Advertisement

Applying MAP-MRI to Identify the WHO Grade and Main Genetic Features of Adult-type Diffuse Gliomas: A Comparison of Three Diffusion-weighted MRI Models

Open AccessPublished:November 05, 2022DOI:https://doi.org/10.1016/j.acra.2022.10.009

      Rationale and Objectives

      Currently, there is no noninvasive method to effectively judge the genotype of diffuse gliomas. We explored the association between mean apparent propagator-MRI (MAP-MRI) and WHO grade 2/3, IDH 1/2 mutations, and chromosome 1p/19q combined deletion genotypes in adult-type diffuse gliomas and compared it with the diagnostic efficiency of diffusion tensor imaging (DTI) and diffusional kurtosis imaging (DKI).

      Materials and Methods

      We prospectively recruited 67 participantshistopathologically diagnosed with adult-type diffuse gliomas. Routine MRI, DKI, and DSI were performed before surgery. The extreme and average partial diffusion indexes of solid tumors were measured. A comprehensive assessment of statistically significant diffusion parameters was performed after Bonferroni correction, including ROC curves, correct classification percentage (CCP), integrated discrimination improvement (IDI), net reclassification improvement (NRI), and k-fold cross validation.

      Results

      For differentiating WHO grade 2/3, q-space inverse variance (QIV), mean kurtosis (MK), non-Gaussianity (NG), and return to the origin probability (RTOP) were different (p’ < .05), with the mean QIV exhibiting the best diagnostic efficacy and stability (AUC = 0.973, CCP = 0.906). We observed significant differences in mean diffusivity (MD), mean square displacement, QIV, MK, and RTOP between the IDH wild-type and IDH mutant groups (p’ < .001) (AUC, 0.806–0.978) and MAP-MRI showed a higher IDI than DTI and DKI (0.094–0.435, NRI > 0, respectively). For the chromosome 1p/19q combined deletion, the minimum QIV was different between the overall (p’ < .05) and no significant differences  in MD and MK was observed.

      Conclusion

      MAP-MRI effectively predicts the WHO grade 2/3, IDH 1/2 mutations, and chromosome 1p/19q combined deletion in adult-type diffuse gliomas, and it may perform better than DTI and DKT.

      Abbreviations:

      1p/19q (synchronous deletion of the short arm of chromosome 1 and long arm of chromosome 19), CCP (correct classification percentage), DKI (diffusional kurtosis imaging), DSI (diffusion spectrum magnetic resonance imaging), FA (fractional anisotropy), IDH (isocitrate dehydrogenase), IDHmut/1p19qdel (IDH mutant and chromosome 1p19q synchronous deletion), IDHmut/1p19qint (IDH mutant and chromosome 1p19q intact), IDHwt (IDH wild-type), IDI (integrated discrimination improvement), MAP-MRI (mean apparent propagation diffusion MRI), MD (mean diffusivity), MK (mean kurtosis), MSD (mean square displacement), NG (non-Gaussianity), NRI (net reclassification improvement), QIV (q-space inverse variance), ROI (region of interest), RTOP (return to the origin probability)

