. These findings can potentially enhance reinforcement learning techniques., Comment: Accepted at ESANN 2024

Physics-inspired Generative Models (GMs), in particular Diffusion Models (DMs) and Poisson Flow Models (PFMs), enhance Bayesian methods and promise great utility in medical imaging. This review examines the transformative role of such generative methods. First, a variety of physics-inspired GMs, including Denoising Diffusion Probabilistic Models (DDPMs), Score-based Diffusion Models (SDMs), and Poisson Flow Generative Models (PFGMs and PFGM++), are revisited, with an emphasis on their accuracy, robustness as well as acceleration. Then, major applications of physics-inspired GMs in medical imaging are presented, comprising image reconstruction, image generation, and image analysis. Finally, future research directions are brainstormed, including unification of physics-inspired GMs, integration with Vision-Language Models (VLMs), and potential novel applications of GMs. Since the development of generative methods has been rapid, this review will hopefully give peers and learners a timely snapshot of this new family of physics-driven generative models and help capitalize their enormous potential for medical imaging.

Neural networks typically generalize well when fitting the data perfectly, even though they are heavily overparameterized. Many factors have been pointed out as the reason for this phenomenon, including an implicit bias of stochastic gradient descent (SGD) and a possible simplicity bias arising from the neural network architecture. The goal of this paper is to disentangle the factors that influence generalization stemming from optimization and architectural choices by studying random and SGD-optimized networks that achieve zero training error. We experimentally show, in the low sample regime, that overparameterization in terms of increasing width is beneficial for generalization, and this benefit is due to the bias of SGD and not due to an architectural bias. In contrast, for increasing depth, overparameterization is detrimental for generalization, but random and SGD-optimized networks behave similarly, so this can be attributed to an architectural bias. For more information, see https://bias-sgd-or-architecture.github.io ., Comment: Accepted at ICML 2024