Prospective longitudinal analysis of physiologic MRI-based tumor habitat predicts short-term patient outcomes in IDH-wildtype glioblastoma

Abstract

Background

This study validates MRI-based tumor habitats in predicting time-to-progression (TTP), overall survival (OS), and progression sites in isocitrate dehydrogenase (IDH)-wildtype glioblastoma patients.

Methods

Seventy-nine patients were prospectively enrolled between January 2020 and June 2022. MRI, including diffusion-weighted and dynamic susceptibility contrast imaging, were obtained immediately postoperation and at three serial timepoints. Voxels from cerebral blood volume and apparent diffusion coefficient maps were grouped into three habitats (hypervascular cellular, hypovascular cellular, and nonviable tissue) using k-means clustering. Predefined cutoffs for increases in hypervascular and hypovascular cellular habitat were applied to calculate the habitat risk score. Associations between spatiotemporal habitats, habitat risk score, TTP, and OS were investigated using Cox proportional hazards modeling. Habitat risk score was compared to tumor volume using time-dependent receiver operating characteristics analysis. Progression sites were matched with spatial habitats.

Results

Increases in hypervascular and hypovascular cellular habitats and habitat risk scores were associated with shorter TTP and OS (all P < .05). Hypovascular cellular habitat and habitat risk scores 1 and 2 independently predicted TTP (hazard ratio [HR], 4.14; P = .03, HR, 4.51; P = .001 and HR, 10.02; P < .001, respectively). Hypovascular cellular habitat and habitat risk score 2 independently predicted OS (HR, 4.01, P = .003; and HR, 3.27, P < .001, respectively). Habitat risk score outperformed tumor volume in predicting TTP (12-month AUC, 0.762 vs. 0.646, P = .048). Hypovascular cellular habitat predicted progression sites (mean Dice index: 0.31).

Conclusions

Multiparametric physiologic MRI-based spatiotemporal tumor habitats and habitat risk scores are useful biomarkers for early tumor progression and outcomes in IDH-wildtype glioblastoma patients.

Graphical Abstract

Graphical Abstract.

Key Points.

1. Hypovascular cellular habitat and habitat risk score were predictors of time-to-progression and overall survival.

2. Habitat risk score outperformed tumor volume in predicting time-to-progression.

3. An increase in hypovascular cellular habitat predicted tumor progression sites.

Importance of the Study.

Spatial heterogeneity in glioblastoma indicates a poor prognosis and contributes to treatment resistance. This study prospectively validates the clinical utility of physiologic MRI-based tumor habitat analysis in predicting time-to-progression (TTP), overall survival (OS), and progression sites in isocitrate dehydrogenase (IDH)-wildtype glioblastoma patients post concurrent chemoradiotherapy (CCRT). An increase in the hypovascular cellular habitat, characterized by both low apparent diffusion coefficient and cerebral blood volume values, showed the most significant association with shorter TTP and OS, and predicted the site of progression. Notably, the habitat risk score outperformed traditional tumor volume assessments in predicting short-term outcomes, offering a valuable tool for patient risk stratification. Our results underscore the potential of multiparametric physiologic MRI-derived spatiotemporal tumor habitat and habitat risk score as useful biomarkers for early tumor progression and clinical outcomes, assisting in patient risk assessment and treatment decision-making in glioblastoma management.

Isocitrate dehydrogenase-wildtype glioblastoma of WHO grade 4 is known for its intratumoral heterogeneity, displaying complex spatial variations in gene expression, histopathology, and macroscopic structure.¹ This heterogeneity contributes to a poor prognosis with diverse treatment responses and the development of resistance in different tumor regions.² Early and accurate prediction of tumor progression can lead to the timely adoption of alternative treatments, such as reoperation or bevacizumab, potentially offering a survival benefit.^3–6 However, the frequent coexistence of tumor recurrence and radiation injury in posttreatment glioblastoma presents a challenge to the prediction of progression,⁷ along with the tumor’s inherent heterogeneity. Physiological MRI techniques, including cerebral blood volume (CBV) and apparent diffusion coefficient (ADC) mapping, enable the identification of distinct tumor regions with variations in metabolism, vascularity, and cellularity.^8,9 ADC provides information on cell density and necrosis,¹⁰ while CBV indicates vessel density.^11,12 Although various imaging techniques, including histograms, texture analysis, and radiomics analysis of normalized CBV (nCBV) or ADC^10,13 values, have been employed to quantify intratumoral heterogeneity on the basis of imaging parameters, these methods often overlook spatial information, restricting tissue characterization in posttreatment glioblastoma.

