Ultra-selective radiation segmentectomy for early-stage hepatocellular carcinoma

Abstract

Background & Aims

Radiation segmentectomy (RS) is an emerging curative-intent therapy for early-stage hepatocellular carcinoma (HCC) when resection or ablation is not feasible. In this study, we evaluated the safety, efficacy, and dosimetric correlates of ultra-selective RS, defined as glass ⁹⁰Y-radioembolization delivered to vessels at least one order beyond parent segmental arteries, targeting <1 Couinaud segment.

Methods

This retrospective study included 38 patients with 42 early-stage (BCLC 0–A) HCCs treated with ultra-selective RS from December 2022 to July 2024. All treatments used glass ⁹⁰Y-microspheres with perfused treatment volumes assessed by cone-beam CT. Post-treatment voxel-based dosimetry was conducted using ⁹⁰Y single-photon emission CT (SPECT)/CT. Tumor response and progression-free survival were assessed by modified RECIST. Explant pathology was used to evaluate treatment effect in transplant recipients, and albumin-bilirubin (ALBI) scores were tracked longitudinally.

Results

The median tumor size was 2.4 cm with a median perfused treatment volume of 66.3 cc (4.5% of total liver volume). The median administered activity was 1.36 GBq (median absorbed dose 837 Gy). Complete response (CR) was achieved in 87% (n = 33), with only one local progression. Median local progression-free survival was not reached. Among 16 tumors with explant data, 69% showed complete necrosis and 25% extensive necrosis (median 88%). Tumor D₉₅ >300 Gy predicted CR (97% CR vs. 0% CR with D₉₅ <300 Gy; p <0.001), with logistic regression yielding an AUC of 0.98. Models incorporating tumor alignment within high-activity ⁹⁰Y-SPECT regions improved predictive accuracy. Over 82% of patients retained or improved ALBI grade during follow-up, with only 2 of 15 patients with baseline ALBI 2b declining to grade 3.

Conclusions

Ultra-selective RS is a feasible and liver-sparing therapy for early-stage HCC. Voxel-based dosimetry confirms dose-response relationships and underscores the importance of tumor coverage.

Impact and implications

This study demonstrates that ultra-selective radiation segmentectomy (uRS) with glass ⁹⁰Y microspheres can achieve high rates of complete imaging and pathologic response in early-stage HCC while treating very small liver volumes. These results replicate and extend prior radiation segmentectomy studies, underscoring that the ablative potential of ⁹⁰Y can be maintained with even greater liver parenchymal preservation. These findings are important for patients who are not candidates for surgery or ablation, especially given low rates of liver function decline after uRS. In practice, uRS may serve as a definitive therapy or as a bridge to liver transplantation, while voxel-based dosimetry provides a framework for ensuring adequate tumor coverage and identifying incomplete responders.

Graphical abstract

Highlights

• •Ultra-selective radiation segmentectomy treats <1 liver segment in HCC.
• •Ultra-selective radiation segmentectomy achieved 87% complete response and 69% complete necrosis on explant.
• •Voxel-based dosimetry confirmed tumor D₉₅ >300 Gy predicts complete response.
• •Liver function was preserved in >80% of patients following ultra-selective radiation segmentectomy.
• •Results replicate radiation segmentectomy outcomes while treating smaller volumes.

Introduction

Hepatocellular carcinoma (HCC) is a rapidly increasing cause of cancer-related death worldwide.¹ For patients with early-stage disease (Barcelona Clinic Liver Cancer [BCLC] stage 0–A) not eligible for liver transplantation, curative options include surgical resection and thermal ablation.² However, these therapies are typically limited to patients with preserved liver function without portal hypertension and favorable tumor anatomy. More recently, radiation segmentectomy (RS) involving the selective transarterial delivery of Yttrium-90 (⁹⁰Y)-microspheres to two Couinaud hepatic segments or less has emerged as an efficacious ablative treatment option, particularly for tumors unsuitable for thermal ablation.^3,4 Contemporary RS studies have demonstrated outcomes comparable to surgical resection or ablation, expanding curative-intent options for early-stage HCC.[3], [4], [5], [6], [7], [8], [9]

With growing technical expertise and advances in intraprocedural imaging, it is now feasible to routinely perform ultra-selective catheterization of sub-segmental arteries, i.e. at least one order beyond the parent segmental artery, perfusing very small liver volumes less than one Couinaud segment. Such an approach to deliver ⁹⁰Y-microspheres, i.e. ultra-selective RS (uRS), has the potential to maintain the ablative potency of RS while further sparing uninvolved normal liver parenchyma. While contemporary RS has demonstrated long-term safety, there is a persistent hazard of liver decompensation when treating greater than 15% total liver volume in those with baseline liver dysfunction.¹⁰ Given that traditional RS often exceeds these volumes, there is a clear need for more conformal therapies that minimize collateral liver damage, especially for patients with limited transplant access or eligibility.^4,11,12 Achieving greater conformality may also extend curative-intent therapies to patients with narrower therapeutic indices and limited hepatic reserve, such as those with portal hypertension or Child-Pugh B status.

vWhile uRS is conceptually attractive, its technical feasibility, safety, and clinical efficacy have not been formally evaluated. In this investigation, outcomes of uRS using glass ⁹⁰Y microspheres in patients with early-stage HCC were determined through longitudinal evaluation of tumor response, local progression-free survival (PFS), and liver function. In addition, voxel-based, post-therapy dosimetry was performed to explore dose-response relationships.

