Introduction
Recent research showed radiation for breast cancer can increase heart risks (1, 2).
In Ref. (2), it has been noted that for every Gy of radiation a women’s heart risk
rises 7.4%. However, the correlation between radiation dose and heart tissue damage
is still an open problem. A more accurate model of heart damage will significantly
improve the heart safety for patients underwent radiotherapy.
Modern radiation treatment planning systems (TPS) use computed tomography (CT) images
for dose calculation and evaluation. For evaluation of heart toxicity from radiotherapy,
the dose-volume histogram (DVH), which is generated by overlying radiation dose distribution
on heart delineations in CT images, is widely used. However, there are three major
factors that deteriorate the accuracy of TPS-calculated heart dose distribution. First
conventional CT is a fundamentally static imaging modality without the capability
to capture and depict the cardiac motion. Instead, heart is usually blurred in CT
images due to the motion artifacts. Second, without special contrast dye, CT provides
limited contrast between blood in heart chambers and the surrounding myocardium. The
heart region in TPS is actually a mixture of myocardium and blood, although only the
radiation dose to the myocardium is accountable for heart risks. Finally, there is
significant intra- and inter-fractional heart motion. As heart beats involuntarily
during and between radiation treatments, myocardium deforms and moves non-rigidly
against the fixed radiation beam so that the static dose distribution calculated in
CT based TPS does not reflect the accurate radiation dose distribution in heart.
There is also concern on the choice of the heart function for the evaluation of radiation
damage. Based on radiation beam geometry, only part of the heart will receive clinically
significant level of radiation during breast cancer treatment. It is possible that
the global heart function remains stable temporarily while cells in the irradiated
part of the myocardium lose part or all of their functions. In this case, regional
heart function, which can be derived from regional heart wall motion and strain analysis,
is a better indication of heart damage corresponding to radiation dose.
Although cardiac MRI is widely used in radiology for the diagnosis of heart disease,
its application in radiation treatment planning is limited. For multiple reasons,
it is not practical to use MRI directly for radiation treatment planning of breast
cancer patients. However, via multimodality deformable image registration (DIR) between
MRI and CT, MRI images may play a more critical role in the evaluation of the heart
damage from whole breast radiation.
Tagged MRI (tMRI) (3) is a relatively new imaging protocol that has been implemented
in the detection and diagnosis of regional heart functional loss. tMRI methods record
regional heart wall motion information as they create identifiable landmark bands
(tags) in the myocardium to establish dense point to point correspondence between
images. ECG-gated tMRI image sets can be acquired at different phases of the cardiac
cycle using the corresponding pulse sequence. The 4D (3D plus time) cardiac motion
model can be retrieved by image registration between tMRIs at different phases.
In the following sessions, we use tMRI as an example to explain how additional heart
function information in MRI is retrieved. It is our objective to demonstrate the additional
information retrieved from MRI can help the evaluation and protection of heart risks
for breast cancer patients, and we want to discuss the possibility of using MRI to
establish a more accurate correlation between regional heart functional loss and radiation
dose.
Method
Heart motion uncertainty in CT
First, we analyze the uncertainty in the CT based TPS-calculated radiation dose distribution
of heart. The cardiac motion artifacts in CT acquisition has been previously studied
(4) so we focus on the uncertainty related to the intra- and inter-fractional cardiac
motion and location variation.
We used kV fluoroscopy imaging to monitor the intra-fractional cardiac motion during
breast cancer treatment (experiment A). For a group of 10 left breast cancer patients
without breath holding or external breath suppression, fluoroscopy was acquired weekly
at the gantry angle of the treating beam for 15 s at 8 fps. The fluoroscopy radiation
dose to the patient was clinical insignificant.
To estimate the inter-fractional heart location variation, we registered the weekly
CBCT of two t-spine patients (experiment B). CBCT images were registered to match
the left breast and the variation of the heart location was evaluated by measuring
the average distance of the heart surface in the registered image.
Cardiac motion retrieval from MRI
For preliminary research purpose, we retrospectively studied two sets of anonymous
tMRI data acquired using the Spatial Modulated Magnetization (SPAMM) pulse sequence.
Both tMRI sets were ECG-gated and acquired at 24 phases during the cardiac cycle.
Each tMRI sets included three long axis (LA) image sets (corresponding to the two
chamber, three chamber, and four chamber view), a short axis (SA) image set, and an
ECG-gated non-tMRI image set acquired at the end of diastole. The slice thickness
of SA tMRI was 5 mm. The spacing between tags was 8 mm. There were both horizontal
and vertical tags in the image.
Given tMRI and the corresponding CT images of the breast cancer patient, the work
flow to estimate the correlation between radiation dose and regional heart functional
loss is illustrated in Figure 1.
Figure 1
The work flow of the cardiac motion tracking and myocardium dose evaluation method.
The part encircled by the dashed line is the current CT based heart dose calculation
and evaluation in the treatment planning system.
The tMRI images went through preprocessing first to remove the intensity non-uniformity
introduced by the surface coils used in the MRI process, and to reduce the impact
of the decay of image intensity between different phases of the cardiac cycle.
The epi- and endo-myocardial contours were generated from tMRI using frequency domain
analysis. The modulated tags corresponded to high frequency components in the frequency
domain and can be effectively removed from the image using frequency filtering. We
segmented the myocardium automatically in the SA image using the method in Ref. (5).
The myocardium contours could be automatically or manually generated in the LA images.