To address this limitation, the concept of tumor habitat analysis has emerged as a promising strategy,^14,15 with this involving the grouping of similar voxels and the delineation of distinct subregions within a heterogeneous tumor through the identification of voxels that share common tumor biology.¹⁶ Through the application of voxel-wise clustering, this approach parcellates distinct tumor habitats,¹⁶ which may reflect the spatial habitats of posttreatment glioblastoma following concurrent chemoradiation therapy (CCRT), offering valuable clinical insights into subregions associated with tumor progression and treatment resistance, and identifying potential therapeutic targets.¹⁷ Previous research used tumor habitat analysis to identify three distinct spatial habitats within posttreatment glioblastoma: hypervascular cellular, hypovascular cellular, and nonviable tissue habitats.¹⁸ This study defined a habitat risk score based on discrete increases in the hypervascular and hypovascular cellular habitats and demonstrated that this score could be used to stratify patients into low-, intermediate-, and high-risk groups. However, the actual clinical efficacy of physiological MRI-based spatiotemporal tumor habitats and habitat risk score for predicting early tumor progression and patient outcomes has not been validated in a prospective study.

In this study, we aimed to prospectively validate physiologic MRI-based tumor habitat analysis for prediction of time-to-progression (TTP) and overall survival (OS), as well as for identifying the site of progression after CCRT in patients with IDH-wildtype glioblastoma.

Methods

This prospective single-center study was registered at ClinicalTrials.gov (ClinicalTrials.gov identifier: NCT02613988) and received approval from the Institutional Review Board of BLINDED (local approval number: 2019-1259). The study was conducted in compliance with the U.S. Health Insurance Portability and Accountability Act regulations and the Declaration of Helsinki. Written and signed informed consent was obtained from each participant prior to enrollment.

Study Population

The inclusion process for the study patients is shown in Figure 1. Consecutive patients who received a histopathologic diagnosis of IDH-wildtype glioblastoma at our institution between January 2020 and June 2022 and gave written informed consent were considered eligible. The patients were enrolled according to the following inclusion criteria: (i) age ≥18 years; (ii) histopathologic diagnosis of IDH-wildtype glioblastoma according to the World Health Organization classification 2021¹⁹; (iii) underwent the current standard treatment of maximum safe surgical resection followed by CCRT and adjuvant temozolomide according to the Stupp protocol²⁰; (iv) underwent structural and physiologic MRI, including diffusion-weighted imaging (DWI) and dynamic susceptibility contrast (DSC) imaging; and (v) presence of a measurable contrast-enhancing lesion (CEL) of more than 1 × 1 cm on the first post-CCRT (#1) examination. The exclusion criteria were as follows: (i) absence of a measurable CEL on the first post-CCRT (#1) examination; (ii) failure to obtain at least three successive imaging examinations or pathologic confirmation to monitor treatment response; and (iii) an examination with inadequate image quality for quantitative analysis. This study aimed to predict early tumor progression in patients with newly diagnosed IDH-wildtype glioblastoma, not in recurrent tumors; thus, MRI images from patients on bevacizumab were not included in this study.

Figure 1.

Flow diagram of the patient inclusion process. CCRT = concurrent chemoradiotherapy; IDH = isocitrate dehydrogenase.

Reference Standard for Final Diagnosis and Endpoints

Progression was pathologically confirmed through second-look operations if clinically indicated. In cases where second-look operations were not performed, consecutive clinico-radiological diagnoses were established by consensus between two experts: BLINDED, with 26 years of experience in neuro-oncology practice, and a neuroradiologist, BLINDED, with 21 years of experience in neuro-oncologic imaging. These diagnoses were based on the modified RANO criteria.²¹ Both the modified RANO criteria²¹ and the recently published RANO 2.0 criteria²² share the following: (i) the first postradiotherapy MRI, rather than the postsurgical MRI, is used as baseline imaging; and (ii) repeat MRI is mandatory to confirm progression within 12 weeks after radiotherapy. Progression was defined according to (i) an additional increase ≥25% in the area on current imaging compared with baseline imaging, and (ii) any new measurable enhancing lesion (≥10 × 10 mm) outside the RT field. Pseudoprogression was defined as the presence of CEL without a need for treatment change at least 6 months after the end of CCRT. This diagnosis allows for a mild increase in CEL, provided there is no change in treatment during this period.²³ Cases of pseudoprogression were classified as “non-progression.” For these cases, the time of first progression was recorded as the date of true progression. DWI and DSC images are included in our routine brain tumor MRI protocol and are automatically uploaded to the Picture Archiving and Communication System for clinical readers to access during their clinic-radiological assessment. During the assessment for the reference standard, tumor habitat analysis was not provided.

