Skip to main content
NCBI home page
Neuro-Oncology logoLink to Neuro-Oncology
. 2025 Feb 17;27(7):1899–1909. doi: 10.1093/neuonc/noaf044

Multi-site retrospective analysis of diffusion and perfusion magnetic resonance imaging correlates to glioma characteristics derived from radio-pathomic maps

Samuel A Bobholz 1, Daniel Aaronsen 2, Aleksandra Winiarz 3, Savannah R Duenweg 4, Allison K Lowman 5, Michael Flatley 6, Fitzgerald Kyereme 7, Jennifer Connelly 8, E Kelly S Mrachek 9, Max O Krucoff 10,11, Anjishnu Banerjee 12, Peter S LaViolette 13,14,✉
PMCID: PMC12417833  PMID: 39960860

Abstract

Background

This study determines the relationship between diffusion and perfusion-based magnetic resonance imaging signatures and radio-pathomic maps of tumor pathology in a large, multi-site cohort.

Methods

This study included perfusion imaging from presurgical relative cerebral blood volume (rCBV) images from the UPenn-GBM dataset and presurgical arterial spin labeling (ASL) imaging from the UCSF-PDGM dataset. Diffusion imaging included fractional anisotropy (FA) estimates derived from diffusion tensor imaging for each subject from each institution. A previously validated autopsy-based model was applied to the structural images from each patient to generate quantitative radio-pathomic maps of cell density and extracellular fluid (ECF). Mean cell density, ECF density, FA, rCBV calculated from dynamic susceptibility contrast imaging, and rCBF calculated from ASL were computed for each patient and statistically compared within contrast-enhancement (CE) and the non-enhancing peritumor FLAIR hyperintensity (FH).

Results

Both rCBV and ASL showed a positive correlation with cell density within CE (rCBV: R = 0.280, P < .001; ASL: R = 0.117, P = .023). However, both perfusion metrics also showed no association with cell density within the FH region at the group level (rCBV: R = 0.0162, P = .731; ASL: R = −0.020, P = .652). Negative correlations were observed between FA and ECF density across both CE and FH in both the UPenn-GBM (CE: r = −.204, P < .001, FH: r = −.332, P < .001) and the UCSF-PDGM (CE:r = −.179, P < .001, FH:−0.355, P < .001). Additionally, a positive ASL-cell density association per subject within FH was associated with a worse survival prognosis (HR = 5.58, P = .022).

Conclusions

These results suggest that radio-pathomic maps of tumor pathology provide complementary information to other MR signatures that reveal prognostically valuable signatures of the non-enhancing tumor margin.

For the podcast associated with this article, please visit ‘https://soc-neuro-onc.libsyn.com/radio-pathomic-maps-and-mr-perfusion’

Key Points.

  • Cell density estimates associated with perfusion within but not beyond enhancement.

  • Extracellular fluid density is negatively correlated with fractional anisotropy.

  • Positive peritumoral perfusion-cellularity associations show worse patient outcomes.

Importance of the Study.

Recently developed radio-pathomic maps of tumor pathology have shown promise in providing noninvasive signatures of occult tumor invasion. However, the relationship between these peritumoral signatures of glioma infiltration and other MR-based characteristics of tumor biology such as perfusion and diffusion remains unknown. This study highlights novel characteristics of the cellularity-perfusion and diffusion-ECF density relationships in two large, publicly available datasets external to the data used to train the model. Perfusion was found to associate positively with cellularity within but not beyond the contrast-enhancing region, indicating that it may be less sensitive to non-angiogenic tumor infiltration, though subjects with hyperperfused, hypercellular non-enhancing tumors showed worse survival outcomes. Fractional anisotropy showed consistent negative associations with ECF, indicating colocalization of necrosis and disruptions of white matter integrity. This study provides novel, repeatable insights into the interpretation of multimodal signatures of tumor pathology within and beyond contrast enhancement.

Gliomas are the most common type of primary brain tumor, with an average incidence of 6 per 100 000 persons.1,2 The World Health Organization (WHO) classifies these tumors by integrating histological features and molecular markers into different categories, which then are ascribed 1 of the 4 grades.3 Unlike pediatric-type diffuse grade 1 gliomas, which are potentially curable, adult type grade 2, 3, and 4 gliomas represent “diffuse” gliomas that can be heterogenous and spread within healthy brain parenchyma. Despite not being curable, treatment for virtually all diffuse glioma subtypes begins with maximal safe surgical resection, which can play a significant role in extending overall survival (OS).3–5 When radiographic changes develop postoperatively, in or adjacent to the operative cavity, it can be difficult to delineate true tumor progression from pseudo-progression, a reactive inflammatory response to chemoradiation.6–8 Pseudo-progression is treated very differently from tumor recurrence and, therefore, there has been significant investment made on behalf of the neuro-oncology community into noninvasive methods of differentiating pseudo-progression from true tumor progression.9,10

