EUREKA ITEA ASSIST -Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems

Important results from the project

One of the main goals of the project was to train AI models via federated learning, with clinical data from different cities. This objective has been achieved with both Scaleout and Inovia AI, and for two types of models (segmentation and dose prediction). Federated training has been carried out between Linköping, Lund and Umeå. Through a final demonstrator, we have shown that models trained via federated learning can be integrated with clinical software from Spectronic and RaySearch. The demonstrator shows that the time to plan treatment can be reduced from 3-4 hours to 15 minutes.

Expected long term effects

The project has led to enhancements to Scaleout´s federated learning framework FEDn, a new iFusion interface for federated learning by Inovia AI, enhancements and expansion of Spectronics platform SAINT, a major increase in clinical use of RaySearch´s radiation therapy software via new and improved models for segmentation and dose prediction, as well as improved sensitivity for the detection of metastases via methods developed by Eigenvision. LiU and Spectronic have developed new models for synthetic images. LiU has developed new methods for diffusion-weighted images.

Approach and implementation

The implementation of the project and the collaboration between all parties has worked well, especially the final demonstrator which was a collaboration between 5 Swedish parties. Cooperation with parties in other countries has contributed to the results being even better, and contributed with a wider perspective. For the part about federated learning, ethical permission and data export took more time than expected, which led to federated training mainly being done between two cities instead of several.

External links

https://itea4.org/project/assist.html