Zuse Institute Berlin

This article reports the results of the second iteration of the autoPET challenge on automated lesion segmentation in whole-body PET/CT, held in conjunction with the 26th International Conference on Medical Image Computing and Computer Assisted Intervention in 2023. In contrast to the first autoPET challenge, which served as a proof of concept, this study investigates whether machine learning-based segmentation models trained on data from a single source can maintain performance across clinically relevant variations in PET/CT data, reflecting the demands of real-world deployment. Methods: A comprehensive biomedical segmentation challenge on PET/CT domain generalization was designed and conducted. Participants were tasked to train machine learning models on annotated whole-body 18F-FDG data (n = 1,014). These models were then evaluated on a test set of 200 samples from 5 clinically relevant domains, including variations in institutions, pathologies, and populations and a different tracer. Performance was measured in terms of average dice similarity coefficient, average false-positive volume, and average false-negative volume. The best-performing teams were awarded in 3 categories. Furthermore, a detailed analysis was conducted after the challenge, examining results across domains and unique instances, along with a ranking analysis. Results: Generalization from a single-source domain remains a significant challenge. Seventeen international teams successfully participated in the challenge. The best-performing team reached an average dice similarity coefficient of 0.5038, a mean false-positive volume of 87.8388 mL, and a mean false-negative volume of 8.4154 mL on the test set. nnU-Net was the most commonly used framework, with most participants using a 3-dimensional U-Net. Despite competitive in-domain results, out-of-domain performance deteriorated substantially, particularly on pediatric and prostate-specific membrane antigen data. Detailed error analysis revealed frequent false-positives due to physiologic uptake and decreased sensitivity in detecting small or low-uptake lesions. A majority-vote ensemble offered minimal performance gains, whereas an oracle ensemble indicates hypothetical gains. Ranking analysis showed no single team consistently outperformed all others across ranking schemes. Conclusion: The second autoPET challenge provides a comprehensive evaluation of the current state of automated PET/CT tumor segmentation, highlighting both progress and persistent challenges of single-source domain generalization and the need for diverse public datasets to enhance algorithm robustness.

How embryos adapt their internal cellular machinery to reductions in cell size during development remains a fundamental question in cell biology1–11. Here, we use high-resolution lattice light-sheet fluorescence microscopy and automated image analysis to quantify lineage-resolved mitotic spindle and chromosome segregation dynamics from the 2- to 64-cell stages in Caenorhabditis elegans embryos. While spindle length scales with cell size across both wild-type and size-perturbed embryos, chromosome segregation dynamics remain largely invariant, suggesting that distinct mechanisms govern these mitotic processes. Combining femtosecond laser ablation12,13 with large-scale electron tomography14, we find that central spindle microtubules mediate chromosome segregation dynamics and remain uncoupled from cell size across all stages of early development. In contrast, spindle elongation is driven by cortically anchored motor proteins and astral microtubules, rendering it sensitive to cell size12,13,15–17. Incorporating these experimental results into an extended stoichiometric model for both the spindle and chromosomes, we find that allowing only cell size and microtubule catastrophe rates to vary reproduces elongation dynamics across development. The same model also accounts for centrosome separation and pronuclear positioning in the one-cell C. elegans embryo18, spindle-length scaling across nematode species spanning ~100 million years of divergence17, and spindle rotation in human cells19. Thus, a unified stoichiometric framework provides a predictive, mechanistic account of spindle and nuclear dynamics across scales and species.

This paper introduces a comprehensive approach to building natural-sounding Urdu Text-to-Speech (TTS) and voice cloning systems, addressing the lack of computational resources for Urdu. We developed a large-scale dataset of over 100 h of Urdu speech, carefully cleaned and phonetically aligned through an automated transcription pipeline to preserve linguistic accuracy. The dataset was then used to fine-tune Tacotron2, a neural network model originally trained for English, with modifications tailored to Urdu’s phonological and morphological features. To further enhance naturalness, we integrated voice cloning techniques that capture regional accents and produce personalized speech outputs. Model performance was evaluated through mean opinion score (MOS), word error rate (WER), and speaker similarity, showing substantial improvements compared to previous Urdu systems. The results demonstrate clear progress toward natural and intelligible Urdu speech synthesis, while also revealing challenges such as handling dialectal variation and preventing model overfitting. This work contributes an essential resource and methodology for advancing Urdu natural language processing (NLP), with promising applications in education, accessibility, entertainment, and assistive technologies.