Perfusion magnetic resonance imaging (MRI) has emerged as one such noninvasive adjunct in the differentiation of tumor progression from pseudo-progression. Various techniques exist for performing perfusion MRIs, including contrast-based methods such as dynamic susceptibility contrast (DSC)11,12 used to measure relative cerebral blood volume and non-contrast-based methods such as arterial spin labeling (ASL),13 used to calculate relative cerebral blood flow (rCBF). Perfusion MRI techniques such as DSC and ASL have been shown to have significant utility in detecting angiogenic glioma presence, and likewise have been shown to increase diagnostic accuracy in differentiating tumor progression from pseudo-progression when evaluating treatment response in high-grade gliomas.13–15 However, infiltrative tumors also can extend beyond the angiogenic margin, which may result in reduced tumor sensitivity for perfusion-based markers. Diffusion tensor imaging (DTI) is often used clinically to map white matter tract disruptions resulting from glioblastoma invasion, as well as for mapping eloquent tissue to aid in preserving cognitive function and quality of life post-surgery. Fractional anisotropy (FA) is a metric derived from DTI that indicates the degree to which diffusion occurs directionally where reductions are thought to indicate disruptions in white matter integrity.16 This could aid in identifying areas where either active tumor presence or tumor-related necrosis are present, and it could help detect abnormal pathology outside the contrast-enhancing region.17,18

Recent advances in radio-pathomics have sought to bridge the gap between imaging phenotypes and underlying cellular pathology. In recent work, we developed a radio-pathomic mapping tool by correlating whole-brain autopsy tissue from glioma patients with premortem MRI slices.19,20 A machine learning model was trained to detect cell density, extracellular fluid (ECF) density, cytoplasm density, and tumor probability, all while using conventionally acquired MRI sequences such as T1 pre- and post-contrast (T1, T1C), T2 fluid-attenuated inversion recovery (FLAIR), and apparent diffusion coefficient (ADC) maps. These maps identified quantitative pathological features within a standard deviation of the ground truth value and reliably identified areas of high tumor probability outside the contrast-enhancing region in both the test set and external data. While identifying these regions is critical for directing and monitoring localized treatment, it remains largely unknown how tumor and non-tumor portions of the non-enhancing tumor margin differ in their physiological properties, particularly with regard to commonly acquired advanced imaging signatures such as diffusion and perfusion MRI that were not included in the model development. These imaging modalities have a robust literature exploring their uses in characterizing enhancing tumors, but without prior methodology for identifying non-enhancing tumors noninvasively, it is unclear how these relationships are expressed in non-angiogenic tumor areas. Therefore, in this study, we correlate diffusion- and perfusion-derived metrics with radio-pathomic maps of cellularity and ECF as a technical feasibility study for a better understanding of how regions of high tumor probability interact with structural and functional imaging metrics. Specifically, we test the hypothesis that increased cell density correlates with hyperperfusion in contrast-enhancement (CE) and test whether the relationship holds in non-enhancing FLAIR hyperintense regions (FH). We also tested the hypothesis that FA would associate with ECF regardless of enhancement status and that prognostically, patients who underwent gross total resection were worse off when tumor invasion and hypervascularity co-localized outside CE.

Materials and Methods

Patient Population

This study was approved by the Institutional Review Board of the Medical College of Wisconsin. This retrospective study used 2 publicly available datasets from 2 different institutions19,21: (1) 630 glioma patients from the UPenn-GBM cohort, including 456 with DSC contrast-based perfusion metrics via relative Cerebral Blood Volume (rCBV) images21 and (2) 501 glioma patients from the UCSF-PDGM dataset, including 426 with non-contrasted perfusion metrics via ASL sequences and contrast-enhancing tumors.22 A graphic overview of the study design and analyses is presented in Figure 1. Patients included were all surgically diagnosed with glioma, with the most common diagnosis being glioblastoma across both datasets. Incomplete information on IDH status for patients prevented the limitation to glioblastomas as defined by the 2021 WHO criterion, therefore this study examined the broad spectrum of gliomas available across the 2 datasets.

Figure 1.

Example imaging and tumor probability maps depicting the workflow described in the caption. The coregistration step takes approximately 2 minutes, with intensity normalization taking 15 seconds. Map generation takes approximately 3 minutes per patient.

Open in a new tab

Overview of study methodology. Radio-pathomic maps of cell density and extracellular fluid (ECF) density are generated from T1, T1C, FLAIR, and ADC images that have been aligned to the FLAIR and intensity normalized (aside from ADC). A 5 × 5 sliding frame is used as input to the trained radio-pathomic model for generating maps, which uses a bagging random forest architecture to predict cell/ECF density from ground truth autopsy pathology. These maps are then compared to fractional anisotropy and perfusion-based metrics within the contrast-enhancing region (orange, area surrounding tumor core) and the non-enhancing FLAIR hyperintense region (blue, area surrounding contrast enhancement).

