Bi-level multi-criteria optimization for risk-informed radiotherapy.

Objective.In radiation therapy (RT) treatment planning, multi-criteria optimization (MCO) allows physicians to find the best plan efficiently. MCO is conventionally solved for a set of generic (population-wide) dosimetric criteria, ignoring patient-specific biological risk factors and compromising clinical outcomes in high-risk groups. We propose a one-shot method-risk-guided MCO-for integration of biological risk factors within conventional MCO, enabling interactive plan navigation between dosimetric and biological endpoints.Approach.A cohort of non small cell lung cancer patients receiving proton/photon RT was retrospectively analyzed. The clinical endpoint was the risk of symptomatic (grade 2+) radiation pneumonitis (RP), modeled using bootstrapped stepwise logistic regression (with interactions) accounting for baseline lung function, smoking history, and conventional dosimetric factors. We utilize an appropriately chosen order relation to fuse the conventional MCO sandwiching algorithm with bi-level optimization, restricting the (infinite) Pareto set to plans with substantial gain in the secondary risk objective for acceptable loss in primary (clinical) objectives. Thus, risk-guided MCO computes risk-optimized counterparts to clinical plans in a single run (rather than a sequential/lexicographic approach) within user-defined trade-offs. Performance was assessed in terms of clinical objectives and predicted RP risk.Main results.Across 19 patients, the risk-guided plans yielded an average reduction of 8.0 percentage points in total lung V20Gy and 9.5 percentage points in right lung Gy, translating into an average RP risk reduction of 7.7 percentage points (range = 0.3%-20.1%), with small changes in target coverage (mean-1.2 D98[%] for CTV) and modest increase in heart dose (mean +1.74 Gy).Significance.This study presents, to our knowledge, the first proof-of-concept for integrating biological risk models directly within multi-criteria RT planning, enabling an interactive balance between established population-wide dose protocols and individualized outcome prediction. Our results demonstrate that the risk-informed MCO can reduce the risk of RP while maintaining target coverage.