Cardiac substructure dose distributions in node-positive and node-negative breast cancer patients undergoing 3D-CRT: comparing the predictive accuracy of mean heart dose and mean left ventricular dose

Abstract

Background and purpose

Cardiotoxicity is a concern, especially in left breast cancer (BC) radiotherapy (RT), and accurate dosimetry is essential for minimizing cardiac exposure. This study evaluated the radiation exposure of cardiac substructures in node-positive and node-negative BC patients who underwent three-dimensional conformal therapy (3D-CRT) and compared the predictive accuracy of mean heart dose (MHD) and mean left ventricular dose (MLVD) in estimating dose distribution to cardiac substructures.

Materials and methods

This study included 55 patients with left-sided breast cancer, comprising 39 with node-positive and 16 with node-negative disease. All underwent adjuvant whole-breast irradiation using 3D-CRT. The heart, ventricles, atria, right coronary (RC), left anterior descending coronary (LADCA), and left circumflex (LCx) arteries were contoured. Dosimetric distributions were evaluated, and Pearson’s correlation and linear regression analyses were used to assess the relationship between cardiac substructures.

Results

The distribution of doses to cardiac substructures was heterogeneous, with LADCA receiving the highest doses: 15.6 Gy in node-positive and 13.2 Gy in node-negative breast cancer patients. Linear regression analysis revealed a weak to moderate predictive ability of MHD/MLVD to predict doses received by the cardiac substructure in both groups, with MLVD demonstrating marginally better results. For node-positive patients, the analysis revealed an R² of 0.40 (p < 0.001) for the association between MHD and LADCA and an R² of 0.45 (p < 0.001) for MLVD and LADCA. In node-negative patients, the R² values were 0.27 (p < 0.001) for MHD versus LADCA and 0.30 (p < 0.03) for MLVD versus LADCA. Pearson’s correlation analysis for node-positive patients indicated r = 0.63 (p < 0.001) for MHD versus LADCA and r = 0.67 (p < 0.001) for MLVD versus LADCA. For node-negative patients, the correlation coefficients were r = 0.52 (p < 0.001) for MHD versus LADCA and r = 0.54 (p < 0.001) for MLVD versus LADCA.

Conclusion

Radiation exposure to cardiac substructures during 3D-CRT for left breast cancer was heterogeneous, with the LADCA receiving the highest mean dose, followed by the LV. MLVD demonstrated superior predictive accuracy over mean heart dose (MHD) for estimating doses to critical substructures, particularly in node-positive patients.

Introduction

Breast cancer (BC) is the most prevalent malignancy among women worldwide, with radiotherapy (RT) playing a pivotal role as an adjuvant therapy to improve local control and overall survival outcomes following breast-conserving surgery or mastectomy [1,2,3,4]. BC RT is among the most widespread therapies in radiation oncology [5]. It could be based on more sophisticated methods, such as intensity-modulated radiation therapy (IMRT) and volumetric-modulated arc therapy (VMAT), or on traditional three-dimensional conformal radiotherapy (3D-CRT) [6]. The more advanced and technology-driven RT techniques have reduced acute and late toxicities and improved cosmetic outcomes. However, they are only available in some centres and have inherent shortcomings [5].

Although various studies have shown that the use of modern RT techniques has reduces cardiac exposure during RT of the left breast, epidemiological data show an increased risk of cardiovascular complications even after exposure to lower doses [3, 7,8,9]. Exposure to low doses does not mean absolute safety because there is no threshold dose limit below which the best curative outcomes are accomplished with the least or zero tendency toward cardiotoxicity [8, 10, 11].

Node-positive BC patients undergoing RT present additional complexity because a larger RT field is needed to accommodate affected lymph nodes. This results in a larger heart volume being irradiated, which can increase the dose received by cardiac substructures, potentially leading to cardiotoxicity [12].