      INTRODUCTION

      Gliomas account for 80% of all primary central nervous system malignancies (
      GBD 2016 Brain and Other CNS Cancer Collaborators
      Global, regional, and national burden of brain and other CNS cancer, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016.
      ). Even when surgery, concurrent radiotherapy and chemotherapy are actively used, patients usually have a poor prognosis (
      • Reni M
      • Mazza E
      • Zanon S
      • et al.
      Central nervous system gliomas.
      ). Previous studies have clarified part of the genetic basis of gliomas, information that has helped classify gliomas and has provided prognostic and predictive data for patients (
      • Ceccarelli M
      • Barthel FP
      • Malta TM
      • et al.
      Molecular profiling reveals biologically discrete subsets and pathways of progression in diffuse glioma.
      ,
      • Yan H
      • Parsons DW
      • Jin G
      • et al.
      IDH1 and IDH2 mutations in gliomas.
      ,
      • van den Bent MJ
      • Brandes AA
      • Taphoorn MJ
      • et al.
      Adjuvant procarbazine, lomustine, and vincristine chemotherapy in newly diagnosed anaplastic oligodendroglioma: long-term follow-up of EORTC brain tumor group study 26951.
      ). The 2021 edition of the World Health Organization (WHO) classification of central nervous system tumors increases the importance of the genetic features of diffuse gliomas, compared to the 2016 edition, and may enable reclassification of the histologic grading in certain characteristic cases (
      • Louis DN
      • Perry A
      • Wesseling P
      • et al.
      The 2021 WHO classification of tumors of the central nervous system: a summary.
      ). Advanced knowledge of the histological and genetic characteristics of the tumor is a fundamental and necessary step to optimize the treatment process, because the scope of surgical resection and subsequent concurrent radiotherapy and chemotherapy are usually based on this information (
      • Jiang T
      • Nam DH
      • Ram Z
      • et al.
      Clinical practice guidelines for the management of adult diffuse gliomas.
      ). Currently, histopathological analysis and genetic testing of tumor specimens are regarded as the gold standards for glioma analysis. However, in some cases, tumor tissues may not be effectively obtained, or genetic testing may not be possible. Because of the heterogeneity of these tumors, a small amount of pathological material may lead to a misdiagnosis in the classification, grading and genetic feature detection of gliomas (
      • McGirt MJ
      • Woodworth GF
      • Coon AL
      • et al.
      Independent predictors of morbidity after image-guided stereotactic brain biopsy: a risk assessment of 270 cases.
      ) or incorrect results caused by the detection technology itself (
      • Woehrer A
      • Hainfellner JA.
      Molecular diagnostics: techniques and recommendations for 1p/19q assessment.
      ). These factors affect the diagnosis, treatment and prognosis of this disease. Therefore, a feasible technology or imaging method is required that can effectively provide genotype information.
      The application of diffusion models in neuro-oncology is extensive (
      • Smits M.
      MRI biomarkers in neuro-oncology.
      ). Several studies have used various diffusion models to genotype gliomas (
      • Maynard J
      • Okuchi S
      • Wastling S
      • et al.
      World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
      ,
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ,
      • Hempel JM
      • Schittenhelm J
      • Brendle C
      • et al.
      Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
      ,
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ,
      • Chu JP
      • Song YK
      • Tian YS
      • et al.
      Diffusion kurtosis imaging in evaluating gliomas: different region of interest selection methods on time efficiency, measurement repeatability, and diagnostic ability.
      ). Some conventional imaging features are thought to be helpful in predicting the hereditary features of gliomas, such as MRI anatomical imaging (
      • Maynard J
      • Okuchi S
      • Wastling S
      • et al.
      World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
      ). However, the accurate manual extraction of imaging features relies on radiologists, which may lead to errors in prediction due to subjective bias, particularly for some patients whose imaging features are not prominent. Thus, an imaging method with high robustness and accuracy is needed to predict the molecular genotype of diffuse gliomas. Diffusion tensor imaging (DTI) is a technique based on DWI that expresses the approximate Gaussian distribution of water molecules in tissues. Tan et al. (
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ) reported that DTI distinguishes the IDH genotype of gliomas. However, of particular concern is that the movement of water molecules in the actual tissue is affected by various factors, and the distribution is non-Gaussian. Diffusion kurtosis imaging (DKI) is believed to describe the non-Gaussian distribution of water protons in the brain (
      • Jensen JH
      • Helpern JA
      • Ramani A
      • et al.
      Diffusional kurtosis imaging: the quantification of non-Gaussian water diffusion by means of magnetic resonance imaging.
      ). However, to date, there has been little agreement on whether DKI can distinguish between the IDH 1/2 mutation and chromosome 1p/19q combined deletion genotypes in diffuse gliomas (
      • Hempel JM
      • Schittenhelm J
      • Brendle C
      • et al.
      Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
      ,
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ).
      Mean apparent propagator-MRI (MAP-MRI) is a newer computational framework based on the acquisition of q-space data for diffusion spectrum magnetic resonance imaging (DSI) (
      • Özarslan E
      • Koay CG
      • Shepherd TM
      • et al.
      Mean apparent propagator (MAP) MRI: a novel diffusion imaging method for mapping tissue microstructure.
      ,
      • Avram AV
      • Sarlls JE
      • Barnett AS
      • et al.
      Clinical feasibility of using mean apparent propagator (MAP) MRI to characterize brain tissue microstructure.
      ,
      • Fick RHJ
      • Wassermann D
      • Caruyer E
      • et al.
      MAPL: tissue microstructure estimation using Laplacian-regularized MAP-MRI and its application to HCP data.
      ). The technology includes DTI, which may be used to not only evaluate the non-Gaussian distribution of water molecules in brain tissue but also effectively measure the probability density function of spin displacement and quantify useful indicators of the probability density function. These indexes reflect the diffusion of protons in complex microstructures (such as diffusion confinement and multiple chambers). MAP-MRI has been reported to be superior to DTI (
      • Le H
      • Zeng W
      • Zhang H
      • et al.
      Mean apparent propagator MRI is better than conventional diffusion tensor imaging for the evaluation of Parkinson's disease: a prospective pilot study.
      ) and has been used to detect changes in diffusion parameters caused by different grades or different genetic features of diffuse gliomas (
      • Wang P
      • Weng L
      • Xie S
      • et al.
      Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma.
      ,
      • Sun Y
      • Su C
      • Deng K
      • et al.
      Mean apparent propagator-MRI in evaluation of glioma grade, cellular proliferation, and IDH-1 gene mutation status.
      ), but the old grouping method may lead to incorrect clinical practical applications. The aim of this study was to describe the association between MAP-MRI and the histological and main genetic features of diffuse gliomas by applying the latest WHO classification criteria method, including WHO grade 2/3, IDH1/2 mutations and 1p/19q combined deletion genotypes, and to further compare its diagnostic efficacy with DTI and DKI.