The primary endpoint of the study was TTP, calculated from the day of each MRI examination (post-CCRT #1, #2, or #3) to the day of first documented progression. The secondary endpoint was OS, calculated from the day of initial habitat analysis that was used as an imaging biomarker in this study (post-CCRT #1) until the day of death, as obtained from the national health care data linked to our hospital. Patients who were alive at the time of analysis were included in the analysis with right-censored data.

MRI Acquisition, Mask Segmentation, and Image Processing

Details of the MRI acquisition, mask segmentation, and advanced image processing are summarized in Supplementary Materials. To analyze changes across consecutive scans, the three-dimensional contrast-enhanced T1-weighted imaging (3D CE T1WI) obtained from each patient was co-registered and resampled to have isometric-voxel sizes. Subsequently, FLAIR, nCBV, and ADC images were co-registered and resampled to the iso-voxel CE T1WI images using rigid transformations with six degrees of freedom in the SPM package (version 12, www.fil.ion.ucl.ac.uk/spm/). The final voxel classifications based on nCBV and ADC values were implemented using a k-means clustering module in the scikit-learn python package.

Multiparametric Physiologic MRI-Based Spatiotemporal Habitat Analysis

Population-level clustering based on previous research.— Using two distinct feature maps, three clusters were established: cluster 1 represented “hypervascular cellular tumor” with high CBV values and low ADC values, cluster 2 represented “hypovascular cellular tumor” with low CBV values and low ADC values, and cluster 3 represented “nonviable tissue” with low CBV values and high ADC values. The ranges for the boundaries of the pretrained and retrospectively validated spatial physiologic habitats were previously reported as 4.37–4.44 for nCBV and 150–187 (× 10^-6 mm²/s) for ADC in a study on 97 patients.¹⁸ The co-registered imaging data and tumor habitat analysis are available at: https://github.com/enaech/Prospective.

Calculation of spatiotemporal habitats and habitat risk score.— Four consecutive MRI examinations were used in the analysis: immediate postoperation (examination 1, postop), the first visit after CCRT (examination 2, post-CCRT #1), the second visit after CCRT (examination 3, post-CCRT #2), and the third visit after CCRT (examination 4, post-CCRT #3). There was a 3-month interval between each examination. First, changes in the number of voxels within the entire CEL and within each habitat were calculated between sequential examinations. Second, the habitat risk score, determined by discrete increases in hypervascular and hypovascular cellular habitats between sequential examinations, was calculated. This habitat risk score was calculated by adding 1 point when a temporal change in each habitat exceeded a predefined cutoff. The cutoffs for the discrete scores were defined as an increase of ≥1 voxel for hypervascular cellular habitat and an increase of more than 130 voxels for hypovascular cellular habitat, as established in a previous study.¹⁸ For example, when a patient showed an increase in both hypervascular cellular habitat (≥1 voxel) and hypovascular cellular habitat (>130 voxels), then the habitat risk score was 2. When a patient showed an increase in either hypervascular cellular habitat (≥1 voxel) or hypovascular cellular habitat (>130 voxels), the habitat risk score was 1. If a patient did not show an increase in either hypervascular cellular habitat (≥1 voxel) or hypovascular cellular habitat (>130 voxels), the habitat risk score was 0.

Analysis of site of progression.— The site of progression was analyzed on the follow-up examination at the time of progression. The volume of CEL at the time of progression was matched with the habitats identified in post-CCRT#1 or post-CCRT #2 examination. The overlap between each spatiotemporal habitat and the CEL volume at the time of progression was quantified using the Dice similarity coefficient, Dice coefficient = 2|P ∩ R|/(|P|+|R|),²⁴ where P represents each spatiotemporal habitat and R represents the CEL volume at the time of progression. The Dice coefficient ranges between 0 (no overlap) and 1 (perfect agreement). When tumor progression was diagnosed by surgical excision without a follow-up MRI examination, the site of tumor progression could not be determined.

The overall process of the tumor habitat analysis is shown in Supplementary Figure 1.

Statistical Analysis

Statistical analysis was performed by an expert biostatistician (BLINDED, with 15 years of experience in biostatistics). All statistical analyses were conducted using the R Statistical Package (version 3.6.3, Institute for Statistics and Mathematics, http://www.R-project.org). A P-value < .05 was considered statistically significant.