The heart is typically considered a singular organ-at-risk (OAR) during RT planning, with the mean heart dose (MHD) frequently utilized to quantify the exposures received by the heart and its substructures [10, 13, 14]. However, MHD has limitations in capturing the detailed radiation exposure of critical cardiac substructures, such as the left anterior descending coronary artery (LADCA), left ventricle (LV), and left atrium (LA), which are more proximally located to the treatment fields [2, 9, 14,15,16,17,18,19,20,21]. This limitation is particularly significant in left-sided BC patients, where the heart’s anterior and apical regions of the heart receive relatively higher doses. Recent studies have highlighted the shortcomings of MHD as a surrogate for substructure-specific doses, demonstrating that it is not a reliable indicator of the dose received by cardiac substructures [14, 19, 22].

While delineating these substructures as distinct organs-at-risk (OARs) is essential for precision heart-sparing therapy, practical challenges including time constraints, non-contrast CT visibility, and clinical workflow demands often limit routine implementation.

Therefore, there is a need for a detailed radiation dose distribution received by the cardiac substructures, and doing so requires contouring of each critical substructure. The heart is an intricate organ with many vital substructures, and the gold standard for cardiac and substructure delineation is manual segmentation following international contouring guidelines [13, 23]. Moreover, delineating the heart and its substructures is not a routine part of the RT planning procedure, considering the number and intricacy of the structures concerned. It is also tedious and time-intensive, especially when the patient throughput is high [13, 23, 24]. For example, the critical OARs during left breast 3D-CRT, the coronary arteries, may not be entirely visible because of their tortuous nature and the non-contrast, thick-slice CT simulation images used in delineation [7, 9, 10, 25,26,27].

The LV is the largest and most muscular cardiac chamber, making it the easiest cardiac substructure to delineate. Regarding proximity to the treatment field during 3D-CRT of the left breast, the LV is the closest cardiac substructure [28, 29]. The LADCA (a radiosensitive substructure and a typical site frequently implicated in ischemic heart disease) is located within the interventricular groove [13, 30]. According to one study, radiation exposure to a volume of the LV receiving 5 Gy predicted significant coronary events compared with MHD [31].

These findings regarding the LV and its ease of delineation highlight the necessity of using it as a predictor of doses received by cardiac substructures, possibly even surpassing the MHD.

This study aimed to evaluate the radiation exposure of cardiac substructures in node-positive and node-negative BC patients who underwent 3D-CRT and compare the predictive accuracy of MHD and MLVD in estimating the dose distribution to cardiac substructures.

Materials and methods

Study population

This retrospective study included 55 adult female patients diagnosed with left BC, comprising thirty-nine node-positive and sixteen node-negative patients. All patients received adjuvant radiotherapy using 3D-CRT, with whole-breast irradiation for those who underwent breast-conserving surgery and chest wall (CW) irradiation for post-mastectomy patients. This was performed using an Elekta TM 160 Agility Leaf Linear Accelerator at AMDI, Universiti Sains Malaysia, between January and December 2017. Patients treated for other malignancies, who received a different RT technique, or who had unavailable computed tomography (CT)-based RT planning data in the archive during the study period were excluded. The university’s ethics committee approved the study (USM/JEPeM/16110523).

CT simulation

Each patient underwent a 3 mm slice computerized axial tomography simulation scan while the Deep breath-holding technique was employed. The patients were positioned supine on a tilted simulation table (with a breast board) with their arms raised above the head. The inferior border of the scan field was centered 1.5 cm inferior to the right sub-mammary crease, whereas the superior border was centered at the sternum and 2nd rib joint. No contrast agents were administered. The acquired CT data were imported in DICOM format into the Monaco 5.1 treatment planning system (TPS) (Elekta Medical System, Crawley, UK).

Target volume and organs at risk (OAR) delineation

At Monaco 5.1 TPS (Elekta Medical System, Crawley, UK), the skin is semi-automatically delineated the skin by the software. The target volume (TV) was defined using the Radiation Therapy Oncology Group Breast Contouring Atlas (RTOG), which adheres to consensus standards for defining anatomical borders and clinical CWs. For OAR, the contours of the complete heart include the ventricle’s infundibulum, right auricle, apex, and all visible myocardia. The spinal cord and left lung were among the contoured OARs. To achieve the objective of this study, each patient’s cardiac chambers (LA, LV, RA, and RV), LADCA, left circumflex artery (LCx), and right coronary artery (RCA) were contoured on the Monaco 5.1 TPS workstation (Fig. 1). Contouring was achieved after dedicated cardiac substructure segment identification and contouring training by an experienced cardiac radiologist and oncologist. The atlas for cardiac delineation published by Duane et al., [7], served as a valuable reference. A cardiac radiologist reviewed all contours and made any necessary amendments before approving the generation of dose-volume histograms (DVHs).

