Factors modulating 99mTc‐MAA planar lung dosimetry for 90Y radioembolization

Abstract Purpose To investigate the accuracy and biases of predicted lung shunt fraction (LSF) and lung dose (LD) calculations via 99mTc‐macro‐aggregated albumin (99mTc‐MAA) planar imaging for treatment planning of 90Y‐microsphere radioembolization. Methods and materials LSFs in 52 planning and LDs in 44 treatment procedures were retrospectively calculated, in consecutive radioembolization patients over a 2 year interval, using 99mTc‐MAA planar and SPECT/CT imaging. For each procedure, multiple planar LSFs and LDs were calculated using different: (1) contours, (2) views, (3) liver 99mTc‐MAA shine‐through compensations, and (4) lung mass estimations. The accuracy of each planar‐based LSF and LD methodology was determined by calculating the median (range) absolute difference from SPECT/CT‐based LSF and LD values, which have been demonstrated in phantom and patient studies to more accurately and reliably quantify the true LSF and LD values. Results Standard‐of‐care LSF using geometric mean of lung and liver contours had median (range) absolute over‐estimation of 4.4 percentage points (pp) (0.9 to 11.9 pp) from SPECT/CT LSF. Using anterior views only decreased LSF errors (2.4 pp median, −1.1 to +5.7 pp range). Planar LD over‐estimations decreased when using single‐view versus geometric‐mean LSF (1.3 vs. 2.6 Gy median and 7.2 vs. 18.5 Gy maximum using 1000 g lung mass) but increased when using patient‐specific versus standard‐man lung mass (2.4 vs. 1.3 Gy median and 11.8 vs. 7.2 Gy maximum using single‐view LSF). Conclusions Calculating planar LSF from lung and liver contours of a single view and planar LD using that same LSF and 1000 g lung mass was found to improve accuracy and minimize bias in planar lung dosimetry.