Sample size.

—The methodology described here adheres to the REMARK (Reporting Recommendations for Tumor Marker Prognostic Studies) recommendations.²⁵ The sample size was calculated according to a Cox proportional hazards regression model with nonbinary covariates.²⁶ Death rates of 0.5–1.0 represent 51 deaths, with a sample size between 51 and 102 patients, yielding an expected power of 80% and an alpha error of 5%.

Use of spatiotemporal habitat and habitat risk score to predict TTP and OS.

—Baseline characteristics including sex, age, Karnofsky performance score (KPS; binary, score >70 or ≤70), initial tumor volume, O6-methylguanine-DNA-methyltransferase (MGMT) promoter status, and extent of surgery were analyzed using descriptive statistics.

To analyze the associations of spatiotemporal habitats and habitat risk score with TTP and OS, univariable and multivariable analyses were conducted using a Cox proportional hazards regression model. The spatiotemporal habitat and habitat risk scores between two consecutive MRI examinations were considered as time-dependent covariates for Cox proportional hazards regression analysis because these predictor variables varied over time (post-CCRT #1, #2, and #3).²⁷ For the calculation of TTP and OS, time zero was set as post-CCRT #1. Hazard ratios (HR) for spatiotemporal habitat indicate the relative change in hazard incurred by a 1-unit increase in each parameter, with 20,000 voxels defining a single unit in a previous study.¹⁸

Diagnostic performance for predicting pseudoprogression.

—We performed additional analysis for patients who experienced a newly appeared or enlarging (>25%) measurable contrast-enhancing mass greater than 1 × 1 cm at post-CCRT #1, raising clinical suspicion of tumor progression or treatment-related change difference (pseudoprogression), and who had available post-CCRT #1 and #2 imaging. The habitat risk score between post-CCRT #1 and #2 was used to calculate diagnostic performance for differentiating tumor progression from pseudoprogression using receiver operating characteristics (ROC) curve analysis.

Risk stratification using habitat risk score.

—Risk stratification into high-, intermediate-, and low-risk categories for TTP and OS on the basis of the habitat risk scores between two consecutive MRI examinations was analyzed using the Kaplan–Meier method (log-rank test). Time-dependent ROC curve analyses were performed to compare the value of habitat risk score and change in CEL volume for predicting TTP and OS at 12, 18, and 24 months.

Results

Patient Characteristics

Of the 247 patients who received a histopathologic diagnosis of IDH-wildtype glioblastoma between January 2020 and June 2022 at our institution, 98 patients refused to participate. Of the remaining 149 potentially eligible participants, 40 without a measurable CEL on post-CCRT #1 examination, 27 with insufficient follow-up, and three with inadequate image quality were excluded. Finally, 79 participants (40 [50.6%] women) with a mean age of 59.4 years ± 11.7 (SD) (range, 28–78 years) were included in the analysis. The baseline clinical characteristics of the 79 patients are summarized in Table 1. The final diagnosis was based on pathologic confirmation for 18 (22.8%) patients and clinico-radiologic follow-up for 61 (77.2%) patients. The clinical characteristics of the excluded study patients are shown in Supplementary Table 1. There were no statistically significant differences in clinical characteristics between the included and excluded patients.

Table 1.

Clinical Characteristics of the Study Patients

Clinical characteristics (n = 79)
Sex, n, male/female	39/40
Age, years, mean ± SD	59.4 ± 11.7
KPS at baseline, n (%)
>70	64 (81.0%)
≤70	15 (19.0%)
MGMT promotor status, n (%)
Methylated	33 (41.8%)
Unmethylated	35 (44.3%)
NA	11 (13.9%)
Extent of resection, n (%)
Gross total resection	53 (67.1%)
Subtotal resection	16 (20.3%)
Biopsy	10 (12.7%)
Secondary treatment after recurrence
Surgery	18 (22.8%)
Bevacizumab	30 (38.0%)
Temozolomide	2 (2.5%)
Others (immunotherapy or radiosurgery)	3 (3.8%)

Abbreviations: KPS = Karnofsky performance score; MGMT = O6-methylguanine DNA methyltransferase gene methylation status; NA = information not available.