Patients and methods

Patient population

This study was approved by the Institutional Review Board at the authors’ institution, with the requirement for written informed consent waived. Treatment decisions were made through the multidisciplinary tumor board. In general, ⁹⁰Y radioembolization was selected as a bridge to transplant in patients with unresectable HCC who presented to the authors’ multidisciplinary clinic. In patients who were not eligible for transplantation, ⁹⁰Y radioembolization was chosen as a curative therapy if resection was unsuitable, due to comorbidities or portal hypertension, or ablation was unsuitable, due to unfavorable anatomy.⁴ Consecutive patients with HCC diagnosed by biopsy or multiphase CT or MRI (LI-RADS) who were treated with uRS between December 2022 through July 2024 were included in this study. Baseline characteristics including liver disease etiology, BCLC stage, Eastern Cooperative Oncology Group (ECOG) performance status, Child-Pugh liver function, albumin-bilirubin (ALBI) score, and baseline alpha-fetoprotein were obtained from initial clinical consultation in the electronic medical record. Inclusion criteria were very early or early-stage HCC without evidence of vascular invasion or extrahepatic spread (BCLC 0-A or C only if ECOG >0). Multifocal disease was allowed as long as patients were within Milan criteria (i.e. up to 3 tumors, <3 cm each). Patients who had received prior liver-directed therapy for the target tumor or systemic therapy (i.e. immune checkpoint inhibitors or tyrosine kinase inhibitors) were excluded. Patients were not excluded based on baseline liver function. uRS was defined as the intentional delivery of glass ⁹⁰Y microspheres (Therasphere, Boston Scientific, Marlborough, MA) into arteries at least one order beyond a parent segmental artery to achieve perfused treatment volumes less than one Couinaud segment (Fig. S1). Patients who received concurrent treatment with other liver-directed therapies including ⁹⁰Y to larger treatment volumes not fulfilling uRS criteria were excluded.

Treatment protocol

All treatments were performed according to well described and established protocols by four board-certified interventional radiologists with at least 5 years’ experience.¹³ All patients underwent mapping hepatic angiography with determination of lung shunt fraction (LSF) by lobar (per consensus recommendations) Tc-99m-labelled macroaggregated albumin injection followed by planar imaging.¹⁴ For each tumor, the intended perfused treatment volume(s) was obtained using intraprocedural, contrast-enhanced cone-beam CT (CBCT). Simplict90Y software (Mirada Medical, Oxford) was utilized to obtain whole liver volumes from the most recent pretreatment multiphase CT or MRI for dosimetry planning. Perfused treatment volumes (PV) from mapping CBCT were segmented using guidance from Hounsfield unit thresholds. Intended PV doses (MIRD_rx) were determined using the Committee on Medical Internal Radiation Dose (MIRD) single compartment method, and adhered to contemporary guidelines for ablative radioembolization of >400 Gy to the PV, with required adjustments for planar-based LSF included.³ Particle specific activity was defined as activity per microsphere (Bq/sphere) based on an updated estimation of 4,354 Bq/sphere at the time of calibration.¹⁵

Post-therapy voxel dosimetry

All post-therapy segmentation, registration, and voxelized dosimetry analysis was performed in MIM (Beachwood, OH; v7.3.7). Tumors were segmented from the most recent pretreatment diagnostic MRI or multiphase CT. For voxelized dosimetry analysis, PVs from clinical treatment planning were not explicitly used. Instead, PVs were re-segmented from CBCT performed during the planning procedure using a method guided by Hounsfield unit thresholding (seed-growing method). Tumor and PV contours were rigidly registered and transferred to post-therapy ⁹⁰Y single-photon emission computed tomography (⁹⁰Y-SPECT)/CT, using both SPECT and CT to guide optimal registration. Post-therapy, voxelized dosimetry was computed from ⁹⁰Y-SPECT/CT using the local deposition method and activity normalization to the PV with MIM SurePlan ⁹⁰Y. For voxelized dosimetry, no adjustment for LSF was included due to the low LSF values observed for patients in this study with small HCC (median planar LSF: 5.4%), and the well-established overestimation bias for planar LSF.¹⁶ Dose-volume metrics were computed from the voxelized dosimetry data, including V_y (y = 200, 400, 600, etc.) where V is the % treatment volume receiving at least y dose in Gy, and D_x (x = 5, 10, 15 … 95) where D is the minimum dose in Gy received by x% of the treatment volume. Treatment parameters and volume and dose metrics for PVs and tumors were compared with response and pathological necrosis (where available).

For further analysis, threshold-based segmentation of the post-therapy ⁹⁰Y-SPECT/CT was used to compute three modified perfused volumes at 10, 20, and 30% of the maximum ⁹⁰Y-SPECT counts value. As an example, the 10% contour included all regions within the liver in the ⁹⁰Y-SPECT with counts values ≥10% of the maximum value. The registered and transferred tumor contours were then used to calculate the percentage of the tumor volume intersecting with threshold-based ⁹⁰Y-SPECT perfused regions.

