An automated planning optimization framework for cervical intensity-modulated radiation therapy using voxel dose...

Knowledge-based planning typically lacks mechanisms to manage prediction uncertainties. We developed an automated framework integrating voxel-dose prediction with adaptive optimization to overcome this limitation. For 27 cervical intensity-modulated radiation therapy (IMRT) cases, voxel-dose predictions initialized an end-to-end optimizer. A fuzzy-inference system dynamically adjusted objectives to correct clinical guidance violations and enhance sparing of organs of interest (OOIs). Results demonstrated significantly improved OOI sparing (e.g., rectum volume receiving 45 Gy was lower by 5.5 ± 2.2%, p < 0.005) while maintaining target coverage comparable to clinical plans. These findings indicate that the framework successfully generated high-quality IMRT plans by addressing prediction uncertainties through adaptive optimization.