Incorporating sensitive cardiac substructure sparing into radiation therapy planning

Abstract Purpose Rising evidence suggests that cardiac substructures are highly radiosensitive. However, they are not routinely considered in treatment planning as they are not readily visualized on treatment planning CTs (TPCTs). This work integrated the soft tissue contrast provided by low‐field MRIs acquired on an MR‐linac via image registration to further enable cardiac substructure sparing on TPCTs. Methods Sixteen upper thoracic patients treated at various breathing states (7 end‐exhalation, 7 end‐inhalation, 2 free‐breathing) on a 0.35T MR‐linac were retrospectively evaluated. A hybrid MR/CT atlas and a deep learning three‐dimensional (3D) U‐Net propagated 13 substructures to TPCTs. Radiation oncologists revised contours using registered MRIs. Clinical treatment plans were re‐optimized and evaluated for beam arrangement modifications to reduce substructure doses. Dosimetric assessment included mean and maximum (0.03cc) dose, left ventricular volume receiving 5Gy (LV‐V5), and other clinical endpoints. As metrics of plan complexity, total MU and treatment time were evaluated between approaches. Results Cardiac sparing plans reduced the mean heart dose (mean reduction 0.7 ± 0.6, range 0.1 to 2.5 Gy). Re‐optimized plans reduced left anterior descending artery (LADA) mean and LADA0.03cc (0.0–63.9% and 0.0 to 17.3 Gy, respectively). LV0.03cc was reduced by >1.5 Gy for 10 patients while 6 cases had large reductions (>7%) in LV‐V5. Left atrial mean dose was equivalent/reduced in all sparing plans (mean reduction 0.9 ± 1.2 Gy). The left main coronary artery was better spared in all cases for mean dose and D0.03cc. One patient exhibited >10 Gy reduction in D0.03cc to four substructures. There was no statistical difference in treatment time and MU, or clinical endpoints to the planning target volume, lung, esophagus, or spinal cord after re‐optimization. Four patients benefited from new beam arrangements, leading to further dose reductions. Conclusions By introducing 0.35T MRIs acquired on an MR‐linac to verify cardiac substructure segmentations for CT‐based treatment planning, an opportunity was presented for more effective sparing with limited increase in plan complexity. Validation in a larger cohort with appropriate margins offers potential to reduce radiation‐related cardiotoxicities.


| INTRODUCTION
Cardiac toxicity is a major complication of cancer treatment and can occur during, shortly after, and even many years after treatment has been delivered. Long-term follow-up of patients undergoing thoracic radiation, such as lymphoma, lung, breast, and esophageal cancers, has shown that in particular, radiation therapy (RT) can lead to radiation-induced cardiac toxicities such as congestive heart failure, pericardial effusion, coronary artery disease, and myocardial infarction. [1][2][3] Yet, when a patient's RT plan is created, only simple whole heart metrics (i.e., mean heart dose (MHD)) are routinely considered for cardiac risk assessment in the current standard of care. The Quantitative Analysis of Normal Tissue Effects in the Clinic (QUANTEC) report assesses dose to the heart as a whole and recommends <10% of it receives >25 Gy as the endpoint of long-term cardiac mortality. 4 Importantly, these whole-heart dose metrics do not provide any information on where dose is distributed.
The heart is a complex organ and dose to its substructures (e.g., coronary arteries, ventricles, atria, great vessels) have been strongly associated with radiation-induced cardiac morbidity 5 and future acute coronary events. 6,7 For example, dose to the left anterior descending artery (LADA) has been linked to an increased risk of myocardial infarction 8 and development of coronary artery calcifications. 9 Similarly, higher doses at the base of the heart (i.e., ascending aorta, superior vena cava, and pulmonary artery) are associated with lower rates of patient survival. 10 Importantly, recent RTOG 0617 subanalyses suggest that dose to the atrial and ventricular cardiac substructures are more strongly associated with survival than assessing dose/volume relationships to the entire heart volume. [11][12][13] In a recent study by van den Bogaard, 6 dose to the left ventricular volume receiving 5 Gy predicted major coronary events better than MHD. A study by Hoppe et al. highlighted the importance of quantifying substructure dose as the MHD becomes less correlated to substructure dose with increasingly conformal delivery. 14 Furthermore, a study by Jacob et al. outlines how the MHD does not accurately predict dose to the left ventricle (LV) and coronary arteries. 15 To date, reducing dose to sensitive cardiac substructures has been severely limited because they are not readily visible on standard x-ray-based imaging used for both RT planning (i.e., computed tomography simulation (CT-SIM)) and RT delivery (i.e., cone-beam CT (CBCT)). Thus, leveraging the superb soft-tissue contrast of magnetic resonance imaging (MRI) may be advantageous as MRI improves cardiac substructure visibility. 16,17 Furthermore, the recent introduction of MRI guided linear accelerators (MR-linacs, Fig. 1 left) has yielded improved tumor and critical structure visualization at 0.35T MRI as compared to CBCT. 18 MRgRT allows for continuous anatomical visualization of the patient's heart and target volume throughout treatment which may offer advantages for improved cardiac sparing. breath-hold conditions. One patient with a left chest wall lesion could not tolerate breath-hold and thus underwent a 175-second free-breathing MRI for treatment planning. TrueFISP is commonly used in cardiac imaging due to high signal-to-noise ratio and imperviousness to motion artifacts. 19,20 All treatment planning was conducted and dose was calculated on a non-contrast CT-SIM in a manner similar to what has been reported for MR-guided RT of thoracic lesions. 21 All CT-SIMs were acquired on a Brilliance Big Bore CT Simulator (Philips Medical Systems, Cleveland, OH) with a 3-mm slice thickness. MR and CT-SIM sessions were conducted on the same day and patients were immobilized in the supine position using molded vacuum cushions. While automatic segmentation methods (i.e., multi-atlas and deep learning methods) provided initial substructure contours on the CT-SIM datasets, a radiation oncologist consulted the co-registered lowfield MRI to modify and confirm the final contours used for treatment planning. As shown by the lack of contrast in the planning CT ( Fig. 1, center), the enhanced soft tissue contrast from the MRI assisted the generation of more reliable cardiac substructure delineations on the corresponding planning CT. Co-registration involved an automatic rigid registration based off a manually drawn, local, cardiac confined bounding box. Normalized mutual information was used as the similarity metric as it has been shown to accurately align multimodality images. 24