Spatiotemporal Habitats Associated with TTP and OS

The results of the univariable and multivariable analyses to evaluate the associations between changes in the three spatial habitats and TTP and OS are summarized in Table 2. In univariable analysis, increases in both hypervascular cellular habitat and hypovascular cellular habitat were significantly associated with a shorter TTP (HR, 250.93; 95% CI, 11.13–5658.40, P = .001; and HR, 3.23 [1.85–5.64], P < .001, respectively) and a shorter OS (HR, 9.32 [1.48–58.64], P = .02; and 3.34 [2.03–5.49], P < .001, respectively). An increase in nonviable tissue was not associated with shorter TTP or OS (Figures 2 and 3).

Table 2.

Univariable and Multivariable Analysis of Spatial Habitats and Habitat Risk Score in Prediction of Time-to-Progression and Overall Survival in Patients With Posttreatment IDH-Wildtype Glioblastoma

	Time-to-progression						Overall survival
	Univariable		Multivariable 1		Multivariable 2		Univariable		Multivariable 1		Multivariable 2
	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P
Age	1.02 (0.99–1.04)	.26	1.02 (0.99–1.05)	.11	1.01 (0.98–1.04)	.57	1.03 (1.00–1.05)	.02	1.03 (1.01–1.06)	.01	1.04 (1.01–1.06)	.004
Sex	1.05 (0.59–1.88)	.86	1.23 (0.66–2.31)	.52	0.77 (0.39–1.41)	.44	1.09 (0.67–1.78)	.72	1.50 (0.89–2.5)	.13	1.48 (0.86–2.54)	.16
Extent of resection		.94		.38		.89		.62		.78		.49
Gross total resection	Ref		Ref		Ref		Ref		Ref		Ref
Subtotal resection	1.02 (0.48–2.17)	.95	0.92 (0.40–2.10)	.84	1.22 (0.54–2.80)	.63	0.75 (0.41–1.38)	.36	0.82 (0.43–1.54)	.53	0.73 (0.39–1.40)	.35
Biopsy	1.18 (0.49–2.85)	.72	2.22 (0.71–6.91)	.17	1.06 (0.37–3.04)	.92	1.03 (0.48–2.18)	.95	0.81 (0.30–2.17)	.67	0.69 (0.28–1.71)	.43
KPS at baseline		.29		.43		.56		.10		.42		.07
>70	Ref		Ref		Ref		Ref		Ref		Ref
≤70	1.48 (0.71–3.08)		1.39 (0.61–3.18)		1.30 (0.54–3.10)	.56	2.07 (1.14–3.78)		1.88 (0.85–3.71)		1.35 (0.65–2.78)
MGMT		.90		.76		.56		.10		.10		.14
Methylated	Ref		Ref		Ref		Ref		Ref		Ref
Unmethylated	1.08 (0.58–2.03)	.80	1.11 (0.54–2.27)	.77	1.37 (0.67–2.83)	.39	2.32 (1.09–4.94)	.03	2.65 (1.06–6.55)	.04	2.47 (1.06–5.77)	.04
NA	1.22 (0.51–2.91)	.65	0.75 (0.25–2.23)	.61	0.88 (0.29–2.71)	.82	1.37 (0.81–2.33)	.24	1.23 (0.67–2.26)	.51	1.58 (0.88–2.85)	.13
Increased in spatial habitata,b
Hypervascular cellular habitat	250.93 (11.13–5658.40)	.001	6.76 (0.03–1440.72)	.49			9.32 (1.48–58.64)	.02	0.46 (0.01–16.80)	.61
Hypovascular cellular habitat	3.23 (1.85–5.64)	<.001	4.14 (1.17–14.57)	.03			3.34 (2.03–5.49)	<.001	4.01 (1.59–10.24)	.003
Nonviable tissue	2.60 (0.74–9.00)	.135	0.33 (0.03–3.81)	.38			2.81 (0.86–9.14)	.09	0.79 (0.09–6.77)	.83
Habitat risk scorea		<.001				<.001		<.001				.001
0	Ref				Ref		Ref				Ref
1	4.02 (1.73–9.37)	.001			4.51 (1.88–10.83)	.001	1.56 (0.82–2.95)	.17			1.72 (0.87–3.42)	.12
2	9.23 (4.33–19.69)	<.001			10.02 (4.45–22.59)	<.001	3.43 (1.93–6.08)	<.001			3.27 (1.80–5.93)	< 0.001

Abbreviations: IDH = isocitrate dehydrogenase; HR = hazard ratio; CI = confidence interval. The bold values indicate P values less than .05.

^aIncreases in spatial habitat and habitat risk score between two consecutive MRI examinations were considered as time-dependent covariates.