MR Imaging Acquisition and Preprocessing

Clinical imaging was collected from each patient’s pretreatment MRI for inclusion in this study. Perfusion images were used in their publicly released form, collected according to the respective protocol for the UCSF-PDGM22 and UPenn-GBM21datasets. Pre-calculated FA maps calculated from DTI included with the dataset were used for this study. Information regarding acquisition parameters and processing steps for perfusion data can be found in the citations for each dataset. T1, T1C, FLAIR, and ADC maps calculated from diffusion-weighted imaging acted as the input for radio-pathomic map generation. Images were preprocessed following the standard used for model generation, which involved alignment of all images to the FLAIR image by using SPM12’s co-registration tools,23 as well as intensity normalization by dividing each voxel by the whole brain intensity standard deviation for each non-quantitative image (T1, T1C, FLAIR).19,24–27

Cell density and Extracellular Fluid Mapping

Preprocessed MR images for each patient were then used to generate radio-pathomic maps of cell density and ECF density using a previously developed algorithm.24 Briefly, large format autopsy samples were collected from brains sliced axially from areas of suspected tumor and non-tumor and aligned to clinical MRI near death. Bagging forest algorithms were used to predict computed features of the pathology including cell density (cells/mm2), ECF density, and cytoplasm density using 5 by 5 voxel tiles from the T1, T1C, FLAIR, and ADC images as input, providing voxel-wise maps of tissue characteristics previously only available via biopsy. These maps were trained on 43 patients and tested on 22 held-out patients, showing high accuracy on internal test data and impressive generalizability to external data. The maps were further converted to tumor probability maps via an additional algorithm. Each map was visually inspected to ensure sufficient quality predictions for qualitative annotation and segmentations.

Statistical Analyses

The mean values for predicted cellularity, ECF, and FA, as well as z-scored ASL-based relative cerebral blood flow (rCBF) perfusion-weighted values in the UCSF-PDGM dataset and z-scored rCBV values in the UPenn-GBM dataset, were computed for both the non-necrotic CE region and the FH region, as defined by the semi-automated, radiologist-verified segmentations included with each dataset. The Pearson correlations between cellularity and perfusion values as well as FA and ECF values were computed for each dataset within each ROI separately. As a follow-up analysis, we also computed the per-subject voxel-wise association between ASL perfusion values and cellularity as well as between FA and ECF within the non-enhancing FLAIR region. The UCSF dataset had surgical information included, specifically identifying those who underwent gross total and sub-total resection. These groups were used to compare survival differences associated with the per-subject cellularity-perfusion association amongst UCSF-PDGM glioblastoma patients who had undergone gross total resection to understand how these relationships impact prognosis.

Results

There was a notable positive correlation between both rCBV and ASL with mean cell density in CE regions of glioma. Specifically, for rCBV derived from DSC perfusion data within the UPenn-GBM dataset, there was a modest positive correlation with mean cell density (R = 0.322, P < .0001; Figure 2) within CE regions. Conversely, in FH regions, the association between rCBV and mean cellular density was not correlated (R = −0.021, P = .650). Example cases revealed hypercellular portions of the FH region with no distinguishing increase in rCBV (Figure 2). ASL perfusion images from the UCSF-PDGM dataset showed a similarly subtle positive correlation within CE regions (R = 0.118, P = .015; Figure 3). However, within FH regions, there was no correlation (R = 0.015, P = .766). Again, examples revealed non-enhancing regions of hypercellularity in the absence of distinguishing ASL signals. These results were also reflected in 2 examples of MCWBB patients for which ground truth histology was available, where areas of hypercellularity beyond the contrast-enhancing margin existed in the absence of hyperperfusion (Figure 4). Comparatively, for both perfusion metrics, these findings were consistent across datasets, indicating a significant relationship with cellular density within areas of CE. The lack of a significant correlation in FH regions was observed for both rCBV and ASL measures. Negative correlations were observed between FA and ECF density across both CE and FH in both the UPenn-GBM (CE: r = −.204, P < .001, FH: r = −.332, P < .001) and the UCSF-PDGM (CE:r = −.179, P < .001, FH:−0.355, P < .001) (Figure 5). No association between cell density and FA was observed within both CE (R = −0.02, P = .680) and FH (R = −0.093, P = .056). Results for these findings stratified by tumor grade and IDH mutation status are included in Supplementary Material. Similar patterns were observed in grade 4 patients and IDH-wild-type patients, with smaller group sample sizes limiting the interpretability of the weaker results in lower-grade and IDH-mutant tumors.

Figure 2.

Scatterplots showing the positive association observed between rCBV and cellularity, as well as example imaging demonstrating the trends observed.

