From probability to practice: demystifying when lung shunt fraction matters in 90Y selective internal radiation therapy

Summary Building on the commendable work by Thomas et al., this editorial proposes a simplified workflow for pretreatment identification of patients needing macroaggregated-albumin-based lung shunt fraction prior to the 90Y treatment. Key Results Thomas et al. introduced a boundary lung shunt fraction metric and provided a quantitative framework for determining whether patient-specific macroaggregated-albumin-based lung shunt fraction is needed prior to the 90Y treatment. Building on Thomas et al.’s foundational work, this editorial proposes a simplified, 2-variable calculation applicable in most clinical scenarios. The simplified workflow produces equivalent results in a small local streamlined 90Y dataset and lowers the barrier to clinical adoption. We read with great interest the article by Thomas et al., “When is patient-specific lung shunt fraction necessary in 90Y selective internal radiation therapy of liver cancer.”1 Conventional 90Y selective internal radiation therapy (SIRT) workflow involves an asynchronous mapping session with selective arteriography and 99mTc macroaggregated albumin (MAA) administration to calculate the lung shunt fraction (LSF) prior to the 90Y treatment session. Recent evidence has suggested that a streamlined 90Y SIRT workflow without the mapping session is safe with equivalent oncologic outcomes in selected patients. Thomas et al. provided a crucial quantitative framework for a decision typically made qualitatively—determining which patient requires a 99mTc-MAA mapping session to calculate LSF. The authors’ work is commendable. They introduced the LSFbound metric: the minimum LSF value at which lung dose limits achievable target volume dose. By combining this new metric with SPECT/CT-derived LSF distribution data from a sizable cohort of 297 patients with 354 90Y SIRT cases, the authors showed that MAA-based SPECT/CT-derived LSF rarely exceeds LSFbound and restricts the target volume dose (median ≤ 1%, maximum ≤ 4%) for most patients with liver malignancy (eg, hepatocellular carcinoma [HCC] ≤ 8 cm and non-HCC tumors without macrovascular invasion). Furthermore, using receiver operator characteristics analyses and clinically relevant probability thresholds, the authors demonstrated that prospective use of LSFbound can maintain 100% sensitivity and > 80% specificity across a wide range of whole liver doses. Their results provided additional quantitative evidence for streamlining 90Y SIRT by adopting the “no-MAA” workflow.2,3 The reduced time-to-treatment4 and cost along with improved patient access and satisfaction is making 90Y SIRT more attractive across different stages of HCC5 and non-HCC tumors amenable to locoregional options. Although the authors’ approach was ingenious, the proposed clinical workflow using LSFbound necessitates estimation of whole-liver dose limit (Livermax) for the calculation of toxicity threshold ratio (TTR) and measurement of the masses of the liver and lungs. Requiring these variables adds complexity and may prove cumbersome in clinical practice. Building on Thomas et al.’s foundational work, we propose a simplification to promote its clinical adoption. This simplified equation obviates the estimation of whole-liver dose limit and measurement of liver/lung masses. In cases where Lungsmax may be different because of prior radiation deposition (eg, occupational exposure, external radiation, SIRT), baseline pulmonary disease restricting lung radiation, or LSFplanar is used instead of LSFSPECT, a different Lungsmax should be used in the denominator. In patients where masslung may be significantly different from 1 kg (eg, chest wall deformities, prior lung resection, small stature), patient-specific masslung should be calculated from cross-sectional imaging.9 The final simplification of the workflow involves interpretation of the calculated LSFbound. The probabilistic density functions10 for each tumor category may be used to calculate the exact PLSF>LSFbound. Alternatively, a standard probability threshold could be adopted. For example, by adopting a Pthresh of 3%, direct comparison of the calculated LSFbound can be made to the 97th percentile LSF in each of the tumor categories (3.8% for HCC < 3 cm, 7.6% for HCC between 3 and 8 cm, 8.3% for HCC > 8 cm, 26.4% for macrovascular invasion [MVI], 7.3% for non-HCC).1,10 If LSFbound is greater than the 97th percentile LSF in the specific tumor category, the risk that LSF would limit the treatment plan is < 3% and MAA-based LSF determination is unnecessary. On the contrary, if LSFbound is less than the 97th percentile LSF in the specific tumor category, the risk of LSF limiting the treatment plan is > 3% and MAA-based patient-specific LSF should be obtained. In clinical scenarios involving greater risk of radiation pneumonitis, LSF99 may be used for stricter patient selection for the “no-MAA” workflow. This logic leads to a straightforward workflow for clinical adoption: (1) determine if MAA administration is required for multicompartment dosimetry; (2) choose a desired target dose and estimate target perfused volume for dose and LSFbound_simplified calculation; (3) calculate LSFbound_simplified; and (4) if LSFbound_simplified is greater than LSF97 of the appropriate tumor category, proceed with “no-MAA” workflow. All patients not amenable to the “no-MAA” workflow should follow the conventional “map-treat” workflow (Figure 1). Simplified 90Y SIRT treatment planning workflow based on the