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: Professor Issam El Naqa, University of South Florida, USA
- Second opponent: Professor Ewen Harrison, University of Edinburgh, UK
- Third member and chair of the evaluation committee: Professor Åslaug Helland, University of Oslo
Chair of the Defence
Professor Emeritus Trond Buanes, University of Oslo
Principal Supervisor
Group Leader Ilangko Balasingham, Oslo University Hospital
Summary
Cancer treatment decisions rely heavily on predicting how long patients might survive or how quickly their disease might progress. Traditionally, doctors make these predictions manually—a time-consuming process that can vary between specialists. This thesis explores how artificial intelligence can transform this approach.
By teaching computers to "see" patterns in medical scans that humans might miss, we've created tools that automatically predict cancer outcomes. Our journey began with kidney cancer, where we trained AI to recognize four levels of tumor aggressiveness in CT scans. This success led us to ask: could these same image patterns predict survival time?
Indeed, they could. We extracted hidden patterns from CT images to forecast kidney cancer survival with impressive accuracy. Taking this further, we then combined these image patterns with key patient information from medical records, creating a more powerful prediction system that outperformed either approach alone.
Turning to brain tumors, we discovered that specific MRI techniques (T2 sequences) contain the most valuable clues about survival. This insight helped us develop a system for childhood brain tumors that predicts when the disease might return after surgery—without requiring doctors to manually outline tumors in the images.
Throughout our work, combining different types of information (images plus medical records) consistently produced the best predictions. While these tools aren't quite ready for hospitals, they represent important steps toward faster, more consistent cancer prognosis that could help doctors plan better treatments for their patients.
Additional information
Contact the research support staff.