Fig. 1

Axial CT simulation image of the manual cardiac substructures and other OAR delineation (coloured lines are drawn around the whole heart, substructures, and lungs). PTV = planning target volume, LADCA = left anterior descending coronary artery, LCX = left circumflex artery, and RCA = right coronary artery)

Treatment planning

Conventional 3D-CRT tangential photon fields conforming to the delineated treatment volume per the RTOG atlas were employed. The tangential field borders extend caudally (1.5 cm), less than 2 cm cranially from the humeral head, medially at the Centre, and laterally at the anterior edge of the serratus anterior. Most patients were treated with 10 or 6 megavolt photons at a dose of 42.56 Gy (range: 41.8–43.2 Gy). The dose was prescribed to the mid-plane at two-thirds the distance along a tangential line extending from the midpoint of the half-beam blocked tangential fields to the skin surface. Treatment comprised 16 daily fractions delivered over 3.5 weeks, with five fractions administered weekly. The dose was initially prescribed to the mid-plane at a two-thirds distance from a tangential line that links the midpoint of ¹/₂ -beam-blocked tangents to the skin. The plan was subsequently optimized using beam angles, collimator angles, and wedges. The dose limits to the heart were, per RTOG references, as follows: <5% of the cardiac volume should receive 40 Gy and less than 10% of the cardiac volume should receive 25 Gy.

Patient characteristics, clinical information, CT-based radiation therapy planning data, and follow-up information were retrospectively collected from patient histories. Dose distributions were calculated, and DVHs were generated for all the contoured structures. The equivalent dose in fractions of 2 Gy (EQD2) was calculated using the linear quadratic model with a cardiac alpha/beta ratio 2.5 to evaluate the cardiac dosimetric parameters [19, 32].

Statistical analysis

A descriptive statistical analysis of the radiation doses was conducted, presenting the values of the mean, standard deviation, and range. Mean dose differences between node-positive and node-negative groups for the heart and cardiac substructures were analysed using independent (unpaired) t-tests. Pearson’s correlation analysis was performed to determine the associations between MHD, and the doses received by the contoured cardiac substructures. Similarly, the research investigated the relationship between the MLVD, and the doses received by contoured heart substructures. Linear regression analysis was employed to evaluate the predictive accuracy of MHD and MLVD in estimating the dose to each substructure. One-sample t-test was used to compare MHD and MLVD against the mean doses received by individual cardiac substructures. Statistical significance was set at p < 0.05. All statistical analyses were conducted using JASP software (Version 0.18.1) [33].

Results

Retrospective dosimetric analysis was available for 55 female left-BC patients who underwent 3D-conformal hypo-fractionated RT using 6 or 10 MV photons. The mean age of the patient cohort was 50 years (range: 24–80 years). Most of the BC patients were diagnosed with invasive ductal carcinoma (IVC) and underwent mastectomy. The radiation dose prescribed for the patients was 42.56 Gy in 16 fractions. Table 1 summarizes the patients’ baseline demographics and their tumour and treatment characteristics.

Table 1 Patients’ baseline demographics and tumour characteristics

Fig. 2

Mean values of the dose received by the heart and contoured substructures in node-positive and node-negative BC patients. (BC = breast cancer, MHD = mean heart dose, LV = left ventricle, RV = right ventricle, LA = left atrium, RA = right atrium, RC = right coronary artery, LADCA = left anterior coronary descending artery, LCx = left circumflex artery)

The dose constraints to the heart were met for all patients (V25Gy < 10%, V40Gy < 5%). Significant interpatient dose variability was observed across the contoured cardiac substructures. The MHD for the entire cohort was 3.4 ± 0.9 Gy, ranging from 2.0 Gy to 6.3 Gy. The mean MHD in node-positive patients was marginally higher than that in node-negative patients (3.5 ± 0.9 Gy vs. 3.2 ± 0.9 Gy, p = 0.12).