      MATERIALS AND METHODS

      Participants and Clinical Data

      We conducted a prospective study and recruited participants who visited our hospital between June 2018 and September 2021. The study was conducted in accordance with the Declaration of Helsinki. The ethics committee of our hospital approved the research protocol (Number: WZ 2022019), and all participants signed an informed consent form before the examination. The inclusion criteria for this study were as follows: (1) adult-type diffuse gliomas were pathologically diagnosed according to the 2021 WHO standards, (2) scan sequences included conventional MRI and diffusion scans (the diffusion imaging scans included at least one DKI or DSI sequence due to long scan times or a poor participant status; when the participant's cooperation was low, the principle of randomness was adopted) with ideal image quality, (3) surgery was performed within 3 months after the scan, and (4) genotyping tests for tumor correlation were performed. The exclusion criteria were participants with glioma who had received preoperative treatment (including steroids, radiotherapy, chemotherapy, or concurrent radiotherapy and chemotherapy) and gliomas in which it was difficult to draw the region of interest (ROI). An overview of the participants selection process is shown in Figure 1.
      In a previous study, 36 participants with diffuse glioma were included, and the study compared the differences in MAP-MRI parameters between groups of participants with different grades of diffuse glioma (
      • Wang P
      • Weng L
      • Xie S
      • et al.
      Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma.
      ). In the present study, 27 participants from this previous study were included. We classified IDH wild-type participants with histologic grade 2/3 and TERT promoter mutation as the grade 4 group according to the 2021 WHO criteria, Finally, three such participants were included in the study, and the remaining hereditary features that might increase the histologic grade will be examined in future studies. According to the IDH 1/2 mutation and 1p/19q combined deletion genotypes, the participants were divided into three groups: participants were first divided into the IDH wild type (IDHwt) group and IDH mutant type (IDHmut) group, and then the IDHmut group was divided into IDH mutant and 1p19q intact (IDHmut/1p19qint) group and IDH mutant and 1p19q synchronous deletion (IDHmut/1p19qdel) group.

      MRI

      All participants underwent preoperative MRI using a 3T scanner (MAGNETOM Skyra; Siemens Healthcare, Erlangen, Germany) equipped with a 32-channel head/neck coil. The conventional MRI sequences included axial T1-weighted, axial T2-weighted, axial T2-weighted FLAIR, axial DWI and 3D contrast-enhanced T1-weighted images after the intravenous administration of 0.1 mmol/kg gadobutrol (Gadovist, Bayer AG, Berlin, Germany).
      The DKI sequence was obtained in the axial plane using the following parameters: TR/TE = 4200/101 ms, FOV = 342 mm × 342 mm, GRAPPA = 2, layer thickness = 5.0 mm, voxel size = 2.7 × 2.7 × 5.0 mm3, 20 layers, b value = 0, 1000, and 2000 s/mm2, and total scanning time = 9 min and 12 s. The DSI sequence was obtained in the axial plane using a half q-space Cartesian grid sampling procedure with the following parameters: TR/TE = 7000/107 ms, FOV = 260 mm × 260 mm, GRAPPA = 2, layer thickness = 3.0 mm, voxel size = 2.2 × 2.2 × 3.0 mm3, 50 layers, b value = 0–3000 s/mm2, and total scanning time = 15 min and 40 s. Detailed information on the parameters is provided in Supplementary Table E1.