Outcomes

Target tumor response was evaluated on multiphase CT or MRI obtained at standard-of-care follow-up intervals, using modified RECIST (mRECIST) criteria, by board-certified radiologists.¹⁷ Local progression was defined as at least a 20% increase in the treated lesion from its best response, according to mRECIST criteria. In general, clinical and imaging follow-up occurred every 3 months per local institutional practice, or sooner at the discretion of the treating physician. Given the early-stage nature of the cohort and the concurrent use of other therapies for non-target lesions, local PFS was selected as the endpoint and censored at liver transplantation.¹⁸ Local tumor follow-up was censored at the time of any subsequent liver-directed therapy to that same tumor, if applicable. Baseline and subsequent ALBI grade and scores after treatment were determined through last documented follow-up and censored to initiation of any subsequent cancer therapy. Transitions in Child-Pugh class from baseline to last follow-up after treatment were visualized by a Sankey plot diagram created in RStudio (v2024.12.0) with ggplot2 and ggalluvial. Treatment-related adverse events (AEs) were monitored for at least 90 days after treatment and were graded according to the National Cancer Institute’s CTCAE version 5.0.¹⁹ In patients undergoing liver transplantation, pathologic tumor response was ascertained from standard-of-care liver explant pathology reports performed by the institution’s pathologists. Complete pathologic necrosis (CPN) was defined as the presence of 100% treatment effect with no viable tumor cells observed. Extensive tumor necrosis was defined as 50-99% tumor necrosis, and partial necrosis as <50%, consistent with prior large studies of ⁹⁰Y radioembolization before liver transplantation.²⁰

Statistical analysis

Continuous variables were reported as median with IQR or total range. Categorical variables were reported as percentages. Comparisons between continuous variables were performed using Mann-Whitney U test when applicable. Local PFS was calculated using the Kaplan-Meier method. Logistic regression models and receiver-operating characteristic curve analysis were applied to predict complete response using treatment planning and dosimetric parameters. Statistical significance was set at an alpha of 0.05. Statistical analyses were performed using GraphPad Prism v 10.5.0 (Boston, MA).

Results

Baseline characteristics

A total of 42 tumors in 38 patients met criteria for uRS and were included in this analysis. Baseline patient characteristics are summarized in Table 1. Most patients were male (27, 71.1%) and had metabolic dysfunction-associated steatotic liver disease as their underlying liver disease (18, 47.3%). Most patients had preserved functional status (ECOG 0; 27, 64.3%) and Child-Pugh A5 liver function (24, 63.2%), with the majority having ALBI grade 1 (20, 52.6%) or ALBI grade 2b (16, 42.1%) liver function at baseline. The majority of patients were early-stage BCLC A (19, 50%), with the remainder being very early-stage BCLC 0 (5, 13.2%), or BCLC C (14, 36.8%) due to ECOG status.

Table 1.

Patient characteristics.

	Treated population (n = 38)
Age, median [IQR], (range)	69 [64-74] (49-85)
Sex (%)
Male	27 (71.1%)
Female	11 (28.9%)
Liver disease etiology (%)
MASLD	18 (47.3%)
HCV	12 (31.6%)
Alcohol	5 (13.2%)
Other	3 (7.0 %)∗
BCLC (%)
0	5 (13.2%)
A	19 (50%)
C	14 (36.8%)
ECOG performance status (%)
0	24 (63.2%)
1	12 (31.2%)
2	2 (5.3%)
Child-Pugh score (%)
A5	24 (63.2%)
A6	7 (18.4%)
B7	5 (13.2%)
B8	2 (5.3%)
ALBI score (%)
1	20 (52.6%)
2a	2 (5.3%)
2b	16 (42.1%)
3	0
Baseline AFP (ng/ml), median [IQR], (range)	5.2, [3.2-9.6], (2.1-256)
Last clinical follow-up in months, median, [IQR], (range)	18.3, [12.8-21.7], (7.5-29.7)
Last imaging follow-up in months, median, [IQR], (range)	9.9, [6.5-15.0], (1.6-28.4)
Time to transplant in months, median, [IQR], (range)	7.3, [2.8-9.0], (2.2-14.6)

^∗Included one HBV, one primary biliary cholangitis, and one unknown. AFP, alpha-fetoprotein; ALBI, albumin-bilirubin; BCLC, Barcelona Clinic Liver Cancer; ECOG, Eastern Cooperative Oncology Group; LDT, liver-directed therapy; MASLD, metabolic dysfunction-associated steatotic liver disease.

Tumor and treatment characteristics are summarized in Table 2. The median tumor size was 2.4 cm (IQR 1.7-3.0 cm, total range 1.0-4.5 cm). Tumors were evenly distributed throughout the liver, with the highest percentage located in the Couinaud segment 8 (13, 30.9%). Most treatments involved a single perfused volume (e.g. angiosome) (37, 80.4%), with a median total perfused treatment volume of 66.3 cc (IQR 44.4-94.6 cc, total range 15.9-205 cc) resulting in a median total treated liver volume of 4.5% (IQR 3.0-6.3%, range 1.3-16.8%). Treated liver volume was >10% in three cases. Median total ⁹⁰Y administered activity and MIRD_Rx per tumor (single compartment MIRD) were 1.36 GBq (IQR 0.83-1.67, total range 0.3-3.61 GBq) and 837 Gy (IQR 600-1330 Gy, range 272-2,360 Gy), respectively. The majority of treatments utilized glass microspheres within the first week of calibration (Table 2), with most having an initial calibration size of less than 5.0 GBq (3.0 GBq n = 40, 3.5 GBq n = 1, 4.0 GBq n = 2, 4.5 GBq n = 3, 5.0 GBq n = 2, 6.0 GBq n = 2, 7.0 GBq n = 2, 9.5 GBq n = 1).

