Trial lecture
“Explainable AI in Medical Diagnostics”
Time and place: May 12, 2025 11:15. Kristen Nygaards sal (5370), Ole-Johan Dahls hus.
Main research findings
This thesis focuses on improving the diagnosis of deep vein thrombosis (DVT) by integrating machine learning (ML) techniques with multi-objective evolutionary optimization.
Conventional ML models were initially evaluated for their effectiveness in classifying DVT, revealing the most capable ML algorithms for this task. A hybrid diagnostic approach was introduced, combining state-of-the-art ML algorithms with a traditional medical approach. The study then introduced a multi-objective evolutionary algorithm (MOEA) to optimize the predictive performance of ML models in diagnosing DVT.
A key contribution of this thesis work is the development of an ML-based framework that integrates multiple performance enhancement techniques, optimizing predictive accuracy and clinical relevance. The framework employs a non-dominated genetic algorithm (NSGA-II) MOEA to fine-tune ML model hyperparameters, select the most relevant predictors in a dataset, and adjust classification thresholds, ultimately delivering a spectrum of Pareto-optimal solutions. These solutions balance overall prediction accuracy and diagnostic sensitivity, ensuring that more DVT cases are correctly identified while maintaining diagnostic reliability.
The findings suggest that combining evolutionary optimization with machine learning can lead to better diagnostic tools for DVT. This methodology extends beyond DVT, providing a robust framework for improving ML-based decision support in other areas of medicine.
Adjudication committee
- Professor Ibrahim A. Hameed, NMBU, Norway
- Assistant Professor Clara Maathuis, Open University, The Netherlands
- Professor Nils Gruschka, University of Oslo, Norway
Supervisors
-
Professor Martin Steffen, University of Oslo, Norway
-
Associate Professor Lars Vidar Magnusson, Østfold University College, Norway
-
Associate Professor Toktam Ramezanifarkhani, University of Oslo, Norway
Chair of defence
- Professor Carsten Griwodz, Department of Informatics, University of Oslo
Contact information at Department: Mozhdeh Sheibani Harat