      Data Processing

      The images were corrected for eddy and motion in advance using DiffusionKit eddy tools (
      • Xie S
      • Chen L
      • Zuo N
      • et al.
      DiffusionKit: a light one-stop solution for diffusion MRI data analysis.
      ). Diffusion kurtosis estimator (DKE) software (
      • Tabesh A
      • Jensen JH
      • Ardekani BA
      • et al.
      Estimation of tensors and tensor-derived measures in diffusional kurtosis imaging.
      ) was used to postprocess the DKI image to obtain the mean diffusivity (MD) map, fractional anisotropy (FA) map of DTI and mean kurtosis (MK) map of DKI. MAP-MRI parameters were calculated using NeuDiLab, a software developed in-house with Python based on the open-source tool DIPY (Diffusion Imaging in Python) (
      • Garyfallidis E
      • Brett M
      • Amirbekian B
      • et al.
      Dipy, a library for the analysis of diffusion MRI data.
      ). The MAP-MRI parameters investigated included the mean square displacement (MSD), non-Gaussianity (NG), q-space inverse variance (QIV) and return to the origin probability (RTOP). Three diffusion models can account for multiple diffusion indicators, and we selected several of these indicators that may be more representative of diffusion characteristics for analysis. Image coregistration was performed using 3D-Slicer. The ITK-SNAP software was used to manually delineate the tumor parenchymal area (
      • Yushkevich PA
      • Piven J
      • Hazlett HC
      • et al.
      User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability.
      ), and the ROI was selected to avoid cystic degeneration, necrosis, hemorrhage, and calcification as much as possible. If the tumor was enhanced (contrast agent uptake was present), the tumor-enhancing area was selected, and T2 FLAIR imaging was used to exclude adjacent normal tissues and edema around the tumor; if the tumor was not enhanced and the signal in the parenchymal area was not uniform, the area with the lowest ADC image signal in the tumor parenchyma visible with the naked eye was selected as the ROI (Fig 2) (
      • Maynard J
      • Okuchi S
      • Wastling S
      • et al.
      World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
      ,
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ,
      • Chu JP
      • Song YK
      • Tian YS
      • et al.
      Diffusion kurtosis imaging in evaluating gliomas: different region of interest selection methods on time efficiency, measurement repeatability, and diagnostic ability.
      ). The analysts adopted the blinding principle for the basic information and clinical history of the participants in this study, and two neuroradiologists (Jinlong He and Shenghui Xie, with 12 and 11 years of experience, respectively) analyzed and measured all the images. The value of each parameter in any substantial part of the tumor was measured ten times, and the total average value of each parameter was calculated as the average value. The three lowest or largest values among the values for 10 parameters were chosen, and the ROI-added area was greater than 20 mm2 to avoid systematic errors as much as possible. Next, the average of the three values was calculated as the extreme value of the parameters. The intraclass correlation coefficient (ICC) was calculated using a two-way random effects model to analyze the agreement between two neuroradiologists. Correlation analysis was performed on the diffusion indicators with sufficient consistency (ICC > 0.8), and the average value of the parameters measured by the two radiologists was recorded as the final value.
      Figure 2
      Figure 2Selection method of region of interest.
      (a–h) The axial T2, ADC map, FLAIR, and T1-CE show the method used to select the ROI and determine the mean ADC (area of tumor parenchymal enhancement, blue and red) and minimum ADC (area with the lowest ADC among mean ADC, blue, also black arrow). Note that the pictures only show one layer of the images and round ROIs were chosen because this method can be replicated in most picture archiving and communication systems (Color version of figure available online).

