Volumetric endpoints in diffuse intrinsic pontine glioma: comparison to cross-sectional measures and outcome correlations in the International DIPG/DMG Registry

Abstract

Background

Cross-sectional tumor measures are traditional clinical trial endpoints; however volumetric measures may better assess tumor growth. We determined the correlation and compared the prognostic impact of cross-sectional and volumetric measures of progressive disease (PD) among patients with DIPG.

Methods

Imaging and clinical data were abstracted from the International DIPG Registry. Tumor volume and cross-sectional product (CP) were measured with mint Lesion™ software using manual contouring. Correlation between CP and volume (segmented and mathematical [ellipsoid] model) thresholds of PD were assessed by linear regression. Landmark analyses determined differences in survival (via log-rank) between patients classified as PD versus non-PD by CP and volumetric measurements at 1, 3, 5, 7, and 9 months postradiotherapy (RT). Hazard ratios (HR) for survival after these time points were calculated by Cox regression.

Results

A total of 312 MRIs (46 patients) were analyzed. Comparing change from the previous smallest measure, CP increase of 25% (PD) correlated with a segmented volume increase of 30% (R² = 0.710), rather than 40% (spherical model extrapolation). CP-determined PD predicted survival at 1 month post-RT (HR = 2.77), but not other time points. Segmented volumetric-determined PD (40% threshold) predicted survival at all imaging timepoints (HRs = 2.57, 2.62, 3.35, 2.71, 16.29), and 30% volumetric PD threshold predicted survival at 1, 3, 5, and 9 month timepoints (HRs = 2.57, 2.62, 4.65, 5.54). Compared to ellipsoid volume, segmented volume demonstrated superior survival associations.

Conclusions

Segmented volumetric assessments of PD correlated better with survival than CP or ellipsoid volume at most time points. Semiautomated tumor volume likely represents a more accurate, prognostically-relevant measure of disease burden in DIPG.

The prognosis for patients with diffuse intrinsic pontine glioma (DIPG), the most common brainstem tumor in childhood, remains dismal.^1,2 With limited treatment options (aside from radiation), many patients enrolled on clinical trials of investigational agents, which rely heavily on radiographic endpoints to assess efficacy. Accurate, reproducible, and prognostically-relevant imaging criteria are critically needed to compare treatment response within and across DIPG studies.^1,3–5 Appropriate discontinuation of ineffective therapy similarly demands reliable radiographic biomarkers of disease progression.

Adopted from the Response Assessment in Neuro-Oncology (RANO) criteria,⁶ endpoints using cross-sectional tumor measures are ubiquitous in clinical trials of DIPG, yet confer many challenges.^1,4 DIPG have irregular, indistinct borders, signal heterogeneity, and variable growth patterns which impede precise evaluation of two-dimensional (2D) size. Selection of the largest tumor dimension can be subjective and influenced by technical factors, including differences in slice thickness, patient position, and head placement.⁷ Recently, the Response Assessment in Pediatric Neuro-Oncology (RAPNO) DIPG working group provided tailored guidance for measuring these tumors.³ However, current radiographic endpoints continue to rely on the cross-product (CP) of perpendicular diameters within a single plane, which may not capture meaningful changes in disease burden. In prior DIPG studies, 2D MRI measures failed to predict survival,⁸ and were complicated by high interobserver variability.⁹

Emerging evidence suggests volumetric measures may offer a more accurate assessment of tumor growth than 2D methods.^7,10–15 Warren et al compared one-, two-, and three-dimensional (3D) measurements of pediatric CNS tumors, including brainstem gliomas, demonstrating agreement across methods in detecting partial response, yet considerable discordance in classifying progressive disease (PD).¹⁰ Within this cohort, 3D measures had the smallest coefficient of variation, indicating a potentially lower likelihood of response misclassification compared to 2D methods. Differences between volumetric and CP measurements were also observed in pediatric low-grade gliomas, with findings suggesting higher accuracy of semiautomated tumor volume assessments.¹¹ Volumetric and 2D measures similarly yielded disparate results in a small sample of DIPG patients.¹² In a study identifying MRI biomarkers in DIPG, early reduction in tumor volume following radiotherapy (RT) correlated with improved progression-free and overall survival.¹³ Finally, when compared to an ellipsoid model for tumor volume calculation, manually segmented volumetric measurements showed greater inter- and intra-reviewer agreement and improved sensitivity for detecting disease progression in DIPG.¹⁴

