Ludwig-Maximilians-Universität München
Open Access LMU ( Ludwig-Maximilians-Univ. München)Not a member yet
40914 research outputs found
Sort by
Personalized deep learning auto‐segmentation models for adaptive fractionated magnetic resonance‐guided radiation therapy of the abdomen
Get PDFBackground:
Manual contour corrections during fractionated magnetic resonance (MR)-guided radiotherapy (MRgRT) are time-consuming. Conventional population models for deep learning auto-segmentation might be suboptimal for MRgRT at MR-Linacs since they do not incorporate manual segmentation from treatment planning and previous fractions.
Purpose:
In this work, we investigate patient-specific (PS) auto-segmentation methods leveraging expert-segmented planning and prior fraction MR images (MRIs) to improve auto-segmentation on consecutive treatment days.
Conclusion:
Personalized auto-segmentation models outperformed the population BMs. In most cases, PS(BM) delineations were judged to be directly usable for treatment adaptation without further corrections, suggesting a potential time saving during fractionated treatment
Exact Computation of Any-Order Shapley Interactions for Graph Neural Networks
Get PDFAlbeit the ubiquitous use of Graph Neural Networks (GNNs) in machine learning (ML) prediction tasks involving graph-structured data, their interpretability remains challenging. In explainable artificial intelligence (XAI), the Shapley Value (SV) is the predominant method to quantify contributions of individual features to a ML model’s output. Addressing the limitations of SVs in complex prediction models, Shapley Interactions (SIs) extend the SV to groups of features. In this work, we explain single graph predictions of GNNs with SIs that quantify node contributions and interactions among multiple nodes. By exploiting the GNN architecture, we show that the structure of interactions in node embeddings are preserved for graph prediction. As a result, the exponential complexity of SIs depends only on the receptive fields, i.e. the message-passing ranges determined by the connectivity of the graph and the number of convolutional layers. Based on our theoretical results, we introduce GraphSHAP-IQ, an efficient approach to compute any-order SIs exactly. GraphSHAP-IQ is applicable to popular message passing techniques in conjunction with a linear global pooling and output layer. We showcase that GraphSHAP-IQ substantially reduces the exponential complexity of computing exact SIs on multiple benchmark datasets. Beyond exact computation, we evaluate GraphSHAP-IQ’s approximation of SIs on popular GNN architectures and compare with existing baselines. Lastly, we visualize SIs of real-world water distribution networks and molecule structures using a SI-Graph
Who is at risk of bias?
Get PDFThe motivated reception of science in line with one’s preexisting convictions is a well-documented, pervasive phenomenon. In two studies (N = 743), we investigated whether this bias might be stronger in some people than others due to dispositional differences. Building on the assumptions that motivated science reception is driven by perceived threat and suspicion and higher under perceived ambiguity and uncertainty, we focused on traits associated with such perceptions. In particular, we tested the impact of conspiracy mentality and victim sensitivity on motivated science reception (as indicated by ascriptions of researchers’ trustworthiness and evidence credibility). In addition, we explored the role of broader personality traits (generalized mistrust and ambiguity intolerance) in this context. None of the investigated dispositions modulated the motivated science reception effect. This demonstrates once again, that motivated science reception is a ubiquitous challenge for the effective dissemination of science and everyone seems to be at risk of it
Parental origin of transgene modulates amyloid-β plaque burden in the 5xFAD mouse model of Alzheimer’s disease
Get PDFIn Alzheimer’s disease (AD) research, the 5xFAD mouse model is commonly used as a heterozygote crossed with other genetic models to study AD pathology. We investigated whether the parental origin of the 5xFAD transgene affects plaque deposition. Using quantitative light-sheet microscopy, we found that paternal inheritance of the transgene led to a 2-fold higher plaque burden compared with maternal inheritance, a finding consistent across multiple 5xFAD colonies. This effect was not due to gestation in or rearing by 5xFAD females. Immunoblotting suggested that transgenic inheritance modulates transgenic protein expression, potentially due to genomic imprinting of the Thy1.2 promoter. Surprisingly, fewer than 20% of 5xFAD studies report breeding schemes, suggesting that this factor might confound previous findings. Our data highlight a significant determinant of plaque burden in 5xFAD mice and underscore the importance of reporting the parental origin of the transgene to improve scientific rigor and reproducibility in AD research