2.C | Treatment planning
For all patients, the CT-SIM was used as the primary image set for treatment planning as has been reported in the literature for MRgRT of thoracic lesions. 21 The co-registration of the low-field MR image to the CT-SIM to elucidate the cardiac substructures was a critical step in allowing the physician to verify the cardiac substructure autosegmentations.
Step-and-shoot intensity modulated radiation therapy (IMRT) planning was used to generate all 16 RT plans at a dose rate of 600 cGy/min. The MR-linac utilizes a fast Monte Carlo dose calculation algorithm 25 and plans were calculated using a 1 × 1 mm dose grid with 1% dose uncertainty. 26 Plans were prescribed to 95% of the planning target volume with total doses for the original treatment plans varying from 30 to 70 Gy delivered in 4-35 fractions. The original treatment plans for all patients included clinical dose constraints for whole heart endpoints. All clinical treatment plans met physician objectives using standard QUANTEC 27,28 and TG-101 29 dosimetric endpoints for OARs.
Along with adding substructure segmentations retrospectively to the original clinical treatment plans for dose assessment, all plans were re-optimized to spare cardiac substructures (SPARE plan).
Strategies for substructure sparing included evaluating the original plan to identify which cardiac substructures were near the planning target volume (PTV) and thus received the most dose. Optimization objectives were then added with increased priority on the substructures receiving higher doses. If the dose limit was unachievable, constraints were relaxed with the overall objective to minimize substructure dose. If the dose to a particular substructure was minimal in the original plan, an additional objective was added in the IMRT optimization to ensure consistency was maintained.
In addition to adding substructures to the optimization, possible further cardiac sparing improvement was also assessed through modifying the beam arrangement (New Angles plan) after the substructures had already been incorporated into the optimization. For plans with lesions that are particularly close in proximity to the heart, it was evaluated whether beams entering or exiting the heart could be potentially removed or modified to further spare the heart and substructures. IMRT techniques were used for all SPARE and New Angles plans with the substructures integrated into the optimization while maintaining tumor volume coverage and minimizing organ at risk (OAR) dose. Table 1  Minimize may be taken as "as low as reasonably achievable" (ALARA). Abbreviations defined in the text.
during plan optimization, derived from the literature, when cardiac substructures were included. All plans were converted to standard fractionation using the equivalent dose to 2 Gy fractions (EQD2, α/ β = 2) to allow for uniform evaluation.

2.D | Dosimetric and statistical assessment
Original, SPARE, and when applicable, New Angle plans were exported from the ViewRay planning system and imported into MIM

3.A | Contour generation and plan complexity
The treatment time per fraction (a metric of plan complexity) across the 16 patients after plan re-optimization was 6.57 ± 3.50 minutes (range 2.60-12.41) for the clinical treatment plan and was 6.93 ± 3.27 minutes (range 2.75-11.99) after re-optimizing (P > 0.05). The mean percent difference in the delivered MUs between the original and re-optimized plans was 1.7 ± 11.3% (range −21.6 to 15.8%) which did not yield a statistically significant difference (P > 0.05).
Four patients benefited from New Angles plans where the number of original treatment beams (range 7-11) shifted by anywhere from −1 to + 3 (range [8][9][10][11][12][13][14]. For two of the four patients, lesions were directly adjacent to the heart (i.e., a pericardial lymph node and a malignant neoplasm of the lung (Fig. 5)). The other two patients presented with upper lung lobe lesions that were greater than 9 cm away from the heart. The average treatment time for these patients after beam angle modification was 6.12 ± 3.68 minutes, which was not significantly different (P > 0.05) from the original treatment time for these 4 patients (6.54 ± 3.31 minutes). Lastly, the mean percent difference in the delivered MUs between the original and re-optimized plans for these patients was 9.5 ± 16.8% (range −16.6 to 23.8%, P > 0.05).

