Lung cancer is the leading cause of cancer-related death worldwide. Early detection greatly increases the chances of successful treatment, but assessing suspicious pulmonary nodules on CT scans remains challenging. As a result, many patients undergo unnecessary follow-up examinations, while the workload for radiologists continues to increase. This thesis investigates how artificial intelligence (AI) can support radiologists in reliably estimating the probability of lung cancer. The results show that AI can indicate when it is uncertain about a prediction and detect when a CT scan differs from the data on which it was trained, helping to ensure the safe use of AI in clinical practice. In addition, an international benchmarking study demonstrated that an AI model outperformed the average of 65 radiologists in assessing a challenging group of pulmonary nodules on screening CT scans. Finally, this thesis shows that AI can accelerate the development of new AI models by automating part of the time-consuming annotation work performed by radiologists.
About the PhD candidate
Dré Peeters (1997) obtained his Master’s degree in Biomedical Sciences at Maastricht University in 2021, specializing in Medical Imaging and Artificial Intelligence. Afterwards he worked as data scientist and started his PhD research at the Diagnostic Image Analysis Group at the department of Medical Imaging at Radboudumc in 2022. Currently, he is working as an AI engineer at Fellowmind.