Clinical outcomes and quality of life in glioblastoma patients treated with MR-guided adaptive radiotherapy using a 1.5 T MR-Linac in an Australian setting

Highlights

• •MR-guided adaptive radiotherapy on a 1.5 T MR-Linac was shown to be feasible.
• •All patients completed treatment with no major treatment-related toxicity.
• •Daily MRI enabled adaptive replanning for tumor and edema volume changes.
• •Quality of life remained stable during therapy, declining only post-treatment.
• •Findings support integrating MRgRT into glioblastoma management workflows.

Abstract

Purpose

The magnetic resonance linear accelerator (MRL) enables daily adaptation of radiation therapy plans. This study presents our institutional experience with the clinical outcomes and quality of life (QoL) of glioblastoma patients treated using a 1.5 Tesla (T) MRL.

Materials and Methods

Eligible patients with glioblastoma were enrolled in a study and treated with MRL between August 2021 and May 2024. Acute toxicity was assessed. QoL was evaluated using the EORTC QLQ-C30 and QLQ-BN20 questionnaires. Cumulative overall survival (OS) and progression-free survival (PFS) were estimated using Kaplan-Meier analysis.

Results

27 patients were included, with a median follow-up of 12.2 months (range 0.5–30.6 months). The mean age was 63 years; 14 were male and had an ECOG performance status of 0–1. Eleven patients underwent gross total resection, 9 had subtotal resection, and 7 had biopsy only. 11 were MGMT-methylated, 15 nonmethylated, and 1 had unknown status. Twenty-five received their first radiotherapy course on the MRL. The mean interval from CT/MRI simulation to radiotherapy start was 12.6 days. 26 patients completed all prescribed fractions on the MRL. The median on-table treatment time per fraction was 28.2 (range: 11.7–66.5) minutes. Eleven patients (39.3%) required Adapt-To-Shape or re-planning within the first week due to tumor progression, indicated by increased T2/FLAIR signal beyond the initial CTV. No Grade 3–4 toxicities were reported. At 3 months, treatment was associated with declines in QoL domains at 3 months with recovery by 6 months. The median OS and PFS were 14.1 and 11.2 months.

Conclusion

Treatment of glioblastoma with MR-guided adaptive radiotherapy is feasible, with manageable acute toxicities and adaptive capabilities to address tumor changes.

Introduction

Glioblastoma (GBM) is the most common primary malignant brain tumor and is associated with a poor prognosis, with a median survival of 15–18 months [1], [2]. The current standard of care consists of maximal safe surgical resection followed by radiotherapy (RT) with concomitant and maintenance temozolomide (TMZ) chemotherapy. The radiation target typically encompasses the residual tumor, the surgical cavity, and a 1.5–2 cm margin including adjacent edema [3], as supported by autopsy studies and stereotactic biopsies [4], [5].

The introduction of magnetic resonance imaging (MRI)-guided radiotherapy (MRgRT) using MRI-linear accelerators (MRLs) has revolutionized treatment planning and delivery for GBM. This technology enables the acquisition of high-resolution MRI scans before each treatment fraction, providing superior soft-tissue visualization [6], [7]. Real-time adaptive radiotherapy facilitated by MRLs allows for adjustments to account for interfractional changes in tumor volume and migration [8], [9], potentially enabling reduced treatment margins and more individualized target volumes.

The 1.5 Tesla (T) MRI-linear accelerator (Unity, Elekta AB, SE) has become increasingly utilized for the treatment of brain tumors, including GBM [10], [11]. However, despite its growing adoption, limited data exist regarding its clinical outcomes and impact on quality of life (QoL) for GBM patients. At our center, the 1.5 T MRL was initially employed for treating brain tumors before expanding to other anatomical sites. Here, we report clinical outcomes and QoL data for GBM patients treated with the 1.5 T MRL in the public Australian healthcare setting.

Materials and Methods

Patients with GBM treated on the 1.5 T MRL at XXX were enrolled through the Feasibility of Imaging and Radiation Treatment Delivery on the MR-Linac (FIRM) trial. Eligible patients had histopathologically confirmed WHO grade 4, IDH-wildtype glioblastoma. Both newly diagnosed and recurrent cases were included. Patients were required to have no contraindications to MRI and be suitable for treatment on the 1.5 T MRL.

From August 2021, when the MRL became operational for brain tumors, to May 2024, all patients received online adaptive MRgRT using the 1.5 T MRL. Prospective collection of clinical, technical, and patient-reported outcome data was conducted with informed consent as part of the FIRM study.

Treatment

All patients underwent computed tomography (CT) and MRI simulation scans, with CT scans acquired at a slice thickness of 2 mm. MRI simulation sequences included pre- and post-gadolinium T1-weighted 3D fat-suppressed black blood (FS BB), T2-weighted 3D, T2-weighted fluid-attenuated inversion recovery (FLAIR), and diffusion-weighted imaging (DWI). MRI datasets were co-registered to the CT datasets, and the treating radiation oncologist performed target delineation.