Taken together, volumetric measurements—especially via semiautomated techniques which more precisely capture the growth of irregularly-shaped tumors¹⁴—may be superior to traditional 2D methods for evaluating response among patients with DIPG. Data comparing outcome associations between CP and volumetric assessments of DIPG are lacking, yet will be essential for optimizing endpoints in future clinical trials. Herein, we sought to (1) determine the correlation between cross-sectional and volumetric measures of tumor extent and response, (2) compare the association of PD classification by these measures with survival postprogression, and (3) characterize tumor growth patterns that may impact imaging response assessment in a representative cohort of patients with DIPG.

Materials and Methods

Study Population

This retrospective study was performed at Cincinnati Children’s Hospital Medical Center following approval by the International DIPG/DMG registry scientific review board and the local IRB. Patients were referred to the registry as previously described,¹⁶ after DIPG was diagnosed locally based upon clinicoradiologic features. A convenience sample of well-characterized patients abstracted from a prior study in the registry who met the following inclusion criteria was included: (1) patient received upfront RT, (2) baseline and serial (≥2) post-RT MRIs were submitted to the registry and met imaging protocol requirements for evaluation described below, and (3) diagnostic MRI was centrally reviewed and assessed as characteristic of or likely DIPG (arising from and involving ≥50% of the pons with a diffuse pattern of involvement).¹⁷ Tumor tissue was not required, but patients with available tissue (biopsy and/or autopsy) were preferentially included; in these cases, histopathology needed to be consistent with infiltrating grade II–IV gliomas (diffuse midline glioma, H3K27M-mutant, glioblastoma, anaplastic astrocytoma, diffuse astrocytoma). Patients with extensive infiltrative disease in the cerebrum or with the majority of tumor involvement in the cerebellum at diagnosis were excluded, due to the difficulty of radiographically measuring tumor size by CP or volumetric methods in these cases. Patients with subventricular tumor spread or noncontiguous metastatic disease denoting progression in the setting of primary brainstem tumor stability or response were excluded. Clinical data (age at diagnosis, sex, treatment, overall survival [time from diagnosis to death]) were abstracted.

Radiographic Evaluation

Imaging protocol: As this study analyzed registry-supplied data, available MRIs were acquired using various protocols. All evaluated MRIs had axial T2-weighted images at a slice thickness between 3–5 mm, with a 0–1 mm interslice gap. Tumor outlines were traced on axial T2-weighted images, visually referencing T2/FLAIR images and T1-weighted sequences (when available).

Tumor measurements

Tumor margins were manually segmented using proprietary, clinically-available software (mint Lesion^TM, v3.4.5; Mint Medical), with extrapolation of CP and volumetric measures as previously described.¹² Manual segmentation was initially performed by a radiology trainee (MTN), with a review, modification as needed, and final approval of each case by a pediatric neuroradiologist with >25 years of experience (JLL), both blinded to clinical data. Training involved the pediatric neuroradiologist performing tumor outlines in the first five cases with the trainee in attendance, with detailed education provided on how to define tumor margins as well as the use of the volumetric system, and thorough review and corrections as needed for all subsequent cases. From the outlined tumor, the largest axial diameter (long axis) and its perpendicular measurement (short axis) were automatically derived; semiautomated CP was calculated by multiplying the long and short axes. Segmented tumor volume was automatically derived from the outlined tumor. The greatest cranial-caudal disease extent was manually measured by the senior reviewer (JLL). Ellipsoid volume was calculated as (π/6)*(semiautomated CP)*(manual cranial-caudal measurement).