The LADCA received the highest dose among the cardiac substructures, with a mean value of 15 ± 6.6 Gy (range: 10.6–21.8 Gy) for the entire cohort. Patients with node-positive BC received a mean LADCA dose of 16.7 ± 6.4 Gy (range: 10.6–21.8 Gy)., whereas node-negative patients received a mean LADCA dose of 14.2 ± 5.9 Gy (range: 12.6–21.2 Gy)., all of which significantly exceeded the MHD.

It is followed by LV with a mean dose of dose of 4.5 ± 2.3 Gy (range: 3.3–9.8 Gy) across all participants. Node-positive patients had an MLVD of 5.8 ± 3.1 Gy (range: 3.3–9.8 Gy), whereas node-negative patients had an MLVD of 3.2 ± 1.2 Gy (range: 3.4–6.3 Gy).

In contrast, the right atrium (RA) received the lowest radiation exposure, with a mean dose of 0.7 ± 0.2 Gy (range: 0.3–1.5 Gy) for node-positive patients and 0.6 ± 0.1 Gy (range: 0.4–0.9 Gy) for node-negative patients.

LCx received the lowest dose among the contoured coronary arteries, with a mean value of 0.9 ± 1.1 Gy (range: 2.0–5.3 Gy) for the entire cohort. Node-positive patients received a mean LCx dose of 1.1 ± 1.1 Gy (range: 2.0–5.3 Gy), whereas node-negative patients received a mean LCx dose of 0.8 ± 0.2 Gy) (range: 2.0–5.3 Gy). (Figures 2 and 3).

Fig. 3

DVH of the contoured cardiac substructures for a single patient. The arrows point to the dose volume curves of the LADCA and LV

One sample t-test revealed significant differences between MHD and the mean doses administered to the delineated cardiac substructures, as well as between MLVD and the mean doses received by the contoured cardiac substructures in both node-positive and node-negative breast cancer patients (p < 0.05).

Pearson’s correlation analysis revealed a range of correlations involving MHD and cardiac substructure doses, as well as MLVD and cardiac substructure doses. Notably, a moderate positive correlation was identified between MHD and RV in node-positive BC patients (r = 0.69, p < 0.001), while a strong correlation was found in node-negative patients (r = 0.73, p < 0.001). Strong positive correlations were found between MLVD and RV in node-positive patients (r = 0.90, p < 0.001) and node-negative patients (r = 0.76, p < 0.001). Moderate positive correlations were noted between MHD and mean LADCA doses in these groups (node-positive: r = 0.63, p < 0.001; node-negative: r = 0.52, p = 0.039). There were also moderate correlations between MLVD and LADCA doses, with r = 0.67 (p < 0.01) in node-positive patients and r = 0.53 (p < 0.031) in node-negative patients. In contrast, weak positive correlations were observed between MHD and mean RA doses, with r = 0.17 (p = 0.51) in node-positive patients and r = 0.23 (p = 0.36) in node-negative BC patients. A negative correlation was noted between MLVD and LA in node-negative patients with r = − 0.03 (p = 0.9) (Figs. 4 and 5).

The linear regression analysis of MHD and the mean LADCA dose revealed moderate correlations for node-positive BC patients (R² = 0.40, p < 0.001) and weak correlations for node-negative patients (R² = 0.27, p < 0.01). Similarly, the analysis between MLVD and the mean LADCA dose indicated moderate correlations in both node-positive BC patients (R² = 0.45, p < 0.001) and node-negative BC patients (R² = 0.30, p < 0.03) BC patients.