Open in a new tab

Associations between perfusion-based rCBV values and predicted cell density within the UPenn-GBM dataset demonstrated a positive association within contrast enhancement (R = 0.28, P < .0001) but not beyond the enhancing margin (R = 0.0162, P = .731). Examples from individual subjects indicate areas of hypercellularity occurring beyond the contrast-enhancing margin that do not show signs of hypervascularity, despite good concordance within the enhancing margin. CPM = cellularity prediction map.

Figure 3.

Scatterplots showing the positive association observed between CBF and cellularity, as well as example imaging demonstrating the trends observed.

Open in a new tab

Associations between arterial spin labeling values and cell density within the UCSF-PDGM dataset, show positive associations within enhancement (R = 0.110, P = .023) and no relationship beyond enhancement (R = −0.020, P = .652), similar to the contrast-based perfusion results seen in the UPenn-GBM dataset. Individual examples also show similar trends of hypercellularity existing in and beyond the FLAIR hyperintense margin that do not show distinguishing perfusion signatures. CPM = cellularity prediction map.

Figure 4.

Example imaging, perfusion maps, cellularity maps, and autopsy histology aligned to the MRI that indicate the cellularity maps tend to more accurately reflect hypercellularity observed histologically than ASL and rCBV.

Open in a new tab

Examples of perfusion-based metrics acquired near death compared to predicted cellularity maps (CPM), as well as actual cellularity from autopsy pathology (Hist.). Both the cellularity predictive maps and the aligned ground truth histology highlight areas of hypercellularity beyond the contrast-enhancing region with non-elevated perfusion in the region, potentially indicating areas of non-angiogenic tumor infiltration. For example, patients have been included in the training of the CPM model and thus do not necessarily represent prediction behavior on held-out data.

Figure 5.

Scatterplots showing the negative association observed between FA and ECF density, as well as example imaging demonstrating the trends observed.

Open in a new tab

Correlations between fractional anisotropy (FA) and extracellular fluid (ECF) density indicate a consistent negative association between the 2 variables both within contrast-enhancement and in non-enhancing FLAIR hyperintensity. Examples from individual subjects show reduced FA within areas of high predicted ECF, particularly near the primary tumor mass.

The per-subject perfusion-cellularity correlation (PCC) was shown to be evenly distributed between groups (positive: n = 134, negative: n = 124, neutral: n = 168). Higher PCC was observed for IDH mutant patients relative to IDH wild-type patients (t = 2.416, P = .016), with no differences observed based on MGMT methylation status. A positive ASL-cell density correlation per subject within FH was associated with worse survival prognosis in patients who underwent gross total resection (HR = 5.58, P = .022, Figure 6). Hypercellularity occurring in areas with increased perfusion was observed in patients with low survival, with subtle non-angiogenic hypercellularity seen in patients with longer survival.

Figure 6.

Kaplan Meier Curves showing better survival for patients with negative perfusion-cellularity correlations at the subject level, with example imaging showing a patient with an overal survival of 60 days that shows high PCC and a patient with an overall survival of 845 days that shows negative PCC.

Open in a new tab

Kaplan–Meier curve showing different survival outcomes for patients with positive (R > 0.1), negative (R < −0.1), or no (−0.1 < R < 0.1) within-FLAIR perfusion-cellularity correlation (PCC) that had undergone gross total resection in the UCSF-PDGM dataset, indicating that patients with negative PCC survive longer than patients with high PCC. Examples from individual subjects highlight this trend, where a short-term survival subject shows regions of non-enhancing tumor highlighted by both the arterial spin labeling and cell density map, whereas a long-term survivor shows reduced perfusion in an area with moderately increased cellularity.

Discussion

Here we present the spatial association between diffusion and perfusion imaging metrics and estimated cell density in gliomas, underscoring how imaging techniques may reflect underlying tumor pathology. Within contrast-enhancing regions, both rCBV and ASL metrics showed a weak but positive and statistically significant correlation with mean radio-pathomic cell density, suggesting concordance between perfusion-based and radio-pathomic methods for tumor detection in the presence of angiogenic tumor activity. Beyond the enhancing margin, no association was observed at the cohort level between perfusion estimates and cellularity, suggesting that hypercellular tumor invasion exists in the pre-angiogenic state beyond the extent of increased perfusion. However, patients that underwent a gross total resection from the UCSF cohort and had non-enhancing regions of hypercellularity with high ASL, saw worse survival outcomes, indicating that hyperperfused, hypercellular tumor in the non-enhancing region may reflect more severe tumor infiltration and was missed during surgery. Additionally, FA was seen to be negatively associated with ECF maps across both enhancing and non-enhancing portions of the brain, indicating a robust relationship across a range of underlying pathology. These results have clinical implications in terms of understanding how angiogenic activity relates to the presence of tumors both within and beyond the contrast-enhancing region and provide a framework for understanding the prognostic implications of interpreting multimodal imaging coupled with radio-pathomic maps of tumor pathology. Additionally, radio-pathomic maps of cell density trained on autopsy samples as ground truth have shown robust utility in detecting regions of non-enhancing tumors and may provide a measurable improvement in clinically delineating the full extent of the tumor.