Progressive disease assessment

PD by CP was defined according to DIPG RAPNO criteria,³ as ≥25% increase compared to the smallest CP measurement from an earlier timepoint. PD by volume, using both segmented and ellipsoid methods, was defined as an increase in volume compared to the smallest prior measure, with two thresholds evaluated: (1) ≥40% increase, predicted from a spherical model of disease progression,¹⁸ previously studied as the volumetric threshold for PD in adult and pediatric gliomas^{11,12,19–21} and (2) increase by segmented or ellipsoid volume that corresponded to 25% increase by CP in linear regression analyses correlating CP and volumetric changes, detailed below.

Psuedoprogresion

Timepoints with imaging consistent with pseudoprogression (by CP and/or volumetric measures) were not considered to represent PD. Pseudoprogression was defined according to collective criteria from recent DIPG studies^3,19,20,22 as an increase in tumor size that would otherwise be considered PD within 6 months of RT completion, subsequent improvement in tumor size to at least stable disease (compared to smallest prior size) on the next MRI (generally within 4–8 weeks), and no change to ongoing anti-tumor treatment (ie, systemic therapy was not discontinued) if applicable. Pseudoprogression by each measurement method (CP and volumetric) was excluded independently.

Partial response assessment

In exploratory analyses evaluating partial response (PR), decreases in tumor CP by both ≥50% and ≥25% were studied, according to RANO and DIPG RAPNO criteria, respectively.^3,6 Volumetric thresholds for PR extrapolated from a spherical model (≥65% and ≥40% decrease, respectively) as well as percent decrease by volume (segmented and ellipsoid) corresponding to 50% and 25% decrease by CP in linear regression analyses were studied.

Statistical Analysis

To identify volume thresholds (segmented and ellipsoid) corresponding to the CP threshold for PD (25% increase), linear regression was performed comparing percentage change from smallest prior tumor size by CP and volumetric methods. Because this relationship is not expected to be linear at large volumes, data after the first timepoint classified as PD by CP was excluded. If PD by CP occurred at the first post-RT timepoint, MRI data through the next PD timepoint was included. Next, discordant timepoints (where CP and volumetric methods disagreed on PD) were identified; survival following these discordant timepoints was compared using a log-rank test. Differences in cranial-caudal tumor growth at discrepant timepoints were analyzed using a Mann-Whitney test. To further determine if differences in radiographic endpoint (CP versus volumetric [segmented and ellipsoid]) predict outcomes, landmark analyses were performed. Following the methodology used in adult glioblastoma studies comparing 2D and volumetric measurements,^21,23–25 differences in survival (from MRI date to death) between patients classified as PD versus non-PD according to CP and volumetric measurements were compared using a log-rank test. Because the timing of PD is uncertain, multiple post-RT time points were selected to show the overall sensitivity of each threshold: 1 month (0.4–<2 months), 3 months (2–<4 months), 5 months (4–<6 months), 7 months (6–<8 months), and 9 months (8–<10 months) after RT completion. These time points were analyzed as they represent common disease evaluation timepoints in DIPG trials and include response assessments beyond 6 months post-RT (outside the traditional pseudoprogression window). To evaluate whether there were significant differences between methods in predicting survival, hazard ratios (HRs) and 95% confidence intervals were determined via Cox proportional hazards regression, using PD status according to different measures at respective timepoints as a single covariate. Multivariable Cox proportional hazards regression analyses were subsequently performed, adjusting for prior bevacizumab use or re-irradiation status. For exploratory PR analyses, similar linear regression and univariate Cox proportional hazards regression analyses as described above were performed, the latter at the first three post-RT timepoints. Statistical tests were performed in SPSS (v26, IBM) and were two-sided, with P-values <.05 considered significant.

