The FDA has given RaySearch permission to begin distribution of its RayStation 8B platform to U.S. providers.

Launched in December 2018, the solution is the first treatment planning system on the radiation oncology market to incorporate machine learning applications. The platform offers applications for organ segmentation and automated treatment planning.

"These capabilities help automate the existing workflows of organ segmentation and plan generation which has the potential to improve efficiency and treatment consistency at the clinics," Fredrik Löfman, head of machine learning at RaySearch, and Emil Ekström, chief functionality owner for RayStation, told HCB News. "In addition, the machine learning framework in RayStation supports model sharing to facilitate knowledge sharing between clinics."

Composed of a state-of-the-art neural network, the system can train models on the available data within the practice, or rely on pre-trained models developed by RaySearch.

The application for organ segmentation utilizes deep neural networks to automate generation of organs for all body sites, while clearance for the machine learning planning application is based on the same model level of approval used to assess a prostate VMAT planning model from Princess Margaret Cancer Centre.

Another offer is a module for evaluating the robustness of treatment plans, including the influence of uncertainties in patient setup and density interpretation of CTs. This allows clinicians to form and analyze multiple scenarios with distinct uncertainty settings at the same time and base them on different robustness metrics connected to the clinical goals of treatment.

Additional enhancements include support for boron neutron capture therapy (BNCT) planning; and improvements for photon planning, which include a Monte Carlo dose engine. The dose engine relies on an algorithm that provides improved accuracy and uses GPU technology for fast dose computation.

"RayStation 8B also adds various improvements and enhancements, including support for directly deliverable MCO for VMAT, collimation of individual energy layers with the Adaptive Aperture of the Mevion S250i HYPERSCAN proton therapy system, and handling of relative biological effectiveness for proton dose," said Ekström.

Clinics with a service contract may access the new version for free. Those wishing to use the machine learning capabilities and tools for Boron neutron capture therapy planning require additional licenses.

The machine learning application for organ segmentation was developed by RaySearch's in-house machine learning department, while the automated treatment planning function was developed in collaboration with Princess Margaret Cancer Centre in Toronto. BNCT planning was developed in collaboration with Sumitomo Heavy Industries and Neutron Therapeutics.

Health IT Homepage