^bHR reported here indicates the relative change in hazard that a 1-unit (20 000 voxels) increase in each imaging parameter incurs.

Figure 2.

Representative case of tumor habitat analysis from a 64-year-old male patient. The hypervascular cellular habitat (cluster 1) shows high nCBV and low ADC, the hypovascular cellular habitat (cluster 2) shows low nCBV and low ADC, and the nonviable tissue habitat (cluster 3) shows low nCBV and high ADC. (Left) Spatial mapping shows an increase in both hypervascular and hypovascular cellular habitat in post-CCRT examinations. The confirmatory scan after 4 weeks indicates tumor progression. (Right) Spatial habitats are defined by clustered voxels. Between postop and post-CCRT #1 examinations, the hypervascular cellular habitat increased by 104 voxels and the hypovascular cellular habitat increased by 166 voxels, resulting in a habitat risk score of 2 points. Between post-CCRT #1 and #2 examinations, the hypervascular cellular habitat increased by 332 voxels and the hypovascular cellular habitat increased by 1256 voxels, resulting in a habitat risk score of 2 points. nCBV = normalized cerebral blood volume; ADC = apparent diffusion coefficient; CCRT = concurrent chemoradiotherapy.

Figure 3.

Representative case of tumor habitat analysis from a 37-year-old male patient. The hypervascular cellular habitat (cluster 1) shows high nCBV and low ADC, the hypovascular cellular habitat (cluster 2) shows low nCBV and low ADC, and the nonviable tissue habitat (cluster 3) shows low nCBV and high ADC. (Left) Spatial mapping in the post-CCRT #1 examination shows a newly developed measurable CEL, but the majority of the lesion is considered nonviable tissue, showing low nCBV and high ADC values. Subsequent follow-up examinations demonstrate a gradual regression of the enhancing lesion, indicating pseudoprogression. (Right) Spatial habitats defined by clustered voxels demonstrate a predominant increase in nonviable tissue between postop and post-CCRT #1 examinations. CCRT = concurrent chemoradiotherapy; nCBV = normalized cerebral blood volume; ADC = apparent diffusion coefficient.

In a multivariable analysis considering spatiotemporal habitats, age, sex, extent of resection, KPS scores at baseline, and MGMT methylation status, only hypovascular cellular habitat was identified as an independent predictor of TTP (HR, 4.14 [1.17–14.57], P = .03). For OS, unmethylated MGMT status (HR, 2.65 [1.06–6.55], P = .04) and hypovascular cellular habitat (HR, 4.01 [1.59–10.24], P = .003) were independent predictors of shorter OS.

Risk Stratification Based on the Habitat Risk Score for TTP and OS

The results of the univariable and multivariable analysis to evaluate the association between habitat risk score and TTP are summarized in Table 2. For TTP, both habitat risk scores 1 and 2 were significantly associated with a shorter TTP in the univariable analysis (HR, 4.02 [1.73–9.37], P = .001; and HR, 9.23 [4.33–19.69], P < .001, respectively). In a multivariable analysis considering habitat risk score, age, sex, extent of resection, KPS scores at baseline, and MGMT methylation status, habitat risk scores 1 and 2 were both identified as independent predictors of TTP (HR, 4.51; 95% CI, 1.88–10.83; P = .001; and HR, 10.02; 95% CI, 4.45–22.59; P < .001, respectively). For OS in the multivariable analysis, both unmethylated MGMT (HR, 2.47 [1.06–5.77], P = .04) and habitat risk score 2 (HR, 3.27 [1.80–5.93], P < .001) were identified as independent predictors of shorter OS.

Kaplan–Meier survival curves between postop and post-CCRT #1, post-CCRT #1 and #2, and post-CCRT #2 and #3 examinations for habitat risk scores for TTP and OS are shown in Supplementary Figure 2. The habitat risk scores between postop and post-CCRT #1, between post-CCRT #1 and #2, and between post-CCRT #2 and 3 examinations stratified patients into low-, intermediate-, and high-risk groups for TTP (log-rank test; P = .001, P < .001, and P < .001, respectively).

Prediction Performance of Habitat Risk Score Compared with CEL Volume

The time-dependent ROC curves of habitat risk score and change in CEL volume for predicting TTP and OS at 12, 18, and 24 months are presented in Figure 4. For predicting TTP at 12 months, the habitat risk score achieved a significantly higher AUC than a change in CEL volume (0.762 vs. 0.646, P = .048). Habitat risk score also demonstrated a higher AUC than change in CEL volume for predicting TTP at 18 months, but the difference did not quite reach statistical significance (0.787 vs. 0.642, P = .054).