Abstract Background Searching large genomic data sets for local alignments poses a computational challenge. A particular obstacle is the handling of repetitive sequences that appear in various contexts and incur a high runtime cost. For practical homology search, it is important to develop a specific but sensitive filter. Good filters reduce the search space before alignment without missing significant matches. Results We introduce DREAM-Stellar, a parallelized, updated version of the pairwise local aligner Stellar. The new aligner, DREAM-Stellar, is composed of four steps: preprocessing the queries and references, building a data structure for distributing the queries, computing in parallel the results and finally combining them. For distributing the queries we use the IBF data structure and a new prefilter for local alignments. We present our comparison of five local aligners on simulated and real genomic data and conclude that heuristic tools like BLAST miss a large percentage of significant local alignments or “drown” them in millions of less significant matches. The new version of Stellar is up to 900 times faster than its predecessor and can find all alignments between a pair of genomes in minutes. With that, the runtime of DREAM-Stellar is on par with tools like BLAST etc. Conclusions DREAM-Stellar is very practical and fast on very long sequences which makes it a suitable new tool for finding local alignments between genomic sequences under the edit distance model. The software is freely available for Linux and Mac OS X at https://github.com/seqan/dream-stellar

In this chapter, we introduce the simplest variant of the family of Frank–Wolfe algorithms, which serves as a foundation for all later versions. The final section of the chapter provides a minimal number of improvements, which are the most influential for further variants.

Abstract Excitonics is an emerging field focused on exploiting and manipulating excitons generated through light-matter interactions. Advancing the field into X-ray excitonics requires precise energy and time control of core-exciton resonances, enabling non-linear X-ray phenomena such as element-specific X-ray transient gratings, and advancing material characterization. To achieve these objectives, it is essential to comprehend the role of many-body effects governing core-exciton dynamics. In this work, we address this challenge by combining experiments with an ab initio approach specifically developed to interpret pump-probe excitations. Applied to the prototypical wide-bandgap semiconductor ZnO, first-principles calculations reproduce experimental results and unveil how the density and distribution of photoexcited carriers dynamically tune Coulomb screening, thereby controlling core-exciton binding energies, while Pauli blocking remains negligible. These insights inform a method for dynamically controlling core-exciton resonances at absorption edges, achieving either a uniform spectral blue shift caused by thermalized carrier distributions on picosecond timescales, or distinct blue shifts for individual resonances, driven by time-dependent carrier distributions on femtosecond timescales.

Abstract Quantum entanglement is the essential resource for quantum communication and distributed information processing in a quantum network. However, the remote generation over a network suffers from inevitable transmission loss and other technical difficulties. This paper introduces the concept of entanglement buffers as a potential primitive for preparing long-distance entanglement. We investigate the filling of entanglement buffers with either one Bell state or a stream of Bell states via SWAP interactions. We illustrate their resilience to imperfect interactions, noise, and losses, making the buffers suitable for a realistic quantum network scenario. Additionally, larger entanglement buffers can always enhance these benefits.

We explore how warm-starting strategies can be integrated into scalarization-based approaches for multi-objective optimization in (mixed) integer linear programming. Scalarization methods remain widely used classical techniques to compute Pareto-optimal solutions in applied settings. They are favored due to their algorithmic simplicity and broad applicability across continuous and integer programs with an arbitrary number of objectives. While warm-starting has been applied in this context before, a systematic methodology and analysis remain lacking. We address this gap by providing a theoretical characterization of warm-starting within scalarization methods, focusing on the sequencing of subproblems. However, optimizing the order of subproblems to maximize warm-start efficiency may conflict with alternative criteria, such as early identification of infeasible regions. We quantify these trade-offs through an extensive computational study.

We present a unified framework for multipartite entanglement characterization based on the conditional gradient (CG) method, incorporating both fast heuristic detection and rigorous witness construction with numerical error control. Our method enables entanglement certification in quantum systems of up to ten qubits and applies to arbitrary entanglement structures. We demonstrate its power by closing the gap between entanglement and separability bounds in white noise robustness benchmarks for a class of bound entangled states. Furthermore, the framework extends to entanglement robustness under general quantum noise channels, providing accurate thresholds in cases beyond the reach of previous algorithmic methods. These results position CG methods as a powerful tool for practical and scalable entanglement analysis in realistic experimental settings.