Events Calendar
Yi Wang (MGH): Machine learning optimization (MLO) auto planning in RayStation
Tuesday 14 April 2020, 12:00pm - 01:00pm
Artificial intelligence (AI) has shown great potentials on promoting the quality, efficiency and consistency of cancer care. Auto planning is among the most promising application of AI technologies to radiation oncology. RaySearch Laboratories (Stockholm, Sweden) introduced the world’s first commercial machine learning auto planning tool – machine learning optimization (MLO) in RayStation 8B. In collaboration with RaySearch Machine Learning Department, the MGH AI lab developed RayStation’s first three MLO models trained by MCO plans, starting with liver SBRT, expanding to lung SBRT, and then the more complex fractionated pancreas treatment with simultaneous integrated boost (SIB). Each model employs multiple strategies which can provide the user multiple auto plans with just one click. Each strategy represents a different prioritization of clinical goals (e.g., target vs. organ), or a different clinical decision (e.g., heating the center of a lung SBRT lesion). The two SBRT models can provide high-quality auto plans that are clinically acceptable either upright or with minimal post processing in standard optimization. The more complex pancreas model can also create high-quality auto plans very close to the original clinical plan. The pancreas auto plans often need some standard post processing steps to suppress the point hotspots in GI organs, which is very difficult for any machine learning algorithm to manage. The recent introduction of a “fine-tuning” function with a quick MCO in RayStation 10A will make it much easier to post process these complex plans, allowing achieved clinical goals to be preserved while further optimizing those unachieved.
Location : Goitein Room
