What is AutoPlanning?
AutoPlanning is a new artificial intelligence (AI) engine that learns from a database of gold standard treatment plans to automatically create the best radiation therapy plan for a new patient, reducing planning times from days to minutes.
How does AutoPlanning help?
Generating a radiation therapy plan can take days of work by a team of highly trained experts. AutoPlanning increases efficiency by reducing that time to a matter of minutes.
The most time-consuming portion of the treatment planning process is manually adjusting plan parameters and goals to direct enough radiation to the cancer while avoiding the nearby healthy organs. Minimizing radiation to healthy organs speeds recovery, reduces the chance of side-effects and treatment-related toxicities, and improves both the quality and duration of the patient’s life after treatment. At the same time, maximizing radiation to the cancer target helps ensure the cancer will be controlled and also prevent cancer from returning later. The AI-driven approach of AutoPlanning rapidly optimizes this trade-off without needing manual guidance by the user.
The time-consuming and manual nature of plan development also creates quality variations between planners, and between cancer care centres. Through automatization the AI engine reduces this treatment variability, and by drawing upon a database of gold-standard plans from world experts the AutoPlanning system can export the expertise from leading centres to less experienced clinical teams, providing high-quality care to all patients regardless of which centre they are treated at.
How is AutoPlanning different?
Traditional planning approaches focus strictly on broad treatment objectives, e.g. the total radiation dose to the lung. However, many undesirable plans can satisfy these broad criteria, as it takes expertise to incorporate factors such as the shape and deliverability of the resulting treatment. AutoPlanning learns the relationship between the patient’s anatomical image and the gold-standard treatment, which enables it to model not only the treatment objectives but the shape of the radiation dose distribution throughout the patient.
The technology is enabled through a novel machine-learning algorithm that learns to measure the similarity between patients, between patients and plans, and between different plans. When a new patient is admitted for treatment, the technology automatically retrieves the most relevant patients from a database of gold standard treatments. A key feature is that the notion of ‘relevant’ is learned in a way that best predicts reflects the expert treatment decisions for the patient, enabling the algorithm to ignore irrelevant data and zero in on what affects patient care (e.g. tumor location in the body). The learned relationships then adjust the treatment characteristics from the gold standard database and adapt them to the new patient to be planned that only includes trade-offs when necessary and eliminates the requirement for the user to specify objectives for each individual patient. The AutoPlanning technology rapidly and automatically generates highly personalized, expert-quality plans for an efficient, standardized treatment planning process requiring minimal to no user interaction.