These results indicate a robust negative association between ECF density and FA across two large publicly available datasets, indicating that increased ECF density may function as a surrogate marker for decreased FA and may indicate disruptions in white matter integrity.28–30 These disruptions may indicate directionality for active tumor propagation or areas of tumor-related necrosis, which could aid in parsing the FLAIR hyperintense region into its tumor and non-tumor components. Furthermore, pseudopalisading necrosis is a hallmark pathological feature of high-grade gliomas, and colocalization of ECF with other tumor signatures such as hypercellularity observed on cell density maps, particularly in hyperperfused areas, may predict the presence of pseudopalisading necrosis.31,32 A somewhat stronger relationship seen in the non-enhancing FLAIR hyperintense region over contrast-enhancing regions could indicate that these disruptions are easier to detect in pre-angiogenic tumor infiltration, though this relationship was observed to be generally stable across both regions of interest. While these ECF maps do not directly indicate tumor presence, these maps do show the potential to detect disruptions in white matter integrity from traditional imaging alone, which could improve scan times clinically by not having to acquire multi-directional DTI. This could improve treatment planning and response monitoring by highlighting areas of non-tumor necrosis and edema within the FLAIR-enhancing region and beyond, detecting the furthest extent of tumor-related white matter disruption. Future research examining pathology from ECF-FA mismatch regions may further highlight the biological underpinnings of this relationship. Additionally, studies of more advanced white matter tract disruption metrics may reveal radio-pathomic signatures that are able to detect pre-angiogenic tumor invasion.

This study reports largely similar results when comparing 2 large datasets as well across contrast-based and non-contrast-based estimates of perfusion. Both ASL and rCBV showed positive associations with cell density within contrast enhancement, but not within non-enhancing FLAIR hyperintensity. Notably, the lack of a significant correlation between perfusion metrics and cell density in non-enhancing FLAIR hyperintense regions suggests a limitation in the ability of current perfusion imaging to detect diffuse infiltration of tumor cells where the blood-brain barrier may remain intact. As this result is held in both exogenous and endogenous contrast-based perfusion techniques, it is likely that these regions reflect areas of the tumor that have yet to generate a hypoxic environment required to induce angiogenesis and may reflect the most regions of infiltration for the growing tumor. This distinction may point to a potential diagnostic gap in non-enhancing tumor regions, which are often challenging to delineate and assess for tumor activity. This could have implications for treatment planning, as regions of non-enhancing tumors may not be adequately targeted due to their underrepresentation in perfusion-based imaging as well as traditional imaging signatures. Recent work has highlighted the promise of supramaximal resection into the FLAIR hyperintense region to improve patient outcomes following surgery,33,34 making these radio-pathomic maps a particularly valuable tool for more precisely targeting areas of infiltration in this region while sparing eloquent tissue in conjunction with intraoperative fluorescence guidance35,36

The differences in survival based on non-enhancing per-subject perfusion-cell density associations highlight the importance of improving our understanding of occult tumor invasion. Patients who had undergone gross total resection with stronger perfusion-cellularity coupling showed worse survival outcomes, suggesting that supramarginal tumors with signs of angiogenic activity impact prognosis. Therefore, a positive association between perfusion values and cellularity in the non-enhancing margin may be seen as a more mature infiltrative tumor that has already depleted the oxygen in the surrounding tissue enough to induce angiogenesis. Understanding the impact of non-enhancing infiltration on disease progression is critical as, in many cases, these tumor regions do not receive the full extent of treatment, including being untargeted by both surgical resection and radiation despite full resection of contrast-enhancing tumor. This result supports the notion that hypercellular, hyperperfused tissue tends to have a measurable impact on patient outcomes and that this signal is detectable via combining radio-pathomic mapping techniques with perfusion-based imaging. Future research into regions of perfusion-cellularity concordance will be critical in understanding how this feature responds dynamically to different treatment strategies later in the clinical history of glioma, as anti-angiogenic treatments such as bevacizumab may impact the strength of this relationship. Particularly, examining how these perfusion-cellularity relationships relate to the furthest extent of tumor spread may reveal how functional imaging correlates to the spatial infiltration of gliomas, making it a strong candidate for future longitudinal studies of non-enhancing proliferation.