Figure 4.

Time-dependent ROC curves comparing habitat risk score (red) and change in CEL volume (blue) for predicting TTP at (A) 12 months and (B) 18 months, and OS at (C) 18 months and (D) 24 months. At 12-month follow-up, the habitat risk score exhibited a significantly higher AUC for TTP than the change in CEL volume. ROC = receiver operating characteristics; TTP = time-to-progression; OS = overall survival; AUC = area under the curve; CEL = contrast-enhancing lesion.

For predicting OS, the habitat score demonstrated a higher AUC than change in CEL volume at 12 months (0.717 vs. 0.628, P = .111), 18 months (0.698 vs. 0.639, P = .217), and 24 months (0.701 vs. 0.682, P = .697), but did not reach statistical significance.

Additional Analysis for Predicting Early Progression from Pseudoprogression

In our cohort, 49 patients experienced a newly appeared or enlarging (>25%) measurable contrast-enhancing mass greater than 1 × 1 cm at post-CCRT #1. Among them, 14 patients were diagnosed with pseudoprogression. For the differentiation of local progression from pseudoprogression, the tumor habitat score had an AUC of 0.85 (95% CI, 0.72–0.94). The sensitivity, specificity, and accuracy were 64.3% (9/14), 88.6% (31/35), and 81.6% (40/49), respectively.

Spatiotemporal Habitats and Site of Progression

Of the 49 patients who demonstrated progression, 25 (51%) had a Dice index calculated for the site of progression. The mean Dice index was 0.02 (range, 0–0.15; SD 0.03) for hypervascular cellular habitat, 0.31 (range, 0.06–0.66; SD 0.16) for hypovascular cellular habitat, and 0.20 for nonviable tissue (range, 0–0.64; SD, 0.15). The Dice index was highest for hypovascular cellular habitat in 17 of the 25 cases (68%). Supplementary Figure 3 demonstrates representative cases illustrating the relationship between temporal changes in hypovascular cellular habitat and the site of progression.

Discussion

In this prospective study, we validated the ability of multiparametric physiologic MRI-based tumor habitat analysis to predict time-to-progression, overall survival, and the site of progression in patients with isocitrate dehydrogenase-wildtype glioblastoma who demonstrated measurable contrast-enhancing lesions after undergoing concurrent chemoradiotherapy. An increase in the hypovascular cellular habitat, represented by both low ADC and CBV values, showed the most significant association with time-to-progression and overall survival, and predicted the site of progression. The habitat risk score successfully stratified patients into low-, intermediate-, and high-risk groups in respect to early tumor progression. The habitat risk score showed benefit over tumor volume for predicting short-term outcomes at 12 months. Our study prospectively validated spatiotemporal habitats defined by both ADC and CBV values, enabling the prediction of patient outcomes by quantifying the initial pathophysiologic changes in posttreatment glioblastoma.

Our results are in accord with a previous retrospective study conducted on a sample of 97 patients with IDH-wildtype glioblastoma.¹⁸ Although the hypervascular cellular habitat characterized by high CBV and low ADC values represents the most malignant region of a tumor, only an increase in the hypovascular cellular habitat with both low ADC and CBV values showed a significant association with TTP and OS in the multivariable analysis. The hypovascular cellular habitat identified in our study may indicate a hypoxic microenvironment resistant to treatment, featuring changes in gene and molecular expression and evolution towards increased malignancy and a more aggressive phenotype.²⁸ Moreover, low CBV values may represent a decrease in the patent vessels delivering the chemotherapeutic agent to the tumor bed, consequently indicating resistance to treatment.

Analysis of spatiotemporal changes in habitats is more effective than relying on a singular snapshot when predicting TTP in posttreatment glioblastomas, especially given the inherent variability in the histopathology of these tumors.²⁹ The habitat risk score, based on discrete increases in hypervascular and hypovascular cellular habitats, successfully stratified patient risk and was identified as an independent predictor of TTP. We employed previously defined cutoffs identified in a retrospective study¹⁸ to calculate the habitat risk score, and prospectively validated both the cutoff of the habitat risk score and its predictive value for patient outcomes. Utilizing the habitat risk score, changes in spatiotemporal habitats can be easily determined, potentially aiding in the stratification of patients with glioblastoma according to their risk of early progression. Furthermore, through time-dependent ROC curve analysis, we confirmed that the habitat risk score exhibited higher predictive power for progression at 12 months than change in tumor volume. This suggests that the habitat risk score may be particularly useful for predicting early tumor progression.