Table 2.

Tumor and treatment characteristics.

	Treated tumors (n = 42)
Tumor size (cm), median, [IQR], (range)	2.4, [1.7-3.0], (1.0-4.5)
Tumor location, Couinaud segment (%)
1	0 (0%)
2	3 (6.5%)
3	7 (15.2%)
4	5 (10.9%)
5	7 (15.2%)
6	7 (15.2%)
7	3 (6.5%)
8	14 (30.4%)
Angiosomes (vials) treated per tumor
1	37 (80.4%)
2	8 (17.4%)
3	1 (2.2%)
Vial calibration day
2	4 (9.5%)
3	7 (16.7%)
4	9 (21.4%)
5	16 (38.1%)
8	1 (2.4%)
9	2 (4.8%)
10	2 (4.8%)
11	1 (2.4%)
Total treated volume per tumor (cc), median, [IQR], (range)	66.3, [44.4-94.6], (15.9-205)
Percent total treated liver per tumor (%), median, [IQR], (range)	4.5, [3.0-6.3], (1.5-16.8)
Total activity delivered per tumor (GBq), median, [IQR], (range)	1.36, [0.83-1.67], (0.3-3.61)
Specific activity (Bq/sphere), median [IQR], (range)	1,189, [1,189-1,999], (251-2,591)
MIRD perfused volume dose∗ (Gy), median, [IQR], (range)	837, [600-1,330], (272-2,362)
Lung shunt fraction (%), median, [IQR], (range)	5.4, [4.6-6.4], (1.6-10.5)

^∗Single compartment MIRD. MIRD, medical internal radiation dose.

Radiographic response and survival

Radiographic response was available for 39 tumors in 35 patients, with the remaining three undergoing liver transplantation before first imaging follow-up (2.2-2.8 months after treatment). Of these three tumors, two demonstrated CPN and one 40% tumor necrosis on explant pathology. Details of best imaging and pathology response are outlined in Table 3. The median last imaging follow-up time was 9.9 months (IQR 6.5-15.0 months, total range 1.6-28.4 months). At first imaging follow-up (median 3.1 months, IQR 3.0-3.3 months, total range 0.8-5.6 months), complete response (CR) by mRECIST was observed in 26 (66.6%) tumors and partial response (PR) in the remaining 13 (33.3%). With further follow-up, a best localized mRECIST CR was seen in 34 (87%) of 39 evaluable tumors, and PR in the remaining 5 (13%). The median time to best overall response was 3.2 months (IQR 3.0-3.9 months, total range 0.8-17.2 months). The median time to CR was 3.3 months (IQR 3.0-5.1 months, total range 0.8-17.2 months), which is likely reflective of the 3-month time interval being the most common point for first follow-up (Fig. S2). Of note, the four cases where first imaging follow-up was earlier than 2 months demonstrated CR. Only one localized progression occurred at 9.2 months in a patient with biopsy proven poorly differentiated HCC in the setting of extrahepatic spread. Three tumors with sustained PR were subsequently treated with percutaneous ablation at the discretion of the treating physician (irreversible electroporation in two at 4.5 and 12.5 months, microwave ablation in one at 8.1 months) due to suspicion of residual disease, all with resulting CR. With a median time to last clinical follow-up of 18.3 months (range 7.5-29.7 months), the median localized PFS was not reached in the entire cohort and when censoring those that underwent liver transplant (Fig. 1). The 18-month local PFS rate was 88%. A total of four deaths were observed in the entire cohort, with none in those who underwent liver transplantation.

Table 3.

Treatment response per tumor.

	n = 39 tumors in 35 patients¶
mRECIST best overall response (%)
CR	34 (87)
PR	5 (13)
SD	0
PD	0∗
Treatment effect on explant pathology	n = 16 tumors in 14 patients
Complete tumor necrosis (100%)	11 (68.8%)
Extensive tumor necrosis (51-99%)	4 (25%)∗∗
Partial tumor necrosis (<50%)	1 (4%)∗∗∗

^¶Three patients underwent liver transplant prior to first imaging follow-up with available pathology reports.

^∗One local progression occurred at 9.2 months after initial PR in a patient with poorly differentiated HCC in the setting of extrahepatic spread.

^∗∗85%, 85%, 90%, and 95% tumor necrosis for each.

^∗∗∗40% tumor necrosis. CR, complete response; HCC, hepatocellular carcinoma; mRECIST, modified RECIST; PD, progressive disease; PR, partial response; SD, stable disease.

Fig. 1.

Local PFS, censored to last follow-up, liver transplant, or additional local treatment (if applicable).

Median local PFS was not reached. PFS, progression-free survival.

