Radiomics.

Top Line: More and more studies are looking at teasing out not who needs radiation but who needs what dose of radiation.

The Study: Rather than relying on tumor genomics, this group is basing their efforts on readily available tumor radiographic characteristics. They used 849 pre-radiation CT scans paired with treatment details and post-treatment surveillance scans to not only predict local failures after lung SBRT but also predict what dose would have resulted in a <5% chance of local failure at 24 months. This model was then prospectively validated on an external cohort. Indeed, those receiving a dose that met or exceeded the predicted dose saw a 0% chance of local failure at 3 years, those receiving slightly less than predicted dose had 13% rate of failure and those receiving much less than predicted dose had a 40% rate of failure. These local failures were associated with both nodal (HR 3.69) and distant (HR 3.64) failures and preceded them 80% of the time. Of note, predicted doses ranged from 73 to 267 Gy BED, meaning some recommended doses exceeded standard dose recommendations, particularly for centrally-located tumors. Therefore, there’s no way to know at this time if local control would have actually been improved with delivery of predicted doses in the much higher range.

TBL: Artificial intelligence pre-treatment review of radiographic tumor characteristics may help predict on which lung tumors we should push the limits with ablative radiation prescription doses. | Randall, JCO Clin Cancer Inform 2023