Building sovereign AI infrastructure for Norwegian hospitals. Developing speech-to-text and LLM pipelines that automate clinical documentation, running on Norwegian soil with no data leaving the country. In production at multiple hospitals.
Ph.D. research on AI-driven cardiac function monitoring using transesophageal echocardiography. 7 publications.
Developed sensor-equipped glove for drone control.
Computerized Artificial Intelligence for Automated Monitoring of Left Ventricular Function by Ultrasound.
Control and supervision of complex dynamic systems. Exchange at University of Trento, Italy.
Sovereign AI platform for Norwegian hospitals. Automates clinical documentation by transcribing doctor-patient conversations into structured medical notes. Runs on Norwegian infrastructure with open European language models — no patient data leaves the country. In production at multiple hospitals.
An AI-driven method utilizing 2D transesophageal echocardiography (TEE) for assessing left ventricular (LV) function in critical care patients post-cardiac surgery. We used state-of-the-art machine learning (ML) architectures for the detection and tracking of the mitral annulus in 2D TEE sequences, for the estimation of mitral annular plane systolic excursion (MAPSE).
A fully automatic pipeline for the estimation of MAPSE in 3D TEE, including volume alignment, cardiac view classification, and mitral annulus segmentation in 3D. The comprehensive view provided by 3D TEE, coupled with the efficiency of automation, offered a more accurate and holistic assessment of the LV function. The use of 3D TEE reduced the need for manual intervention, thus streamlining clinical workflows and potentially improving patient outcomes in critical care settings.
Generation of synthetic 2D TEE data with ground truth motion fields of myocardial contraction. Method was developed by CREATIS, INSA, Lyon, France and adopted to TEE in collaboration with the research group through a research stay. This simulation pipeline generated synthetic images by mimicking ultrasound wave propagation and reflection through a physical simulator named SIMUS, coupled with post- process calculations. The pipeline also simulated various myocardial contraction patterns, including myocardial infarction in different cardiac segments. This was achieved by locally modifying the scatter map deformations to simulate reduced longitudinal contraction in affected areas, ensuring that global contraction remained consistent and that the endocardial and epicardial borders were unaffected.
Optical flow and point trajectory estimation in 2D TEE for the approximation of regional LV function in critical care patients by segmental longitudinal strain (SLS). AutoStrain held potential to transform the current practice by providing a more objective and automated approach to strain quantification. Accurate prediction of deformation by estimation of myocardial motion was vital for assessing regional LV function and detecting conditions such as ischemic cardiomyopathy.
Reach out via email or social media.