The gross tumor volume (GTV) included the residual tumor and surgical cavity as visualized on T1-weighted post-contrast MRI. An additional 1.5 cm margin, including T2/FLAIR hyperintense regions adjusted for anatomical barriers and routes of spread, was added to define the clinical target volume (CTV), and an isotropic 3 mm margin was applied to create the planning target volume (PTV). Treatment planning utilized a predefined Monaco TPS template that included prescription dose, intensity-modulated radiotherapy (IMRT) optimization parameters, beam arrangement, dosimetric criteria, and calculation and sequencing parameters. Beam arrangements and IMRT constraints were also adjusted per patient as needed to meet all dosimetric requirements.

Plans were calculated using the GPUMCD Monte Carlo algorithm with [parameters: 1% statistical uncertainty per plan and either a 2–3 mm dose grid dependent on target size and position relative to critical OARs.] A dose reference point was placed within the high dose target volume to enable an independent secondary MU check during online planning. Patient-specific quality assurance (QA) was performed using the Arccheck-MR device (Sun Nuclear, Melbourne, Fl) before treatment initiation. As our site has only one MRL, backup VMAT treatment plans for conventional linear accelerators were prepared to ensure continuity of care during potential MRL downtime or maintenance. Before initiating treatment, patients underwent a non-treatment session (Fraction 0) on the MRL. The aim of this session was to evaluate patient tolerability and feasibility of the setup, image quality and robustness of the reference treatment plan to online plan adaptation.

Three adaptive workflows were employed for daily treatment adjustments: Adapt-to-Position (ATP), Adapt-to-Shape (ATS), and full CT-based replanning. The ATP workflow involved rigidly shifting the reference treatment plan to align with daily anatomy based on translational adjustments, followed by re-optimization of the plan. In contrast, the ATS workflow required manual adaptation of target and organ-at-risk (OAR) contours to reflect the anatomy of the day using the online MRI, after which a new treatment plan was re-optimized to account for these changes. This facilitated online replanning in the event of progression of the tumor seen on localization scans. Full CT-based replanning was primarily used during the early phase of MRL implementation, before the ATS workflow was fully established at our institution. In these cases, anatomical or target changes detected on daily MRI necessitated contour modification and creation of a new treatment plan on the original CT simulation dataset, which was then used for subsequent fractions. An online plan check was performed which includes verifying the electron density grid and performing a secondary MU check using RadCalc software (Lifeline Software, Tyler, TX). Any total dose discrepancy greater than 3% or beam discrepancy greater than 10% between Monaco TPS and RadCalc required further investigation online. FLAIR images were acquired during the treatment planning stage for patient position verification purposes and DWI MRI images were acquired during delivery for offline review. The radiation oncologist reviewed and approved the newly generated online plan. Workflow timings were also recorded.

Data collection

Data on patient characteristics included age, sex, and performance status according to the Eastern Cooperative Oncology Group (ECOG) scale. Tumor characteristics were limited to location in the brain and MGMT methylation status. Treatment details included the type of resection, use of concurrent temozolomide, whether it was a primary or re-irradiation course, total radiation dose, interval from CT/MRI simulation to radiotherapy initiation, therapy completion rate on the MRL, workflow durations, and the utilization rates of ATP and ATS strategies.

Acute toxicities were defined as those occurring from the start of treatment up to three months of follow-up, and were reported according to the Common Terminology Criteria for Adverse Events (CTCAE) version 5. Fatigue, headache, nausea, and dermatitis were routinely assessed during treatment and subsequent follow-up.

Patients also completed the EORTC QLQ-C30 and BN20 questionnaires to assess QoL at baseline and at 3-, 6-, 12-, and 24-month follow-up visits. Based on the QLQ-C30 results, global health status (GHS), along with functional and symptom scales (range 0–100), were calculated according to the EORTC scoring manual. Higher scores on GHS and functional scales indicate better QoL or functioning, whereas higher scores on symptom scales suggest greater symptom severity. The QLQ-BN20 questionnaire, specific for patients with brain tumors, contains 11 scales which assess symptoms and neurologic deficits, with higher score indicating worse QoL.

Progression was determined using the Response Assessment in Neuro-Oncology (RANO) criteria [12] or based on changes in clinical management prompted by imaging or clinical findings. Progression-free survival (PFS) was defined as the interval from the completion of RT to either disease progression or death, whichever occurred first. Overall survival (OS) was defined as the time from the completion of RT to death. Patterns of progression were classified based on their location relative to the CTV. Marginal progression involved 20–80% of the recurrent GTV within the 15 mm CTV margin, out-of-field progression was defined as < 20%, and in-field progression as > 80% within the CTV.