INTRODUCTION
calculated from an estimated lung shunt fraction (LSF), the planned 90 Y-microsphere administered activity, and the estimated lung mass (M). The instructions for use (IFU) for the two commercially available 90 Y-microsphere products, SIR-Spheres (Sirtex Medical Inc., Woburn, USA) and TheraSphere (Boston Scientific, Marlborough, USA), each define procedures for estimating LSF and LD using 99m Tc-macroaggregated albumin ( 99m Tc-MAA) planar scintigraphy with varying details. 1,2 The SIR-Spheres IFU outlines administering ∼140 MBq (∼4 mCi) of 99m Tc-MAA, acquiring anterior and posterior views of chest and abdomen bed positions for 0.7-1.0 million counts per image, contouring the lungs and liver in both views, and finally estimating LSF from the geometric mean of the lung and liver contour counts. The TheraSphere IFU, on the other hand, only states administering a tracer dose of 99m Tc-MAA and calculating LSF using a ratio of the lung and total image counts without defining the views and bed positions imaged. Neither IFU provides detailed guidance on lung mass calculation beyond the TheraSphere IFU recommending a 1000 g M for all patients. Finally, both SIR-Spheres and TheraSphere IFUs define 30 and 50 Gy LD limits for single and cumulative radioembolization treatments, but only the SIR-Spheres IFU imposes an additional LSF maximum limit at 20%.
Although planning lung dosimetry with 99m Tc-MAA planar scintigraphy is standard-of -care (SOC) across most clinical practices, there are various limitations affecting both the accuracy and reproducibility of the estimated planar LSF and LD values (LSF planar , LD planar ). 3 First, the diagnostic 99m Tc-MAA biodistribution does not perfectly replicate the therapeutic microsphere biodistribution and will overestimate the microsphere LSF and LD primarily as a result of radiotracer degradation into free 99m Tc-pertechnetate in the blood pool. 4,5 Second, planar scintigraphy does not provide patient-specific lung mass estimates, so standard man/woman values must be applied. Finally, accurate and replicable quantification of activity distribution with planar imaging is difficult due to variable scatter and attenuation effects in the chest and abdomen, organ overlap in the projected 2D views, lack of anatomical landmarks for organ delineation, relatively low signal in the lungs, poor spatial resolution, and patient respiratory motion. 3 In a recent study with a digital XCAT phantom, Bastiaannet et al. estimated the SOC LSF planar geometric mean calculation overestimates LSFs in the typical clinical range of 0-20 percentage points (pp) by at least 25% and identified that the error is driven primarily because of the differences in lung and liver attenuation. 6 Fortunately, from a patient safety standpoint, the combination of all these factors result in the predicted MAA-based LSF planar and LD planar overestimating the true MAA-based LSF and LD which in turn over-estimates true 90 Y-microsphere LSF and LD values, as has been shown in both phantom and patient data. [6][7][8][9][10][11][12] In fact, the incidence of radiation pneumonitis following radioembolization in current practice is extremely low, even in patients with high lung shunting estimated with 99m Tc-MAA planar imaging receiving LDs above the IFU single-and multiple-treatment planar-based lung dosimetry limits. [13][14][15][16] Unfortunately, without established SPECT-based LD limits and without reliable techniques to translate SPECT-based dosimetry values to planarbased dosimetry values, the use of pre-therapy 99m Tc-MAA SPECT imaging for lung dosimetry, if any, will vary between practices. However, the overestimation of the true LSF using SOC planar imaging has two potentially major consequences in patient care. It can result in the patient becoming ineligible to receive radioembolization altogether or receiving a lower (potentially less beneficial) tumor dose to satisfy the planar-based LD limits.
In a 2018 survey of members of the Cardiovascular and Interventional Radiological Society of Europe (CIRSE), 82% of the 71 responding centers stated that lung shunting was the primary reason patients were excluded from radioembolization treatment. 17 Once eligible candidates proceeded to therapy, 61% of centers reported 2%-10% of all patients required a dose reduction due to the high estimated LD. The impact of lung shunting can also be quantified by reviewing patient exclusions in the recent SARAH and DOSISPHERE clinical trials. 18,19 In these trials, 15 of 52 (29%, SARAH) and 6 of 18 (33%, DOSISPHERE) of all patients excluded after 99m Tc-MAA imaging were because of high lung shunts. If lung shunts estimated with planar imaging were not used as an exclusion criterion, an additional 8% (15+174, SARAH) and 10% (6+60, DOSI-SPHERE) of patients could have participated in the study and, more importantly in the case of the SARAH trial, received radioembolization treatment with demonstrated improved tumor response and quality of life over the sorafenib treatment arm.
Recent studies have confirmed that the current treatment planning with 99m Tc-MAA planar scintigraphy overestimates the actual 90 Y-microsphere LSF and LD and have demonstrated the improvement of 99m Tc-MAA-based LSF and LD accuracy and reproducibility by using SPECT/CT imaging instead of planar scintigraphy in both phantoms and patients. 6-12 One publication, for example, found that SPECT-based LSF and LD (LSF spect , LD spect ) were on average 63% and 53% lower than the SOC LSF planar and LD planar values, respectively. 12 Unfortunately, the over-estimated SOC LSF planar and LD planar cannot be used to reliably predict the more accurate and lower LSF spect and LD spect values, as seen by the broad range of errors reported between the respective planar-and SPECT-based values. [10][11][12] While the accuracy improvement of SPECT/CT over planar imaging has been demonstrated in phantoms and patients, planar imaging remains prevalent as SOC for lung dosimetry and, in many clinical practices, the only possible imaging after 99m Tc-MAA administration. However, the two sets of instructions outlined with varying amount of detail in the device IFUs are only two of the many possible ways of calculating LSF planar and LD planar . The purpose of this work is to evaluate the accuracy and variability of various planar imaging algorithms for calculating LSF planar and LD planar at our institution.
This work does not seek to establish a de facto LSF planar and LD planar calculation to be implemented across all practices for all glass-and/or resin-microsphere radioembolization patients. Instead, this retrospective analysis focuses on how different approaches in the dosimetry calculations from the same set of planar imaging inputs can change the predicted LSF planar and LD planar . To this end, analysis of the patient cohort consisted of two components: first, determining which LSF planar and LD planar methodology resulted in values closest to LSF spect and LD spect , and second, quantifying the relative impact of planar view, region-of -interest (ROI) contour, and liver shine-through correction on LSF planar accuracy.