For various cardiac substructures, the regression analysis examining the relationship between MHD and the mean doses received by the RV, LA, and LCx arteries showed R² values of (R² = 0.48, p < 0.001), (R² = 0.15, p < 0.005), and (R² = 0.10, p = 0.06) for node-positive patients, and (R² = 0.54, p < 0.001), (R² = 0.04, p < 0.001), and (R² = 0.12, p = 0.037) for node-negative patients. In the same vein, the regression analysis between MLVD and the mean doses received by the RV, LA, and LCx arteries revealed R² values of (R² = 0.60, p < 0.001), (R² = 0.16, p = 0.012), and (R² = 0.10, p = 0.131) for node-positive patients, whereas for node-negative patients, the R² were (R² = 0.58, p < 0.001), (R² = 0.01, p = 0.013), and (R² = 0.04, p < 0.001) respectively (Table 2).

Table 2 Linear regression coefficient values between MHD and MLVD versus the cardiac substructures for node-positive and node negative BC patients

These results indicate that the doses received by the LADCA and various important cardiac structures can be inferred from MLVD, with MLVD showing greater accuracy in predicting the doses received by LADCA and these critical cardiac substructures than MHD.

Fig. 4

Scatter plots (a-h) showing the relationship between MHD and MLVD vs. dose to cardiac substructures in Gy for node-positive BC patients. (MHD = mean heart dose, MLVD = mean left ventricular dose, RV = right ventricle, LV = left ventricle, RA = right atrium, LA = left atrium, LADCA = left anterior descending coronary artery)

Fig. 5

Scatter plots (a-h) showing the relationship between MHD and MLVD vs. dose to cardiac substructures in Gy for node-negative BC patients. (MHD = mean heart dose, MLVD = mean left ventricular dose, RV = right ventricle, LV = left ventricle, RA = right atrium, LA = left atrium, LADCA = left anterior descending coronary artery)

Discussion

Adjuvant RT, along with surgery, chemotherapy, and hormonal therapy, is crucial for BC therapy; however, it increases the risk of radiation-induced cardiotoxicity [2, 3, 25]. Modern RT techniques have decreased cardiac radiation exposure; however, the risk of RT-induced cardiotoxicity remains [3, 10]. The risk increases with increasing cardiac radiation exposure, with intra-individual heterogeneity of cardiac dose-volume parameters being a concern [10, 14, 34, 35]. MHD is a commonly used surrogate for assessing potential cardiotoxic effects after BC RT. It is influenced by factors affecting its predictive efficacy in cardiac substructure doses [3, 10, 35]. Studies suggest that doses to individual cardiac substructures should be considered separately during RT rather than using MHD as a proxy for all cardiac substructures [2, 22, 35]. This study assessed the accuracy of the MLVD in predicting the doses received by cardiac substructures. Our results show that the predictive accuracy of the MLVD is greater than that of the commonly used MHD in predicting the doses received by the cardiac substructures, as evidenced in Table 2.

Radiation dose distribution to the cardiac substructures was heterogeneous with the LADCA receiving the highest amount of radiation followed by the LV, which received a much higher dose than some of the cardiac substructures. These research outcomes align with those reported by Van Den Bogaard et al. [31] and Costin et al. [36]. Similarly, Prunaretty et al. [37] demonstrated in their study that the dose distribution to the heart and its substructures varied significantly. Additionally, a study by Nilsson et al. [38], assessed the distribution of coronary artery stenosis among BC patients to determine the correlation between the doses received during RT and the location of stenosis. They discovered that LV and LADCA received the highest doses, with the middle and lower segments of LADCA receiving even higher doses than the upper segment.

The high doses received by LADCA and LV correspond to their anatomical closeness to the left breast/CW target volume. LADCA situated anteriorly in the interventricular groove, received the highest mean dose in both node-positive and node-negative BC patients, followed by the LV (Fig. 2). This shows the direct exposure of these substructures to tangential photon fields. In contrast, RA and LCx, which are located posteriorly, received lower doses.

While proximity generally correlates with higher doses, interpatient variability (as seen in LADCA ± 6.4 Gy) shows that individual anatomical factors such as heart position, breast size, and beam arrangement influence the dose distribution.

Node-positive BC patients received higher mean doses across all substructures compared to node-negative BC patients; however, this difference is not statistically significant (p = 0.12), possibly due to limitations in sample size and uneven distribution in the sample sizes between the groups (n = 39 vs. n = 16). Furthermore, the difference may be attributed to larger treatment volumes encompassing regional nodes, which have led to increased cardiac exposure. Nonetheless, the ranking of substructure doses remained consistent across both node-positive and node-negative subgroups, underscoring proximity as a primary driver.

