Dosimetric Impact and Correction of Lipiodol Deposition on Photon-Beam Radiotherapy for Hepatocellular Carcinoma Using a Novel Simulation Model

Introduction

Liver cancer is the third leading cause of cancer-related mortality worldwide, with approximately 750,000 deaths annually.¹ Hepatocellular carcinoma (HCC) accounts for 75–85% of cases of liver cancer.² Surgical resection is used as curative therapy, but only 25–40% of HCC tumors are cured by hepatectomy.^3,4 Transcatheter arterial chemoembolization (TACE) is the preferred treatment for unresectable HCC and is the first option for palliative therapy in patients with preserved liver function.^4–7 Combining TACE with radiotherapy improves survival.^8,9 A meta-analysis revealed that in patients with unresectable HCC, treatment with TACE combined with radiotherapy extended the median survival by 13.5 to 22.7 months compared with that of patients treated with TACE alone.¹⁰

In the photon-beam treatment planning system (TPS), the radiotherapy dose distribution is calculated based on conversion of the Hounsfield unit (HU) values on CT images to relative electron density (RED). HU values are converted to RED using predefined calibration curves specific to the scanning parameters, including tube voltage (kVp) and reconstruction algorithm.¹¹ Therefore, CT scanning parameters, particularly kVp, can influence HU values and consequently affect the HU-RED calibration curve used in the TPS.¹² Lipiodol is an iodine compound that is widely used as an embolic agent during TACE treatment in patients with HCC. In contrast to iodine contrast agents, Lipiodol selectively deposits in HCC lesions without “washing out” and remains visible with elevated HU values on CT images over an extended period. The dosimetric effect of iodine contrast agents is well documented,^13–15 but few studies have investigated the dosimetric influence of Lipiodol. The Lipiodol area leads to an elevation of HU values in tumor lesions on CT images and potential uncertainty of the dose evaluation in the TPS. Moreover, heterogeneous Lipiodol deposition results in spatially heterogeneous HU changes, further increasing the complexity of dose calculation. The TPS has been reported to overestimate the blocking ability of Lipiodol on the proton beam, and the elevated HU values of the Lipiodol area lead to a deeper beam depth, resulting in a higher dose administered to the tumor area and organs at risk (OARs) than planned.¹⁶ Liquid iodine markers have been reported to introduce a 4.8% dose perturbation in proton radiotherapy for lung cancer.¹⁷ However, to our knowledge, the dosimetric influence and correction of Lipiodol area in photon-beam radiotherapy for HCC has not previously been investigated.

CT scanning parameters, particularly kVp, can influence HU values and consequently affect the HU-RED calibration curve used in TPS. In photon-beam TPS, HU values are converted to RED using predefined calibration curves that are specific to the acquisition kVp and reconstruction algorithm. In this study, CT simulation was performed using fixed scanning parameters to minimize potential variability in HU-RED conversion.

To simplify the initial investigation, this study first evaluated the dosimetric impact under the assumption of homogeneous Lipiodol distribution. Therefore, we developed a novel Lipiodol deposition simulation model (LDSM) and evaluated the influence of Lipiodol area on photon-beam dose distribution by comparing the dose calculated by the TPS (D_TPS) and the detected dose (D_DET). We evaluated the RED of the Lipiodol area in the TPS and the dosimetric influence based on simulation CT images of patients with HCC who had undergone TACE to serve as a source of reference for radiotherapy planning. In this study, as CT scanning parameters were standardized in this study, their potential impact on HU–RED conversion was not specifically investigated.

Material and Methods

Development of the Lipiodol Deposition Simulation Model (LDSM)

Lipiodol samples were prepared by mixing churned porcine liver (100 g, Supplementary Figure 1a) with Lipiodol (Libodo, 480 mg I/mL, 10 mL/branch, Supplementary Figure 1b) in six different ratios (0 mL/100 g, 2 mL/100 g, 5 mL/100 g, 10 mL/100 g, 15 mL/100 g, and 20 mL/100 g), which were loaded into 60 mL polyethylene sample tubes (Supplementary Figure 1c) with a density similar to that of water (Supplementary Figure 1d and e). To enable CT scanning and dosimetry, the model was assembled with the ionization chamber (A1SL) channel in the middle and the porcine liver-Lipiodol samples loaded into the abdominal simulation phantom (CIRS_008A, Supplementary Figure 1f).

