Search

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

Recent studies have underscored the capabilities of natural imaging foundation models to serve as powerful feature extractors, even in a zero-shot setting for medical imaging data. Most commonly, a shallow multi-layer perceptron (MLP) is appended to the feature extractor to facilitate end-to-end learning and downstream prediction tasks such as classification, thus representing the de facto standard. However, as graph neural networks (GNNs) have become a practicable choice for various tasks in medical research in the recent past, we direct attention to the question of how effective GNNs are compared to MLP prediction heads for the task of 3D medical image classification, proposing them as a potential alternative. In our experiments, we devise a subject-level graph for each volumetric dataset instance. Therein latent representations of all slices in the volume, encoded through a DINOv2 pretrained vision transformer (ViT), constitute the nodes and their respective node features. We use public datasets to compare the classification heads numerically and evaluate various graph construction and graph convolution methods in our experiments. Our findings show enhancements of the GNN in classification performance and substantial improvements in runtime compared to an MLP prediction head. Additional robustness evaluations further validate the promising performance of the GNN, promoting them as a suitable alternative to traditional MLP classification heads. Our code is publicly available at: https://github.com/compai-lab/2024-miccai-grail-kiechle, Comment: Accepted at MICCAI 2024 - GRAIL Workshop

In this work, we introduce Progressive Growing of Patch Size, a resource-efficient implicit curriculum learning approach for dense prediction tasks. Our curriculum approach is defined by growing the patch size during model training, which gradually increases the task's difficulty. We integrated our curriculum into the nnU-Net framework and evaluated the methodology on all 10 tasks of the Medical Segmentation Decathlon. With our approach, we are able to substantially reduce runtime, computational costs, and CO2 emissions of network training compared to classical constant patch size training. In our experiments, the curriculum approach resulted in improved convergence. We are able to outperform standard nnU-Net training, which is trained with constant patch size, in terms of Dice Score on 7 out of 10 MSD tasks while only spending roughly 50% of the original training runtime. To the best of our knowledge, our Progressive Growing of Patch Size is the first successful employment of a sample-length curriculum in the form of patch size in the field of computer vision. Our code is publicly available at https://github.com/compai-lab/2024-miccai-fischer., Comment: Accepted at MICCAI2024; Changes for Camera-Ready-Version for MICCAI2024 (missing in this arxiv submission): Replaced T-Test with Wilcoxon Signed Ranked Test, as DSC samples are not normally distributed => now only significant improvements and no significant decreases in performance for PGPS/PGPS+

Pathological lymph node delineation is crucial in cancer diagnosis, progression assessment, and treatment planning. The MICCAI 2023 Lymph Node Quantification Challenge published the first public dataset for pathological lymph node segmentation in the mediastinum. As lymph node annotations are expensive, the challenge was formed as a weakly supervised learning task, where only a subset of all lymph nodes in the training set have been annotated. For the challenge submission, multiple methods for training on these weakly supervised data were explored, including noisy label training, loss masking of unlabeled data, and an approach that integrated the TotalSegmentator toolbox as a form of pseudo labeling in order to reduce the number of unknown voxels. Furthermore, multiple public TCIA datasets were incorporated into the training to improve the performance of the deep learning model. Our submitted model achieved a Dice score of 0.628 and an average symmetric surface distance of 5.8~mm on the challenge test set. With our submitted model, we accomplished third rank in the MICCAI2023 LNQ challenge. A finding of our analysis was that the integration of all visible, including non-pathological, lymph nodes improved the overall segmentation performance on pathological lymph nodes of the test set. Furthermore, segmentation models trained only on clinically enlarged lymph nodes, as given in the challenge scenario, could not generalize to smaller pathological lymph nodes. The code and model for the challenge submission are available at \url{https://gitlab.lrz.de/compai/MediastinalLymphNodeSegmentation}., Comment: Accepted for publication at the Journal of Machine Learning for Biomedical Imaging (MELBA) https://melba-journal.org/2024:008