Although the RANO criteria based on tumor size changes serve as primary endpoints in glioblastoma, early identification of tumor progression through tumor habitat analysis may enable swift adjustments to secondary treatments such as surgery, bevacizumab administration, or participation in clinical trials, thereby potentially enhancing patient survival rates.^3,6 This emphasis on early detection of tumor progression is particularly aligned with recent discussions within the RANO resect group,³⁰ which advocate for supramaximal resection and nonenhancing tumor resection, further emphasizing the significance of early intervention.

The spatiotemporal habitat is valuable not only for predicting patient outcomes but also for identifying tumor subregions associated with tumor progression and treatment resistance. Through quantitative analysis of the Dice index, we observed a correlation between tumor progression foci and regions that exhibited a short-term increase in the hypovascular cellular habitat. In the context of recurrent glioblastoma, re-resection is limited,³¹ and the presence of spatial heterogeneity in posttreatment glioblastoma poses challenges for obtaining adequate lesion samples for effective histological analysis. Even with sufficient tissue sampling, there are currently no explicit standards for histologic diagnosis of pseudoprogression, residual glioma, and recurrent glioma.³² In this study, we demonstrated the use of hypovascular cellular habitat to noninvasively pinpoint the site of progression. Furthermore, it can be anticipated that utilizing radiographic guidance based on this method will enhance the effectiveness of tissue sampling in cases of recurrent glioblastoma in the future.

The biological hypothesis for hypovascular cellular habitat is as follows. Hypoxia is a hallmark of glioblastoma, creating an essential environmental cue for the maintenance of glioma stem cell-like cells (GSCs), the cell population believed to be responsible for tumor resistance to chemotherapy and radiation.^33,34 The glioblastoma microenvironment activates a variety of cellular programs in GSCs, which in turn remodel the microenvironment architecture.³⁵ Three major microenvironments have been described in glioblastoma—the hypoxic-necrotic core, the perivascular niche, and the invasive edge.³⁶ Based on our results, hypovascular tumor habitats indicate the high-cellular portion of the tumor microenvironment, closely located near to the hypoxic-necrotic core, which interacts with glioma GSCs and promotes heterogeneity and treatment resistance.

This study has several limitations. First, there was a lack of precise pathological correlations with image-based tumor habitat. Spatial mapping of MRI tumor habitat analysis ultimately requires biologic validation with topological mapping of pathology. Recently, this issue was addressed using the Ivy Glioblastoma Atlas (http://glioblastoma.alleninstitute.org/) public data, which has a comprehensive pathology-molecular map of glioblastoma with en bloc resection, as well as DWI and DSC perfusion imaging.³⁷ In an analysis based on imaging-pathologic co-registered cubes, hypovascular cellular habitat in CEL showed significant correlation with cellular tumor on pathology (r = 0.238, P = .005). This topological biologic validation further supports the current study findings that hypovascular cellular habitat is associated with tumor aggressiveness and shorter TTP. Second, this study was conducted at a single center. Therefore, it is essential to apply different protocols from other institutions and include a relatively large number of patients to provide broader validation.

In conclusion, multiparametric physiologic MRI-based spatiotemporal tumor habitats and the habitat risk score derived from postconcurrent chemoradiotherapy imaging were validated as useful biomarkers for early tumor progression and clinical outcomes in patients with isocitrate dehydrogenase-wildtype glioblastoma. An increase in the hypovascular cellular habitat can be considered a robust imaging biomarker for identifying the site of progression.

Supplementary material

Supplementary material is available online at Neuro-Oncology (https://academic.oup.com/neuro-oncology).

Funding

This research was supported by a grant from the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIP) (grant number: RS-2023-00208227 and RS-2023-00305153) and a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: HI22C1723).

Conflict of interest statement

One of the authors of this manuscript (NakYoung Kim) is an employee of DYNAPEX LLC. The remaining authors declare no financial or nonfinancial competing interests.

Author contributions

H.H.M. contributed to data analysis and writing the manuscript; J.E.P. contributed to data analysis, conceptual design, and writing the manuscript; N.K. contributed to image analysis; S.Y.P. contributed to statistical analysis; Y.K., S.W.S., S.K.H., J.H.K. provided patient samples; H.S.K. contributed to conceptual design and project integrity. All authors read and approved the final manuscript.

Data availability

Data generated or analyzed during the study are available at https://github.com/enaech/Prospective.