Limitations

This study is subject to several limitations inherent to the radio-pathomic model and the imaging techniques utilized. First, the accuracy of perfusion MRI metrics, including rCBV and ASL, is dependent on the precision of image acquisition and processing. Variability in scanner hardware, imaging protocols, and software algorithms can introduce inconsistencies that affect the reproducibility of results across different institutions and patient populations. Additional errors may result from using a model trained on post-treatment tissue as ground truth, as many but not all the patients included in the brain bank have undergone numerous therapies known to impact the presence of tumors on imaging. Furthermore, the previously validated radio-pathomic model that establishes the ground truth includes a time between imaging and tissue collection at autopsy which remains a source of potential error in generating the prediction maps. While some limited pathological confirmation has been assessed at the time of surgery in untreated high-grade gliomas, we intend to further validate the predictive maps within a more tightly constrained temporal window. Despite these limitations, the findings highlight critical relationships between advanced imaging signatures and radio-pathomic maps of tumor pathology that remain stable across 2 of the largest publicly available GBM datasets, as well as their impact on prognosis. Future research both in our own autopsy data and studies using biopsy tissue as ground truth should focus on refining radio-pathomic models, as well as trying to ascertain the clinical utility of such models, specifically how they can be used in challenging clinical situations such as the accurate delineation of tumor boundaries in the post-treatment setting. Additionally, fractional tumor burden quantification has been used in other work to more precisely quantify what portion of the hyperperfused region represents tumor burden and could be a future avenue for comparative research with our radio-pathomic maps of tumor burden.37,38

Conclusions

In conclusion, in the largest radio-pathomic mapping study to date, our study demonstrates that perfusion MRI metrics, specifically rCBV and ASL, show a significant association with tumor cell density within contrast-enhancing regions of glioma. This suggests their potential utility as surrogate markers for cellularity in angiogenic tumor areas. However, their limited correlation with cell density in non-enhancing FLAIR hyperintense regions highlights the need for more sensitive imaging techniques to fully capture the extent of glioma infiltration. Our future research will focus on enhancing the capabilities of radio-pathomic models by exploring multimodal imaging approaches that integrate advanced imaging signatures (perfusion, diffusion, MR spectroscopy, chemical exchange saturation transfer imaging, etc.) to provide a more accurate and comprehensive assessment of tumor biology. The findings from this study contribute to the evolving landscape of glioma imaging and underscore the importance of continued innovation in this field.

Supplementary material

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

noaf044_suppl_Supplementary_Materials
noaf044_suppl_supplementary_materials.docx (973.5KB, docx)

Acknowledgments

We would like to thank our colleagues at the University of California San Francisco and the University of Pennsylvania for curating and disseminating their glioma imaging datasets, without which this research would not be possible. We are also deeply indebted to our brain bank participants and their families for their commitment to research which has resulted in the radio-pathomic models tested in this study.

Contributor Information

Samuel A Bobholz, Department of Radiology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Daniel Aaronsen, Department of Neurosurgery, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Aleksandra Winiarz, Department of Biophysics, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Savannah R Duenweg, Department of Biophysics, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Allison K Lowman, Department of Radiology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Michael Flatley, Department of Radiology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Fitzgerald Kyereme, Department of Radiology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Jennifer Connelly, Department of Neurology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

E Kelly S Mrachek, Department of Pathology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Max O Krucoff, Department of Biomedical Engineering, Medical College of Wisconsin, Milwaukee, Wisconsin, USA; Department of Neurosurgery, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Anjishnu Banerjee, Department of Biostatistics, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Peter S LaViolette, Department of Biomedical Engineering, Medical College of Wisconsin, Milwaukee, Wisconsin, USA; Department of Radiology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA.

Funding

PSL: American Brain Tumor Association Grant DG160004, Froedtert Foundation, Strain for the Brain 5K Run, Milwaukee, WI, Advancing a Healthier Wisconsin, the Ryan M. Schaller Foundation, NIH/NCI R01CA290631, R01CA218144, R01CA218144-02S1, R21CA231892, and R01CA249882. SB: Strain for the Brain 5K Run.

Conflict of interest statement

P.S.L. holds US Patent 12171542 that protects portions of the intellectual property used in this study. The authors have no conflicts of interest relevant to this work to disclose.

Authorship statement

Study Design: S.A.B., P.S.L., and J.C.; Data Collection: S.A.B., P.S.L., A.K.L., J.C., D.C., K.M., and F.K.; Data analysis and interpretation: S.A.B., D.A., P.S.L., and S.R.D.; Manuscript drafting: D.A. and S.A.B.; Manuscript review: All authors.

Data availability

All non-publicly available data will be made available upon reasonable request to the corresponding author.