There were multiple means to track the movement of the tags during the cardiac cycle,
such as active contours (6), B-Spline (7), physics deformable models (8), and meshless
deformable model (9).
The reconstructed cardiac motion model had two uses. First, the myocardial strain
distribution was derived from the myocardium motion, and the abnormality in the distribution
was used as the indication of local heart tissue damage. Second, the cardiac model
was integrated with the CT-MRI image registration to calculate the accumulative radiation
dose distribution in the myocardium during cardiac cycle. The radiation dose to the
blood was ignored, as it would not directly cause heart risks.
Given the motion distribution, the myocardial strain was computed as the derivative
of the displacement vectors at image pixels. Strain depicted the variation in motion
between different parts of the heart. Abnormalities (either high or low value) in
strain distribution reflected local myocardial motion abnormalities, which was a direct
indication of regional heart functional loss.
The CT-MRI image registration was conducted to build a connection between the TPS
and the tMRI image domain. First, we register CT to the end-of-diastole non-tMRI using
mutual information based multimodality image fusion. The 4D heart model (including
both the cardiac motion and the volumetric myocardium model) in the MRI with regard
to the beam geometry in the CT images was determined after the registration. In the
next step, we accumulated the myocardial dose distribution at different cardiac phases
to reconstruct the cardiac-motion-adjusted accumulative myocardium dose distribution.
The final step was to align the strain distribution to the motion adjusted dose distribution.
The correlation between the strain and radiation dose was calculated to enable us
to establish the radiation-dose-to-heart-risk model in future research.
Discussion
The uncertainty in the TPS dose distribution caused by CT imaging was not considered
explicitly in previous studies. The cardiac motion artifacts in CT imaging can be
reduced after using modern technology such as the multi-detector computed tomography
(MDCT), however currently the high cost limited wide use of such techniques at radiation
oncology clinics. The low blood-to-tissue contrast in CT can be increased by injecting
contrast dye during patient simulation, although this requires longer preparation
time and the improvement is limited if the imaging motion artifact was not well addressed.
Based on experiment A, the average infra-fractional motion of the heart wall, as projected
to the beam eye view in kV fluoroscopy, was 1.3 ± 0.3 cm. Average inter-fractional
heart location variation can be as much as 1.5 cm as measured in the CBCT images acquired
in experiment B.
To address the uncertainties in heart location and the corresponding dose distribution,
we proposed to use MRI in the evaluation of heart risks for breast cancer patients.
As demonstrated by the tMRI-based cardiac analysis framework, MRI had less motion
artifacts, higher blood-to-tissue contrast (by using appropriate pulse sequence),
and provided infra-fractional cardiac motion information.
The accuracy of the MRI-based cardiac analysis was determined by the accuracy of fundamental
image processing modules such as registration, segmentation, and motion tracking.
Multimodality image registration between CT and MRI was a well-studied problem and
commercial software is now available to generate satisfactory registration results.
However, it should be noted that the couch top used in radiation oncology CT simulator,
and diagnosis MRI, were different. A DIR should be conducted to correct for the variation
of anatomy caused by different couch tops. It was also critical that the registration
should align the surface markers in the CT and MRI images since they determined the
radiation beam geometry in breast cancer radiation treatment. Effective approaches
to automatically delineate the myocardium and to derive the strain from MRI images
have been proposed in previous studies. The image registration and motion tracking
uncertainties have been discussed in previous research efforts (9, 10). The overall
uncertainty in the proposed methodology also depended on the interpolation and extrapolation
error during the projection process to transfer the displacement and the radiation
dose distribution between different image domains. Interpolation and extrapolation
errors were hard to quantify or validate directly. We can use the inverse minimization
procedure to reduce the error, at the cost of extra processing time.
The major technical challenge in MRI-based heart risk analysis was the reconstruction
of the cardiac-motion-adjusted accumulative radiation dose distribution. To get the
accumulative dose, one needed to deform the original CT image to regenerate CT images
at different cardiac phases using the cardiac motion derived from the tMRI images.
The difficulty increased as the CT and the tMRI imaging planes were not the same and
intersected each other at oblique angles. The accuracy of the regenerated CT images
needed further validation before using for dose recalculation.
Given adequate information, a polynomial fit can be generated to describe the correlation
between regional heart function loss and the radiation dose. The fitted model can
be used to quantitatively estimate the heart risk based on accumulative radiation
dose.
Finally, it should be noted that although the motion adjusted radiation dose distribution
is more accurate and specific than the currently used heart DVH in treatment planning
CT, it was still an approximation to the actual dose distribution. The method we proposed
did not consider inter-fractional heart location variation and the impact of respiration
on heart location. Moreover, the patient heart beat pattern may change during the
course of radiation treatment. All these factors caused extra uncertainties in the
calculated accumulative myocardium dose.
Conclusion
We discussed the uncertainties of using CT calculated dose to evaluate the radiation
damage to the heart. To improve the quality of heart risk analysis for breast cancer
patients, we proposed a tMRI-based framework to derive the cardiac motion, the myocardium
strain, and eventually the regional heart function loss. The proposed framework demonstrated
the possibility and technical challenge of establishing a correlation between myocardium
damage and radiation dose for breast cancer patients using MRI. By using MRI, regional
heart function loss could be detected and the radiation dose can be adjusted by generating
the accumulative dose during cardiac cycle. We plan to collect tMRI data from more
patients to improve the accuracy, efficiency, and statistical robustness of the proposed
framework in future studies.
Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial
or financial relationships that could be construed as a potential conflict of interest.