Results

Cohort

Forty-eight patients with DIPG from the convenience sample met the initial inclusion criteria. Two patients were excluded due to the majority of tumors involving the cerebellum (n = 1), and subventricular tumor spread denoting tumor progression in the setting of primary brainstem tumor stability (n = 1). Therefore, 46 patients were included in the analysis, with clinical and demographic features summarized in Table 1. The median age at diagnosis was 6.8 years (Range: 2–27 years), 22 (48%) were female, and the median overall survival was 15 months (Range: 5–34 months). All patients received upfront RT. Forty-one (89%) patients received systemic therapy, concurrent with RT in 24 (52%) and/or adjuvant in 40 (87%). Eight (17%) patients underwent re-irradiation. Among 46 analyzed patients, seven (15%) had pseudoprogression by CP (n = 4) and/or volumetric (n = 5) measures. Of these seven patients, four received systemic therapy concurrent with RT and/or were receiving adjuvant therapy at the time of imaging demonstrating pseudoprogression (agents included temozolomide, vorinostat, and ribociclib); the remaining three patients received only radiation.

Correlation Between CP and Volumetric Thresholds of PD

Manual segmentation was performed on 312 MRIs from 46 patients (Figure 1A). A total of 158 response timepoints between diagnosis and progression were used to compare CP and volumetric measures (percent change from smallest prior tumor size) via linear regression (Figure 1B and C). In this analysis, PD by CP (25% increase) corresponded to an increase of approximately 30% by segmented volume (R² = 0.710) and 50% by ellipsoid volume (R² = 0.877). As the calculated segmented and ellipsoid volumetric thresholds for PD varied from the 40% threshold predicted by a spherical model, all thresholds (30% and 40% for segmented volume; 40% and 50% for ellipsoid volume) were included in the analyses below.

Discordant Classification of First PD Between CP and Volume Measures

In 19 (41%) of 46 patients, there was a discordant classification of the first PD between CP and segmented volume (using 40% and/or 30% thresholds) (Table 2). Using a segmented volumetric threshold of 40%, patients classified as PD by volume but not by CP (example: Figure 1A [5 months post-RT]) had significantly shorter survival (from time of first discordant scan) than patients classified as PD by CP but not by volume (median: 3 versus 6 months, P = .011). Among patients classified as PD by segmented volume (40% threshold) but not by CP, there was significantly greater cranial-caudal tumor growth: specifically, the median increase in cranial-caudal dimension from smallest prior value was 38% (Range: 8% to 84%), compared to 16% (Range: -7% to 23%) among the tumors classified as PD by CP but not by volume, excluding two cases of pseudoprogression (Mann-Whitney U = 48, P = .019). Similar trends were seen when using a segmented volumetric threshold of 30%, but results did not reach statistical significance. There were 12 cases of discrepant PD classification between ellipsoid volume (using 40% and/or 50% thresholds) and CP. No differences in survival were observed when comparing discordant cases of PD classification between ellipsoid volume using 40% threshold and CP. However, when using an ellipsoid volume 50% threshold, a trend toward worse survival was observed among patients classified as PD by volume but not by CP compared to patients classified as PD by CP but not volume (P = .13); the former had significantly greater cranial-caudal growth (median: 37.5% versus 13.1%; Mann-Whitney U = 34, P = .005).