The electronic density of Lipiodol samples was calibrated to avoid bias in HU values between different CT scanners and TPS. The sample tubes loaded with Lipiodol samples were assembled in groups of three and inserted into the abdominal simulation CT phantom (Supplementary Figure 1g) with the ionization chamber detector placed at the center of each of the six sample tubes. The CT scans (Figure 1) were performed using a CT simulator (Siemens SOMATOM Definition AS CT; tube voltage, 120 kV; tube current, 300 mA), and transferred directly to the TPS (Pinnacle³ version 16.2; Philips Healthcare, Cambridge, MA, USA). The HU-RED calibration curve, established under identical 120-kV parameters and reconstruction algorithm, was applied to derive mean RED values for Lipiodol samples in the TPS.

Figure 1 Computed tomography (CT) scans of the lipiodol samples based on the Lipiodol deposition simulation model (LDSM). (a) The ionization chamber detector was located at the center of the 2 mL/100 g sample tube; (b) the ionization chamber detector was located at the center of the 15 mL/100 g sample tube. The blue square box labeled A indicate the anterior-posterior (A-P) direction.

Dose Comparison and Correction

The Lipiodol area, ionization chamber detectors, and OARs were delineated for each CT image. In the Pinnacle³ TPS with the collapsed cone convolution (CCC) algorithm, 6 MV X-rays were set to irradiate Lipiodol samples in different fractional doses (1–10 Gy/f), with flattening-filter (FF) or flattening-filter-free (FFF) beams, and the D_TPS was recorded. The charge numbers of the ionization chamber at the center of each sample tube were detected using an electrostatic meter, and D_DET was recorded when using the linear accelerator to irradiate the CT phantom inserted with sample tubes with the same beam as in the TPS. The relationship between the relative dose deviation and RED in the Lipiodol area was assessed using the dose deviation factor (δ), which was calculated using the equation:

To facilitate the initial study, a homogeneous Lipiodol deposition was assumed as the prerequisite for RED correction. The RED correction of the Lipiodol area was performed in three steps (Supplementary Figure 2). First, the Lipiodol area was identified on CT images based on the gross tumor volume (GTV) boundary, and mean HU values were extracted. Second, RED of Lipiodol area was estimated using phantom-derived Lipiodol concentration-RED calibration curves. Finally, the corresponding δ was applied to manually adjust RED in the TPS to ensure consistency with D_DET.

Patients and Re-Evaluation of Radiotherapy Plans

The records of patients with HCC who had been treated with TACE between May 2019 and September 2024 were retrospectively reviewed. The eligibility criteria included treatment with photon-beam radiotherapy and >50% uptake of Lipiodol after TACE. The cancer was staged according to the American Joint Commission on Cancer Staging Manual (eighth edition).¹⁸ The sample size (n = 60) was determined based on data availability within our institution and is comparable to that reported in similar dosimetric studies. This sample size was considered sufficient to demonstrate the observed effects and to support the objectives of the present study. This study protocol was approved by the Independent Ethics Committee of the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences, and Peking Union Medical College (approval number 22/094-3295). The requirement for informed consent was waived because of the retrospective study design.

The CT images of the patients were corrected as described above. To evaluate the influence of the Lipiodol area on the dose distribution of the photon-beam, we compared the planning strategies using the uncorrected and corrected CT images. The volume for RED correction on corrected CT images was determined by the GTV with a homogenous distribution of Lipiodol assumed. The dose distribution was then recalculated based on the corrected RED values using the same planning settings with uncorrected CT images. The influence of the Lipiodol area was quantified using the prescription dose coverage (V100%) and conformity index (CI) of the GTV, 50% prescription dose coverage (V50%) of normal liver, and maximum dose (D_MAX) of gastrointestinal tissue.