Patient cohort and imaging protocols
The cohort in this institutional review board-approved retrospective analysis consisted of all consecutive patients with hepatocellular carcinoma (HCC) who underwent treatment planning for 90 Y radioembolization with glass microspheres at our institution between 1 January 2015 and 31 December 2016.Forty-six consecutive patients (9 women, 37 men) underwent 52 planning procedures (7 women once, 2 women twice, 33 men once, 4 men twice) and 39 patients (85%) underwent 44 treatment procedures (7 women once, 1 woman twice, 27 men once, 4 men twice). There were no strict patient selection criteria applied for data used in the study because the comparisons between LSF and LD were based on matched inputs between various approaches and therefore independent of the specifics of the patient population or disease status.
All LSF and LD values were derived from three imaging series acquired as part of SOC treatment planning: diagnostic chest CT, 99m Tc-MAA planar scintigraphy, and 99m Tc-MAA SPECT/CT. Planar (140 keV center and 15% width photopeak, 40 × 52 cm FOV, LEHR collimator, 7 min/view) and SPECT (140 keV center and 15% width photopeak, 15% width lower-scatter window, 40 × 52 cm FOV, LEHR collimator, 16 seconds/view, 3D-OSEM iterative reconstruction with attenuation, scatter, resolution recovery corrections, 5 mm FWHM Gaussian filter) imaging protocols followed previously published methodologies. 12 Analysis of LSF planar and LD planar in this work was not differentiated based on patient sex, age, treatment volume (lobar or whole liver), or treatment number as each unique planning and treatment procedure served as its own control in all calculations.

99m Tc-MAA SPECT/CT dosimetry
Following the recommendations in previous publications, 10-12 LD spect was calculated using patientspecific lung mass (M pat ) from densitovolumetry of a diagnostic chest CT, LSF spect from liver and lung 99m Tc-MAA SPECT counts, and the administered 90 Y activity during the treatment procedure (A 90Y ).
Left lung (L), right lung (R), and liver (H) SPECT counts were measured in contours manually drawn on the rigidly registered CT imaging using MIM Software. However, because of differences in spatial resolution and in respiratory and patient motion between SPECT and CT acquisitions, SPECT signal originating in the liver is often found outside of the registered CT liver boundaries. To account for this liver 99m Tc-MAA signal recorded outside CT liver boundaries, the liver CT contour was morphologically expanded by 2 cm, and all SPECT counts within the expanded liver CT contour were attributed to the liver (C spect H ). LSF spect was calculated as the ratio of the estimated total lung counts (C spect LR ) to the estimated total liver counts (Equation 2).
However, to minimize the likelihood of liver 99m Tc-MAA signal shine-through effect inflating the SPECT counts within the CT lung boundaries, the total lung counts were estimated by first calculating the left lung SPECT counts (C spect L ) and the patient-specific left lung mass (M pat L ) only within the left lung CT contour superior to the 2 cm liver CT expansion to estimate the average lung count density. Based on our previous work, the left lung is generally farther away from the superior liver and thus will be less contaminated by the liver shine through. The total lung counts (C spect LR ) were then calculated as the product of the average lung count density and the total lung mass from the diagnostic chest CT (Equation 3).
By using the total lung mass from the diagnostic chest CT, this approach will also compensate for any signal from the apex of any lungs that may have been truncated in the SPECT/CT field of view. 12

99m Tc-MAA planar dosimetry
This work investigated four factors involved in planar dosimetry: (1) planar view(s) selected, (2) ROI combination selected, (3) lung ROI count liver shine-through correction applied, and (4) Left lung (L), right lung (R), and liver (H) ROIs were contoured free-hand on anterior (ant) and posterior (post) planar scintigraphy views by trained nuclear medicine technologists (each with 3+ years of experience) and verified by nuclear medicine physician (with 12+ years of experience) to define the lung (C LR = C L + C R ) and liver (C H ) counts in each view ( Figure 1). Additionally, the total measured counts (C T ) in each individual view were recorded for each patient.

Best planar dosimetry
The best planar dosimetry methodology in the study cohort was defined as the LSF planar and LD planar calculations (LSF planar best and LD planar best , respectively) that most closely estimated the corresponding LSF spect and LD spect values. SPECT/CT-based values were selected as the benchmark for comparison as they have been shown to most accurately quantify the true 99m Tc-MAA distribution and therefore most accurately represent the post-radioembolization 90 Y lung shunt and dose. More specifically, LSF planar best was defined as the LSF planar methodology with the lowest median absolute difference (in pp) from LSF spect across all patients undergoing treatment planning, while LD planar best was defined as the LD planar methodology with the lowest median absolute difference (in Gy) relative to LD spect across all patients undergoing treatment procedures.

