Liver cancer presents great challenge for radiation therapy (RT) due to the desired preservation of liver function, tumor proximity to critical organs, and motion uncertainties due to respiration and peristalsis; these confounding factors are further amplified when in the setting of bulky liver tumors (1–3). Radiation dose escalation is limited due to the risk of radiation-induced liver disease (RILD) and the need to spare adjacent gastrointestinal organs, which often leads to conservative dose constraints even with utilization of advanced imaging and motion management techniques during radiation delivery (1, 4). Maintaining functional hepatic reserve is a primary determinant of safety. Classic NTCP/QUANTEC analyses link mean liver dose and spared liver volume to the risk of classic RILD and contemporary guidelines for primary liver cancers recommend preserving a critical volume of uninvolved liver (e.g., ≥700 cm³ kept below 15 Gy for 3 fractions or 21 Gy for 5 fractions in SBRT schemas) and constraining mean liver dose within 28 Gy or 32 Gy depending on the diagnosis, particularly in cirrhotic patients (1, 5, 6). These realities motivate development of radiation therapy modalities that would intensify tumor dose while maintaining mean liver dose and meeting organ-at-risk (OAR) dose constraints.

Spatial Fractionated Radiotherapy (SFRT) has re-emerged as an advanced radiation therapy approach that is particularly well-suited to treat deep-seated lesions (7). Rather than aiming for uniform target coverage, SFRT intentionally delivers a spatially modulated dose with alternating high-dose “peaks/vertices” and low-dose “valleys,” while maintaining OAR dose within tolerance. Historically, GRID, one types of SFRT, used a physical GRID block to create a 2D pattern (8); one drawback to GRID is that the 2D treatment approach is not well suited for deep-seated tumors due to rapid dose falloff as a function of depth. Subsequent advances in multi-leaf collimation and inverse planning has enabled the wide-spread adoption of lattice radiotherapy (LRT), another form of SFRT in which spherical high-dose vertices are distributed throughout the gross tumor volume (GTV) in a lattice pattern, thereby enabling advantageous dose distribution within deep-seated tumor, such as bulky liver cancer (7, 9, 10). Current publications about LRT have established general consensus about vertex coverage, valley dose, peak-to-valley dose ratio (PVDR), and planning approaches, which has enabled the clinical translation of LRT (11, 12).

LRT is typically utilized using volume modulated arc therapy (VMAT) on standard photon-based linear accelerators and has demonstrated its capability of creating high intratumoral vertices and meeting OAR dose constraints (13, 14). However, exit dose and integral dose remain intrinsic to photons, and maintaining low valley dose becomes progressively more challenging as tumor size increases, potentially limiting the extent of dose escalation or fraction numbers achievable in the setting of photon LRT (12, 14–16).

Proton therapy offers dosimetric advantage over photon-based RT due to its sharp distal fall-off, the so-called Bragg peak, which eliminates beam exit dose, as well as overall integral dose. Given the potential immune modulating advantages of SFRT, reduction of integral dose in the tumor volume is a challenge in photon-based SFRT (17, 18). Naturally, this raises the hypothesis that proton-based LRT could significantly lower valley dose and improve PVDR-like metrics relative to photon-based LRT, particularly for large hepatic targets.

To elucidate the potential utility of proton-based LRT in the setting of bulky liver cancers, we developed a practical treatment planning strategy to implement LRT with intensity modulated proton therapy (IMPT) using a clinical pencil beam scanning (PBS) proton beam line. We also performed a paired dosimetric comparison of IMPT-LRT with photon VMAT-LRT in a retrospective cohort of liver cancer patients with bulky disease. The aim of this work was to determine if IMPT-LRT is superior in minimization of valley dose and maximization of PVDR dose characteristics, while maintaining peak vertex dose coverage, as compared to standard photon-based VMAT-LRT, under a common clinically prescribed, single-fraction 20Gy LRT treatment (11, 12, 16, 19, 20).

This work contributes three elements for the clinical SFRT community: (i) a reproducible, liver-specific IMPT-LRT planning workflow with robust optimization; (ii) quantitative, paired evidence against a standard-of-care VMAT-LRT implementation using the same lattice geometry; and (iii) a reporting framework anchored to consensus SFRT/LRT metrics to facilitate cross-study comparison and protocol development.

This study establishes a practical PBS IMPT−LRT planning strategy for large liver tumors and demonstrates that IMPT−LRT can simultaneously deepen valleys and strengthen vertex coverage in a paired comparison with contemporary VMAT−LRT under a single−fraction 20 Gy template. The dosimetric advantages we observed are aligned with the physical properties of protons due to its minimal exit dose and lower integral dose, and with SFRT theory that prioritizes low valley dose and robust PVDR−like metrics for therapeutic index.