Comparison of Different TPS

The consistency of dose deviation was assessed in different TPS, incorporating six distinct dose calculation algorithms, including RayStation version 12B (RaySearch Laboratories, Stockholm, Sweden) with the CCC and Monte Carlo (MC) algorithms, Eclipse version 16 (Varian Medical Systems, Palo Alto, CA, USA) with the anisotropic analytical algorithm (AAA) and Acuros XB (AXB), Monaco version 5.61 (Elekta, Stockholm, Sweden) with the MC algorithm, and TomoTherapy version 7 (Accuray Incorporated, Madison, WI, USA) with the convolution-based algorithm.

Statistical Methods

All statistical analyses were performed using SPSS Statistics, version 20.0 (IBM Corp., Armonk, NY, USA). Paired-samples t-tests were used for dosimetric comparisons of phantom experiments and patients’ radiotherapy plans. Univariate and multivariate analyses were performed using binary logistic regression model to evaluate the factors associated with clinical dosimetric correction. All patient baseline characteristics were initially assessed. Tumor-related and physical parameters were evaluated for their association with clinical dose deviation, while demographic variables and treatment parameters were excluded. Furthermore, the mean density values of normal liver tissue were excluded due to minimal variability across patients. All variables considered were included in the multivariate model. Statistical significance was set at p < 0.05.

Results

Baseline RED Assessment of LDSM

The calibration curves of the relationship between the mean RED of the Lipiodol samples and the Lipiodol concentration are shown in Supplementary Figure 3 (R² = 0.9994). The mean RED of the sample without Lipiodol was 1.062 g/cm³, consistent with the RED of normal human liver tissue (1.017–1.077 g/cm³),¹⁹ enabling the LDSM to accurately simulate intrahepatic Lipiodol deposition in humans.

Lipiodol Area Leads to a Dose Underestimation in the TPS

In the Pinnacle³ TPS with the CCC algorithm, the D_TPS values were less than the D_DET values, revealing an underestimation of dose in the photon-beam TPS due to the uncorrected RED of the Lipiodol area. The relationship between δ and RED of Lipiodol area in different fractional dose (range, 1–10 Gy/f) and by FF and FFF beams is shown in Figure 2. The other δ values were derived through linear interpolation, with average linearity coefficients of R² = 0.9985 for FF beams and R² = 0.9975 for FFF beams. The δ value of samples without Lipiodol mixing (0 mL/100 g) was close to 0% (0.06 ± 0.04%), confirming that the calculated TPS values were consistent with the measured values without the influence of the Lipiodol area. The δ value of each specific fractional dose increased gradually as the Lipiodol concentration increased (range, 0.5–5.7%). The overall δ value of the FFF beam was higher than that of the FF beam with the same fractional dose (range, 0.5–5.7% vs. 0.3–4.3%). The detailed data was shown in Supplementary Table 1.

Figure 2 Dose deviation based on the phantom study. The dose deviation factor (δ) between the calculated dose (DTPS) and the detected dose (DDET) within Lipiodol area of varying relative electron density (RED) under different fractional dose (1–10 Gy/fraction) and for both flattening-filter (FF) and flattening-filter-free (FFF) beams. (The other δ values were derived through linear interpolation, with average linearity coefficients of R2 = 0.9985 for FF beams and R2 = 0.9975 for FFF beams.).

Patient Characteristics

The records of 60 patients who underwent photon-beam radiotherapy with intrahepatic Lipiodol deposition after TACE were reviewed. Their characteristics are shown in Table 1. They were predominantly male (n = 52, 87%) with a median age of 60 years (range, 29–81 years), median GTV of 491.6 mL (range, 53.6–2113.0 mL), and median shortest distance (D) from the gastrointestinal tissue of 2.65 cm (range, 0.1–7.2 cm). The prescription dose to 95% of the GTV and planning target volume was 6000 cGy and 5000 cGy, respectively, in 25 fractions. The majority of patients had tumors located in the right hepatic lobe, and 80.0%) had stage III or IV disease. The mean density values of normal liver tissue and the Lipiodol area based on the conversion of the HU values from the patients’ CT images, were 1.07 g/cm³ (range, 1.05–1.09 g/cm³) and 1.61 g/cm³ (range, 1.23–1.98 g/cm³), respectively.