2.4.1
Best planar LSF A total of 18 different LSF planar methodologies were evaluated per planning procedure. For each planning procedure, six separate LSF planar estimates (Equations 6-11) were initially calculated through the combination of three possible view choices (anterior only, posterior only, or geometric mean) and two possible contour choices [lungs and liver (LR, H); lungs and total frame (LR, T)] with no lung ROI count liver shine-through correction.
For each of the six view and contour combinations, two additional LSF planar values were calculated to test two simple liver shine-through corrections of the lung ROI counts, resulting in a total of 18 different LSF planar calculations. Based on previous work developing a SPECT/CT methodology noting that the right lung counts were primarily affected by liver shine-through effects, the measured right lung counts C R in each of F I G U R E 1 Example contours of the liver, left lung, right lung, and total frame in 99m Tc-macro-aggregated albumin ( 99m Tc-MAA) anterior (a) and posterior (b) views of patient undergoing standard-of -care LSF planar calculations (Equations 6-11) for radioembolization treatment planning, illustrating the variability and uncertainty in delineating the organs of interest with freehand contours.
the Equations (6)-(11) was replaced with "liver-shinethrough-free" right lung counts estimated in one of the two ways.In the first,the corrected right lung counts were calculated as C Rc = 1.15 × C L , where the constant 1.15 factor corresponds to the standard man assumption that the right lung is 15% larger than the left lung. 20 In the second, the corrected right lung counts were calculated as C Rc = (A R ∕A L ) × C L , where A R ∕A L corresponds to the ratio of the drawn right lung to left lung ROI areas. Both of these corrections assume the lungs have similar perfusion and the left lung ROI counts are minimally impacted by liver-originating 99m Tc-MAA signal. 12 The median and range of calculated LSF planar values (in pp) and of pairwise differences between LSF spect and each of the 18 LSF planar methodologies across all planning procedures is reported. LSF planar best was defined as the methodology with the lowest median absolute difference.

2.4.2
Best planar lung dose determined above, while the two lung mass values were either the standard 1,000 g lung mass IFU assumption (M std ) or the patient-specific M pat derived from diagnostic chest CTs used in the LD spect calculation.
In each treatment procedure, all four LD planar calculations (Equation 5) were made using the same actual administered 90 Y activity, where the net administered activity was calculated per the TheraSphere IFU. 2 The median and range of calculated doses (Gy) and of pair-wise differences between LD spect and each of the four LD planar methodologies across all treatment procedures are reported.

Best planar dosimetry
In the 52 planning procedures, the median (range) LSF spect was 2 pp (0-11 pp). Table 1 shows the absolute errors for the 18 possible LSF planar calculations. Figures 2 and 3 Patient-specific lung mass (M pat ) from densitovolumetry of diagnostic chest CT yielded median (range) values of 816 g (548-1178 g). Stated otherwise, the 1,000 g M std assumption over-estimated M pat lung mass by a median (range) 22% (−15% to 82%) in this cohort.
In the 52 treatment procedures, the median (range) LD spect was 2.1 Gy (0.3-25.5 Gy). Table 2 summarizes the distribution of calculated doses and calculated differences relative to LD spect for each of the LD planar calculations. Figures 4 and 5, respectively, show the Bland-Altman and boxplots of the absolute errors reported in Table 2. Based on these results, LD planar best was LD ant|LR,T std (i.e., using the best LSF ant LR,T and 1,000 g standard lung mass) with lowest median absolute errors from LD spect of 1.2 Gy (error range of −3.0 to 14.3 Gy). TA B L E 1 Distribution of median (minimum, maximum) absolute LSF planar errors = LSF planar − LSF spect (percentage points, pp) in 55 planning procedures for 18 different LSF planar methodologies.

Lung counts calculation View
Contours

DISCUSSION
The clinical practice of 90 Y-microsphere radioembolization has evolved in recent years to incorporate advanced 3D functional and anatomical imaging to improve treatment planning and dosimetry. Lung dosimetry with 3D functional/anatomical imaging has been demonstrated to provide the most accurate and reproducible values. [6][7][8][9][10][11][12] In practice, however, 2D imaging remains the SOC for lung dosimetry, so it is important to understand how the biases and errors in 2D lung dosimetry can negatively impact patient management. From first principles, the most accurate planar lung dosimetry was expected to be estimated using LSF planar from anterior view lung and liver contours, and patientspecific lung masses. However, in this work, the 99m Tc-MAA-based LSF planar methodology resulting in the closest values to 99m Tc-MAA-based LSF spect values were calculated as the ratio of lung counts to total image counts in the anterior view of the planar image (i.e., LSF ant LR,T ). Compared to the SOC LSF geo LR,H methodology using the ratio of the geometric mean of lung counts to the geometric mean of liver counts, the more accurate LSF ant LR,T methodology reduced the median (maximum) LSF over-estimation from 5 pp (12 pp) to 2 pp (6 pp). Although a 1,000 g lung mass generally over-estimated patient-specific lung masses, the closest LD planar values to LD spect were calculated using LSF ant LR,T and the 1,000 g lung mass assumption.

