Our research centers on integrating artificial intelligence into vascular imaging, with a particular focus on ischemic and hemorrhagic stroke, interventional neurovascular procedures, and neuro-oncology. We work across the entire translational pipeline, from algorithm development to internal and external validation, with a strong emphasis on clinical applicability.

In parallel, we conduct a broad range of non-AI clinical research, addressing critical gaps in imaging optimization, predictive modeling, outcome studies, and neurointerventional procedures. These efforts include both prospective and retrospective studies designed to improve clinical decision-making and patient outcomes. Many of these projects serve as the foundation for future AI development, providing structured data, expert-annotated cases, and clinically relevant hypotheses.

This diversity enables us to bridge the gap between clinical practice and data science, ensuring that our AI research is informed by real-world clinical challenges and grounded in robust evidence.

We develop, validate and improve deep learning models for segmentation, detection, and classification tasks on CT, MRI, and angiographic imaging. Our goal is to enhance diagnostic precision and support decision-making in acute neurovascular care.

In neuro-oncology, we focus on leveraging imaging features and advanced AI models, including deep learning and foundation models, to predict treatment response and clinical outcomes. By integrating multi-modal imaging data with outcome-driven learning objectives, we aim to develop robust, clinically applicable tools for patient stratification and personalized therapy planning.

In addition to our AI research, we conduct clinical imaging studies with a strong focus on neurointerventional radiology. These projects investigate imaging-based predictors and develop tools to improve treatment selection and outcome estimation. Specific areas of interest include imaging biomarkers for patient stratification and therapy guidance such as in the embolization of subdural hematomas and other minimally invasive neurovascular procedures.

We harness and evaluate large language models to improve information extraction, reporting automation, clinical question answering, and workflow optimization in radiology.

A key area of focus is the extraction and structuring of data from unstructured radiology reports and clinical texts to enable scalable, interoperable research datasets. By integrating LLMs into data pipelines, we aim to build high-quality repositories that support multicenter studies, real-world evidence generation, and the development of robust AI models.