clinical workflow proposed by Thomas et al. published 5 February 2026 in Radiology Advances https://doi.org/10.1093/radadv/umag007. Applying this simplified approach to the example cases in the authors’ supplemental material yields equivalent results to their original calculation (Table 1).1 Case 1’s LSFbound_simplified is 29.4%, greater than the 97th percentile LSF at 3.8% in the specific category of HCC < 3 cm. As a result, the probability of the actual LSF and the Lungsmax restricting the perfused volume dose is less than 3%. Similarly, for cases 2 and 5, the LSFbound_simplified at 12.9% and 15.2%, respectively, and are both greater than the 97th percentile LSF at 7.6% and 7.3% in the respective categories of HCC between 3 and 8 cm and non-HCC tumors. For case 3 with an HCC > 8 cm, the LSFbound_simplified at 7.8% is less than the 97th percentile in that specific tumor category at 8.3%. As a result, the probability of the actual LSF and the Lungsmax restricting the perfused volume dose is greater than 3%. MAA-based patient-specific LSF should be obtained. For case 4 with MVI, the LSFbound_simplified at 19.8% is less than the 97th percentile LSF at 26.4%. Similar to case 3, MAA-based mapping session is recommended. Such results demonstrate that the simplified equation (4) retains the core predictive utility in determining whether MAA-based patient-specific LSF is needed. Comparison of LSFbound Calculations in Example Cases from Thomas et al.* Abbreviations: HCC = hepatocellular carcinoma, LSF = lung shunt fraction, LSFbound = lung shunt fraction boundary where the maximal permitted lung dose starts to restrict the maximal dose to the perfused volume, LSFbound_LiverMax calculated exactly as defined in Thomas et al. using estimated maximum mean dose to the liver, LSFbound_LiverRx = lung shunt fraction boundary calculated using retrospectively determined mean dose to the liver as reported in Thomas et al., LSFbound_simplified = lung shunt fraction boundary computed using the simplified equation (4), MAA = macroaggregated albumin, MVI = macrovascular invasion, PV = perfused volume. Thomas et al. published 5 February 2026 in Radiology Advances https://doi.org/10.1093/radadv/umag007. The proposed workflows, original and simplified, have also been retrospectively applied to a local single-centered experience of “no-MAA” 90Y SIRT. Institutional review board approval was obtained with written consent waived. From July 2024 to April 2026, 16 patients (3 female, 13 male; age 70.5 ± 2.8 [mean ± standard errors of the mean]) underwent 17 “no-MAA” radiation segmentectomy for 17 HCCs (15 < 3 cm, two 3–8 cm) using resin 90Y microspheres. Perfused volumes were 149.0 ± 19.3 mL, accounting for 8.7 ± 1.3% of the total liver volume, with target perfused volume dose at 450 ± 53 Gy. LSFbound_LiverRx, calculated using retrospectively determined mean dose to the liver, was 27.3 ± 2.5%; LSFbound_LiverMax, calculated as defined by Thomas et al. using estimated mean doses to the liver, was 15.5 ± 1.0%; LSFbound_simplified, calculated using (4), was 28.1 ± 2.4% (analysis of variance P < .0001; pairwise comparison demonstrated that LSFbound_LiverRx and LSFbound_simplified are equivalent, P = .5). None of the calculated LSFbound values was smaller than LSF97 for the respective tumor categories (3.8% for HCC < 3 cm, 7.6% for HCC 3–8 cm). Post 90Y treatment SPECT/CT revealed clinical LSF at 6.2 ± 0.7% (maximum: 10.5%) and LungRx 4.0 ± 0.7 Gy (maximum: 12.2 Gy). Pretreatment calculations of LSFbound, original or simplified, would have provided quantitative confidence that MAA-based LSF determination was not necessary in these cases, validating a “no-MAA” approach. It is important to note that this method, original or simplified, has limitations as noted in the authors’ article. The model’s probabilities rely on the authors’ single center dataset, which, although ample, had limited representation of subgroups like HCC with MVI and large HCC > 8 cm. Institutional and population specific probability density functions should be evaluated and used. Furthermore, the proposed workflow does not apply when MAA SPECT/CT is essential for multicompartment dosimetry. Finally, the conventional 90Y SIRT workflow with separate mapping and treatment sessions should remain as the back-up option for patients not amenable to the streamlined workflow. Ultimately, the study by Thomas et al. represents an important evolution in 90Y SIRT. It moves the field from qualitative, population-based heuristics to quantitative, patient-specific probability determination. Instead of asking, “What is the LSF?”, they reframed the question to “Does LSF matter for this patient’s treatment plan?” This editorial builds on their excellent framework by demonstrating that their core insight can be distilled into a simpler, 2-variable calculation in most clinical scenarios. We hope this simplification lowers the barrier to clinical adoption, thereby empowering interventional radiologists to streamline 90Y SIRT workflow for a broader patient population with quantitative confidence. No funding was received during the preparation of this manuscript. N.L. receives funding from the Society of Interventional Radiology and the Radiological Society of North America, serves on the editorial board for Radiology Advances and Journal of Vascular and Interventional Radiology. S.K. reports no conflict of interests. T.F. reports no conflict of interests. N.J.R. receives support from Terumo Medical Corporation, Boston Scientific Corporation, and Inari Medical Incorporated. Data available on request.