While we examined proton LRT in this study, there are currently other proton SFRT modalities under investigation. Proton minibeam radiotherapy (pMBRT) had drawn attention due to its striking normal-tissue sparing in preclinical brain/skin models and early patient-specific planning studies using thick multi-slit collimators at the nozzle exit (21–24). However, pMBRT remains largely preclinical because maintaining PVDR is depth-limited and challenging for deep tumors (e.g., PVDR ≈8–12 at ~20 mm, but ≈1.1–1.6 by ~80–90 mm of depth in tissue) (25). Also, pMBRT clinical delivery either relies on mechanical collimation which reduces efficiency and adds neutron production, or requires an optimized/new nozzle since current PBS nozzles are not suitable for magnetic minibeam generation (26–31). These limitations and sensitivity to motion for sub-millimeter patterns make pMBRT less immediately deployable for deep, moving abdominal targets like the liver (4, 24). In contrast, our PBS-based IMPT-LRT strategy is directly analogous to photon LRT (with growing clinical experience), uses standard hardware and TPS, integrates with robust optimization for motion/range, and is therefore more clinically applicable in the near term.

From a translational perspective, enhanced vertex and valley dose ratio and lower liver dose, support several clinical pathways: (i) single−fraction IMPT−LRT as a boost preceding or following conventionally fractionated EBRT; (ii) multi−fraction IMPT−LRT (e.g., 3–5 fractions) to balance biologic effectiveness and OAR recovery; and (iii) selective dose escalation for very large GTVs where photon lattice plans struggle to keep mean liver dose and central hepatobiliary tract within dose constraints. These strategies are consistent with emerging SFRT consensus documents and proton LRT planning reports emphasizing standardized vertex/valley endpoints (VTVH D80, VTVLmean, PVDR-like ratios) and robust delivery practices (range/setup robustness, gating/DIBH, repainting).

Our work focused on bulky liver cancer due to the complexity of treating this disease, as well as recent clinical results with liver LRT that show it is an emerging treatment approach. A 2023 systematic review concluded that LRT appears safe, with heterogeneous but promising efficacy signals across multiple treatment sites, while emphasizing the need for standardized endpoints and prospective data specific to the liver (32). For hepatocellular carcinoma (HCC), dosimetric feasibility work shows that VMAT-LRT can escalate intratumoral dose when conventional SBRT is OAR-limited, supporting the rationale for spatially modulated boosts in challenging hepatic presentations (8, 33). Clinically, a recent liver-focused case demonstrated rapid palliation and marked volumetric response after single-fraction 16 Gy LRT delivered to ~1% of the metastatic liver burden while keeping the mean liver dose ≈3 Gy and without acute toxicity, highlighting the potential of LRT to relieve symptoms at very low integral dose (15). Considering these circumstances, our PBS proton LRT workflow is designed to enhance valley sparing and lower non-target liver dose relative to VMAT-LRT, which should lead to better protection of functional parenchyma and facilitate dose intensification or fraction reduction in large hepatic targets. Mechanistically, improved valley sparing may better protect functional liver parenchyma, potentially lowering the risk of RILD when combined with following EBRT (1, 5, 34). At the same time, higher vertex doses could enhance intratumoral cytotoxicity and immune priming in heterogeneous hypoxic tumors (35–37). Our findings therefore provide a dosimetric foundation for protocol development in hepatic SFRT/LRT that prospectively measures liver function (e.g., ICG, HIDA, or SPECT/PET) alongside clinical toxicity and local control.

Delivery considerations are paramount for the liver. Proton PBS delivery is more sensitive to interplay effect due to respiration and peristalsis motions, and proton range uncertainty comparing with photon modalities. As a result, proper planning approaches and motion management strategies need to be established for IMPT-LRT to mitigate setup uncertainties, anatomical variations (from intra- and inter-fractional motion), and range uncertainties. In this study, inspiration breath hold gating was used during initial simulation for the patients included in this study. During planning, a 3-mm setup and 3.5% range uncertainty parameters were used in robust optimization, which is consistent with our clinical stereotactic body radiation therapy treatments that utilize protons. The 3mm setup uncertainty takes into account of patient setup variation and spot positioning accuracy. For liver stereotactic treatment, CT-on-Rails (CTOR) which has superior imaging quality than CBCT will be used during patient setup to minimize the setup error. Spot position accuracy is checked daily and monitored by the ProBeat system during treatment, where an averaged 1mm accuracy is expected. Before implementing in clinic, End-to-end tests will be conducted to evaluate the efficiency of the 3mm setup uncertainty.