Pathologic response after liver transplantation

Explant pathology data was available for 16 tumors in a total of 14 patients who underwent liver transplantation at a median 7.4 months (range 2.2-14.6 months) after uRS treatment. Of these 14 tumors, 11 (68.8%) demonstrated CPN. The five tumors that did not exhibit CPN showed extensive treatment effect, with a median 85% tumor necrosis (total range 40-95%). While there was a trend of longer time to transplant in tumors achieving CPN (median 7.8 months, IQR 5.0-10.0 months, total range 2.5-14.6 months) vs. those that did not (median 3.6 months, IQR 2.6-7.2 months, total range 2.2-8.8 months), this was not statistically significant (p = 0.138, Mann-Whitney U test). There was also no statistical difference in tumor size between tumors achieving CPN (median 2.2 cm, IQR 1.7-2.6 cm, total range 1.0-3.0 cm) vs. those that did not (median 2.2 cm, IQR 1.8-2.4 cm, total range 1.6-3.4 cm, p = 0.816, Mann-Whitney U test).

Post-treatment voxel dosimetry

Of the 39 tumors with imaging follow-up, 38 from 35 patients had both imaging response and voxelized dosimetry data available for analysis. One patient had no available post ⁹⁰Y-SPECT/CT. Table 4 outlines both treatment planning parameters (n = 39) and voxelized dosimetry metrics (n = 38) for all tumors with imaging follow-up, with comparisons between those showing CR and PR/PD included. No planning parameter showed a significant difference between CR and PR/PD groups, even the predicted MIRD target dose (MIRD_Rx: p = 0.060). On the other hand, nearly all post-therapy voxel dose metrics showed significant differences between CR and PR/PD groups, including the voxelized version of mean PV dose (PV_voxel). Of note, PV_voxel values were in general higher than MIRD_Rx values from clinical treatment planning due to the removal of the LSF adjustment for PV_voxel.

Table 4.

Tumor response: prescribed MIRD and voxel-based dosimetry.

	All tumors, median (IQR)	CR, median (IQR)	PR/PD, median (IQR)	p value∗
Planning metrics (n = 39)
PV (ml)	66.3 (44.4–94.6)	69.2 (45.0–94.1)	47.7 (41.5–111)	0.958
Tumor (ml)	6.5 (4.3–13.7)	6.2 (4.1–12.5)	12.2 (6.5–14.6)	0.401
Tumor Size (cm)	2.4 (1.7–3.0)	2.4 (1.7–2.9)	3.0 (2.0–3.5)	0.360
PVliver (%)	4.5 (3.0–6.3)	4.7 (3.0–6.4)	4.1 (2.8–8.7)	0.894
TumorPV (%)	13.1 (6.5–21.9)	13.3 (6.1–20.1)	12.1 (8.9–27.9)	0.699
AA (GBq)	1.36 (0.83–1.67)	1.38 (0.96–1.78)	0.80 (0.45–1.40)	0.100
SA (Bq/sphere)	1,189 (1,189–1,999)	1,189 (1,189–1,999)	1,189 (490–1,280)	0.108
PV (spheres/ml)∗∗	14.9 (10.0–25.0)	14.9 (9.8–25.1)	14.9 (9.8–25.0)	0.980
MIRDRx (Gy)	837 (600–1,330)	839 (604–1,360)	538 (330–832)	0.060
Voxel dose metrics (n = 38)
PVvoxel (Gy)	867 (631–1,410)	870 (702–1,410)	542 (362–782)	0.035
Tumorvoxel (Gy)	1,600 (1,050–2,350)	1,720 (1,190–2,530)	726 (671–857)	<0.001
V200 (%)	100 (99.9–100)	100 (100–100)	98.1 (95.6–99.3)	<0.001
V400 (%)	99.8 (95.1–100)	100 (98.0–100)	81.5 (76.8–89.1)	<0.001
V600 (%)	97.9 (85.5–100)	99.5 (92.0–100)	56.6 (53.0–66.0)	<0.001
D70 (Gy)	1,180 (793–2,140)	1,320 (940–2,250)	523 (460–560)	<0.001
D95 (Gy)	707 (401–1,470)	855 (524–1,520)	275 (209–315)	<0.001
Tumor (spheres/ml)∗	23.7 (7.3–45.3)	25.4 (17.6–46.7)	18.7 (13.7–29.8)	0.310
NLvoxel (Gy)	743 (534–1,210)	753 (568–1,230)	450 (289–772)	0.083
TNR	2.0 (1.5–2.8)	2.3 (1.6–2.9)	1.6 (1.2–2.2)	0.202

^∗Mann-Whitney U test.

^∗∗Values in thousands. AA, administered activity; CR, complete response; Dx, minimum dose received by x% of tumor volume; MIRD_Rx, prescribed dose to the PV (MIRD); NL_voxel, mean dose to the non-tumor tissue in the PV; PD, progressive disease; PR, partial response; PV, perfused volume; PV_liver, % liver volume occupied by PVl; PV_voxel, voxel mean dose to PV; SA, specific activity; TNR, tumor to normal dose ratio; Tumor_PV, % of PV volume occupied by tumor; Tumor_voxel, voxel mean dose to tumor; Vy, % of tumor volume receiving y dose.