Statistical analysis

Descriptive statistics were prepared to summarise the patient characteristics and treatment details. Results were expressed as means and standard deviation (±SD) or range for normally distributed continuous variables or median (range) for skewed variables. QoL domains were evaluated following the guidelines outlined in the EORTC manual [13], [14]. Higher scores on functional scales indicated better functioning and overall QoL, whereas higher scores on symptom scales reflected greater symptom severity and poorer QoL. Changes in QoL over time were analyzed using paired t-tests with mean differences (MD) along with 95% confidence intervals (CI) also estimated. Minimally important differences (MID) were determined based on the criteria established by Dirven et al. for QoL domains relevant to glioma patients [15]. In cases where specific thresholds were not available, a change of ≥ 10 points was considered clinically significant. Kaplan-Meier curves were prepared to present the PFS and OS outcomes, with estimated survival and progression-free survival probabilities estimated along with 95% confidence intervals. Statistical analysis was conducted using Stata version 18.0 (StataCorp, College Station, Texas, USA), with a p-value of less than 0.05 considered to indicate statistical significance.

Results

Study Population

Patient characteristics and treatment details of the 27 patients included in the study are listed in Table 1. Mean age at registration was 62.9 years. Of these, 14 (51.9%) were male. The majority of patients had an ECOG performance status of 0–1 (96.2%), and most tumors were predominantly located in the temporal lobe (39.3%). MGMT promoter methylation was identified in 11 patients (40.7%), while 15 (55.6%) were unmethylated, and 1 (3.7%) had unknown status. Eleven patients (40.7%) underwent gross total resection, 9 (33.3%) had subtotal resection, and 7 (25.9%) had biopsy only. Most (89.3%) received concurrent temozolomide. Twenty-five received their first radiotherapy course on the MRL. One patient underwent two MRL treatments for distant failure, while another received a single MRL course for marginal recurrence following prior treatment on a conventional linac.

Table 1.

Baseline characteristics and treatment details, n(%) unless otherwise indicated.

Characteristics	N = 27
Age in years, mean (SD)	62.93 (10.21)
Sex
Male	14 (51.9%)
Female	13 (48.1%)
ECOG grade
0	13 (48.1%)
1	13 (48.1%)
2	1 (3.7%)
Tumor location*
Frontal	9 (32.1%)
Temporal	11 (39.3%)
Occipital	1 (3.7%)
Parietal	7 (25.0%)
MGMT promoter methylation
Methylated	11 (40.7%)
Unmethylated	15 (55.6%)
Unknown	1 (3.7%)
Surgery
Gross total resection	11 (40.7%)
Subtotal resection	9 (33.3%)
Biopsy	7 (25.9%)
Concurrent systemic treatment*
Temozolomide	25 (89.3%)
Lomustine	1 (3.6%)
Temozolomide + ADI-PEG 20	1 (3.6%)
None	1 (3.6%)
Course of radiation in MRL
First only	25 (92.6%)
Second only	1 (3.7%)
First and second	1 (3.7%)
Fractionation scheme*
60 Gy/30 fractions	16 (57.1%)
40 Gy/15 fractions	11 (39.3%)
26 Gy/13 fractions**	1 (3.6%)
Days from CT/MR simulation to start of RT, mean (SD)	12.61 (2.87)
Percentage of fractions completed on MRL, median (range)	100% (96.67–100)
On-table treatment time per fraction (mins), median (range)	28.2 (11.7–66.5)
On-table treatment time per fraction for ATP workflow (mins), median (range)	28.0 (11.7–61.7)
On-table time per fraction for ATS workflow (mins), median (range)	41.0 (27.9–66.5)
Distribution of adaptation techniques*
ATP only	13 (46.4%)
CT replanning and ATP	9 (32.1%)
Mixed ATP/ATS	5 (17.9%)
CT replanning and mixed ATP/ATS	1 (3.6%)
No. of fractions requiring ATS/replanning per patient,*** median (range)	1 (1–7)
Timing of ATS/replanning by fraction interval	N = 38 fractions requiring ATS/replanning
F0-5	12 (31.6%)
F6-10	10 (26.3%)
F11-15	5 (13.2%)
F16-20	4 (10.5%)
F21-25	5 (13.2%)
F26-30	2 (5.3%)

ATP: adapt-to-position; ATS: adapt-to-shape; CT: computed tomography; ECOG: Eastern Cooperative Oncology Group; Fx: fraction, where x = number of fraction; MGMT: O6-methylguanine-DNA methyltransferase; MR: magnetic resonance; MRL: magnetic resonance-linear accelerator; SD: standard deviation

*n = 28, 28 MRL treatments for 27 patients with 1 patient receiving 2 MRL treatments for distant failure.

**Discontinued treatment per patient preference.

***For patients requiring CT replanning or ATS workflow only, n = 15 courses.

Treatment details

Sixteen patients (57.1%) were treated with 60 Gy in 30 fractions, 11 patients (39.3%) with 40 Gy in 15 fractions, and one (3.6%) discontinued treatment after 26 Gy in 13 fractions. The discontinuation was not due to toxicity but because the patient elected not to continue with further chemoradiotherapy. The mean interval from CT/MRI simulation to the start of radiotherapy was 12.6 days. One patient received one fraction on a conventional linear accelerator due to technical issues, while the remaining 27 patients completed all prescribed fractions on the MRL.