Regression testing is crucial in ensuring that pure code refactoring does not adversely affect existing software functionality, but it can be expensive, accounting for half the cost of software maintenance. Automated test case generation reduces effort but may generate weak test suites. Test amplification is a promising solution that enhances tests by generating additional or improving existing ones, increasing test coverage, but it faces the test oracle problem. To address this, we propose a test oracle derivation approach that uses object state data produced during System Under Test (SUT) test execution to amplify regression test oracles. The approach monitors the object state during test execution and compares it to the previous version to detect any changes in relation to the SUT's intended behaviour. Our preliminary evaluation shows that the proposed approach can enhance the detection of behaviour changes substantially, providing initial evidence of its effectiveness., Comment: 8 pages, 1 figure

Identifying and selecting high-quality Metamorphic Relations (MRs) is a challenge in Metamorphic Testing (MT). While some techniques for automatically selecting MRs have been proposed, they are either domain-specific or rely on strict assumptions about the applicability of a pre-defined MRs. This paper presents a preliminary evaluation of MetaTrimmer, a method for selecting and constraining MRs based on test data. MetaTrimmer comprises three steps: generating random test data inputs for the SUT (Step 1), performing test data transformations and logging MR violations (Step 2), and conducting manual inspections to derive constraints (Step 3). The novelty of MetaTrimmer is its avoidance of complex prediction models that require labeled datasets regarding the applicability of MRs. Moreover, MetaTrimmer facilitates the seamless integration of MT with advanced fuzzing for test data generation. In a preliminary evaluation, MetaTrimmer shows the potential to overcome existing limitations and enhance MR effectiveness., Comment: 8 pages, 3 figures

We investigate the relaxation dynamics in a chiral one-dimensional quantum channel with finite range interactions, driven out of equilibrium by the injection of high-energy electrons. While the distribution of high-energy electrons, after dissipation of some of their energy, has been examined previously (arXiv:2108.00685), we study the distribution of charge carriers excited from the channel's Fermi sea during this process. Utilizing a detector to measure the energetic imprint in the Fermi sea downstream of the injection point, we discover an initial symmetry in the distribution of excited electrons and holes relative to the Fermi level. However, this symmetry breaks down with stronger interactions and increased propagation distances, attributed to terms of order four and beyond in the interaction. We provide an intuitive interpretation of these results in terms of interference between states with different numbers of plasmons in the Fermi sea., Comment: 7 pages, 8 figures

Metamorphic Testing (MT) is a testing technique that can effectively alleviate the oracle problem. MT uses Metamorphic Relations (MRs) to determine if a test case passes or fails. MRs specify how the outputs should vary in response to specific input changes when executing the System Under Test (SUT). If a particular MR is violated for at least one test input (and its change), there is a high probability that the SUT has a fault. On the other hand, if a particular MR is not violated, it does not guarantee that the SUT is fault free. However, deciding if the MR is being violated due to a bug or because the MR does not hold/fit for particular conditions generated by specific inputs remains a manual task and unexplored. In this paper, we develop a method for refining MRs to offer hints as to whether a violation results from a bug or arises from the MR not being matched to certain test data under specific circumstances. In our initial proof-of-concept, we derive the relevant information from rules using the Association Rule Mining (ARM) technique. In our initial proof-of-concept, we validate our method on a toy example and discuss the lessons learned from our experiments. Our proof-of-concept demonstrates that our method is applicable and that we can provide suggestions that help strengthen the test suite for regression testing purposes., Comment: 8 pages, 5 figures, 5 tables