Specific dose metrics and the breakdown of CR vs. PR/PD are further demonstrated in Table S1 and Fig. 2. The most predictive voxel dose metric for discerning CR vs. PR/PD was D₉₅ >300 Gy. A total of 34/38 tumors analyzed achieved this D₉₅ threshold with 33/34 (97%) showing CR. All four tumors with D₉₅ <300 Gy showed PR/PD. Among the five PR/PD tumors, only one had MIRD_Rx ∼300 Gy. Three of the other four tumors fell well below the fitting line for D₉₅ vs. MIRD_Rx and had D₉₅ <300 Gy. This shows that these tumors without CR did not achieve comparable dose distributions relative to MIRD_Rx that were observed for the majority of CR tumors. A logistic regression model to predict CR using tumor D₉₅ was found to be highly significant and accurate. This model had a log-likelihood ratio of 21.2 (p <0.0001) and it predicted response in 36/38 tumors correctly (32/33 CR, 4/5 PR/PD). The AUC was 0.98 (p = 0.0006) with a tumor D₉₅ at 50% probability for CR of 324 Gy. Dose metric data for comparing tumors with CPN vs. non-CPN are also included in Table S1, showing that the differentiation between CPN and non-CPN was not as readily observed as for tumor response. Additionally, no statistically significant differences were observed in any parameter analyzed to distinguish between CPN and non-CPN tumors (p ≥0.362).

Fig. 2.

Voxel-based dosimetry data for the tumors in this study, separated by CR and PR/PD.

(A) Vy is percent of tumor volume receiving y dose and (B) Dx is minimum dose received by x percentage of tumor volume. Clear distinctions in the dose metric distributions between CR and PR/PD tumors, but regions of overlap remain where the same range of values led to either CR or PR/PD. The overlapped regions shrink as the minimum dose (y) decreases in Vy and as the volume receiving a minimum dose (x) increases in Dx. (C) Tumor D₉₅ is plotted vs. MIRD_Rx, with CR and PR/PD tumors differentiated in the plot. A linear fit of Tumor D₉₅ = 0.99∗MIRD_Rx (R2 = 0.50) for CR tumors is also plotted. A D₉₅ threshold of 300 Gy (dotted line) is 33/34 predictive for CR and 4/4 predictive for PR/PD. CR, complete response; MIRD_Rx, prescribed dose to the PV (MIRD); PD, progressive disease; PR, partial response.

Treatment-related AEs and liver function

Treatment-related AEs are summarized in Table S2. The most common AE was fatigue, reported in 17 patients (44%), and was limited to grade 1–2 in all cases. The most frequent laboratory abnormality was lymphopenia. Grade 3 AEs occurred in six patients (14%) and included lymphopenia in five (13.2%) and a single case of arterial injury (2.6%) due to closure device failure requiring endovascular repair. All grade 3 lymphopenia cases were transient and resolved with follow-up. There were no grade 4 or 5 treatment-related AEs.

Changes in ALBI score from baseline were tracked after 39 treatments in 38 patients up to the time of transplant or receiving any subsequent cancer therapy. One patient was analyzed twice since they underwent sub-segmentectomy to two different tumors 9 months apart. Fig. 3A demonstrates the trajectory of ALBI score over time after treatment. With a median ALBI score follow-up time of 265 days (range 66-670 days), the majority (32/39, 82%) retained or improved their baseline ALBI grade at last follow-up. Of patients who were ALBI 1 at baseline, only 3 (15%) saw deterioration of grade at last follow-up, which was modest to ALBI 2a in all cases. Importantly, only two patients (13%) who were ALBI 2b at baseline deteriorated to ALBI 3, while two others in this baseline category saw improvements in their ALBI grade at last follow-up. Consistent with these trends, the majority also had stable (n = 28, 71.8%) or improved (n = 4, 10.3%) Child-Pugh score at their last follow-up (Fig. 3B), Notably, all patients who were Child-Pugh A who worsened at least one score were also ALBI 2b at baseline (n = 4), whereas all ALBI 1 or ALBI 2a patients had stable Child-Pugh scores. Among patients who were baseline Child-Pugh B (B7–B8), 4 of 7 (57%) had non-worsening or improved scores at last follow-up.

Fig. 3.

Liver function after uRS.

(A) Changes in ALBI score plotted over time after uRS treatment (day 0). Patients are grouped based on their baseline ALBI score just prior to treatment (blue – ALBI 1, green – ALBI 2a, orange – ALBI 2b). (B) Sankey plot illustrating transitions of Child-Pugh class from baseline to last follow-up. Bar height is proportional to number of patients in that category. Ribbon width reflects number transitioning between categories. Ribbons colored by class at baseline (green = Child-Pugh A, orange = Child-Pugh B). ALBI, albumin-bilirubin; uRS, ultra-selective radiation segmentectomy.

Discussion

This study demonstrates that uRS using glass ⁹⁰Y-microspheres is technically feasible and a highly effective liver-sparing treatment option for patients with early-stage HCC. With a CR rate of 87% by mRECIST and 69% complete tumor necrosis by explant pathology, outcomes were comparable to those reported with traditional RS, despite treating significantly smaller liver volumes.^3,4,21 This response was also durable, with a median local PFS not reached and only a single local progression across all 39 evaluable tumors. The median perfused treatment volume was 66.3cc, representing a median 4.5% total liver volume, which is well below thresholds associated with increased risk of hepatic decompensation for ablative radioemblization.¹⁰ These findings underscore the ablative potential of this highly conformal radioembolization, while highlighting its ability to preserve liver parenchyma. In addition, it confirms the replicability of selective, ablative ⁹⁰Y-radioembolization for the treatment of early-stage HCC across multiple treatment centers and operators.