Table 1 Baseline Patient and Tumor Characteristics and Radiation Therapy Parameters

Re-Evaluation of Patients’ Radiotherapy Plans

The mean density values of corrected Lipiodol area were 1.35 g/cm³ (range, 1.14–1.55 g/cm³). The dosimetric deviation values before and after RED correction of Lipiodol area on CT images are shown in Table 2. When the radiotherapy plans were recalculated using the corrected RED values, the V100% of the GTV increased by 3.42 ± 0.92% (range, 2.00–4.99%, p < 0.001), but the CI did not change significantly (p = 0.52). The V50% of normal liver tissues increased by 0.72± 0.36% (range, 0%–1.3%, p < 0.001) and the gastrointestinal D_MAX increased by 183.22 ± 138.80 cGy (range, 2.00–453.00 cGy, p < 0.001). After RED correction, 26 (39.4%) patients had an increase in the gastrointestinal DMAX of more than 200 cGy.

Table 2 Dosimetric Deviation Before and After Relative Electron Density Correction of the Lipoidal Area

An example of the CT images of a typical patient is shown in Figure 3. Figure 3a and 3b were generated from the same CT slice of the same patient and demonstrate different dose distributions calculated using different assigned RED values of the Lipiodol area. In this example, the original RED of the Lipiodol area was 1.98 g/cm³, and the corrected RED was 1.55 g/cm³. The RED values are not directly displayed on the CT screenshots. These values correspond to the RED assigned within the Pinnacle treatment planning system for dose calculation purposes. Specifically, 1.98 g/cm³ represents the original RED automatically assigned to the lipiodol area based on CT calibration, while 1.55 g/cm³ represents the corrected RED used to more accurately reflect the physical electron density of lipiodol after correction. After RED correction, the percentage of the GTV V100% increased by 4.00%, the CI of the GTV increased by 2.10%, the V50% of the normal liver increased by 0.22%, the gastrointestinal (GI) D_MAX increased 485 cGy from 5573 cGy to 6058 cGy, an increase of 485 cGy. Visually, Figure 3b shows greater overlap between the 5600 cGy (yellow) and 6000 cGy (blue) isodose lines and the adjacent gastrointestine (green colorwash), compared with Figure 3a. In this example, the re-evaluation changed the dose from a clinically acceptable dose to a dose exceeding the maximum recommended dose.

Figure 3 Dosimetric evaluation of patients’ radiotherapy plans. (a) Dose distribution based on computed tomography (CT) image without relative electron density (RED) correction of Lipiodol area; (b) Dose distribution based on CT image with relative RED correction of Lipiodol area. The Orange color arrow indicates the increased gastrointestinal (GI) maximum dose (DMAX) after RED correction.

Factors Affecting RED Correction

An increase in the gastrointestinal D_MAX of >200 cGy was defined as a dose deviation requiring clinical RED correction. The shortest distance (D) between the tumor and gastrointestinal tissue was significantly correlated with the gastrointestinal D_MAX and required clinical dosimetric correction of the Lipiodol area (p < 0.001, Table 3). The predictive value of the D for dose exceeding of the gastrointestinal D_MAX was assessed using receiver operating characteristic (ROC) curve analysis. D had a strong predictive value (area under curve, 0.868; 95% CI, 0.779–0.957), with an optimal cutoff of 2.75 cm (Figure 4).

Table 3 Analysis of Factors Associated with Clinical Dosimetric Deviation

Figure 4 Receiver operating characteristic (ROC) curve analysis. Predictive value of the distance (D) for estimating the necessity of RED correction of the Lipiodol area and D is the shortest distance between the gross tumor volume (GTV) and the gastrointestinal (GI) tissue.