Structured reporting of neuroendocrine tumors in PET/CT using [18F]SiTATE - impact on interdisciplinary communication
Get PDFOur retrospective single-center study aims to evaluate the impact of structured reporting (SR) using a self-developed template on report quality compared to free-text reporting (FTR) in [18F]SiTATE Positron Emission Tomography/Computer Tomography (PET/CT) for the primary staging and therapy monitoring of patients diagnosed with neuroendocrine tumors (NET). In total 50 patients were included. FTRs and SRs were generated post-examination. All reports were evaluated by a radiologist and a surgeon through a questionnaire to determine their contribution to facilitating clinical decision-making and to assess their completeness, linguistic quality, and overall quality. SR significantly increased the capacity of facilitating therapy decision-making from 32% in FTR to 55% in SR (p < 0.001). Trust in the report was significantly higher in SR with a mean of 5.0 (SD = 0.5) vs. 4.7 (SD = 0.5) for FTR (p < 0.001). SR received significantly higher mean ratings regarding linguistic quality with 4.7 for SR vs. 4.4 for FTR (p = 0.004) and overall report quality with a mean of 4.9 for SR vs. 4.6 for FTR (p < 0.001). Concluding that SR enhances the overall quality of reports in [18F]SiTATE-PET/CTs for NET staging, serving as a tool to streamline clinical decision-making and enhance interdisciplinary communication in the future
Serum neurofilament light chain predicts disease severity in axonal variants of acute immune neuropathie
Get PDFBackground and purpose : The purpose was to explore the prognostic utility of neurofilament light chain (NfL) in patients with immune-mediated polyradiculoneuropathies (IMPs). Methods : This retrospective monocentric study analysed serum and cerebrospinal fluid samples from patients diagnosed with IMP collected prior to treatment initiation. NfL concentrations were correlated with clinical outcomes, including F score and hospitalization duration. Results : Amongst 115 IMP patients tested, baseline cerebrospinal fluid and serum NfL (sNfL) concentrations were higher in acute inflammatory axonal polyradiculoneuropathy (AIAP) than other IMP variants. In the AIAP cohort, a positive correlation was observed between baseline sNfL concentrations, F score and hospitalization duration. Multivariate linear regression analysis further supported the predictive relationship between elevated baseline sNfL concentrations and clinical outcomes. Using receiver operating characteristic analysis, a cut-off value for sNfL of 351 pg/mL was found to predict an F score >3 in AIAP with a sensitivity of 40% and specificity of 81.8%. AIAP patients with sNfL concentrations above this threshold required longer hospitalization (extended by 15 days). Discussion : Our findings highlight the potential of baseline sNfL as an effective marker for distinguishing between IMP variants and predicting the prognosis of AIAP. Further validation may facilitate translation of sNfL into clinical practice, potentially identifying high-risk patients for tailored treatment approaches
Diagnosis and treatment of autonomic failure, pain and sleep disturbances in Parkinson’s disease: guideline “Parkinson’s disease” of the German Society of Neurology
Get PDFFemale sex is linked to a stronger association between sTREM2 and CSF p-tau in Alzheimer’s disease
Get PDFProtein misfolding: understanding biology to classify and treat synucleinopathies
No full textProtein misfolding and aggregation is a major pathological hallmark in a variety of human conditions, including cancer, diabetes, and neurodegeneration. However, we still do not fully understand the role of protein accumulation in disease. Interestingly, recent breakthroughs in artificial intelligence (AI) are having a tremendous impact on our ability to predict three-dimensional protein structures and understand the molecular rules governing protein folding/misfolding. This progress will enable us to understand how intrinsic and extrinsic factors trigger protein misfolding, thereby changing protein function. These changes, in some cases, are related to normal biological responses and, in other cases, associated with pathological alterations, such as those found in many neurodegenerative disorders. Here, we provide a brief historical perspective of how findings in the field of prion diseases and prion biology have enabled tremendous advances that are now forming the basis for our understanding of disease processes and discuss how this knowledge is now emerging as central for our ability to classify, diagnose, and treat devastating neurodegenerative disorders such as Parkinson’s and Alzheimer’s diseases
