4π Non-Coplanar Plans Cut Predicted Hematologic Toxicity 23% in Cervical Cancer via Bone Marrow Sparing

The arXiv preprint describes a single-institution retrospective study that extracted radiomic, clinical, and dosimetric features from planning CTs of 114 cervical cancer patients treated with coplanar VMAT between 2021 and 2023. Researchers trained ML models to predict hematologic toxicity, achieving peak AUC 0.79 when radiomics were included, then replanned cases with integrated non-coplanar beam orientation and fluence optimization. The new plans achieved large, statistically significant reductions in bone marrow volumes receiving 10–40 Gy while maintaining target coverage and other OAR constraints.

These dose reductions translated through the validated model into a predicted 23% relative drop in acute HT risk (RR 0.77, OR 0.68). Feature importance ranked CT radiomics and marrow dosimetry highest, confirming that conventional coplanar arcs unnecessarily irradiate pelvic bone. The approach directly addresses a common dose-limiting toxicity that delays chemotherapy and worsens quality of life in locally advanced cervical cancer.

Because the study is retrospective and relies on a surrogate ML endpoint rather than observed toxicity, prospective confirmation is required. A multi-center randomized trial comparing 4π versus VMAT with primary endpoint of grade 3+ neutropenia would establish whether the modeled benefit materializes in real patients and whether delivery time or setup complexity offsets gains.

If validated, 4π planning could be integrated into commercial treatment planning systems within three years, offering a hardware-agnostic way to lower toxicity without new drugs or extended fractionation.