Click here to stream the public defence
Due to copyright issues, an electronic copy of the thesis must be ordered from the faculty. For the faculty to have time to process the order, the order must be received by the faculty at the latest 2 days before the public defence. Orders received later than 2 days before the defence will not be processed. After the public defence, please address any inquiries regarding the thesis to the candidate.
Trial Lecture – time and place
See Trial Lecture.
Adjudication committee
- First opponent: Assistant professor Samaneh Nasiri, Emory University, USA
- Second opponent: Professor Helen Sjöland, University of Gothenburg, Sweden
- Third member and chair of the evaluation committee: Professor Mathis Stokke, University of Oslo
Chair of the Defence
Professor II Ida Gjervold Lunde, University of Oslo
Principal Supervisor
Head of Research, Medical Technology and E-health, Arian Ranjbar, Akershus University Hospital
Summary
Heart failure is common, carries a poor prognosis, and often goes undetected despite effective treatment. Diagnosis relies mainly on clinical assessment, natriuretic peptides, and cardiac imaging. AI applied to the ECG (AI–ECG) may offer a scalable complementary approach.
This thesis aimed to enable development and use of AI–ECG for heart failure detection, via four objectives: converting plotted ECGs into digital time series; developing and validating AI–ECG models across the ejection-fraction spectrum; relating predictions to measures of cardiac structure and function; and supporting reproducibility through open dissemination of software and models.
First, an open-source ECG digitisation pipeline was developed for scanned and photographed ECGs, converting plotted traces into machine-readable signals. Built for heterogeneous images, it enables analysis of paper archives and use where digital infrastructure is limited.
Second, deep-learning models were trained on routine clinical data using pragmatic heart-failure labels from diagnostic codes and NT-proBNP. Evaluated in prospective internal and external cohorts, they showed strong discrimination across the ejection-fraction spectrum, including preserved ejection fraction, indicating AI–ECG is not restricted to reduced systolic function.
Third, AI–ECG probabilities were compared with echocardiographic measures in two cohorts. Predictions correlated with several measures of cardiac structure and function, most strongly with global longitudinal strain, followed by ejection fraction, left atrial volume, and pulmonary pressure. This suggests the model captures meaningful cardiac abnormalities, possibly including early dysfunction.
All principal models and software were released openly. Together, the thesis establishes AI–ECG as a scalable, physiologically grounded tool that can complement existing diagnostics and support earlier detection. Prospective evaluation is needed to guide clinical integration.
Additional information
Contact the research support staff.