Background: Deep learning (DL)-based head and neck lymph node level (HN_LNL) autodelineation is of high relevance to radiotherapy research and clinical treatment planning but still underinvestigated in academic literature. Methods: An expert-delineated cohort of 35 planning CTs was used for training of an nnU-net 3D-fullres/2D-ensemble model for autosegmentation of 20 different HN_LNL. A second cohort acquired at the same institution later in time served as the test set (n=20). In a completely blinded evaluation, 3 clinical experts rated the quality of DL autosegmentations in a head-to-head comparison with expert-created contours. For a subgroup of 10 cases, intraobserver variability was compared to the average DL autosegmentation accuracy on the original and recontoured set of expert segmentations. A postprocessing step to adjust craniocaudal boundaries of level autosegmentations to the CT slice plane was introduced and the effect on geometric accuracy and expert rating was investigated. Results: Blinded expert ratings for DL segmentations and expert-created contours were not significantly different. DL segmentations with slice plane adjustment were rated numerically higher (mean, 81.0 vs. 79.6,p=0.185) and DL segmentations without slice plane adjustment were rated numerically lower (77.2 vs. 79.6,p=0.167) than manually drawn contours. DL segmentations with CT slice plane adjustment were rated significantly better than DL contours without slice plane adjustment (81.0 vs. 77.2,p=0.004). Geometric accuracy of DL segmentations was not different from intraobserver variability (mean, 0.76 vs. 0.77, p=0.307). Conclusions: We show that a nnU-net 3D-fullres/2D-ensemble model can be used for highly accurate autodelineation of HN_LNL using only a limited training dataset that is ideally suited for large-scale standardized autodelineation of HN_LNL in the research setting., Comment: 14 pages, 6 figures, published in Frontiers in Oncology

Metamorphic testing (TM) examines the relations between inputs and outputs of test runs. These relations are known as metamorphic relations (MR). Currently, MRs are handpicked and require in-depth knowledge of the System Under Test (SUT), as well as its problem domain. As a result, the identification and selection of high-quality MRs is a challenge. \citeauthor{PMR1} suggested the Predicting Metamorphic Relations (PMR) approach for automatic prediction of applicable MRs picked from a predefined list. PMR is based on a Support Vector Machine (SVM) model using features derived from the Control Flow Graphs (CFGs) of 100 Java methods. The original study of \citeauthor{PMR1} showed encouraging results, but developing classification models from CFG-related features is costly. In this paper, we aim at developing a PMR approach that is less costly without losing performance. We complement the original PMR approach by considering other than CFG-related features. We define 21 features that can be directly extracted from source code and build several classifiers, including SVM models. Our results indicate that using the original CFG-based method-level features, in particular for a SVM with random walk kernel (RWK), achieve better predictions in terms of AUC-ROC for most of the candidate MRs than our models. However, for one of the candidate MRs, using source code features achieved the best AUC-ROC result (greater than 0.8)., Comment: 8 pages, 6 tables, 3 figures

Due to data privacy constraints, data sharing among multiple centers is restricted. Continual learning, as one approach to peer-to-peer federated learning, can promote multicenter collaboration on deep learning algorithm development by sharing intermediate models instead of training data. This work aims to investigate the feasibility of continual learning for multicenter collaboration on an exemplary application of brain metastasis identification using DeepMedic. 920 T1 MRI contrast enhanced volumes are split to simulate multicenter collaboration scenarios. A continual learning algorithm, synaptic intelligence (SI), is applied to preserve important model weights for training one center after another. In a bilateral collaboration scenario, continual learning with SI achieves a sensitivity of 0.917, and naive continual learning without SI achieves a sensitivity of 0.906, while two models trained on internal data solely without continual learning achieve sensitivity of 0.853 and 0.831 only. In a seven-center multilateral collaboration scenario, the models trained on internal datasets (100 volumes each center) without continual learning obtain a mean sensitivity value of 0.699. With single-visit continual learning (i.e., the shared model visits each center only once during training), the sensitivity is improved to 0.788 and 0.849 without SI and with SI, respectively. With iterative continual learning (i.e., the shared model revisits each center multiple times during training), the sensitivity is further improved to 0.914, which is identical to the sensitivity using mixed data for training. Our experiments demonstrate that continual learning can improve brain metastasis identification performance for centers with limited data. This study demonstrates the feasibility of applying continual learning for peer-to-peer federated learning in multicenter collaboration.

