- Chairs: Wouter van Elmpt & Dirk Verellen
Artificial intelligence (AI) techniques such as advanced machine learning and deep learning are currently finding their way into clinical routine practices. There is still an open question on how properly design, validate and commissioning these algorithms. AI sometimes present as ‘black box’ algorithms and the results may not always be deterministic or predictable upfront due to the large number of degrees of freedom that these algorithms have.
In this workshop we will discuss the principles how to design, validate and implement such algorithms from various perspectives. A) Developer/researcher perspective: what are the possibilities, problems and pitfalls in designing such algorithms and how to quantify their accuracies. B) The (clinical) end-user on how to validate (commission) the output of such techniques for use in clinical practice.
We welcome contributions from the community (both developers as well as users) working with AI that have a focus on radiotherapy or medical imaging: e.g. pseudo CT reconstruction from MRI only imaging, automatic segmentation using deep-learning techniques, automatic dose distribution predicting, image registration, noise reduction in medical imaging; subjects are not limited to these topics but just serve as example.
This being a workshop we want to encourage an active participation and interaction between the participants to foster collaboration and networking. For that reason, participants will be requested to prepare a short presentation (a pitch) to present their research in the field allowing identification of common points of interests and share experiences.
Guideline on how to design, validate and implement machine learning algorithms for clinical use.