The safety of this approach was further reflected not only in the low treatment-related AE profile but also in the preserved hepatic function, measured by both ALBI and Child-Pugh scores. The treatment was generally well tolerated, with most AEs limited to grade 1–2 in severity. Only six patients (14%) experienced grade 3 events, all of which were transient or manageable without long-term consequences. Among patients evaluable for ALBI trajectory, 82% retained or improved their baseline ALBI grade following uRS, and deterioration to ALBI grade 3 was rare. Of note, most patients in our cohort who had baseline ALBI 2 status could be further sub-stratified as ALBI 2b (>-2.270), which was associated with worse liver function prognosis compared to ALBI 2a.²² Despite this, only 2 (13%) of these patients with ALBI 2b baseline liver function deteriorated to ALBI 3 on their last follow-up, while two others actually saw an improvement in their ALBI grade. Concordantly, at last follow-up, 82% had non-worsening Child-Pugh scores. All baseline Child-Pugh A patients who worsened by at least 1 point were ALBI 2b at baseline, whereas all ALBI grade 1 or 2a patients maintained a stable Child-Pugh score. In the small Child-Pugh B subgroup, 57% had non-worsening or improved scores. This preservation of hepatic reserve is particularly relevant for patients ineligible for curative surgery or transplantation. These results reinforce uRS as a viable curative-intent option for early-stage HCC in carefully selected patients.

Interestingly, no planning parameters showed statistically significant differences between CR and non-CR tumors, not even MIRD_Rx. While this may not be ideal for aligning treatment planning with optimized response outcomes, it validates the concept that post-therapy imaging and dosimetry assessments are important for guiding clinical decisions after ⁹⁰Y radioembolization. Voxel-based post-treatment dosimetry confirmed clear dose-response relationships in this study, with tumor D₉₅ >300 Gy strongly predictive of CR. These findings align with prior established work, where dose-volume metrics at the boundary of the DVH (dose-volume histogram) curve (e.g. Vy = 100%, D₉₅, D₉₉) have been shown to be highly predictive of response and pathological necrosis.^21,23,24 While outcomes were largely favorable, the small subset of tumors that failed to achieve CR, including one case of local progression, highlight a potential limitation of excessive treatment selectivity. Fig. 4 shows fused ⁹⁰Y-SPECT/CT axial images from the five PR/PD cases to highlight these observations. In Fig. 4A, the tumor still aligns well with the high-activity regions of the PV and the tumor D₉₅ is comparable to MIRD_Rx. But the PV was likely underdosed, leading to a tumor D₉₅ insufficient to ensure CR. It is worth noting that the case in Fig. 4A had the highest tumor D₉₅ that did not lead to CR. In Figs. 4B-E, it is clear that these tumors did not consistently overlap with the regions in the PV receiving the highest ⁹⁰Y activity and absorbed dose. The tumor D₉₅ values are all much lower than MIRD_Rx, and all <300 Gy as a result. Fig. 4E represents the case with PD, where the misalignment was most severe among all five cases without CR. Fig. 4F shows a case with two CR tumors for comparison: the PV was not underdosed, both tumors consistently overlapped with the highest ⁹⁰Y distributions within the PV, and tumor D₉₅ values were both well above 300 Gy. Therefore, voxel-based post-treatment dosimetry may identify tumors that have received inadequate treatment and will require subsequent therapies.

Fig. 4.

Fused axial ⁹⁰Y-SPECT/CT images.

Fused axial ⁹⁰Y-SPECT/CT images for four tumors with PR (A-D), one with PD (E), and an example case with two CR tumors (F). The liver, PV, and tumor contours are shown in each image, and the MIRD_Rx and tumor D₉₅ dose values are also included. ⁹⁰Y-SPECT, ⁹⁰Y single-photon emission CT; CR, complete response; MIRD_Rx, prescribed dose to the PV (MIRD); PD, progressive disease; PR, partial response; PV, perfused volume.

To further showcase the clinical utility of these observations, we tested two additional logistic regression models that utilized MIRD_Rx, the only dose metric that can be controlled prior to treatment. The first model used MIRD_Rx alone, while the second model combined MIRD_Rx with additional metrics for the percentage of the tumor volume within the 10%, 20%, and 30% ⁹⁰Y-SPECT threshold PV contours. These additional metrics helped to differentiate response between tumors based on their overlap with the highest regions of ⁹⁰Y activity in the PV. In the MIRD_Rx only model, all 38 tumors were predicted as CR with an average probability for tumors with true CR and PR/PD of 0.88 and 0.78, respectively. The second model with additional metrics related to tumor overlap with ⁹⁰Y-SPECT activity distributions accurately predicted response in 37/38 tumors (33/33 CR, 4/5 PR/PD), with mean probability for CR and PR/PD tumors of 0.96 and 0.29, respectively.