DNA-based nanonetworks have a wide range of promising use cases, especially in the field of medicine. With a large set of agents, a partially observable stochastic environment, and noisy observations, such nanoscale systems can be modelled as a decentralised, partially observable, Markov decision process (DecPOMDP). As the agent set is a dominating factor, this paper presents (i) lifted DecPOMDPs, partitioning the agent set into sets of indistinguishable agents, reducing the worst-case space required, and (ii) a nanoscale medical system as an application. Future work turns to solving and implementing lifted DecPOMDPs., Comment: Accepted at the Tenth International Workshop on Statistical Relational AI (StarAI-2021)

Ergodic many-body systems are expected to reach quasi-thermal equilibrium. Here we demonstrate that, surprisingly, high-energy electrons, which are injected into a quantum Hall edge mode with finite range interactions, stabilize at a far-from-thermalized state over a long time scale. To detect this non-equilibrium state, one positions an energy-resolved detector downstream of the point of injection. So far, non-equilibrium distributions in integrable systems were either found not to display relaxation at all, or generically relax to near-thermal asymptotic states. In stark contrast, the here-obtained many-body state comprises fast-decaying transient components, followed by a nearly frozen distribution with a peak near the injection energy., Comment: 11 pages, 4 figures

Nonradiative processes limit optoelectronic functionality of nanocrystals and curb their device performance. Nevertheless, the dynamic structural origins of nonradiative relaxations in nanocrystals are not understood. Here, femtosecond electron diffraction measurements corroborated by atomistic simulations uncover transient lattice deformations accompanying radiationless electronic processes in semiconductor nanocrystals. Investigation of the excitation energy dependence shows that hot carriers created by a photon energy considerably larger than the bandgap induce structural distortions at nanocrystal surfaces on few picosecond timescales associated with the localization of trapped holes. On the other hand, carriers created by a photon energy close to the bandgap result in transient lattice heating that occurs on a much longer 200 ps timescale, governed by an Auger heating mechanism. Elucidation of the structural deformations associated with the surface trapping of hot holes provides atomic-scale insights into the mechanisms deteriorating optoelectronic performance and a pathway towards minimizing these losses in nanocrystal devices., Comment: 17 pages, 4 figures

Software testing helps developers to identify bugs. However, awareness of bugs is only the first step. Finding and correcting the faulty program components is equally hard and essential for high-quality software. Fault localization automatically pinpoints the location of an existing bug in a program. It is a hard problem, and existing methods are not yet precise enough for widespread industrial adoption. We propose fault localization via Probabilistic Software Modeling (PSM). PSM analyzes the structure and behavior of a program and synthesizes a network of Probabilistic Models (PMs). Each PM models a method with its inputs and outputs and is capable of evaluating the likelihood of runtime data. We use this likelihood evaluation to find fault locations and their impact on dependent code elements. Results indicate that PSM is a robust framework for accurate fault localization., Comment: 4 pages, 3 figures, 2 tables, 3rd Workshop on Validation, Analysis and Evolution of Software Tests