Comparisons of Outcome Association

To assess the ability of different radiographic measurements to predict survival (from MRI timepoint to death), landmark analyses were performed comparing survival curves of PD versus non-PD patients at 1, 3, 5, 7, and 9 months post-RT. Using CP measures, a difference in survival was observed at 1 month post-RT (median survival post-MRI examination: 8.2 versus 13.4 months, P = .013, Figure 2A). However, no survival difference was detected by CP classification of PD versus non-PD at 3, 5, 7, or 9 months post-RT (P > .05) (Figures 2F and K, 3A and F). In contrast, differences in survival were observed when PD was defined by volumetric measures. When using a segmented volumetric threshold of 40%, there were significant differences in survival between patients classified as PD versus non-PD at every post-RT imaging timepoint: 1 month (8.2 versus 13 months, P = .035), 3 months (5.7 versus 10.8 months, P = .010), 5 months (4.0 versus 9.1 months, P = .003), 7 months (3.9 versus 8.9 months, P = .009), and 9 months (4.5 versus 8.3 months, P = .001) (Figures 2B, 2G, 2L, 3B, 3G). When using a segmented volumetric threshold of 30%, results were identical to the 40% volumetric threshold at 1 and 3 months post-RT, with similar significant differences in survival at 5 months (3.8 versus 10.4 months, P < .001) and 9 months post-RT (4.7 versus 9.0 months, P = .015) (Figure 2C, 2H, 2M, 3H). A trend toward a survival difference was observed at 7 months post-RT (4.6 versus 9.2 months, P = .067; Figure 3C). When using an ellipsoid volume threshold of 40%, PD classification was not associated with survival at any post-RT timepoint (all P > .05; Figures 2D, 2I, 2N, 3D, 3I). However, classification of PD by ellipsoid volume using a threshold of 50% was significantly correlated with survival at 1, 3, and 7 months post-RT (P < .05), with a trend toward a survival difference at five-months post-RT (P = .070).

Table 3 summarizes the HRs of death for patients classified as PD by CP, segmented volume using 40% or 30% thresholds, and ellipsoid volume using 40% or 50% thresholds at all five post-RT imaging time points. At one month post-RT, there were significant associations with survival for PD classified by CP and by segmented volume using 40% or 30% thresholds (P < .05), with similar HRs (2.57–2.77); correlations between ellipsoid volume using 40% or 50% thresholds were not significant (P > .05). At the four subsequent post-RT timepoints, PD by CP was not predictive of survival (all P > .05). However, PD by segmented volume using a 40% threshold was significantly correlated with survival at all time points, with HRs ranging from 2.57–16.29 (all P < .05). PD by segmented volume using a 30% threshold was significantly associated with survival at 1, 3, 5, and 9 months post-RT (HRs: 2.57, 2.62, 4.65, 5.54), with a trend toward association at seven months post-RT. PD by ellipsoid volume using a 50% threshold was significantly associated with survival at 3 and 7 months post-RT (HRs: 2.67, 2.35). At 5 and 9 months post-RT, HRs for PD by ellipsoid volume using a 50% threshold were greater than corresponding HRs for PD by CP, but did not reach statistical significance. PD by ellipsoid volume using a 40% threshold was not associated with survival at any timepoint (all P > .05).

In multivariable analyses adjusting for prior bevacizumab use or re-irradiation status, results were similar, with segmented volume (40% threshold) remaining significantly correlated with survival at nearly all time points, representing the measurement threshold most consistently and strongly predictive of outcome (Supplementary Materials S1–S2).

Partial Response Exploratory Analysis

Linear regression and Cox proportional hazards regression analyses were conducted to correlate and compare the prognostic impact of PR classification by CP and volumetric (segmented and ellipsoid) measures at the first three post-RT timepoints (Supplementary Material S3). Notably, designation of PR by no method or threshold (CP or volume [segmented or ellipsoid]) was predictive of survival at any time point (P > .05).

Discussion

Advancing the treatment paradigm of DIPG will require precise, reproducible, and clinically meaningful radiographic criteria to assess response. Despite an improved understanding of the inherent challenges in measuring DIPG, including their infiltrative nature with irregular, indistinct borders, as well as the limitations of applying 2D techniques,^1,3,4 imaging endpoints in clinical trials continue to rely on cross-sectional measurements. Given emerging evidence that semiautomated tumor volumetry may provide a more accurate evaluation of disease burden and growth,^7,10–15 we sought to directly compare 2D and volumetric (both segmented and ellipsoid) measures of PD among patients with DIPG. To our knowledge, this is one of the first and largest studies correlating CP and volumetric thresholds of PD, comparing their association with outcomes, and characterizing tumor growth patterns with the potential to impact radiographic response assessments within a representative cohort of patients with DIPG.