The median on-table treatment time per fraction was 28.2 min (range: 11.7–66.5 min). For fractions treated using the ATP workflow, the median treatment time was 28.0 min (11.7–61.7 min), whereas fractions requiring ATS had a median treatment time of 41.0 min (27.9–66.5 min). ATP workflows remained relatively consistent across all fractions, whereas ATS showed greater variability, with on-table treatment times exceeding 50 min in two fractions and 60 min in one fraction (Fig. 1). Stepwise workflow durations for ATP and ATS fractions are summarized in Fig. 2. Of the 11 recorded workflow steps, ATS-specific tasks such as fusion-to-recontour and recontour-to-optimise added 8.5 and 6.1 min, respectively. Consequently, the median fusion-to-optimise interval was 11.6 min for ATS compared to 2.7 min for ATP. Other steps, including QA and treatment delivery were comparable between workflows. These added durations contributed to the overall longer on-table treatment times observed with ATS. Notably, a trend toward shorter treatment times was observed with increased experience, with median durations of 27 min for the first five patients, 31 min for patients 6–10, 29 min for patients 11–15, 27 min for patients 16–20, 28 min for patients 21–25, and 21 min for patient 26 onwards.

Fig. 1.

Total on-table treatment time per fraction on MRL across radiotherapy course.

Fig. 2.

Workflow timings for Adapt-to-Position and Adapt-to-Shape workflows.

A total of 28 radiotherapy courses were delivered to 27 patients, with one patient treated twice on the MRL. Of these, 13 courses were completed entirely using the ATP workflow. In nine courses, at least one full replan on the simulation CT was performed during treatment, resulting in a combination of CT-based replanning and ATP. During the early phase of MRL implementation for GBM, the ATS workflow was not yet fully established, necessitating full replanning in some cases. Five treatment courses employed a mixed ATP and ATS workflow, and one course required both CT replanning and a combination of ATP and ATS. Within the first week of treatment, 11 out of 28 courses (39.3%) required replanning or ATS due to tumor progression, indicated by expansion of T2/FLAIR signal beyond the initial CTV; in four of these, replanning occurred after F0. Across the cohort, 15 courses (53.6%) involved ATS or replanning at some point during treatment, accounting for 38 out of 405 total fractions (including F0). The median number of ATS or replanned fractions was one per course. Most adaptations occurred between F0-5 (12 of 38, 31.6%) (Table 1).

Toxicity

The incidence of acute treatment-related toxicities is summarized in Table 2. Fatigue was the most commonly reported toxicity, with its incidence increasing over the course of treatment. Grade 1 fatigue was observed in 16 patients (57%), while 6 patients (21%) experienced grade 2 fatigue. Headache was infrequent, with only 5 patients (18%) reporting grade 1 headaches. Similarly, nausea was uncommon, with 4 patients (14%) experiencing grade 1 nausea. Grade 1 dermatitis was reported in 7 patients (25%). No grade 3 or 4 toxicities were observed.

Table 2.

Acute treatment-related toxicities graded according to CTCAE v5.

Toxicities	N = 28*
Fatigue
Grade 0	6 (21%)
Grade 1	16 (57%)
Grade 2	6 (21%)
Headache
Grade 0	23 (82%)
Grade 1	5 (18%)
Nausea
Grade 0	24 (86%)
Grade 1	4 (14%)
Dermatitis
Grade 0	21 (75%)
Grade 1	7 (25%)

*n = 28, 28 MRL treatments for 27 patients with 1 patient receiving 2 MRL treatments for distant failure.

Quality of life

For QoL and oncologic outcomes, only data from the first course of treatment were included for the patient who received two separate MRL treatments. The GHS and functional and symptom scales from the EORTC QLQ-C30 and QLQ-BN20 questionnaires are summarized in Table 3 and Table 4, respectively. Response rates were 100% at baseline (28/28 patients), 75% at 3 months (21/28 patients), 54% at 6 months (15/28 patients), 32% at 12 months (9/28 patients), and 7% at 24 months (2/28 patients). At 3 months, treatment was associated with a decline in GHS (MD −15.08 [95% CI −27.98 to −2.18]), physical functioning (MD −17.46 [-29.88 to −5.04]), cognitive functioning (MD −19.05 [-31.35 to −6.75]), as well as worsening pain (MD 15.87 [0.61 to 31.14]). At 6 months, declines in physical functioning persisted (MD −19.11 [-36.48 to −1.75]). By the 12-month follow-up, most domains had returned to baseline levels.

Table 3.

Changes in quality-of-life measures over time based on EORTC QLQ-C30 scores (paired data).