The correlation between MHD/MLVD and cardiac substructures varied significantly between node-positive and node-negative subgroups. In node-positive BC patients, MLVD exhibited stronger correlations with anterior cardiac substructures over MHD (e.g., LADCA: R² = 0.67 vs. MHD R² = 0.63; RV: R² = 0.90 vs. MHD R² = 0.69). Larger treatment volumes (including regional nodes) likely increased cardiac exposure consistency, enhancing MLVD’s correlation with the anterior substructures. Similarly, in node-negative BC patients, MLVD exhibited stronger correlations with anterior cardiac substructures over MHD (e.g., LADCA: MLVD R² = 0.53 vs. MHD R² = 0.52, RV: MLVD R² = 0.76 vs. MHD R² = 0.73) maintaining the same trend as in node-positive BC patients. MLVD outperformed MHD for critical substructures (e.g., LADCA), but it showed diminished correlation for posterior regions (e.g., RA).

The superiority of MLVD over MHD comes from its anatomical proximity to high-dose regions near the treatment field. Unlike MHD, which averages doses across the entire heart, MLVD better captures dose gradients to anterior substructures, aligning with studies that show LV-specific metrics like LV-V5, predict cardiac toxicity more reliably than MHD [31].

Previous studies support the relevance of mean dose in normal tissue complication probability (NTCP) models for specific substructures. Marks et al., [39] demonstrated in their study that the incidence of new perfusion defects increased with larger irradiated LV volumes and higher mean LV doses, showing a significant trend over time. Nilsson et al., [38] discovered that left BC RT with regional nodal irradiation, particularly to the internal mammary chain (IMC), is high-risk and associated with an increased likelihood of coronary artery stenosis in radiation hotspot areas.

However, our findings are contrary to Atkins et al., [26] who showed that mean based metric is insufficient to predict LAD V15 Gy with confidence suggesting that its validity for optimally predicting cardiac events should be reassessed. But, in their study, the mean-based metric assessed was MHD, which MLVD outperformed in the current study. Although no single metric suffices for all substructures, our study shows incremental improvement of MLVD over MHD in estimating doses to high-risk substructures like LADCA.

The linear regression analyses revealed distinct patterns in how MHD and MLVD predict radiation exposure to cardiac substructures, with notable differences between node-positive and node-negative subgroups. In node-positive patients, MLVD showed greater predictive accuracies with anterior and critical substructures. This likely reflects the larger treatment volumes that include regional nodes, which enhance the consistency of cardiac exposure. Meanwhile, in node-negative patients, the predictive accuracies were weaker than in node-positive patients, likely due to smaller fields and greater anatomical variability. However, MLVD still outperformed MHD for critical and anterior substructures. MLVD’s superiority over MHD might be due to its proximity to high-dose regions near the treatment field. This finding aligns with the study Van den Bogaard et al. [31], which shows that LV-specific metrics predict cardiac toxicity more reliably than MHD. While MLVD does not replace substructure-specific delineation, it offers a realistic surrogate for clinics lacking resources for advanced planning.

The study’s limitation lies in its retrospective and purely dosimetric focus, lacking assessments of acute and late cardiotoxicity. Future research with a larger sample size evaluating the effects of both acute and late cardiotoxicity should be explored.

Conclusion

Radiation exposure to cardiac substructures during 3D-CRT for left BC was heterogeneous with LADCA receiving the highest mean dose, followed by the LV. MLVD demonstrated superior predictive accuracy over MHD for estimating doses to the LADCA and other substructures, particularly in node-positive patients. However, neither MHD nor MLVD reliably predicted doses to posterior structures underscoring the need for substructure-specific delineation where feasible. While MLVD offers a realistic improvement over MHD in resource-constrained settings, it does not replace the necessity of individualized substructure optimization for precise cardiac-sparing radiotherapy. Future work should incorporate clinical outcomes and advanced imaging techniques, such as cardiac MRI, to enhance NTCP models by utilizing both mean and volume-based parameters.