In a linear regression analysis incorporating 158 response time points, an increase of 30% by segmented volume corresponded to the current 2D criteria for PD (25% increase).³ This volumetric threshold (30%) is less than the corresponding mathematical prediction (40%) assuming a spherical model with equal progression in all three dimensions.^{11,12,21,23,24} These results are consistent with the recognized nonspherical growth pattern of DIPG secondary to anatomic constraints of the brainstem as well as common lateral extension into the brachium pontis and cerebellum.^4,9,12,26 When analyzing ellipsoid volumetric measurements, an increase of 50% (>40%) corresponded to the established CP threshold of 25%, though as described below, PD classification by ellipsoid volume was inferior to segmented volume in predicting survival. Riley et al also observed discordance between tumor volume measurements calculated by a mathematical (ellipsoid) model versus region-of-interest manual segmentation volumetry among patients with DIPG, with the latter technique similarly exhibiting greater sensitivity for detecting PD.¹⁴ Findings suggest mathematical models for determining volume do not precisely measure disease burden of these invasive, irregularly-shaped tumors, and that 3D changes in DIPG size can be more accurately measured using semiautomated, segmentation methods.^4,14 Importantly, manual volumetric segmentation represents a feasible measurement tool with potential for expansion to clinical practice given that corresponding software is becoming more readily available.^7,10–15 Moreover, semiautomated volumetric analysis has demonstrated high intra- and inter-reviewer reliability (intra-observer and inter-observer agreement indices=92.3% and 89.8% [Riley et al.]; intraclass correlation coefficient = 0.9 [D’Arco et al].),^11,14 and prior reports have shown that tumor tracing is not particularly time-intensive (uniformly performed in under 10 minutes/scan in our study).¹¹

Discordant classification of the first PD by CP and volumetric measures occurred in up to 41% (19/46) of patients with DIPG in our cohort. These results align with significant discordance rates between 2D and 3D techniques in defining PD across other pediatric CNS tumors described by Warren et al. (54% [21/39])¹⁰ and D’Arco et al. (40% [8/20]).¹¹ Among the discrepant MRIs analyzed in the present study, there were differences in cranial-caudal tumor growth, greater in tumors classified by PD by segmented or ellipsoid volumetric measures, but not CP. Furthermore, inferior survival was observed in patients whose tumors were classified as PD by volume, but stable disease by CP compared to patients with the stable disease by volume, but PD by CP (statistically significant when analyzed with segmented volume 40% threshold, trend with ellipsoid volume 50% threshold). These findings suggest that CP measures may underestimate tumor size and disease progression, particularly cranial-caudal extension, compared to volumetric assessments,¹² with clinical and prognostic implications. Patients with PD as determined by volume may be evaluated earlier for alternate therapies. Furthermore, as drug anti-tumor activity is often determined by radiographic response assessments, disparate impressions of PD may confound early measures of efficacy.^10,11