      Pathology

      All the tissue samples were prepared as paraffin blocks and analyzed at our institution's pathology department using the latest methodology consistent with the WHO 2021 guideline on histopathology and immunohistochemistry (
      • Louis DN
      • Perry A
      • Wesseling P
      • et al.
      The 2021 WHO classification of tumors of the central nervous system: a summary.
      ). The pathologist (Lixin Weng) who analyzed the images had 23 years of work experience and was blinded to the clinical information and imaging results. A one-step method (multiplex PCR amplification combined with next-generation sequencing [NGS)) was used to detect the IDH1/2 mutation, 1p/19q combined deletion genotypes and TERT promoter mutation (
      • Higa N
      • Akahane T
      • Yokoyama S
      • et al.
      A tailored next-generation sequencing panel identified distinct subtypes of wildtype IDH and TERT promoter glioblastomas.
      ,
      • Zacher A
      • Kaulich K
      • Stepanow S
      • et al.
      Molecular diagnostics of gliomas using next generation sequencing of a glioma-tailored gene panel.
      ). See the supporting materials for specific steps.

      Statistical Analysis

      The statistical software packages SPSS 24.0 (SPSS, Inc., Chicago, IL, USA), Medcalc Version 19.7.4 (MedCalc Software Ltd, Ostend, Belgium) and R version 4.1.2 were used to analyze the data. For the data that conformed to a normal distribution and equal variances (determined by Levene's test for the homogeneity of variance), an independent-samples T test was used for comparison. If the aforementioned conditions were not met, the Mann–Whitney U test was used for analysis. The Bonferroni correction was used to correct for multiple comparisons, using P' instead of P, and receiver operating characteristic (ROC) curves and the correct classification percentage (CCP) (
      • Li J
      • Jiang B
      • Fine JP.
      Multicategory reclassification statistics for assessing improvements in diagnostic accuracy.
      ) were determined for the parameters that were statistically significant (p' < .05). For significant diffusion parameters, we evaluated Cohen's d. Previous studies have suggested a correlation between the three models (
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ,
      • Avram AV
      • Sarlls JE
      • Barnett AS
      • et al.
      Clinical feasibility of using mean apparent propagator (MAP) MRI to characterize brain tissue microstructure.
      ,
      • Wu YC
      • Field AS
      • Alexander AL.
      Computation of diffusion function measures in q-space using magnetic resonance hybrid diffusion imaging.
      ,
      • Ning L
      • Westin CF
      • Rathi Y.
      Estimating diffusion propagator and its moments using directional radial basis functions.
      ), and thus we compared them by performing Spearman's rank correlation analysis (Supplementary Table E6, Fig 4). Differences in efficacy between the parameters were assessed using the DeLong test, integrated discrimination improvement (IDI) and net reclassification improvement (NRI) (
      • Li J
      • Jiang B
      • Fine JP.
      Multicategory reclassification statistics for assessing improvements in diagnostic accuracy.
      ). We aimed to more robustly evaluate the diagnostic efficacy, stability, and generalizability of diffusion parameters by performing k-fold cross validation (K = 5).

      RESULTS

      Participant Distribution

      During the study period, 291 participants with suspected glioma were evaluated. Finally, 224 participants were excluded, and 67 participants with diffuse glioma (mean age, 50 ± 12 years [standard deviation); 35 men) were included from June 2018 to September 2021. Twenty-four of these participants had undergone only one DKI or DSI scan, and thus the final test set included 55 sets of images. The characteristics of the participants are shown in Table 1.
      Table 1Participant Demographics
      All Adult-Type Diffuse Glioma SubtypesIDHwtIDHmut/1p19qintIDHmut/1p19qdelp
      All (%)6725(37.3)17(25.4)25(37.3)
      Sex (male/female)
      Age and sex from the participant or their family.
      35/3213/128/914/11.852
      Age (mean ± SD, years)
      Age and sex from the participant or their family.
      50±1255±1348±1046±11.010
      2021 WHO grade
      2221111
      320614
      425250
      Participants with images
      Due to long scan times or a poor participant status, diffusion imaging scans included at least one DKI or DSI sequence. When the participant's cooperation was low, adopted the principle of random.
      21578
      317611
      423230-
      1p/19q, synchronous deletion of the short arm of chromosome 1 and long arm of chromosome 19; IDH, isocitrate dehydrogenase; IDHmut/1p19qdel, IDH mutant and 1p19q synchronous deletion; IDHmut/1p19qint, IDH mutant and 1p19q intact; IDHwt, IDH wild-type; WHO, World Health Organization.
      # Age and sex from the participant or their family.
      low asterisk Due to long scan times or a poor participant status, diffusion imaging scans included at least one DKI or DSI sequence. When the participant's cooperation was low, adopted the principle of random.

