SBIR Phase I: A Data-Driven, Knowledge-Based Platform for Peer Review in Radiation Oncology Treatment Planning

This SBIR Phase I project will leverage the data generated routinely by radiation oncology practices to automate the peer-review process in which radiation oncologists evaluate the quality of planned treatments. Each cancer patient possesses a unique layout of normal anatomy relative to their disease and this impacts the characteristics of the treatment delivered. An example is that it is harder to achieve a low dose to an organ that is closer to the irradiated tumor. Today, quality evaluation for a planned treatment effectively assumes that the anatomical geometry of every patient with a given disease is the same. The proposed innovation personalizes the evaluation of radiotherapy treatment plans by comparing a patient's anatomy to thousands of past patients. The highest standards achieved are used as a yardstick for the current patient. Furthermore, plan evaluation is transformed from a subjective, error-prone, procedure to an efficient standardized process, including sophisticated anomaly detection. The benefit to the patient is lower doses to healthy organs and lower risk of their treatment being impacted by errors. This can reduce side effects and improve quality of life. The benefit to clinics is workflow efficiency and improved quality metrics, resulting in cost savings and proof of high practice standards. The proposed software product is an automated peer review system for radiation oncology treatment planning based on a large database of existing patient treatment plans. The unique strength of the innovation lies in it being based on a large and varied dataset that can provide reference patients to which the current patient can be meaningfully compared. During the treatment planning process, the system will allow radiation oncologists to detect anomalies in dose prescription and structure delineation, as well as obtain patient-specific dosimetric objectives against which to evaluate treatment plan quality. Machine learning approaches will be used for anomaly detection, where deviations from the norm will be flagged as requiring examination. Dosimetric objectives will be obtained using a search procedure that considers the geometric relationships between each targeted volume and each organ at risk. Automated peer-review will allow plan assessment during the treatment planning process itself rather than it being a retrospective step after the initiation of a patient's treatment course. Prospective review allows for customization based on individual patient characteristics and thus expands the role of peer review from quality assurance to the personalization of radiation treatments. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.