A5091689170- Osteoarthritis Treatment and Mechanisms
- Radiomics and Machine Learning in Medical Imaging
- Knee injuries and reconstruction techniques
- Total Knee Arthroplasty Outcomes
- Artificial Intelligence in Healthcare and Education
- Lower Extremity Biomechanics and Pathologies
- AI in cancer detection
- Advanced MRI Techniques and Applications
- Bone fractures and treatments
- COVID-19 diagnosis using AI
- Rheumatoid Arthritis Research and Therapies
- Orthopedic Surgery and Rehabilitation
- Hip and Femur Fractures
- MRI in cancer diagnosis
- Spondyloarthritis Studies and Treatments
- Sports injuries and prevention
- Optical Coherence Tomography Applications
- Machine Learning in Healthcare
- Shoulder Injury and Treatment
- Orthopaedic implants and arthroplasty
- Medical Imaging Techniques and Applications
- Advanced X-ray and CT Imaging
- Bone and Joint Diseases
- Generative Adversarial Networks and Image Synthesis
- Spine and Intervertebral Disc Pathology
Universitätsklinikum Aachen
2016-2025
University Hospital Carl Gustav Carus
2025
RWTH Aachen University
2015-2024
Friedrich-Alexander-Universität Erlangen-Nürnberg
2024
Heidelberg University
2024
University Hospital Heidelberg
2024
National Center for Tumor Diseases
2024
Düsseldorf University Hospital
2019-2023
Heinrich Heine University Düsseldorf
2019-2023
Hochschule Düsseldorf University of Applied Sciences
2020
Purpose To determine the diagnostic accuracy for clinically significant prostate cancer achieved with abbreviated biparametric magnetic resonance (MR) imaging in comparison full multiparametric contrast material–enhanced MR men elevated prostate-specific antigen (PSA) and negative transrectal ultrasonography (US)–guided biopsy findings; to detection rate of versus contrast-enhanced between-reader agreement interpretation imaging. Materials Methods In this institutional review board–approved...
Abstract Recent advances in computer vision have shown promising results image generation. Diffusion probabilistic models generated realistic images from textual input, as demonstrated by DALL-E 2, Imagen, and Stable Diffusion. However, their use medicine, where imaging data typically comprises three-dimensional volumes, has not been systematically evaluated. Synthetic may play a crucial role privacy-preserving artificial intelligence can also be used to augment small datasets. We show that...
Abstract Although generative adversarial networks (GANs) can produce large datasets, their limited diversity and fidelity have been recently addressed by denoising diffusion probabilistic models, which demonstrated superiority in natural image synthesis. In this study, we introduce Medfusion, a conditional latent DDPM designed for medical generation, evaluate its performance against GANs, currently represent the state-of-the-art. Medfusion was trained compared with StyleGAN-3 using...
Background Deep learning (DL) models can potentially improve prognostication of rectal cancer but have not been systematically assessed. Purpose To develop and validate an MRI DL model for predicting survival in patients with based on segmented tumor volumes from pretreatment T2-weighted scans. Materials Methods were trained validated retrospectively collected scans diagnosed between August 2003 April 2021 at two centers. Patients excluded the study if there concurrent malignant neoplasms,...
Abstract Large language models (LLMs) have shown potential in various applications, including clinical practice. However, their accuracy and utility providing treatment recommendations for orthopedic conditions remain to be investigated. Thus, this pilot study aims evaluate the validity of generated by GPT-4 common knee shoulder using anonymized MRI reports. A retrospective analysis was conducted 20 reports, with varying severity complexity. Treatment were elicited from evaluated two...
A knowledge gap persists between machine learning (ML) developers (e.g., data scientists) and practitioners clinicians), hampering the full utilization of ML for clinical analysis. We investigated potential ChatGPT Advanced Data Analysis (ADA), an extension GPT-4, to bridge this perform analyses efficiently. Real-world datasets study details from large trials across various medical specialties were presented ADA without specific guidance. autonomously developed state-of-the-art models based...
This study compares 2 large language models and their performance vs that of competing open-source models.
Abstract Deep learning applied to whole‐slide histopathology images (WSIs) has the potential enhance precision oncology and alleviate workload of experts. However, developing these models necessitates large amounts data with ground truth labels, which can be both time‐consuming expensive obtain. Pathology reports are typically unstructured or poorly structured texts, efforts implement reporting templates have been unsuccessful, as lead perceived extra workload. In this study, we hypothesised...
Background Reducing the amount of contrast agent needed for contrast-enhanced breast MRI is desirable. Purpose To investigate if generative adversarial networks (GANs) can recover scans from unenhanced images and virtual low-contrast-enhanced images. Materials Methods In this retrospective study performed January 2010 to December 2019, simulated low-contrast were produced by adding noise existing GANs then trained (approach A) or T1- T2-weighted B). Two experienced radiologists tasked with...
