While artificial intelligence (AI) is automating certain aspects of radiation oncology, the real value will come when it is able to accurately predict patient outcomes. Kristy Brock, PhD, of the University of Texas MD Anderson Cancer Center in Houston, weighed in during the American Society for Radiation Oncology (ASTRO) annual meeting and noted that writing an AI algorithm is just the beginning.
Validating thousands (if not millions) of images will be required. And even then, things are not foolproof. An algorithm developed by MD Anderson became stumped when it thought images of patients were in the prone position when they were supine.
There are a number of potential uses for AI in radiation therapy, including:
- Deriving tumor history without doing a biopsy, then using that to develop a targeted treatment plan.
- Segmenting and identifying toxicity risk to normal tissue on a patient-by-patient basis.
- Predicting patient outcomes long term.
Read the full article here.
As Dr. Brock states, data is the critical factor for algorithms to improve. The availability of curated, annotated images is a real problem. When used in standard clinical practice, HealthMyne creates a wealth of data quickly, with each lesion identified having a full radiomic profile calculated and saved to the database. The data is then available for these types of important research and to support multiple workflows in oncology such as lung cancer screening, tumor conferences and more.