Contact: Sarah Verboom / Ioannis Sechopoulos.
Start date: between October 2026 and February 2027
Duration: 6 to 9 months
Background: Biomedical Engineering or similar
Requirements: Experience with python or an interest to learn python
Breast cancer is the most commonly diagnosed cancer for women in the Netherlands. One in seven women will be diagnosed with breast cancer in their lifetime. To detect breast cancer early, when chances of successful treatment are higher, women between 50 and 75 years old are invited for breast cancer screening. During screening, X-ray images are made of both breasts, called a mammogram. Each mammogram is evaluated by two independent radiologists for the presence of suspicious abnormalities.
In the last years, AI models have been developed that can detect breast cancer from a mammogram with a similar performance as a radiologist. In some hospitals and screening programs in other countries, AI detection models are already used. These models are often used to aid radiologists during their evaluation and sometimes to replace one of the two radiologists in screening. In each case, there is always at least one radiologist involved for every mammogram. However, AI detection models are getting better, and some are surpassing radiologist’s performance. In the future, we might not need to have a radiologist involved for every case.
This is only possible if we trust AI models to only give a result when that result is certain. If a result is uncertain, we should always involve a radiologist. In this project you will develop and test multiple uncertainty quantification methods for this goal. With uncertainty quantification we try to measure how certain an AI result is. In essence, we are trying to develop models that can also say “I am not sure” instead of always giving an answer. This project will build upon our previous work [1]. You will get to work with an improved test model made by Screenpoint Medical and a large set of retrospective data. The goal of this project is set, but there is a lot of room to get creative with the methods and analyses.
Are you a student looking for a place to do a master thesis, do you have experience with python or really want to learn, and are you enthusiastic about challenging AI problems with societal impact, please reach out to us!
More on the information contact: Sarah Verboom,
1. Verboom, S.D., et al., AI Should Read Mammograms Only When Confident: A Hybrid Breast Cancer Screening Reading Strategy. Radiology, 2025. 316(2): p. e242594.