Background/Introduction: Concurrent electroencephalography and resting-state functional magnetic resonance imaging (rsfMRI) have been widely used for studying the (presumably) awake and alert human brain with high temporal/spatial resolution. Although rsfMRI scans are typically collected while individuals are instructed to focus their eyes on a fixated cross, objective and verified experimental measures to quantify degree of vigilance are not readily available. Electroencephalography (EEG) is the modality extensively used for estimating vigilance, especially during eyes-closed resting state. However, pupil size measured using an eye-tracker device could provide an indirect index of vigilance. Methods: Three 12-min resting scans (eyes open, fixating on the cross) were collected from 10 healthy control participants. We simultaneously collected EEG, fMRI, physiological, and eye-tracker data and investigated the correlation between EEG features, pupil size, and heart rate. Furthermore, we used pupil size and EEG features as regressors to find their correlations with blood-oxygen-level-dependent fMRI measures. Results: EEG frontal and occipital beta power (FOBP) correlates with pupil size changes, an indirect index for locus coeruleus activity implicated in vigilance regulation (r = 0.306, p < 0.001). Moreover, FOBP also correlated with heart rate (r = 0.255, p < 0.001), as well as several brain regions in the anticorrelated network, including the bilateral insula and inferior parietal lobule. Discussion: In this study, we investigated whether simultaneous EEG-fMRI combined with eye-tracker measurements can be used to determine EEG signal feature associated with vigilance measures during eyes-open rsfMRI. Our results support the conclusion that FOBP is an objective measure of vigilance in healthy human subjects.

A novel mesoscopic electron spectrometer allows for the probing of relaxation processes in quantum Hall edge channels. The device is composed of an emitter quantum dot that injects energy-resolved electrons into the channel closest to the sample edge, to be subsequently probed downstream by a detector quantum dot of the same type. In addition to inelastic processes in the sample that stem from interactions inside the region between the quantum dot energy filters (inner region), anomalous signals are measured when the detector energy exceeds the emitter energy. Considering finite range Coulomb interactions in the sample, we find that energy exchange between electrons in the current inducing source channel and the inner region, similar to Auger recombination processes, is responsible for such anomalous currents. In addition, our perturbative treatment of interactions shows that electrons emitted from the source, which dissipate energy to the inner region before entering the detector, contribute to the current most strongly when emitter-detector energies are comparable. Charge transfer in which the emitted electron is exchanged for a charge carrier from the Fermi sea, on the other hand, preferentially occurs close to the Fermi level.

Electron microscopy (EM) has been instrumental in our understanding of biological systems ranging from subcellular structures to complex organisms. Although EM reveals cellular morphology with nanoscale resolution, it does not provide information on the location of proteins within a cellular context. An EM-based bioimaging technology capable of localizing individual proteins and resolving protein-protein interactions with respect to cellular ultrastructure would provide important insights into the molecular biology of a cell. Here, we report on the development of luminescent nanoprobes potentially suitable for labeling biomolecules in a multicolor EM modality. In this approach, the labels are based on lanthanide-doped nanoparticles that emit light under electron excitation in a process known as cathodoluminescence (CL). Our results suggest that the optimization of nanoparticle composition, synthesis protocols and electron imaging conditions could enable high signal-to-noise localization of biomolecules with a sub-20-nm resolution, limited only by the nanoparticle size. In ensemble measurements, these luminescent labels exhibit narrow spectra of nine distinct colors that are characteristic of the corresponding rare-earth dopant type.

Engineering shape and interactions of nanoscopic building blocks allows for the assembly of rationally designed macroscopic three-dimensional (3D) materials with spatial accuracy inaccessible to top-down fabrication methods. Owing to its sequence-specific interaction, DNA is often used as selective binder to connect metallic nanoparticles into highly ordered lattices. Moreover, 3D crystals assembled entirely from DNA have been proposed and implemented with the declared goal to arrange guest molecules in predefined lattices. This requires design schemes that provide high rigidity and sufficiently large open guest space. We here present a DNA origami-based tensegrity triangle structure that assembles into a 3D rhombohedral crystalline lattice. We site-specifically place 10 nm and 20 nm gold particles within the lattice, demonstrating that our crystals are spacious enough to host e.g. ribosome-sized macromolecules. We validate the accurate assembly of the DNA origami lattice itself as well as the precise incorporation of gold particles by electron microscopy and small angle X-ray scattering (SAXS) experiments. Our results show that it is possible to create DNA building blocks that assemble into lattices with customized geometry. Site-specific hosting of nano objects in the transparent DNA lattice sets the stage for metamaterial and structural biology applications., Comment: 21 pages; 4 figures; supplementary information is included