It has been shown that cancer patients treated at larger clinics (with specialized radiation oncologists) have better outcomes that those treated at small clinics, and that poor quality radiation therapy treatment planning can have reduce overall chance of survival by as much as 20%. Thus, it is of vital importance to address this issue so that patients across the country can receive the same quality treatment plan if they are treated at their local clinic (60% of which have <10 practitioners) as they would if they were treated at a major academic center. One approach to improve consistency and quality is to have all treatment plans independently reviewed by radiation oncologists experi...

Read More

It has been shown that cancer patients treated at larger clinics (with specialized radiation oncologists) have better outcomes that those treated at small clinics, and that poor quality radiation therapy treatment planning can have reduce overall chance of survival by as much as 20%. Thus, it is of vital importance to address this issue so that patients across the country can receive the same quality treatment plan if they are treated at their local clinic (60% of which have <10 practitioners) as they would if they were treated at a major academic center. One approach to improve consistency and quality is to have all treatment plans independently reviewed by radiation oncologists experienced in each particular disease site (“peer review”). We do this at our own clinic and also with partner institutions, resulting in treatment plan changes for 6-25% of patients depending on the treatment anatomy and the experience of the original clinical team. Peer-review is, however, an extremely time-consuming process taking an average of 15 minutes of radiation oncologist time per patient (and much longer for complex cases). Our proposed project, which will build on our group’s strong background in developing fully automated treatment planning and quality assurance experience, will develop a fully automated peer-review process. To do this, we will use deep learning models to identify and contour the necessary normal tissues and targets for head and neck, breast and cervical cancer; we will also develop AI approaches to predict the expected dose distribution. By training these state-of-the-art artificial intelligence techniques using thousands of patient data from our institution, the predictions will closely mirror the treatment plans that patients receive at our institution. For each new patient, our AI predictions will be automatically compared with the original treatment plan, and any differences highlighted to the clinical team for review and possible change to the plan.

Read Less