A5068645252- Cardiovascular Disease and Adiposity
- Radiomics and Machine Learning in Medical Imaging
- Medical Imaging Techniques and Applications
- Cardiac Imaging and Diagnostics
- Cardiovascular Function and Risk Factors
- AI in cancer detection
- Control Systems and Identification
- Neural Networks and Applications
- Cardiovascular Health and Disease Prevention
- Natural Language Processing Techniques
- Cell Image Analysis Techniques
- Head and Neck Cancer Studies
- Artificial Intelligence in Healthcare and Education
- Lung Cancer Diagnosis and Treatment
- Advanced Image and Video Retrieval Techniques
- Cancer Immunotherapy and Biomarkers
- Linguistics, Language Diversity, and Identity
- Radiology practices and education
- Medical Image Segmentation Techniques
- Fault Detection and Control Systems
- Topic Modeling
- Advanced X-ray and CT Imaging
- Advanced Optimization Algorithms Research
- Cardiovascular, Neuropeptides, and Oxidative Stress Research
- Machine Learning in Healthcare
Maastricht University
2023-2025
Harvard University
2023-2025
Mass General Brigham
2024-2025
University of Applied Sciences Stralsund
2025
University Hospital Cologne
2025
Dana-Farber Brigham Cancer Center
2024
Brigham and Women's Hospital
2023-2024
Dana-Farber Cancer Institute
2024
Massachusetts General Hospital
2023-2024
Intel (United States)
2024
Abstract Foundation models in deep learning are characterized by a single large-scale model trained on vast amounts of data serving as the foundation for various downstream tasks. generally using self-supervised and excel reducing demand training samples applications. This is especially important medicine, where large labelled datasets often scarce. Here, we developed cancer imaging biomarker discovery convolutional encoder through comprehensive dataset 11,467 radiographic lesions. The was...
Abstract Background and Aims Skeletal muscle (SM) fat infiltration, or intermuscular adipose tissue (IMAT), reflects quality is associated with inflammation, a key determinant in cardiometabolic disease. Coronary flow reserve (CFR), marker of coronary microvascular dysfunction (CMD), independently body mass index (BMI), inflammation risk heart failure, myocardial infarction, death. The relationship between SM quality, CMD, cardiovascular outcomes not known. Methods Consecutive patients (n =...
On serial lung cancer screening CT scans in individuals at high risk of developing cancer, atypical epicardial adipose tissue volume change was associated with all-cause mortality, while decrease and density increase were elevated cardiovascular mortality.
Abstract Background Heavy smokers are at increased risk for cardiovascular disease and may benefit from individualized quantification using routine lung cancer screening chest computed tomography. We investigated the prognostic value of deep learning-based automated epicardial adipose tissue compared it to established factors coronary artery calcium. Methods in heavy enrolled National Lung Screening Trial followed 12.3 (11.9–12.8) years. The was segmented quantified on non-ECG-synchronized,...
PURPOSE Current approaches to accurately identify immune-related adverse events (irAEs) in large retrospective studies are limited. Large language models (LLMs) offer a potential solution this challenge, given their high performance natural comprehension tasks. Therefore, we investigated the use of an LLM irAEs among hospitalized patients, comparing its with manual adjudication and International Classification Disease (ICD) codes. METHODS Hospital admissions patients receiving immune...
Artificial intelligence (AI) has great potential to transform the clinical workflow of radiotherapy. Since introduction deep neural networks, many AI-based methods have been proposed address challenges in different aspects Commercial vendors started release tools that can be readily integrated established workflow. To show recent progress AI-aided radiotherapy, we reviewed studies five major radiotherapy including image reconstruction, registration, segmentation, synthesis, and automatic...
Foundation models (FMs) have shown transformative potential in radiology by performing diverse, complex tasks across imaging modalities. Here, we developed CT-FM, a large-scale 3D image-based pre-trained model designed explicitly for various radiological tasks. CT-FM was using 148,000 computed tomography (CT) scans from the Imaging Data Commons through label-agnostic contrastive learning. We evaluated four categories of tasks, namely, whole-body and tumor segmentation, head CT triage,...
A linear programming (LP) based method is proposed for learning from experimental data in solving the nonlinear regression and classification problems. LP controls both number of basis functions a neural network (i.e., support vector machine) accuracy machine. Two different methods are suggested their equivalence discussed. Examples function approximation (pattern recognition) illustrate efficiency method.
Foundation models represent a recent paradigm shift in deep learning, where single large-scale model trained on vast amounts of data can serve as the foundation for various downstream tasks. are generally using self-supervised learning and excel reducing demand training samples applications. This is especially important medicine, large labeled datasets often scarce. Here, we developed imaging biomarker discovery by convolutional encoder through comprehensive dataset 11,467 radiographic...
This paper formulates the learning of support vector machines (SVM) as a linear programming problem. An SVM has property that it chooses minimum number data points to use centres for Gaussian kernel functions in order approximate training within given error. A (LP) based method is proposed solving both regression and classification Examples function approximation class separation illustrate efficiency method. In addition, explores possibility using with radial basis kernels compress an...
The quadratic programming (QP) and the linear (LP) based method are recently most popular methods for learning from empirical data (observations, samples, examples, records). Support vector machines (SVMs) newest models on QP algorithm in solving nonlinear regression classification problems. LP also controls both number of basis functions a neural network (i.e., support machine) accuracy machine. Both result parsimonious network. This results compression. Two different compared terms SVs...
Generative adversarial networks (GANs) have been used to successfully translate images between multiple imaging modalities. While there is a significant amount of literature on the use cases for these approaches, has limited investigation into optimal model design and evaluation criteria. In this paper, we demonstrated performance different approaches task cone-beam computer tomography (CBCT) fan-beam (CT) translation. We examined implications choosing 2D 3D models, size patches, integration...
2638 Background: Immune checkpoint inhibitor (ICI)-induced colitis, hepatitis, and pneumonitis are common immune-related adverse events (irAEs); however, the true incidence for these irAEs remains incompletely understood. Chart review is gold standard their detection but time-consuming cannot be implemented in large cohorts. The use of ICD codes limited sensitivity specificity. Large language models (LLMs) a scalable method answering queries from human-generated text, though there no data on...
Background: Immune-checkpoint inhibitor (ICI)-induced myocarditis is the most fatal immune-related adverse event (irAE). Early recognition and treatment of ICI associated with improved outcomes. However, current approaches to diagnosis have limitations. Open-source large language models (LLMs) are an accessible scalable method answering queries from human-generated text, therefore may assist in real-time early detection among at-risk patients. Research Question: Can a free, open-source LLM...
Research exploring CycleGAN-based synthetic image generation has recently accelerated in the medical community due to its ability leverage unpaired images effectively. However, a commonly established drawback of CycleGAN, introduction artifacts generated images, makes it unreliable for imaging use cases. In an attempt address this, we explore effect structure losses on CycleGAN and propose generalized frequency-based loss that aims at preserving content frequency domain. We apply this...