QoL Measure	Mean (SD), n = 21		Mean Difference (95% CI)	p-value
	Baseline*	3-month follow-up
Global health score	64.29 (20.77)	49.21 (23.56)	−15.08 (−27.98 to −2.18)	0.024
Physical functioning	73.81 (19.73)	66.35 (26.12)	−17.46 (−29.88 to −5.04)	0.008
Role functioning	61.11 (36.64)	49.21 (35.93)	−11.90 (−29.23 to 5.42)	0.167
Emotional functioning	75.40 (21.81)	78.57 (19.82)	3.17 (−7.04 to 13.39)	0.524
Cognitive functioning	71.43 (25.36)	52.38 (34.27)	−19.05 (−31.35 to −6.75)	0.004
Social functioning	53.97 (36.10)	48.41 (39.76)	−5.55 (−24.19 to 13.08)	0.541
Fatigue	36.51 (25.86)	48.68 (31.72)	12.17 (−1.58 to 25.92)	0.080
Nausea and vomiting	1.59 (5.01)	8.73 (16.35)	7.14 (−1.02 to 15.30)	0.083
Pain	4.76 (11.95)	20.64 (32.02)	15.87 (0.61 to 31.14)	0.042
QoL Measure	Mean (SD), n = 15		Mean Difference (95% CI)	p-value
	Baseline*	6-month follow-up
Global health score	68.33 (21.18)	58.33 (26.54)	−10.00 (−28.48 to 8.48)	0.265
Physical functioning	85.33 (15.78)	66.22 (34.96)	−19.11 (−36.48 to −1.75)	0.033
Role functioning	67.78 (36.98)	54.44 (38.56)	−13.33 (−27.37 to 0.71)	0.061
Emotional functioning	77.78 (23.29)	75.00 (22.49)	−2.78 (−14.31 to 8.75)	0.613
Cognitive functioning	76.67 (21.64)	63.33 (31.62)	−13.33 (−31.53 to 4.86)	0.138
Social functioning	61.11 (35.45)	54.44 (35.89)	−6.67 (−24.04 to 10.71)	0.424
Fatigue	33.33 (24.13)	45.93 (30.25)	12.59 (−0.33 to 25.52)	0.055
Nausea and vomiting	2.22 (5.87)	5.56 (10.29)	3.33 (−1.84 to 8.51)	0.189
Pain	2.22 (5.87)	12.22 (27.07)	10.00 (−5.52 to 25.52)	0.189
QoL Measure	Mean (SD), n = 9		Mean Difference (95% CI)	p-value
	Baseline*	12-month follow-up
Global health score	73.15 (13.03)	61.11 (21.24)	−12.04 (–33.31 to 9.23)	0.228
Physical functioning	86.67 (13.33)	71.11 (28.29)	−15.55 (–33.31 to 2.20)	0.078
Role functioning	79.63 (32.03)	62.96 (42.31)	−16.67 (−39.76 to 6.43)	0.135
Emotional functioning	89.82 (13.68)	86.11 (13.18)	−3.70 (−16.95 to 9.54)	0.537
Cognitive functioning	87.04 (16.20)	77.78 (20.41)	−9.26 (−30.61 to 12.09)	0.347
Social functioning	66.67 (32.27)	61.11 (39.09)	−5.55 (−36.27 to 25.17)	0.688
Fatigue	24.69 (15.49)	38.27 (28.39)	13.58 (−10.49 to 37.66)	0.229
Nausea and vomiting	1.85 (5.56)	9.26 (14.70)	7.41 (−1.90 to 16.71)	0.104
Pain	3.70 (7.35)	9.26 (14.70)	5.55 (−5.54 to 16.65)	0.282

CI, confidence interval; QoL, quality of life; SD, standard deviation

Outcomes in bold represent statistically significant differences (p < 0.05) and clinically meaningful differences as defined by Dirven et al.

*Baseline values differ across follow-up analyses as not all patients had baseline measurements available for inclusion at each follow-up time point. Consequently, the baseline means reflect only the subset of patients with paired data for the respective follow-up.

At the 24-month follow-up, paired data were available for only two patients which limits statistical analysis.

Table 4.

Changes in quality-of-life measures over time based on EORTC QLQ-BN20 scores (paired data).

QoL Measure	Mean (SD), n = 21		Mean Difference (95% CI)	p-value
	Baseline	3-month follow-up
Future uncertainty	41.67 (35.26)	32.14 (20.46)	−9.52 (–23.68 to 4.63)	0.176
Visual disorder	6.88 (14.26)	12.17 (20.76)	5.29 (−3.84 to 14.43)	0.241
Motor dysfunction	14.81 (22.31)	27.51 (25.00)	12.70 (−0.62 to 26.01)	0.061
Communication deficit	17.99 (19.71)	35.98 (37.83)	17.99 (2.44 to 33.54)	0.026
QoL Measure	Mean (SD), n = 15		Mean Difference (95% CI)	p-value
	Baseline	6-month follow-up
Future uncertainty	40.56 (33.17)	27.78 (22.86)	−12.78 (–22.47 to −3.09)	0.013
Visual disorder	2.96 (6.60)	19.26 (31.28)	16.30 (1.25 to 31.34)	0.036
Motor dysfunction	17.04 (24.80)	22.22 (27.22)	5.19 (−6.87 to 17.24)	0.372
Communication deficit	13.33 (15.26)	28.89 (34.83)	15.56 (−2.73 to 33.84)	0.089
QoL Measure	Mean (SD), n = 9		Mean Difference (95% CI)	p-value
	Baseline	12-month follow-up
Future uncertainty	33.33 (35.36)	29.17 (30.86)	−4.17 (−35.55 to 27.21)	0.763
Visual disorder	0	5.56 (10.29)	5.56 (−3.04 to 14.15)	0.171
Motor dysfunction	18.06 (22.95)	26.39 (29.66)	8.33 (−13.73 to 30.40)	0.402
Communication deficit	9.72 (13.85)	18.05 (17.75)	8.33 (−5.49 to 22.15)	0.197