      Diffusion Parameters Quantify the Molecular Type of Diffuse Glioma

      Twenty-five enhancements in 25 IDHwt tumors, nine enhancements in 17 IDHmut/1p19qint tumors, and 18 enhancements in 25 IDHmut/1p19qdel tumors were identified. The interobserver reproducibility of all the diffusion parameters was good (ICC, 0.857–0.993) (Online Supplemental Data).
      In the correlation comparison of the three diffusion model parameters, a positive correlation was found between MD, MSD, and QIV with correlation coefficients ranging from 0.485 to 0.902. A positive correlation was found between MK, NG, and RTOP with correlation coefficients ranging from 0.522 to 0.863. A negative correlation was found between MD, MSD, and QIV and MK, NG, and RTOP with correlation coefficients ranging from −0.362 to −0.987. No significant correlation was found between FA and the other parameters (p < .05) (Online Supplemental Data).

      WHO 2 and WHO 3

      In grade 2/3 adult-type diffuse gliomas, the average and minimum QIV values of the grade 3 group were significantly lower than those of the grade 2 group (p’ < .01), and the average and maximum MK, NG, and RTOP values were significantly higher than those in the grade 2 group (p’ < .05). The MD, MSD and FA values showed nonsignificant results (p’ > .05) (Online Supplemental Data and Figure 3, Figure 4). Cohen's d values ranged from 1.835–2.548. The average QIV had both the highest AUC and CCP (0.973 and 0.906, respectively), and was very stable. NG may be superior to MK in predicting the tumor grade (IDI = 0.408 and 0.266, respectively; NRI = 0.063 and 0.031, respectively), but the instability of diagnostic efficacy is a concern.
      Figure 3
      Figure 3Boxplot showing the distribution of the diffusion parameters for IDH wild-type, IDH mutant and 1p19q intact, IDH mutant and 1p19q synchronous deletion, and WHO grade 2/3 adult-type diffuse gliomas.
      IDHmut/1p19qdel, IDH mutant and 1p19q synchronous deletion; IDHmut/1p19qint, IDH mutant and 1p19q intact; IDHwt, IDH wild-type; MD, mean diffusivity; MK, mean kurtosis; MSD, mean square displacement; NG, non-Gaussianity; QIV, q-space inverse variance; RTOP, return to the origin probability (Color version of figure available online).
      Figure 4
      Figure 4ROC curve analysis for diffusion parameters.
      IDHmut/1p19qdel, IDH mutant and 1p19q synchronous deletion; IDHmut/1p19qint, IDH mutant and 1p19q intact; IDHwt, IDH wild-type; MD, mean diffusivity; MK, mean kurtosis; MSD, mean square displacement; NG, non-Gaussianity; QIV, q-space inverse variance; RTOP, return to the origin probability (Color version of figure available online).

      IDHwt and IDHmut

      In all the adult-type diffuse gliomas, the average and minimum MD, MSD, and QIV values of the IDHwt group were significantly lower than those of the IDHmut group (p’ < .001), and the average and maximum MK, NG, and RTOP values were significantly higher than those in the IDHmut group (p’ < .001). The FA values showed nonsignificant results (p’ > .05) (Online Supplemental Data and Fig 3). Cohen's d values ranged from 1.348–2.236. The ROC curve and its characteristics are shown in Figure 4 and Supplementary Table E5, with values ranging from 0.834 to 0.978, where QIV had the highest CCP. MSD, QIV and RTOP showed a higher IDI than MD (0.189–0.435 and NRI > 0, respectively), but seemed to produce the mean and extreme values with the same effect (Online Supplemental Data). The fivefold cross validation results are shown in Supplementary Table E7. The maximum RTOP value had the highest AUC (0.911 ± 0.071). The maximum NG value had the lowest accuracy (0.653 ± 0.364).

      IDHmut/1p19qint and IDHmut/1p19qdel

      In all the adult-type diffuse gliomas, only the minimum QIV value of the IDHmut/1p19qdel group was lower than that of the IDHmut/1p19qint group (p’ < .05), and none of the remaining diffusion parameters showed differences between groups (p’ > .05). Cohen's d was 1.181. However, after excluding the effect of grade, the average QIV and RTOP appeared able to identify the 1p/19q combined deletion genotype (p’ < .05). The AUC and CCP of the minimum QIV were not as high as expected (0.806 and 0.563, respectively), but high robustness (0.870 ± 0.105) was observed after fivefold cross validation.

