While the enormous potential of AI in medical imaging have been demonstrated over the last decade, the vast majority of the presented algorithms have only been evaluated in an artificial, strictly controlled setting. Often, a limited amount of high-quality images are used, acquired in expert centers, which is generally not representative of the much more heterogeneous data that can be expected in clinical practice. This gap between the domain in which these algorithms are developed and the application domain casts uncertainty on the actual clinical performance of such AI algorithms. Especially in the field of endoscopy, this domain gap poses a problem, as the quality of the imagery heavily depends on the skill and experience of the endoscopist. In addition to data heterogeneity, which places demands on system robustness, two additional challenges currently hinder clinical integration of endoscopic AI. First, the clinical application typically requires real-time, embedded operation. Current state-of-the-art endoscopic AI models are generally large and fail to meet a real-time execution speed. Second, while the interface between the algorithm and the endoscopist will be elemental in successful application of supportive endoscopic AI systems, current systems are evaluated in a man-vs-machine experimental setup, in which there is no communication between the medical specialist and the AI system.
In this project, we aim to make the next steps towards clinical application of AI-based supportive systems in endoscopy, by addressing key aspects for this translation, such as model robustness, efficiency and interpretability. The PhD candidate starting on this project will explore novel methods to improve the robustness, efficiency and interpretability of the employed AI models, which are currently convolutional neural networks. Specific tasks towards this goal include:
This project is carried out in an intensive collaboration between Eindhoven University of Technology and the Amsterdam University Medical Centers (A-UMC). The consortium has an unrivaled track record on AI for endoscopy, including publications in top-tier journals in the field, awarded grants and industry collaborations. As a highlight, last year, the consortium published the results of a developed AI system for detecting early Barrett’s cancer that demonstrated very high detection rates, outperforming a large group of international endoscopists and a matching performance in a live clinical pilot study.
The envisioned candidate for the open position holds an MSc degree in a relevant field for this project (e.g. Electrical engineering, Computer science, Applied Mathematics, Artificial Intelligence, …) with a strong background in machine learning, computer vision and image processing and a drive to improve healthcare. The candidate should have:
For this particular project, some additional points are desired and could break the tie for two equally suited candidates. These properties are:
Do you recognize yourself in this profile and would you like to know more? Please contact
dr. Fons van der Sommen, fvdsommen[at]tue.nl or Prof.dr. Peter de With, p.h.n.de.with[at]tue.nl
for more general information.
For information about terms of employment, please contact HR, hrservices.flux[at]tue.nl
Please visit www.tue.nl/jobs to find out more about working at TU/e!
We invite you to submit a complete application by using the 'apply now'-button on this page.
The application should include a:
We look forward to your application and will screen it as soon as we have received it.
Screening will continue until the position has been filled.
The VCA research group has regularly more opportunities (surveillance, industrial AI applications) than only this vacancy. If you largely satisfy the profile, but not specifically interested in healthcare, you are also allowed to apply.
We do not respond to applications that are sent to us in a different way.
Please keep in mind you can upload only 5 documents up to 2 MB each. If necessary please combine files.