Planar LSF bias from view selection
The geometric mean LSF planar calculation, as outlined in the IFUs, does not account for the differences in scatter and attenuation coefficients between the patient's abdomen and chest and therefore does not correct for preferential attenuation between views. Namely, both the liver and lung ROI count estimations lack their respective e x attenuation correction, where is the effective linear attenuation coefficient and x is the body thickness. 21 Because the liver is more attenuating than the lungs, applying the e x factor would increase the "corrected" liver counts more than the "corrected"lung counts,resulting in a lower "corrected"geometric mean LSF planar than the IFU LSF planar (Equation 4). In general, using only the anterior view data resulted in the best LSF planar accuracy. Incidentally, the liver is located preferentially anterior within the torso. As a result, photons emitted from the liver will be attenuated to a greater extent on their longer path toward the posterior detector than toward the anterior detector. Therefore, from first principles, the anterior view will typically contain higher liver counts relative to the posterior view.
Allred et al. overcome the inaccuracies introduced by the preferential signal attenuation by the liver and heart by calculating lung shunt using only the liver ROI counts in the anterior view and the lung ROI counts in the posterior view. 8 In a torso phantom, they report that this approach reduced the LSF overestimation in the SOC geometric mean calculation from up to ∼6 pp down to ∼2 pp across a range of LSFs < 10 pp. Although this exact methodology was not evaluated in this study, our results corroborate their conclusion that using data from TA B L E 2 Distribution of median (minimum, maximum) calculated doses (Gy) and absolute errors (LD planar − LD spect ) in 44 treatment procedures for six different LD planar methodologies. a single view leads to more accurate lung dosimetry than applying geometric mean calculations on data from two views. LSF ant was more accurate than LSF post in 92% of planning procedures (48/52) regardless of contour choice. In the four (8%) planning procedures with more accurate LSF post , LSF ant methodologies without liver shine-through corrections differed anywhere between 4 and 15 pp from LSF spect whereas LSF post methodologies only differed by 1 pp to 5 pp from LSF spect ( Table 3). In the worst case scenario with Patient D, LSF ant LR,H of 24 pp was 170% higher than the assumed true LSF spect of 9 pp but only 40% higher than the more accurate LSF post LR,H of 13 pp ( Figure 6). Upon further inspection, all four cases had primarily posteriorly distributed 99m Tc-MAA on their fused SPECT/CT imaging, resulting in higher posterior than anterior planar liver counts and thus lower and more accurate posterior view-based LSFs. These results indicate that the lower value between LSF ant and LSF post will be the more accurate LSF planar value for treatment planning. Thus, we conclude that planar lung dosimetry accuracy was improved in our patient cohort by not using LSF geo approaches but instead by selecting the most appropriate single view for quantitation after reviewing both LSF ant and LSF post values alongside patient imaging.

Planar LSF bias from contour selection
Contouring is one of the largest sources of uncertainty in calculating LSF planar because the true lung and liver boundaries are hard to visualize and accurately delineate. Not only do the IFUs differ in whether or not to contour the lungs and liver separately, they do not provide a precise methodology to perform their respective contouring. For example, some practices acquire a separate scan with a 99m Tc flood source under the patient to guide with the lung contouring; some adjust the window width and level based on the maximum pixel value to guide liver contouring; some use the same contours for both anterior and posterior views; and some freehand everything. As a result, different institutions and even different individuals at the same institution may estimate different LSF planar values for the same patient.
Our results indicate that LSF LR,T values were always equally close, if not closer, to LSF spect than LSF LR,H values. However, this seeming improvement in accuracy is largely driven by the inclusion of excess extra-hepatic and extra-pulmonary 99m Tc-signal in the denominator, that originates, not from 99m Tc-MAA biodistribution, but rather from 99m Tc-pertechnetate following the dissociation of 99m Tc-MAA. 4,5 The additional 99m Tc signal can be observed in organs with typical intravenous 99m Tcpertechnetate uptake, such as the kidneys, thyroid, and stomach. In Patient D, for example, 99m Tc signal originating from the kidneys and stomach is especially evident in the posterior view ( Figure 6) which lead to lower LSF LR,T versus LSF LR,H values ( Table 3). As a consequence, the higher denominator in LSF LR,T versus LSF LR,H equations for the same numerator lowers the LSF planar estimate and minimizes the known LSF spect overestimation of all LSF planar calculations. However, we found minimal differences in LSF planar with contouring choice for the majority of cases (Table 1 and Figure 2), highlighting the low impact of extra-hepatic signal on planar dosimetry accuracy. Therefore, practices can select which contouring choice works best for their treatment planning workflow as long as they are consistent for all patients and that they continue to review both anterior and posterior planar images for F I G U R E 6 Example patient (Patient D in Table 3) with LSF spect of 0.09 with primarily posterior distribution of 99m Tc-macro-aggregated albumin ( 99m Tc-MAA) within the liver, as seen in the fused SPECT/CT axial slice. As a result, the posterior view exhibited higher liver intensity (both planar images displayed using same window width/level) and resulted in the more accurate 0. 13