References

  • 1. Miller  KD, Ostrom  QT, Kruchko  C, et al.  Brain and other central nervous system tumor statistics, 2021. CA Cancer J Clin.  2021;71(5):381–406. [DOI] [PubMed] [Google Scholar]
  • 2. Ostrom  QT, Cioffi  G, Gittleman  H, et al.  CBTRUS statistical report: Primary brain and other central nervous system tumors diagnosed in the United States in 2012-2016. Neuro Oncol. 2019;21(suppl 5):v1–v100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Skardelly  M, Kaltenstadler  M, Behling  F, et al.  A continuous correlation between residual tumor volume and survival recommends maximal safe resection in glioblastoma patients: A nomogram for clinical decision making and reference for non-randomized trials. Front Oncol.  2021;11(December 2021):748691. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Beiko  J, Suki  D, Hess  KR, et al.  IDH1 mutant malignant astrocytomas are more amenable to surgical resection and have a survival benefit associated with maximal surgical resection. Neuro Oncol. 2014;16(1):81–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Behling  F, Rang  J, Dangel  E, et al.  Complete and incomplete resection for progressive glioblastoma prolongs post-progression survival. Front Oncol.  2022;12(February 2022):755430. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Wen  PY, Macdonald  DR, Reardon  DA, et al.  Updated response assessment criteria for high-grade gliomas: Response assessment in neuro-oncology working group. J Clin Oncol.  2010;28(11):1963–1972. [DOI] [PubMed] [Google Scholar]
  • 7. Wen  PY, Chang  SM, Van den Bent  MJ, et al.  Response assessment in neuro-oncology clinical trials. J Clin Oncol. 2017;35(21):2439–2449. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Ellingson  BM, Chung  C, Pope  WB, Boxerman  JL, Kaufmann  TJ.  Pseudoprogression, radionecrosis, inflammation or true tumor progression? Challenges associated with glioblastoma response assessment in an evolving therapeutic landscape. J Neurooncol.  2017;134(3):495–504. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9. Abdulla  S, Saada  J, Johnson  G, Jefferies  S, Ajithkumar  T.  Tumour progression or pseudoprogression? A review of post-treatment radiological appearances of glioblastoma. Clin Radiol.  2015;70(11):1299–1312. [DOI] [PubMed] [Google Scholar]
  • 10. Young  JS, Al-Adli  N, Scotford  K, Cha  S, Berger  MS.  Pseudoprogression versus true progression in glioblastoma: What neurosurgeons need to know. J Neurosurg.  2023;139(3):748–759. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Sudre  CH, Panovska-Griffiths  J, Sanverdi  E, et al.  Machine learning assisted DSC-MRI radiomics as a tool for glioma classification by grade and mutation status. BMC Med Inform Decis Mak.  2020;20(1):149. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Cha  S, Lupo  JM, Chen  MH, et al.  Differentiation of glioblastoma multiforme and single brain metastasis by peak height and percentage of signal intensity recovery derived from dynamic susceptibility-weighted contrast-enhanced perfusion MR imaging. AJNR Am J Neuroradiol.  2007;28(6):1078–1084. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Choi  YJ, Kim  HS, Jahng  GH, Kim  SJ, Suh  DC.  Pseudoprogression in patients with glioblastoma: Added value of arterial spin labeling to dynamic susceptibility contrast perfusion MR imaging. Acta Radiol.  2013;54(4):448–454. [DOI] [PubMed] [Google Scholar]
  • 14. van Dijken  BRJ, van Laar  PJ, Holtman  GA, van der Hoorn  A.  Diagnostic accuracy of magnetic resonance imaging techniques for treatment response evaluation in patients with high-grade glioma, a systematic review and meta-analysis. Eur Radiol.  2017;27(10):4129–4144. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Prager  AJ, Martinez  N, Beal  K, et al.  Diffusion and perfusion MRI to differentiate treatment-related changes including pseudoprogression from recurrent tumors in high-grade gliomas with histopathologic evidence. Am J Neuroradiol. 2015;36(5):877. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Huber  T, Bette  S, Wiestler  B, et al.  Fractional anisotropy correlates with overall survival in glioblastoma. World Neurosurg. 2016;95(November 2016):525–534.e1. [DOI] [PubMed] [Google Scholar]
  • 17. van den Elshout  R, Scheenen  TWJ, Driessen  CML, et al.  Diffusion imaging could aid to differentiate between glioma progression and treatment-related abnormalities: A meta-analysis. Insights Imaging. 2022;13(1):158. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Fioni  F, Chen  SJ, Lister  INE, Ghalwash  AA, Long  MZ.  Differentiation of high grade glioma and solitary brain metastases by measuring relative cerebral blood volume and fractional anisotropy: A systematic review and meta-analysis of MRI diagnostic test accuracy studies. Br J Radiol.  2023;96(1141):20220052. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Bobholz  SA, Lowman  AK, Brehler  M, et al.  Radio-pathomic maps of cell density identify brain tumor invasion beyond traditional MRI-defined margins. AJNR Am J Neuroradiol.  2022;43(5):682–688. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Bobholz  SA, Lowman  AK, Connelly  JM, et al.  Noninvasive autopsy-validated tumor probability maps identify glioma invasion beyond contrast enhancement. Neurosurgery.  2024;95(3):537–547. https://journals.lww.com/neurosurgery/fulltext/2024/09000/noninvasive_autopsy_validated_tumor_probability.9.aspx [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Bakas  S, Sako  C, Akbari  H, et al.  The University of Pennsylvania glioblastoma (UPenn-GBM) cohort: Advanced MRI, clinical, genomics, & radiomics. Sci Data.  2022;9(1):453. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Calabrese  E, Villanueva-Meyer  JE, Rudie  JD, et al.  The University of California San Francisco preoperative diffuse glioma MRI dataset. Radiol Artif Intell. 2022;4(6):e220058. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Ashburner  J, Friston  KJ.  Voxel-based morphometry—the methods. Neuroimage.  2000;11(6 Pt 1):805–821. [DOI] [PubMed] [Google Scholar]
  • 24. Bobholz  SA, Lowman  AK, Connelly  JM, et al.  Non-invasive tumor probability maps developed using autopsy tissue identify novel areas of tumor beyond the imaging-defined margin. Neurosurgery. 2024;95(3):537–547. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Bobholz  SA, Lowman  AK, Barrington  A, et al.  Radiomic features of multiparametric MRI present stable associations with analogous histological features in patients with brain cancer. Tomography. 2020;6(2):160–169. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Nguyen  HS, Milbach  N, Hurrell  SL, et al.  Progressing bevacizumab-induced diffusion restriction is associated with coagulative necrosis surrounded by viable tumor and decreased overall survival in patients with recurrent glioblastoma. AJNR Am J Neuroradiol.  2016;37(12):2201–2208. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. McGarry  SD, Hurrell  SL, Kaczmarowski  AL, et al.  Magnetic resonance imaging-based radiomic profiles predict patient prognosis in newly diagnosed glioblastoma before therapy. Tomography. 2016;2(3):223–228. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Şahin  S, Ertekin  E, Şahin  T, Özsunar  Y.  Evaluation of normal-appearing white matter with perfusion and diffusion MRI in patients with treated glioblastoma. MAGMA. 2022;35(1):153–162. [DOI] [PubMed] [Google Scholar]
  • 29. Salvalaggio  A, Pini  L, Gaiola  M, et al.  White matter tract density index prediction model of overall survival in glioblastoma. JAMA Neurol. 2023;80(11):1222–1231. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. Li  Y, Wang  J, Song  SR, et al.  Models for evaluating glioblastoma invasion along white matter tracts. Trends Biotechnol.  2023;42(3):293–309. [DOI] [PubMed] [Google Scholar]
  • 31. Rong  Y, Durden  DL, Van Meir  EG, Brat  DJ.  ‘Pseudopalisading’ necrosis in glioblastoma: A familiar morphologic feature that links vascular pathology, hypoxia, and angiogenesis. J Neuropathol Exp Neurol.  2006;65(6):529–539. [DOI] [PubMed] [Google Scholar]
  • 32. Aldape  K, Zadeh  G, Mansouri  S, Reifenberger  G, von Deimling  A.  Glioblastomapathology, molecular mechanisms and markers. Acta Neuropathol.  2015;129(6):829–848. [DOI] [PubMed] [Google Scholar]
  • 33. Aziz  PA, Memon  SF, Hussain  M, et al.  Supratotal resection: An emerging concept of glioblastoma multiforme surgery—systematic review and meta-analysis. World Neurosurg. 2023;179(March 2024):e46–e55. [DOI] [PubMed] [Google Scholar]
  • 34. Rossi  M, Gay  L, Ambrogi  F, et al.  Association of supratotal resection with progression-free survival, malignant transformation, and overall survival in lower-grade gliomas. Neuro Oncol. 2021;23(5):812–826. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35. Bettag  C, Schregel  K, Langer  P, et al.  Endoscope-assisted fluorescence-guided resection allowing supratotal removal in glioblastoma surgery. Neurosurg Focus.  2021;50(1):E3. [DOI] [PubMed] [Google Scholar]
  • 36. Schucht  P, Knittel  S, Slotboom  J, et al.  5-ALA complete resections go beyond MR contrast enhancement: Shift corrected volumetric analysis of the extent of resection in surgery for glioblastoma. Acta Neurochir (Wien).  2014;156(2):305–12; discussion 312. [DOI] [PubMed] [Google Scholar]
  • 37. Herings  SDA, van der Wijk  MW, von Beckerath  V, et al.  Fractional tumor burden maps increase the confidence of reading brain MR perfusion. Eur J Radiol.  2024;178(September 2024):111644. [DOI] [PubMed] [Google Scholar]
  • 38. Anil  A, Stokes  AM, Chao  R, et al.  Identification of single-dose, dual-echo based CBV threshold for fractional tumor burden mapping in recurrent glioblastoma. Front Oncol.  2023;13(January 2023):1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

noaf044_suppl_Supplementary_Materials
noaf044_suppl_supplementary_materials.docx (973.5KB, docx)

Data Availability Statement

All non-publicly available data will be made available upon reasonable request to the corresponding author.

Articles from Neuro-Oncology are provided here courtesy of Society for Neuro-Oncology and Oxford University Press

RESOURCES