Adaptable AI for automated segmentation in radiotherapy
| Reference number | |
| Coordinator | Region Västerbotten - Medicinsk teknik-FoU, Centrum för medicinsk teknik och strålningsfysik |
| Funding from Vinnova | SEK 4 554 618 |
| Project duration | November 2017 - June 2021 |
| Status | Completed |
| Venture | Digital health |
Important results from the project
The overall objective has been to develop a national infrastructure for algorithm development on clinical data and use this to develop automatic AI methods for segmentation in radiotherapy. The prototype for the infrastructure is in place at the end of the project and has led to consequential investments in a national perspective. The intended product developments have also taken place and can, after the end of the project, be distributed within Elekta´s and Peltarion´s regular product portfolio for radiation treatment and machine learning, respectively.
Expected long term effects
The project had wished that finished products could be developed on larger datasets and also more clearly setting demands for infrastructure. However, this has been able to be bridged to reach critical objectives. Furthermore, the project has generated a follow-up project with a national representation and an ambitious goal of making population-representative data available for the development of algorithms in the future. It has also led to increased internal investments by project partners to further explore machine learning applications.
Approach and implementation
The project´s approach generated results which, despite a number of delays, e.g. a global pandemic, of course also led to obstacles to overcome. The collaborative aspect of the project has continued to be important in achieving objectives and concentrating competence as well as differing perspectives on shared problems. A key impact, however, is that the pandemic has minimized the number of physical meetings, which is obviously not desirable both for collaboration within the project and to be able to reach out with results through, for example, scientific meetings and conferences.