In landmark analyses conducted at five post-RT MRI time points (between one and nine months following completion of RT), segmented volumetric assessments of PD demonstrated stronger correlations with the outcome overall than cross-sectional or ellipsoid volume measures. At one month post-RT, both segmented volume (30% and 40% thresholds) and CP definitions of PD were correlated with survival, though PD classification by ellipsoid volume was not prognostic. While the conventional 2D radiographic response to RT has not previously been shown to predict survival in DIPG,^1,8,27 our findings suggest patients whose tumors progress following RT (excluding cases of pseudoprogression) when analyzed by either segmented volume or CP experience worse outcomes and may represent a more biologically aggressive, refractory subgroup. However, at all four later imaging time points, only PD by segmented volume and not by CP was predictive of survival, with variable associations between ellipsoid volume and outcome. Specifically, segmented volumetric measures of PD using a 40% threshold significantly correlated with survival at all time points. At the lower threshold of 30% volumetric increase, similar associations with survival were identified at four of five post-RT timepoints, with a trend toward association at the remaining timepoint. Segmented volumetric assessments of PD (both 40% and 30% thresholds) outperformed both CP and ellipsoid volume in predicting outcomes at the latter four post-RT examination time points, with more consistent associations and generally larger effect sizes observed with PD by segmented volume using a 40% threshold, even after adjusting for prior bevacizumab use or re-irradiation status. PD classification by ellipsoid volume (using 50% threshold) was statistically superior to CP at two post-RT timepoints, though inferior to segmented volume overall. Our findings expand upon a previous report from Poussaint et al., in which early volumetric response following radiation correlated with prolonged survival among patients with DIPG.¹³ Studies comparing outcomes based on volumetric and 2D measurements of PD in adult glioma have produced mixed results,^{15,21,23–25} but did not evaluate tumors of a single, shared location. In contrast, all tumors in our cohort were of pontine origin, and this consistent, confined location may strengthen the association between tumor volume and overall disease burden related to more similar boundaries for expansion. Analogously, volumetric (but not 2D) measurements of optic pathway gliomas limited to the optic nerve in location have been shown to correlate with visual acuity loss.^28,29 Taken together, segmented volumetric assessments of PD for DIPG may not only provide a more accurate method for detecting treatment failure, but also a biomarker of clinical outcome (inferior survival) in these patients.

Although segmented volumetric measures of PD consistently predicted outcomes in our DIPG cohort, no analyzed CP or volume method for designating PR by any threshold correlated with survival at post-RT timepoints. While interpretation may be impeded by the relatively small number of radiographic responses, these findings corroborate previous reports suggesting a lack of prognostic impact of PR (versus non-PR) to RT when defined by bi-dimensional measures in DIPG.^1,8,27 Radiographic determination of anti-tumor activity (aside from prolonged stable disease) therefore remains challenging in DIPG, without CP or volumetric thresholds of PR that reliably predict improved survival. Continued research into meaningful imaging endpoints denoting therapy efficacy for DIPG is critical.

Given notable differences in outcome based on PD classification measurement as well as the aforementioned high rate of discordance between methods, we recommend that semiautomated volumetry, when feasible, be incorporated into prospective DIPG clinical trials (at least to be studied in combination with routine bi-dimensional measurements prospectively), with ongoing research focused on determining the optimal threshold defining PD by volume. Both segmented volumetric thresholds of 40% (spherical model extrapolation) and 30% (corresponding to PD by CP) were better able to predict survival than CP or ellipsoid volume (either 40% or 50% thresholds), further underscoring that bi-dimensional measurements as well as mathematical volume calculations, by any threshold (despite the latter better detecting cranial-caudal extension than the former), are likely insufficient to capture meaningful increases in DIPG growth. While a 30% segmented volumetric threshold offers earlier identification of PD, it is less closely associated with survival compared to a 40% segmented volumetric threshold at later time points, suggesting the latter may represent a more appropriate target for defining PD, which will be critical to continue assessing.

Pseudoprogression presents challenges for evaluating radiographic response to chemoradiotherapy in high-grade gliomas, including DIPG,^{3,19,20,22,30} and has the potential to confound outcome analyses. We attempted to limit misclassification bias by designating cases consistent with pseudoprogression as non-PD. Our definition of pseudoprogression applied criteria adopted from collective reports,^19,20,22 in accord with the RAPNO DIPG recommendations,³ and following specific guidelines utilized in a recent DIPG clinical trial.²² The incidence of pseudoprogression in our cohort by CP and/or segmented volumetric measures (analyzed independently) was 15% (7/46), which is in agreement with previously described frequencies of pseudoprogression among DIPG patients (14–18%).^19,20,22 A higher occurrence of pseudoprogression was reported by Calmon et al. (44%), albeit using broader criteria that didn’t require subsequent improvement in tumor size.³⁰ Although some instances of pseudoprogression may have been missed, potential bias was likely minimized by excluding pseudoprogression from PD using a definition based on clinical use which yielded similar incidence rates as prior studies. Moreover, the improved survival correlation of segmented volume compared to CP and ellipsoid volume methods at both early and late post-RT timepoints alike (ie, observed seven and nine months after RT, outside the traditional pseudoprogression window) provides additional evidence favoring semiautomated volumetric measures despite cases of pseudoprogression.

