- 10.00 – 10.45: Trial Lecture. Title: TBA
- 12.00 - 16.00 Public defence
Ordinary opponents:
- First opponent: Professor Björn Þór Jónsson, Reykjavik University
- Second opponent: Professor João Barbosa Breda, KU Leuven/University of Porto
- Leader of the evaluation committee: Professor Weiqin Chen, OsloMet
Leader of the public defence: TBA, Department of Computer Science, OsloMet - Oslo Metropolitan University, Oslo, Norway
Supervisors:
The main supervisor was Michael Riegler, Head of AI, Research Professor, Simula Research Laboratory & Professor, Faculty of Social Sciences, OsloMet
Co-supervisors:
- The first co-supervisor was Anis Yazidi, Professor, Faculty of Technology, Art and Design, Department of Computer Science, OsloMet
- The second co-supervisor was Professor and Ophthalmologist Tor Paaske Utheim, Eye Health Clinic and OsloMet.
- The third co-supervisor was Associate Professor and Ophthalmologist Xiangjun Chen, University of South-Eastern Norway and Dry Eye Clinic
Abstract
A technological revolution in healthcare
We are living through a technological revolution. Artificial intelligence (AI) and machine learning (ML) are becoming part of everyday life; from the apps we use to the way doctors diagnose disease. This fusion of technology and science is often called the fourth industrial revolution, and one of its most exciting frontiers is healthcare.
Promise and pitfalls of machine learning in medicine
In recent years, ML has shown remarkable potential in medical research. These models can analyse complex data and make predictions that help doctors diagnose diseases faster and more accurately.
As ML models grow more sophisticated, they require large, high-quality datasets to function well.
Medical datasets are expensive to develop, limited in size, often consist of private data, and not always representative of the broader population. This can lead to models that perform well during development but struggle in real-world settings.
Another challenge is transparency: understanding how a model arrives at its decision is crucial, both for scientific progress, and especially when those decisions influence treatment.
Explainable AI: From black box to insight
This is where explainable AI comes in. By revealing which features a model uses to make predictions, researchers and clinicians can better trust and interpret its output.
For example, if a model predicts the presence and severity of a disease, extracting which symptoms or biomarkers it relied on is of great scientific interest. Similarly, being able to audit a model used in a clinical setting is essential for physicians when making treatment decisions.
Makes a difference for dry eyes disease
One area where this technology could make a real difference is dry eye disease. It is one of the most common yet underdiagnosed conditions in medicine, affecting millions and placing a heavy emotional and financial burden on patients and society. Despite its prevalence, it often goes untreated.
This thesis explores how ML can advance research into dry eye disease. From laboratory, clinical, and hybrid studies, we employ different types of ML models for prediction, dimensionality reduction, feature extraction and computer vision then combine them with expert feedback to better understand both the disease and the models themselves.
We also openly publish our models and source-code as well as a comprehensive dataset of ocular goblet cells.
Our findings suggest that ML and explainable AI can be powerful tools in dry eye disease research. Used conscientiously and in concert with domain experts, these instruments can provide novel connections and reveal important discoveries.
