A systematic evaluation of lattice radiation therapy: from automated vertex placement to dosimetric criteria

Objective. This study aims to establish a predictive, standardized lattice radiotherapy (LRT) workflow by systematically evaluating crystallographic close-packing strategies, formulating volume-stratified parameter guidelines, and implementing an integrated dosimetric evaluation framework. Approach. Lattice structures (4,095 configurations) were constructed across 21 patients with bulky tumours (613–1909 cm³) utilising three automated close-packing models (face-centred cubic [FCC], axial hexagonal close-packed [HCPxoy], and coronal rotated [HCPyoz]) parameterised by vertex diameter (Dlat) and the ratio of centre-to-centre distance-to-diameter (Dctc/Dlat). After evaluating geometric scaling, a representative sub-cohort spanning the volume spectrum was selected to generate 33 high-dose VMAT plans (20 Gy in 2 fractions, D0.03cc ≥ 140%). Dosimetric evaluation was conducted using global indices (PVDR1, PVDR2) and an automated 1D local gradient scoring algorithm (PVDR3, Mean ± SD). Main results. While Dlat and Dctc/Dlat governed geometric scaling, lattice volume occupancy (LVO) and vertex count (Nlat) did not differ significantly across placement models and rotational angles (p > 0.05), indicating that geometric placement results were insensitive to placement strategies. In parameter sensitivity plans, local PVDR3 increased from 1.67 ± 0.28 to 5.41 ± 2.91 as Dctc/Dlat expanded from 2.0 to 3.5. A volume-stratified Dctc/Dlat ratio of 3.0–3.5 was optimal for ensuring geometric eligibility (1% ≤ LVO ≤ 10%) while guaranteeing steep physical gradients (PVDR3 ≥ 3.0): a ratio of 3.5 maximises local fall-off (PVDR3 = 5.41 ± 2.91) in ultra-bulky targets (>1200 cm³), whereas 3.0 is the necessary standard for medium-bulky targets (600–1200 cm³, PVDR3 = 3.05 ± 0.63). Although vertical staggering (Δz) physically reduced local dose bridging (p < 0.01), plan-level PVDR3 showed no statistically significant differences across placement models (p > 0.05). Significance. This work transitions LRT from empirical trial-and-error to a deterministic, parameter-driven framework. The proposed volume-stratified guidelines, flexible close-packing models, and automated PVDR3 scoring pipeline provide essential physical benchmarks for clinical trial standardisation.