Artificial intelligence and other automation approaches are set to dramatically improve workflows, consistency and quality in radiation therapy treatment planning and other image-guided cancer therapies. This is important because it should lead to improved patient outcomes and reduced treatment cost. The use of AI may, however, also result in avoidable risk to the patient, specifically related to human error, over-reliance (automation bias), and off-label use. AI workflows should, therefore, be designed to avoid these risks. One example is the influence of imaging protocol and image quality on AI-based auto-contouring (which is used to identify the tumor and normal tissues), as even ro...

Read More

Artificial intelligence and other automation approaches are set to dramatically improve workflows, consistency and quality in radiation therapy treatment planning and other image-guided cancer therapies. This is important because it should lead to improved patient outcomes and reduced treatment cost. The use of AI may, however, also result in avoidable risk to the patient, specifically related to human error, over-reliance (automation bias), and off-label use. AI workflows should, therefore, be designed to avoid these risks. One example is the influence of imaging protocol and image quality on AI-based auto-contouring (which is used to identify the tumor and normal tissues), as even robust auto-contouring algorithms may give incorrect answers if presented with images of insufficient quality or taken with an imaging protocol that was not included in the AI model's training set. Furthermore, with fully-automated systems there are reduced opportunities for clinical staff to catch errors, meaning that poor quality images may result in errors that are passed through the entire treatment planning process, with increased risk of reaching the patient. Furthermore, even with manual review of image quality, this is a subjective process so images with inadequate image quality may still be passed to the AI tool.

Our proposed project will develop a fully automated system to flag patient images that would lead to incorrect auto-contouring, thus mitigating this risk to patients, while facilitating us taking advantage of the potential benefits that AI offers. In this project we will compare several analytic and AI-based tools for assessing image quality based on patient images, and then validate these tools in our own clinic (at a large specialized cancer center) and also through an on-line service that is accessed by diverse radiation therapy clinics across the world.

Read Less