Other limitations to consider include the modest sample size and retrospective nature of this study. Rare cases of extensive cerebellar involvement at diagnosis and PD defined by subventricular tumor spread in the setting of brainstem disease stability were excluded, potentially decreasing generalizability of our findings to patients with more widely disseminated disease. Median overall survival for patients in our cohort was 15 months, which is greater than pooled historical control data (11 months); this difference may be explained by the relatively small number of patients as well as that most received systemic therapy, which has been shown to confer a small, but statistically significant survival benefit in DIPG.² Importantly, further investigation comparing CP and volumetric measurements of PD should be performed prospectively in larger DIPG cohorts, including in combination with clinical status and corticosteroid use data. Although intratumoral necrosis, enhancement, and cysts were uniformly included in overall tumor dimensions (latter were rare in our cohort), we did not separately measure these components or assess the potential impact on outcomes. Future studies should investigate the presence and prognostic significance of these features, and whether advanced imaging modalities, such as MR spectroscopy, can be combined with volumetric measures performed on conventional MRI to guide response evaluation. Finally, tumor manual segmentation results were determined by a single experienced radiologist, precluding determination of interobserver variability of this technique in our cohort, which deserves exploration in larger-scale efforts. However, as introduced above, previous studies have demonstrated high inter-reviewer concordance in similar volumetric analyses.^11,14

Despite these limitations, this study provides valuable insight regarding measurement of DIPG, which can aid in optimizing radiographic response endpoints for this challenging disease. Growth patterns of DIPG do not conform to spherical mathematical models and there may be greater progression along dimensions not captured in a single axial plane and/or irregular extension, such that traditional cross-sectional measures and often ellipsoid volume methods fail to identify meaningful changes in disease burden in a sizable portion of patients. Tumor volume, obtained by semiautomated manual segmentation methods, can more precisely detect progression and reliably predict survival in DIPG across several post-RT timepoints. Continued research in prospective trial and clinical practice settings will be essential, but compared to both bi-dimensional and ellipsoid volume assessments, segmented volumetric measurements likely represent a superior imaging biomarker of PD in DIPG.

Acknowledgments

We thank the patients and families for their invaluable contribution to this research. We acknowledge the support of the Curing Kids Cancer Foundation. For financial support of the IDIPGR, we thank The Cure Starts Now Foundation, The Cure Starts Now Australia, Brooke Healey Foundation, Wayland Villars Foundation, Aidan’s Avengers, Aubreigh’s Army, Austin Strong, Cure Brain Cancer, Jeffrey Thomas Hayden Foundation, Laurie’s Love Foundation, Love Chloe Foundation, Musella Foundation, Pray Hope Believe, Reflections Of Grace, Storm the Heavens Fund, Whitley’s Wishes, Gabriella’s Smile Foundation, The Gold Hope Project, The Isabella and Marcus Foundation, Lauren’s Fight for Cure, Robert Connor Dawes Foundation, Ryan’s Hope, Benny’s World, Lily Larue Foundation, Marlee’s Mission, RUN DIPG, American Childhood Cancer Organization, The DIPG Collaborative, Snapgrant.com, the Kyler Strong Foundation, and Keris Kares.

Funding

This study was financially supported by Curing Kids Cancer (PdB).

Conflict of interest

None

Author contributions

Conception and design: JLL, PdB, MAL, MTN, MD, AL. Data acquisition: All. Data analysis/interpretation: JLL, PdB, MAL, MTN, AL, BC. Drafting/revising critically: All. Final approval: All.

References

Author notes