Background Clinicians consider both imaging and nonimaging data when diagnosing diseases; however, current machine learning approaches primarily from a single modality. Purpose To develop neural network architecture capable of integrating multimodal patient compare its performance to models incorporating modality for up 25 pathologic conditions. Materials Methods In this retrospective study, were extracted the Medical Information Mart Intensive Care (MIMIC) database an internal comprised...
Artificial intelligence (AI) has a multitude of applications in cancer research and oncology. However, the training AI systems is impeded by limited availability large datasets due to data protection requirements other regulatory obstacles. Federated swarm learning represent possible solutions this problem collaboratively models while avoiding transfer. these decentralized methods, weight updates are still transferred aggregation server for merging models. This leaves possibility breach...
Artificial intelligence (AI) models are increasingly used in the medical domain. However, as data is highly sensitive, special precautions to ensure its protection required. The gold standard for privacy preservation introduction of differential (DP) model training. Prior work indicates that DP has negative implications on accuracy and fairness, which unacceptable medicine represent a main barrier widespread use privacy-preserving techniques. In this work, we evaluated effect training AI...
Large language models (LLMs) have shown potential in radiology, but their ability to aid radiologists interpreting imaging studies remains unexplored. We investigated the effects of a state-of-the-art LLM (GPT-4) on radiologists' diagnostic workflow.
To develop and validate a deep learning-based method for automatic quantitative analysis of lower-extremity alignment.In this retrospective study, bilateral long-leg radiographs (LLRs) from 255 patients that were obtained between January September 2018 included. For training data (n = 109), U-Net convolutional neural network was trained to segment the femur tibia versus manual segmentation. validation 40), model parameters optimized. Following identification anatomic landmarks, mechanical...
Computer vision (CV) has the potential to change medicine fundamentally. Expert knowledge provided by CV can enhance diagnosis. Unfortunately, existing algorithms often remain below expectations, as databases used for training are usually too small, incomplete, and heterogeneous in quality. Moreover, data protection is a serious obstacle exchange of data. To overcome this limitation, we propose use generative models (GMs) produce high-resolution synthetic radiographs that do not contain any...
Due to the rapid advancements in recent years, medical image analysis is largely dominated by deep learning (DL). However, building powerful and robust DL models requires training with large multi-party datasets. While multiple stakeholders have provided publicly available datasets, ways which these data are labeled vary widely. For Instance, an institution might provide a dataset of chest radiographs containing labels denoting presence pneumonia, while another focus on determining...
When clinicians assess the prognosis of patients in intensive care, they take imaging and non-imaging data into account. In contrast, many traditional machine learning models rely on only one these modalities, limiting their potential medical applications. This work proposes evaluates a transformer-based neural network as novel AI architecture that integrates multimodal patient data, i.e., (chest radiographs) (clinical data). We evaluate performance our model retrospective study with 6,125...
The success of Deep Learning applications critically depends on the quality and scale underlying training data. Generative adversarial networks (GANs) can generate arbitrary large datasets, but diversity fidelity are limited, which has recently been addressed by denoising diffusion probabilistic models (DDPMs) whose superiority demonstrated natural images. In this study, we propose Medfusion, a conditional latent DDPM for medical We compare our DDPM-based model against GAN-based models,...
Pre-training datasets, like ImageNet, have become the gold standard in medical image analysis. However, emergence of self-supervised learning (SSL), which leverages unlabeled data to learn robust features, presents an opportunity bypass intensive labeling process. In this study, we explored if SSL for pre-training on non-medical images can be applied chest radiographs and how it compares supervised images. We utilized a vision transformer initialized its weights based (i) natural (DINOv2),...
Purpose To determine if multiparametric magnetic resonance (MR) imaging mapping can be used to quantify the response loading of histologically intact human knee cartilage. Materials and Methods Institutional review board approval written informed consent were obtained. Twenty macroscopically cartilage-bone samples obtained from central lateral femoral condyles in 11 patients undergoing total replacement. A clinical 3.0-T MR system was generate T1, T1ρ, T2, T2* maps with inversion recovery,...
Abstract Unmasking the decision making process of machine learning models is essential for implementing diagnostic support systems in clinical practice. Here, we demonstrate that adversarially trained can significantly enhance usability pathology detection as compared to their standard counterparts. We let six experienced radiologists rate interpretability saliency maps datasets X-rays, computed tomography, and magnetic resonance imaging scans. Significant improvements are found our...
Large language models (LLMs) have broad medical knowledge and can reason about information across many domains, holding promising potential for diverse applications in the near future. In this study, we demonstrate a concerning vulnerability of LLMs medicine. Through targeted manipulation just 1.1% weights LLM, deliberately inject incorrect biomedical facts. The erroneous is then propagated model's output while maintaining performance on other tasks. We validate our findings set 1025 This...