This study had several limitations. First, this study used a planning−only study design (lack of delivery QA). The vertex integrity and valley sparing could be compromised during treatment due to variation of spot positioning, range uncertainties, patient’s position variations. End-to-end tests would be needed to establish the treatment protocol. Second, we used a relative biological effectiveness (RBE) factor of 1.1 for proton therapy, which may not sufficiently capture high−dose−per−fraction biology. We did not perform organ−specific statistical testing or subgroup analyses by GTV size. Nevertheless, the consistency of valley sparing across all cases and the magnitude of the PVDR−like improvements support the generalizability of the core finding. The second limitation is lack of strategy for robustness evaluation. During our clinical implementation of IMPT SFRT, for each candidate plan we will compute robustness scenarios including standard setup and range uncertainties and, critically, motion-specific scenarios, then assess three SFRT metrics per scenario: peak coverage within VTVH (e.g., D95), valley dose in the valley volume adjacent to the relevant OAR interface, and the resulting PVDR. For free-breathing deliveries, robustness will explicitly evaluate dose calculated on scans of T0 and T50; for DIBH, we will propagate the plan across multiple repeat breath-hold CTs acquired at simulation to capture breath-hold variability. We will summarize results as both scenario-average and scenario-minimum (“worst-case”) values, and reported as PVDR_robust, peak-coverage robust, and valley-dose robust, and require that all meet predefined acceptance thresholds, with particular emphasis on preserving PVDR_robust at the tumor–OAR interface. When thresholds are not met, we will iterate geometry and delivery (vertex size/placement, repainting, gating/beam timing, or modest beam re-arrangement. The third limitation is that this study only focuses on liver tumors. We focus on bulky liver tumors because (i) hepatic toxicity tightly correlates with the volume of uninvolved liver receiving low–intermediate dose, making valley sparing particularly impactful; (ii) tumors often in close proximity to stomach/duodenum/bile ducts, where sharper dose fall-off may reduce OAR toxicities; (iii) the liver’s respiratory motion is manageable with 4D robust optimization, gating/repainting, fiducial/surface guidance or breath-hold techniques; and (iv) liver IGRT is mature yet imperfect, so range-modulated valleys may better protect parenchyma under uncertainty. This setting offers a clean readout of proton IMPT LRT’s dosimetric advantage. Future work will evaluate proton IMPT LRT in additional sites and histologies that photon LRT has shown feasibility and early clinical signal, which includes non-small cell lung cancer (NSCLC), head and neck, gynecologic (bulky cervix), and soft-tissue sarcoma. In voluminous NSCLC, retrospective data suggest safety and tumor response (38). Palliative head-and-neck series report symptom relief with acceptable toxicity (39). Prospective multi-site data (LITE SABR M1) support short-term safety in large tumors, and time-resolved case series show rapid palliation across sites (40, 41). For gynecologic disease, far-advanced cervical cancer has demonstrated high metabolic responses with good tolerance after LRT (42). In sarcoma, neoadjuvant extremity cases show technical feasibility of integrating high-dose lattice with standard RT (43). These experiences motivate site-specific proton LRT protocols to test whether enhanced valley-sparing can further reduce normal-tissue dose while preserving (or improving) clinical benefit.

Future work will include LET optimization to further enhance the dose contrast in vertex and valley, and thorough study of organ-specific studies for different lattice construction and different planning techniques, as well as understanding the RBE effects of proton LRT in clinical treatment planning and delivery.

Moreover, only DIBH was used as motion management method in this planning study. Based on our experience in proton treatment, respiratory excursions over 5 mm produce phase-dependent hot/cold spots between T0 (end-inspiration) and T50. This effect would amplify and erode planned PVDR in single-fraction SFRT, where minimal temporal averaging leaves interplay unmitigated. We therefore will explore a threshold-based policy for future study: for motion less than 5mm, we create the plan using free breathing CT scan with T0/T50 explicitly included in robust optimization and apply layered repainting during delivery. For motion equal to or higher than 5mm, we treat with deep-inspiration breath-hold (DIBH) and, to account for breath-hold variability between simulation and treatment, acquire more than 3 repeat DIBH CTs at simulation and include all repeats in robustness optimization. If DIBH proves infeasible, we revert to amplitude/phase gating with compression and increase repainting to further mitigate interplay (44, 45). As a future direction, we will prospectively benchmark PVDR erosion vs motion amplitude, test whether multi-CT DIBH planning stabilizes valley dose relative to single-CT approaches, and define adaptive re-simulation triggers (e.g., >2–3 mm drift in observed motion or breath-hold level). Recent developments in proton SFRT include robust multi-field planning strategies (e.g., primary plus robust complementary beams) and proton arc-based lattice (ARC/SPArc) are concepts that seek to balance vertex intensification with valley suppression and OAR sparing (19, 20, 30). Motion management strategies and planning studies could be explored with proton ARC/SPArc lattice concepts to further reduce the valley dose and minimize entrance dose, albeit at the slight increase of integral dose.

In summary, proton-based IMPT−LRT is feasible and shows significant dosimetric advantages over traditional photon-based VMAT−LRT for liver tumors in this paired planning study. IMPT−LRT provided superior valley sparing and higher vertex coverage than VMAT−LRT for large liver tumors, with comparable PRV hotspots under a shared lattice template. In terms of plan quality and dosimetry, our results establish IMPT-LRT as a superior modality to VMAT-based LRT with deeper valleys, stronger peaks, same OAR safety, and provide a quantitative basis for prospective clinical implementation (e.g., boost strategies or multi-fraction LRT).