CI, confidence interval; QoL, quality of life; SD, standard deviation

Outcomes in bold represent statistically significant differences (p < 0.05) and a minimally important difference of ≥ 10-point change.

*Baseline values differ across follow-up analyses as not all patients had baseline measurements available for inclusion at each follow-up time point. Consequently, the baseline means reflect only the subset of patients with paired data for the respective follow-up.

At the 24-month follow-up, paired data were available for only two patients which limits statistical analysis.

PROs for the EORTC QLQ-BN20 domain revealed worsening in communication deficits (MD 17.99 [95% CI 2.44 to 33.54]) at 3 months and visual disorders (MD 16.30 [1.25 to 31.34]) at 6 months. Conversely, the domain of future uncertainty showed improvement at 6 months (MD −12.78 [–22.47 to −3.09]). By the 12-month follow-up, these domains had also returned to baseline levels.

At the 24-month follow-up, paired data were available for only two patients. The small sample size precluded meaningful statistical analysis, and the results at this time point may not be representative of the broader population.

Oncological outcomes

The median follow-up time for the cohort was 12.16 months. Progression-free survival (PFS) rates were 73.7%% (95% CI 52.6%-86.5%) at 3 months, 69.8% (48.6%-83.7%) at 6 months and 47.0% (26.3%-65.3%) at both 12 24 months. Overall survival (OS) rates were 96.3% (76.5%-99.5%) at 3 months, 85.2% (65.2%-94.2%) at 6 months, 65.8% (44.55–80.6%) at 12 months, 27.8% (10.6%-48.2%) at 18 months and 13.9% at 24 months (2.6%-34.4%). The median PFS and OS were 11.23 (95% CI: 3.88 – not estimable) and 14.06 (95% CI: 9.00–17.94) and 9.5 months, respectively. Among patients with available data who exhibited radiologic progression, the distribution of progression patterns included 1 case of marginal progression, 10 cases of in-field progression, and 4 cases of out-of-field progression.

Discussion

This study demonstrates the feasibility and tolerability of MRgRT for GBM in an Australian public setting, with a median of 100% of patients completing treatment and no grade 3–4 toxicities observed. It is also the first to report QoL outcomes in GBM patients treated on the MRL. While declines in GHS, physical and cognitive functioning, and increased pain were observed at 3 months post-treatment, most domains returned to baseline by 6 months. The integration of frequent imaging into treatment workflows was key to detecting dynamic tumor changes that necessitated replanning. These findings support the role of MRgRT in managing GBM and offer insights into imaging frequency and adaptation strategies that may inform workflows even on conventional linear accelerators. While MRgRT has been explored in Australia for other tumor sites such as liver, prostate, bladder, and oligometastases [16], [17], [18], with some ongoing prospective studies further investigating its applications [19], [20], [21], this is the first published report focusing on its use for brain tumors, including GBM.