From our analysis in this study (Fig. 2C), there was variation in how well the key dose metric of tumor D₉₅ correlated with MIRD_Rx. The general fit connecting tumor D₉₅ with MIRD_Rx for CR tumors was tumor D₉₅ ∼ MIRD_Rx, but some CR tumors deviated well below this average correlation. However, these tumors maintained D₉₅ ≥300 Gy due to a much higher MIRD_Rx to compensate. Especially for the smaller PV and tumor volumes associated with the uRS approach, a tumor D₉₅ sufficient to produce CR may be easier to achieve, despite dose non-uniformities. This is because target MIRD_Rx can be increased well above traditional recommendations (e.g. 400 Gy) while maintaining very low liver toxicity. Nevertheless, to increase the likelihood of tumor D₉₅ being correlated with target MIRD_Rx, it is important to ensure sufficient tumor coverage with the identified PV so that regions with high ⁹⁰Y activity distribution overlap with the treated tumor. The findings from the few PR/PD cases in this study highlight that in some circumstances highly focal delivery, while minimizing off-target dose, may risk insufficient coverage of tumor margins or peritumoral microscopic disease. As such, implementation of the uRS approach should mandate careful scrutiny of tumor coverage within the perfused treatment volume on CBCT and the consideration of additional treatment regions with traditional RS methods for adequate tumor margin coverage, especially in lower-risk patients with well-compensated liver function. As shown in Fig. S3, other dose metrics like NL_voxel and Tumor_voxel showed stronger correlations with MIRD_Rx than tumor D₉₅. Unfortunately, these dose metrics that correlated better with the only controllable quantity prior to therapy (MIRD_Rx) were not as consistent at differentiating tumor response.

This study has several limitations. First, it is a single-center retrospective cohort with a modest sample size. However, the outcomes reported here are comparable to those with traditional RS from both prospective and retrospective studies.^3,4,21,25 The limited number of patients who subsequently went on to receive liver transplantation limited the ability to delineate any meaningful dosimetry parameters associated with CPN. However, this was not the primary objective of our study and detailed cohorts describing these optimized parameters with glass ⁹⁰Y-microspheres have been pursued by other groups.²³ It should be noted that limitations in the resolution of ⁹⁰Y-SPECT/CT imaging make accurate assessment of the detailed regions of DVH curves difficult, especially for small treatment volumes and tumors associated with RS and uRS approaches. These effects may contribute to higher uncertainty in reported dose metrics for the smallest tumors analyzed in this study. Caution is also warranted when directly comparing specific, dosimetry-derived thresholds between studies, as the details of voxel dosimetry analysis, time to imaging follow-up, and time to pathology analysis may all influence such thresholds. Nevertheless, integration of ⁹⁰Y-SPECT-based dose distributions and threshold contour alignment may further enhance prediction of incomplete responses and guide post-treatment management or retreatment strategies. Finally, the consistency of imaging response, dosimetry, and pathologic data across multiple endpoints provides robust support for the clinical utility of uRS.

In conclusion, radiation sub-segmentectomy is a promising ablative strategy for early-stage HCC that achieves high rates of tumor control while preserving liver function. With appropriate procedural planning to ensure margin coverage, uRS may serve as a definitive therapy or transplant-bridging modality in patients not amenable to standard curative approaches. Further prospective validation of this technique and integration of real-time perfusion assessment tools will be important to guide its broader application.

Abbreviations

90Y, Yttrium-90; AE, adverse event; ALBI, Albumin-Bilirubin; BCLC, Barcelona Clinic Liver Cancer; CBCT, cone-beam CT; CPN, complete pathologic necrosis; CR, complete response; D_x, minimum dose received by x% of tumor volume; ECOG, Eastern Cooperative Oncology Group; HCC, hepatocellular carcinoma; LSF, lung shunt fraction; MIRD, medical internal radiation dose; MIRD_Rx, prescribed MIRD dose; NL_voxel, mean dose to the non-tumor tissue in the PV; PD, progressive disease; PFS, progression-free survival; PR, partial response; PV, perfused volume; PV_voxel, voxel mean dose to PV; RS, radiation segmentectomy; SPECT, single-photon emission CT; TACE, transarterial chemoembolization; mRECIST, modified RECIST; uRS, ultra-selective radiation segmentectomy; V_y, % of tumor volume receiving y dose.

Financial support

The authors declare no financial support was received for conducting the study or writing the manuscript.

Authors’ contributions

Study concept and design: CM, MAT; acquisition of data: CM, TA, SM, GF, MAT; analysis and interpretation of data: CM, GF, MAT, JK, JDG, NM, DB; drafting of the manuscript: CM, MAT; critical revision of the manuscript for important intellectual content: all authors; statistical analysis: CM, MAT; study supervision: CM, MAT.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the author(s) used ChatGPT (OpenAI) for assistance in phrasing and wording of text. After using this tool, the author(s) reviewed and edited the content as needed and take full responsibility for the content of the publication.

Data availability

The data supporting the findings of this study are available upon reasonable request from the correspdoning author. Due to privacy and ethical restrictions, individual participant data cannot be shared publicly.

Conflict of interest

CM reports research support from the American Cancer Society, Society of Interventional Oncology, and Boston Scientific. CM receives speaking fees from Boston Scientific and advisor fees from AstraZeneca Pharmaceuticals LP and Eisai. AT receives research support from Boston Scientific. The remaining authors have nothing to declare.

Please refer to the accompanying ICMJE disclosure forms for further details.

Supplementary data

The following are the Supplementary data to this article.