AI That Listens to Doctors Could Make Radiation Therapy Safer
New AI segmentation could mean more accurate cancer treatment with fewer side effects.
A review examined prompt-driven deep learning for radiotherapy target volume segmentation, a time-consuming step whose final contour depends on the radiation oncologist’s expertise. It notes that image-only models cannot encode the clinical reasoning behind target volume delineation, especially for the clinical target volume.
Across 13 studies in head and neck, breast, prostate, lung, cervix, and rectal cancers, seven used clinical text prompting, four used geometric prompting, and two used interactive prompting; three combined modalities. Reported Dice similarity coefficient ranged from 0.67 to 0.93, compared with 0.65 to 0.91 for image-only baselines.
None reported sensitivity or precision. Clinical text prompting appears well suited to clinical target volume delineation, but evidence remains limited, and standardized prompt engineering and clinical validation are needed before routine radiotherapy practice.
- AI with prompts improved tumor outlining accuracy by up to 2% compared to standard AI.
- The review included 13 studies covering head and neck, breast, prostate, lung, cervix, and rectal cancers.
- No studies reported sensitivity or precision, so real-world effectiveness is still unknown.
Why It Matters
More accurate tumor outlining could lead to more effective radiation therapy with fewer side effects for cancer patients.