Planar LD biases: LSF versus lung mass estimations
LD planar accuracy improved when the accuracy of LSF planar improved (by using the appropriate LSF ant or LSF post view instead of LSF geo ), but worsened when the accuracy of lung mass was improved (by using patientspecific lung masses M pat instead of 1,000 g M std ). Based on Equation (5), LD planar values will approach the lower, more accurate, LD spect values as the ratio of LSF planar to the estimated lung mass (M est ) approaches the ratio of LSF spect to M pat .So,while anatomically incorrect, the 1000 g M std assumption, which, on average, over-estimates the true lung mass M pat , results in more accurate LD planar estimates than using M pat .
To date, there is no established guidance on using patient-specific lung masses for treatment planning nor are there updated lung shunt and dose limits based on the more accurate and personalized measurements possible with 3D functional and anatomical imaging. Therefore, at this time, the value of calculating patientspecific lung mass in planar dosimetry is unknown, and at least within our patient cohort, the additional burden of this calculation is not even warranted. Nevertheless, practices should beware the possible discrepancies of calculated LDs with 1,000 g versus patientspecific lung masses, especially in patients with smaller lungs.

Limitations and future directions
Shortcomings of the work include the use of clinical data from a single institution and the limited range of lung shunts observed in the patient population. Of the 52 planning procedures, 28 (54%,) all had an LSF planar < 10 pp regardless of planar views, contours, or liver shine-through correction combination calculated. Finally, all calculations in this work are based on imaging 99m Tc-MAA biodistribution, which does not always reflect the eventual 90 Y-microsphere biodistribution, especially in the lung compartment. The impact of scanner type, 99m Tc-MAA activity, image acquisition time, and inter-reader contour variability and other factors on LSF planar and LD planar accuracy and precision were not investigated. The variability in contouring and the relatively low lung signal are likely the major reasons why our liver shine-through correction attempts did not improve LSF planar accuracy as they have LSF spect accuracy. Nevertheless, the concepts presented in this work could be used in clinical practice to improve our protocols for 90 Y-microsphere treatment planning with 99m Tc-MAA. The fact that low LSF geo values (<10%) are typically observed, along with the scarce evidence for radiation pneumonitis above the current SOC LD limits, have led to the proposition of an algorithm whereby high LSF planar or high LD planar alone are not used as a contraindication for radioembolization. 13 Rather, patients with LD planar expected to exceed 20 Gy would be assessed for risk of radiation pneumonitis by calculating LD spect and consideration of other comorbidities. 13 Future work is necessary to optimize the algorithm and establish new safety thresholds for the updated, more accurate, LD planar and LD spect calculations.

CONCLUSIONS
In this retrospective study, 99m Tc-MAA-based lung dosimetry accuracy with 2D planar imaging: (1) improved by only using a single view to calculate LSF instead of applying a geometric mean between anterior and posterior views,(2) improved slightly when using lung and total image contours to calculate LSF because of extra-hepatic 99m Tc-MAA signal, and (3) worsened when using patient-specific lung masses to calculate dose instead of assuming 1,000 g lungs for all patients.

AC K N OW L E D G M E N T S
This work was supported in part by UTMDACC Cancer Center Support Grant CA016672. James P. Long was partially supported by a grant from the National Institute of Health (UL1TR003167). We thank Editing Services, Research Medical Library at UT MD Anderson Cancer Center for assistance with manuscript preparation.

AU T H O R C O N T R I B U T I O N S
Benjamin P. Lopez and S. Cheenu Kappadath were responsible for study design, data collection and analysis, and manuscript preparation. James P. Long assisted in statistical analysis. Armeen Mahvash and Marnix G. E. H. Lam provided clinical feedback. All authors read and approved the final manuscript.