The potential of MRL for treatment adaptation was evident in this study, with 53.6% of patients requiring some form of adaptation, either through the ATS workflow or full replanning, based on changes detected on daily MRL imaging. Notably, 39.3% of all patients required adaptation within the first week of treatment, including 14% at fraction 0, with a median interval of 12.6 days (including weekends) between CT/MRI simulation and treatment start due to progression during planning time, suggesting shorter time between simulation and treatment start is indicated in this subgroup of patients. In the absence of symptoms, these subgroup of patients would have a marginal miss if treated on conventional linear accelerator, resulting in subsequent early marginal recurrence. In contrast, Tseng et al. reported a replanning rate of 30% mostly occurring between fractions 9 and 12 [11]. This supports existing evidence on the dynamic nature of GBM during treatment, including tumor shrinkage and migration, which has been extensively documented in prior studies. Stewart et al. observed significant tumor volume reductions over a 6-week chemoradiation course, with median GTV volumes decreasing from 18.4 cm³ at baseline to 14.7 cm³ at fraction 10, 13.7 cm³ at fraction 20, and 13 cm³ one month post-treatment. Tumor migration exceeded 5 mm in more than half of patients by fractions 10 and 20, underscoring the risk of geographic misses with static treatment margins [8]. Similarly, Cullison et al. reported lesion growth in 64% of patients and cavity shrinkage in 46% using a 0.35 T MRL. Migration distances reached up to 4.1 cm for intact lesions and 2.1 cm for cavities, with average migration distances of 1.3 cm and 0.6 cm, respectively [9]. Their findings suggest that shrinking cavities may benefit from smaller treatment margins to spare healthy brain tissue, while growing lesions may require margin expansions to maintain adequate coverage. Dooley et al. further highlighted the complexity of tumor dynamics in high-grade glioma patients treated with chemoradiotherapy on an MRL, reporting that 36% experienced tumor volume increases greater than 10% during treatment, particularly in weeks 5 and 6. These changes may reflect pseudoprogression, true progression, or edema-related effects [22]. Collectively, these studies illustrate the limitations of conventional radiotherapy techniques in addressing tumor changes during treatment. Although specific to MR-guided workflows, our findings may inform practice in centers without MRL systems. Nearly 40% of patients required adaptation within the first week and over half within two weeks, suggesting that repeat imaging during the first 1–2 weeks of chemoradiation, especially when simulation-to-treatment intervals are prolonged, could help reduce the risk of marginal miss. Although daily MR imaging provides the most detailed assessment of tumor and edema dynamics, the optimal imaging frequency for all patients remains unclear. In parallel with this study, we are conducting an ongoing analysis of tumor dynamics in this MRL-treated cohort to assess whether early on-treatment changes correlate with early outcomes and may identify patients who benefit from closer imaging surveillance. Such findings may support a risk-adapted imaging approach, reserving more frequent imaging for patients with early or clinically significant changes. MRgRT offers a significant advantage by enabling daily imaging to monitor tumor dynamics and optimize treatment delivery. Current clinical practice involves a 1.5–2 cm expansion from the GTV to create the CTV, often including T2/FLAIR hyperintense regions. In our workflow, T2/FLAIR signal was used as the primary surrogate for daily adaptation in GBM, as gadolinium contrast administration is not feasible on a daily basis and T1-contrast–enhanced sequences cannot be routinely employed. While practical, this approach has inherent limitations, since T2/FLAIR expansion may represent either true progression or treatment-related effects such as edema or pseudoprogression. However, the use of daily adaptive imaging with MRL systems has the potential to reduce treatment margins, minimizing unnecessary dose exposure to healthy brain tissue while maintaining tumor coverage. In addition to improving target accuracy, MRL facilitates the monitoring of cerebral edema, allowing for better correlation with clinical symptoms and evaluation of corticosteroid response. The ongoing UNITED study, which is investigating reduced CTV margins of 5 mm [23], along with evidence showing reduced doses to the brain and hippocampus through adaptive radiotherapy [24], highlights the potential of MRL systems to enhance oncologic outcomes and preserve neurocognitive function in GBM patients.

Oncologic outcomes in this cohort align with contemporary benchmarks for GBM. The PFS and OS rates in this study are comparable to those reported for patients in the EORTC-NCIC trial, which remains the cornerstone study establishing the standard of care for GBM [25], [26]. Our findings suggest that the integration of MRgRT does not compromise survival outcomes while offering potential advantages in treatment precision and toxicity management.

This study is the first to evaluate QoL outcomes in GBM patients treated with the MRL. A decline in GHS, physical functioning, cognitive functioning, and worsening pain was observed at 3 months post-treatment, with most domains returning to baseline by 6 months, except for persistent deterioration in physical functioning. Compared to Taphoorn et al., who evaluated QoL in a larger cohort of 573 patients treated with radiotherapy alone or in combination with temozolomide, our findings showed more pronounced early declines [27]. Taphoorn et al. reported minimal QoL variations and improvements in baseline GHS, whereas our study identified a temporary decline in GHS at 3 months with an increasing trend at 6 and 12 months. However, both studies similarly noted stable emotional functioning, improved future uncertainty, and worsening communication deficits. Minniti et al., in a study of elderly patients treated with short-course radiotherapy and temozolomide, reported consistent improvements or stability in most QoL domains, including GHS, social functioning, and cognitive functioning [28]. These findings of improved GHS and cognitive functioning also differ from our results. The temporary decline in GHS observed in our study may appear unexpected, given the MRL's potential to reduce treatment toxicity through precise radiation delivery. This discrepancy could be explained by the inclusion of patients with advanced disease or prior treatments, which likely influenced baseline QoL and treatment response. Furthermore, the small sample size limits the generalizability of these findings. Nevertheless, the increasing trend in GHS scores at 6 and 12 months highlights the potential long-term benefits of MRgRT, warranting further research in larger, more homogenous cohorts.