3.B | Cardiac Substructure Sparing
The radiation dose to the whole heart after plan re-optimization met all clinical objectives. 27,28 All sparing plans significantly reduced the MHD (P < 0.05) with an average reduction of 0.7 ± 0.6 Gy (range 0.1 to 2.5 Gy). Furthermore, D 0.03cc to the heart was reduced by 8.6 ± 12.1 Gy (range −8.6 to 39.9 Gy) across all patients after plan re-optimization (P < 0.05). presented in Table 1), 4 were brought below 10 Gy after re-optimization (average reduction for these patients was 13.4 ± 7.0 Gy). D 0.03cc for the remaining patient was reduced from 29.0 to 11.2 Gy.
Similarly, D 0.03cc to the LV was reduced in 14 cases (range 0.05 to 12.85 Gy) with 10 patients having >1.5 Gy reductions. There was a large reduction (>7%) in LV-V5 for 6 patients with an initial LV-V5 greater than 10%. LA mean dose (Fig. 2, center) was either equivalent or reduced (average reduction 0.9 ± 1.2 Gy) for all SPARE plans.
For Patient 3, the left atrial mean dose was reduced to <8.5 Gy which has been shown to be a threshold associated with decreased survival, 10 and highlights the importance of optimizing plans while considering these thresholds. Lastly, the left atrial maximum dose, which has been significantly associated with non-cancer death, 30 was reduced by 2.3 ± 6.4 Gy across all 16 patients.   Table 2 Fig. 3.  We have also shown here that negligible increases in treatment time per fraction and MUs delivered after plan re-optimization were observed, suggesting similar complexity of the radiation treatment plan. Moreover, even though the modified beam angles plans involved either adding or removing beams in the revised treatment plan, the differences in treatment time per fraction and MUs delivered were negligible (P < 0.05). This shows that there will be a negligible practical penalty at the machine for incorporating cardiac substructures in the treatment planning process.

3.D | Individual patient results
We also found that modifying the beam angle and number of beams used to consider cardiac substructures after the plan had been re-optimized also had the potential to increase cardiac substructure radiation sparing. However, much like the findings by Lester et al., 36 the results were patient specific as lesion location and proximity to the heart and its substructures played a role in if the patient would benefit from plan re-optimization and beam modification. Patients that benefited from beam angle modification varied in both the number of beams added or removed and in the proximity of the lesion to the heart (i.e., directly adjacent). So, although beam angle modification was shown to provide improvements over solely re-optimizing the plan for select cases (4/16 cases), re-optimization alone provided the majority of cardiac substructure sparing, and thus we have shown that simply including substructures in the optimization will provide benefit to a large portion of patients. We found that tumor location also plays a role in the extent a substructure is able to be spared, regardless of plan geometry. For example, the LA for Patient 2 was directly adjacent to the tumor volume yet the mean dose difference after re-optimization of the LA as shown in It is worth noting that although there was a statistically significant sparing of mean dose to the heart achieved after plan re-optimization, this may be due to the added weight in the optimizer for when all the substructures are included. However, Fig. 4 highlights that standard whole heart dose metrics were not sensitive to a cardiac sparing treatment planning approach, whereas individual substructure endpoints clearly identified dosimetric, and clinically meaningful gains (i.e., associated with clinical outcomes). Furthermore, the insufficiency of quantifying the MHD alone has been recently affirmed by studies recommending the inclusion of cardiac substructures as RT treatments become more conformal (i.e., intensity modulated RT). 15,37 For example, the LV-V5, which has been shown to be more predictive of acute cardiac events than mean heart dose, 6 was reduced~15% and the mean dose to the AA was reduced by~6 Gy, suggesting that with confirmation in a larger cohort, further sparing may offer potential for improved survival. 10 This underscores the importance of using more sensitive metrics for dose evaluation and not simple whole-heart evaluations that are currently being implemented. Finally, increasing the size of the patient cohort with varied target locations will help identify the patient geometries that will benefit most from cardiac substructure sparing, as discussed above. However, the size of the patient cohort in the current study is consistent with the previously mentioned studies where 7-8 patients were used. 36,39 An increase of size such as this could be completed through applying this work to a prospective clinical trial, like that of Jacob et al, 41 or be applied to multi-institutional studies, such as the study recently completed by Dess et al., 42 and could also help to determine if cardiac substructure dosimetric sparing has an effect on clinical outcomes.

| CONCLUSION
This work applied a multimodality imaging and contouring workflow to showcase the possibility of providing robust dose sparing of cardiac substructures with MRgRT. New treatment plans maintained PTV and OAR doses and did not substantially increase delivery time or required monitor units, suggesting stable plan quality and a negligible increase in plan complexity when cardiac substructure sparing was introduced. This study emphasized how high-quality cardiac substructure segmentations and sparing plans may be generated at lowfield MRI, which offers strong potential for lower substructure doses at initial planning and the ability to further maintain that condition via daily online MR-guided adaptive radiation therapy. Validation in a larger cohort with appropriate margins will offer the potential to reduce radiation-related cardiac toxicities and the dose assessment of currently overlooked radiosensitive substructures.