Comparison of Dose Deviations Using Different TPS

The dose calculation accuracy of LDSM differed by TPS type. RayStation with the CCC algorithm underestimated the dose (Figure 5a); RayStation with the MC algorithm showed minimal dose deviation (Figure 5b); Eclipse with the AAA showed moderate dose deviation (Figure 5c); Eclipse with the AXB algorithm showed minimal dose deviation (Figure 5d); TomoTherapy with the convolution-based algorithm showed marked dose deviation (Figure 5e); and Monaco with the MC algorithm showed minimal dose deviation (Figure 5f).

Figure 5 Dose deviation in different TPS with different algorithms. (a) RayStation with collapsed cone convolution (CCC) algorithm; (b) RayStation with Monte Carlo (MC) algorithm; (c) Eclipse with anisotropic analytical algorithm (AAA); (d) Eclipse with Acuros XB (AXB) algorithm; (e) TomoTherapy with convolution-based algorithm; (f) Monaco with MC algorithm.

Discussion

To our knowledge, this is the first study to quantify the dosimetric influence and correction of Lipiodol on photon-beam radiotherapy after TACE by comparing the calculated value in the TPS with the detected value. We developed a novel LDSM using different ratios of Lipiodol and porcine liver mixtures to simulate the intrahepatic Lipiodol area after TACE with different Lipiodol densities and re-evaluated the dose distribution based on patients’ radiotherapy plans with RED correction. ROC curve analysis was used to provide a reference for clinical RED correction of the Lipiodol area. To prevent underestimation of the GI dose in photon-beam TPS, RED correction of the Lipiodol area is recommended when D is less than 2.75 cm, which may serve as a practical cutoff for clinical application.

In photon-beam TPS, the HU values of the CT images were converted to RED accurately, using predefined calibration curves. Although HU-RED conversion is precise, dose calculations based solely on RED values introduce systematic errors in regions containing high atomic number (Z) materials, such as Lipiodol. This is because most convolution-based TPS algorithms primarily account for electron density while neglecting enhanced photon interactions, particularly photoelectric absorption and secondary-electron generation, associated with high-Z elements such as iodine (Z = 53). The dose deviation was particularly evident in TPS algorithms that simplified scatter modeling and secondary electron effects. Devices such as Pinnacle and RayStation which use the CCC algorithm, and Eclipse which use the AAA, do not fully capture the complexity of photon interactions in high-Z materials. As a result, these systems tend to underestimate the localized dose in Lipiodol-rich regions. This underestimation is more pronounced in the FFF-beam mode, in which the increased proportion of low-energy photons amplifies the photoelectric effect, than that in the FF-beam mode. In contrast, MC-based algorithms, such as those in Monaco and RayStation Monte Carlo mode, provide more accurate simulations of photon interactions. By tracking individual photon and secondary electron paths, these algorithms effectively model photoelectric effects, Compton scattering, and electron cascades, significantly reduce dose discrepancies in high-Z environments.

Dosimetric correction methods for iodine contrast agent and metallic artifacts in TPS have been investigated and applied in clinical radiotherapy planning.^13,20–22 However, to our knowledge, correction for Lipiodol used during TACE in photon-beam TPS has not been reported previously. The RED of the Lipiodol area is typically either uncorrected or approximated to the density of water during post-TACE radiotherapy planning. However, neither of these methods assesses the dose of radiation accurately, leading to dose uncertainty during radiotherapy planning. Because current TPS platforms do not allow direct modification of the Z value, we propose a correction method of adjusting RED values to create an equivalent RED that incorporates the effects of atomic number, thereby compensating for Z-related discrepancies without altering the TPS structure. These equivalent RED values do not reflect true electron density but are modified to ensure accurate dose calculations in Lipiodol-rich areas, offering a practical solution to overcome the inherent limitations of current TPS algorithms, enabling more precise dose distribution in both target volumes and OARs, providing a theoretical basis for precise post-TACE radiotherapy planning in patients with HCC, and improving the safety of radiotherapy.

