3129 - Lattice Peak Optimization: Geometry-Adaptive Lattice Radiotherapy Using Mixed-Integer Optimization

Purpose/Objective(s): Lattice radiotherapy (LATTICE) is a spatially fractionated treatment strategy that delivers high radiation doses to discrete high-dose “peak” regions within the tumor while maintaining lower doses in surrounding “valley” regions, resulting in a high peak-to-valley dose ratio (PVDR). Traditional LATTICE planning typically relies on manual or heuristic peak placement subject to geometric constraints, including minimum inter-peak spacing and required distance from organs-at-risk (OAR). These strategies restrict the number of feasible peaks and may limit target coverage or OAR sparing. In this study, we propose a lattice peak optimization (LPO) framework that simultaneously optimizes peak selection and dose distribution to enhance LATTICE plan quality while maximizing the number of deliverable peaks within the target.

Materials/Methods: Proton LATTICE planning is formulated as a mixed-integer optimization model that selects an optimal subset of peaks from a large pool of candidate locations distributed throughout the target. Binary decision variables represent peak selection, and continuous variables define proton spot weights. The formulation incorporates geometric feasibility constraints between peaks while optimizing dosimetric objectives to improve PVDR and reduce OAR dose exposure. The resulting nonconvex problem is addressed using iterative convex relaxation within an alternating direction method of multipliers (ADMM) framework.

Results: The LPO method was evaluated on three clinical cases containing 150-400 candidate peak locations, from which 4-13 peaks were selected. Compared to 50-90 randomly generated LATTICE configurations per case, LPO consistently produced higher PVDR and improved OAR sparing. In an abdominal case, the composite objective value was 2.93 (worst random), 2.40 (median random), 1.90 (best random), and 1.95 (LPO), with similar performance trends observed in the other cases.

Conclusion: We present a geometry-aware mixed-integer optimization framework for lattice peak placement that improves PVDR and OAR sparing compared with manual and random LATTICE approaches. The framework is modality-independent and can be readily extended to photon-based LATTICE planning.