Treatment time is a key operational metric in MRgRT, reflecting the balance between personalized treatment and workflow efficiency. In our study, the median on-table treatment time was 28.2 min per fraction, with ATP being shorter (28.0 min) than ATS (41.0 min) due to added recontouring steps. These findings are consistent with prior reports across tumor sites. Tseng et al. noted 37.3 min for GBM patients treated with ATP [11], while Tetar et al. reported 45-minute median durations for prostate cancer and up to 64 min for pancreatic tumors using ATS [29], [30]. The shorter and more consistent times in our GBM cohort likely reflect the anatomical stability of intracranial targets, largely rigid image registration and lack of motion management, though rotational corrections during image fusion may still be required. Notably, median treatment times improved with experience, reflecting institutional learning and workflow optimization. This has clinical relevance for GBM patients, who may be vulnerable to fatigue and cognitive effects from prolonged sessions. Automation in contouring and plan adaptation may further reduce ATS durations while preserving the advantages of adaptation.

The integration of advanced MRI sequences also provides valuable insights into the physiology and metabolism of GBM, enhancing treatment planning and monitoring. Diffusion-weighted imaging (DWI) measures random movement of water molecules in tissue, reflecting alterations in tumor cellularity and necrosis, and has shown promise in predicting progression, recurrence, and survival outcomes through metrics such as apparent diffusion coefficient (ADC) values and diffusion changes [31], [32]. Perfusion imaging, which measures blood flow and volume within tumors, offers complementary information on vascularity and metabolism, with studies demonstrating significant correlations between perfusion parameters and survival [33]. Other sequences of potential value include chemical exchange saturation transfer, magnetic transferase, and spectroscopy sequences. Another potential advantage of MRgRT is its capacity for personalized dose escalation. In GBM, most recurrences are local and often occur within areas receiving the highest RT dose [34]. Unfortunately, prior studies on dose-escalated RT have not shown significant benefit, largely because they were conducted without temozolomide or with older nitrosourea-based chemotherapy regimens [35], [36], [37]. MRgRT offers a unique opportunity to implement individualized dose escalation by leveraging advanced MRI sequences to target areas with increased cellularity or metabolism. A phase 2 trial evaluated dose-intensified chemoradiation targeting hypercellular (TVHCV) and hyperperfused (TVCBV) tumor regions in GBM, achieving a 12-month OS rate of 92% and a median OS of 20 months in patients receiving combined TVHCV/TVCBV boosts. Tumor reduction at 3 months was associated with superior survival (29 vs. 12 months), with minimal impact on neurocognitive function, symptom burden, and QoL compared to standard therapy [38].

This study has several strengths, including its prospective design, rigorous follow-up, and comprehensive collection of toxicities, oncologic outcomes, and QoL data. The inclusion of only glioblastoma patients ensures a homogeneous study population. However, the findings should be interpreted cautiously due to certain limitations. The small sample size and inclusion of patients treated with varying fractionation regimens, including re-irradiation, may have influenced outcomes. Selection bias may also exist, given the criteria for selecting patients treated on the MRL. Additionally, incomplete QoL data due to lost to follow-up and mortality may introduce potential biases.

Another important limitation of this study is the lack of systematically collected data on corticosteroid use and tapering during treatment. Corticosteroids are well known to influence MRI appearances in GBM by reducing peritumoral edema, altering contrast enhancement, and affecting diffusion metrics [39], [40]. Prior studies have demonstrated that corticosteroid administration can result in substantial reductions in contrast-enhancing tumor volume and changes in ADC values, producing imaging changes that may not reflect true tumor response but rather steroid-related effects [41]. Such steroid-induced imaging alterations could potentially influence the interpretation of tumor dynamics on the MRL and, consequently, adaptive decision-making. In routine clinical practice at our institution, most patients treated with MRgRT for GBM are either not receiving corticosteroids at treatment initiation or are weaned off dexamethasone prior to or within the first week of RT, unless clinically indicated. However, because steroid dosing and tapering were not prospectively recorded, we were unable to evaluate whether corticosteroid exposure influenced MRI-observed changes, adaptation rates, or treatment workflows. Conversely, the frequent imaging provided by MRgRT may offer an opportunity to monitor tumor and edema dynamics in real time and potentially inform corticosteroid management, particularly when imaging changes are interpreted alongside clinical symptoms. Future prospective studies incorporating standardized documentation of corticosteroid use will be important to clarify the interaction between steroid therapy, MRI-observed changes, and adaptive radiotherapy strategies in GBM.

Despite these limitations, this study establishes the feasibility of MR-guided adaptive radiotherapy for GBM. Rather than demonstrating a survival advantage, the primary value of MRgRT in this setting lies in improving treatment accuracy and workflow in this dynamic disease. Our findings highlight the importance of minimizing delays between imaging, treatment planning, and delivery of treatment to reduce the risk of early geographic miss. Future research should focus on direct comparisons of outcomes between MR-guided and conventional computed tomography-guided techniques, as well as exploring the long-term benefits of adaptive radiotherapy in this challenging patient population.

Conclusion

Treatment of GBM with MRgRT is feasible, with manageable acute toxicities and adaptive capabilities to address tumor changes. QoL outcomes suggest overall tolerability, with most QoL domains recovering by six months. These findings support further investigation of MRgRT in larger, prospective studies.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.