This study aims to facilitate radiotherapy planning and provide a theoretical foundation for more accurate radiotherapy for post-TACE patients in HCC, enhancing the precision and safety of radiotherapy. The uncorrected RED of Lipiodol area led to dose underestimation in the photon-beam TPS, leading to increased risk of complications in OARs. Clinical RED correction of Lipiodol area had a significant effect on the dose calculation in both the tumor and OARs. Although an increase in the average dose had minimal influence on normal liver with a larger volume, the uncorrected RED of Lipiodol area introduced dose uncertainty and underestimated the radiation dose delivered to the critical OARs, such as the gastrointestinal tissue, which is proximal to the tumor, potentially leading to radiation toxicity. Gastrointestinal injury is a dose-limiting toxicity for patients with HCC treated with radiotherapy; therefore, RED correction of the Lipiodol area is particularly important for accurate dose evaluation of the target volume and gastrointestinal tissue.

Using an aqueous model containing Lipiodol based on the MC algorithm, Kawahara et al found a bias in the dose calculation of Lipiodol in TPS, in which the calculated dose in TPS was lower than the simulated dose of a stereotactic body radiotherapy photon-beam stream²³ The dosimetric bias was similar to that found in this study. A subsequent study evaluating the dose-enhancing effect of Lipiodol indicated an increased local absorbed dose and relative biological effect in Lipiodol area, found greater bias with FFF beams than with FF beams,²⁴ consistent with the findings of this study. In patients treated using the 10-MV FFF beam, the dose inside the tumor was increased by approximately 11% and 10% due to the biological and physical effects of Lipiodol, respectively.^25,26 In our institution, FFF beams are used in the majority of patients; however, the FF beam results are relevant for centers that use FF beams. However, these previous studies primarily relied on MC-based simulations or theoretical modeling to evaluate dose deviations. In contrast, we quantified the dosimetric influence and correction of Lipiodol by integrating phantom-based measurements with evaluation of patients’ radiotherapy plans in TPS, enabling a more refined and clinically grounded validation of dose deviations. Moreover, this study proposes a clinical method for correcting dose deviations, which can be directly integrated into clinical radiotherapy planning workflows.

Establishing the LDSM ensures the precision and repeatability of this study. In previous dosimetry studies, pure Lipiodol was typically used to replace intrahepatic Lipiodol deposition instead of considering the different Lipiodol concentrations in patients.^16,23 Currently, no models are available to simulate the Lipiodol area with different densities after TACE. In preliminary experiments, we explored the optimal mixing ratio of Lipiodol and porcine liver by blending different concentration gradients of Lipiodol with the porcine liver for CT scan, and designed and manufactured a specialized sample tube for a phantom capable of internal dose detection, comprising multiple hollow polyethylene sample tubes that could house the detectors, assembled securely using threaded connections, forming the basis for the LDSM.

This study has some limitations. First, as a single-center retrospective study, inherent selection bias may exist and multicenter studies are warranted to further validate the robustness and clinical applicability of our results. Second, the Lipiodol area in post-TACE patients did not have uniform density, and correcting the entire region to a single density represents a simplification of the heterogeneous intratumoral Lipiodol deposition in clinical practice. Future studies incorporating automatic identification and density assignment, potentially using artificial intelligence, are warranted to better account for heterogeneous deposition patterns. The potential for automatic identification and assignment of density with the help of artificial intelligence warrants further investigation. Third, owing to the heterogeneity of the patients with HCC, we were unable to fully evaluate the effects of tumor size and location on the dose distribution in photon-beam radiotherapy. Therefore, development of a more comprehensive predictive model is required.

Conclusion

The LDSM developed in this study was used to evaluate the influence of the Lipiodol area on the dose distribution of photon-beam radiotherapy. The radiodensity of the Lipiodol area was not accurately estimated using convolution-based or anisotropic analytical algorithms, leading to clinically significant dose underestimation in the TPS. We recommend dosimetric correction of the Lipiodol area to achieve accurate dose evaluation of both the target volume and the OARs, especially in the TPS using convolution-based or anisotropic analytical algorithms. The applicability of the proposed method to heterogeneous Lipiodol deposition warrants further investigation, and long-term clinical outcomes require additional follow-up studies.