      DISCUSSION

      In this study, the average and extreme values of the diffusion parameters of solid tumors were measured simultaneously, and the diagnostic performance of three diffusion-weighted models (including MAP-MRI, DTI, and DKI) was compared for adult-type diffuse gliomas (AUC, 0.806–0.978; CCP, 0.563–0.909) with histological grade 2/3, IDH 1/2 mutation, and 1p/19q combined deletion genotypes. Our results show that the multishell MAP-MRI model effectively distinguishes the histological and main genetic features of adult-type diffuse gliomas based on the latest WHO classification criteria, especially QIV. MD and FA are not able to be used to discriminate histological grade 2/3, while the stability of MK is poor. According to the present study, MD and MK do not effectively predict the 1p/19q combined deletion genotype. These results may be facilitated by the assumption that MAP-MRI does not rely on a priori models, unlike the Gaussian and non-Gaussian distribution models of DTI and DKI.
      In most studies, the rationale for standardization is to eliminate individual differences (
      • Maynard J
      • Okuchi S
      • Wastling S
      • et al.
      World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
      ,
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ,
      • Hempel JM
      • Schittenhelm J
      • Brendle C
      • et al.
      Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
      ,
      • Sun Y
      • Su C
      • Deng K
      • et al.
      Mean apparent propagator-MRI in evaluation of glioma grade, cellular proliferation, and IDH-1 gene mutation status.
      ). We believe that the parameters corresponding to the solid part of the tumor depend mainly on the tumor itself rather than the damaged white matter of the brain. In our study, no difference was observed in the FA values among the groups, which demonstrated this hypothesis. However, some researchers have revealed residual fiber bundles within tumors (
      • Seow P
      • Hernowo AT
      • Narayanan V
      • et al.
      Neural Fiber Integrity in High- Versus Low-Grade Glioma using Probabilistic Fiber Tracking.
      ). Standardization still needs to be discussed. In addition, we do not know whether the white matter of the contralateral hemisphere is infiltrated by tumor cells (
      • Price SJ
      • Jena R
      • Burnet NG
      • et al.
      Improved delineation of glioma margins and regions of infiltration with the use of diffusion tensor imaging: an image-guided biopsy study.
      ,
      • Pallud J
      • Varlet P
      • Devaux B
      • et al.
      Diffuse low-grade oligodendrogliomas extend beyond MRI-defined abnormalities.
      ), or whether diffuse imaging is affected by the spatial resolution and cannot be used to effectively describe small changes in the solid part of a tumor (
      • Zhang J
      • van Zijl PC
      • Laterra B
      Unique patterns of diffusion directionality in rat brain tumors revealed by high-resolution diffusion tensor MRI.
      ). Therefore, we did not correct for these parameters. Shortening the scanning time is necessary. However, simultaneous multislice inspection may bias the extraction of diffusion features (
      • Muftuler LT
      • Nencka AS
      • Koch KM
      Diffusion propagator metrics are biased when simultaneous multi-slice acceleration is used.
      ), and singleshell diffusion scanning may be less affected than multishell model. The selection of rationalized scanning parameters requires further investigation.
      Our study revealed that water molecules in IDH wild-type adult-type diffuse gliomas were more restricted than those in IDH mutants, consistent with previous studies (
      • Maynard J
      • Okuchi S
      • Wastling S
      • et al.
      World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
      ,
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ,
      • Hempel JM
      • Schittenhelm J
      • Brendle C
      • et al.
      Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
      ,
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ,
      • Sun Y
      • Su C
      • Deng K
      • et al.
      Mean apparent propagator-MRI in evaluation of glioma grade, cellular proliferation, and IDH-1 gene mutation status.
      ). Combined with the findings reported by Wang et al. (
      • Wang P
      • Weng L
      • Xie S
      • et al.
      Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma.
      ), we postulate that MSD effectively assesses the IDH status irrespective of rank because it explicitly calculates second-order moment tensors of the mean apparent propagator (
      • Wu YC
      • Field AS
      • Alexander AL.
      Computation of diffusion function measures in q-space using magnetic resonance hybrid diffusion imaging.
      ,
      • Ning L
      • Westin CF
      • Rathi Y.
      Estimating diffusion propagator and its moments using directional radial basis functions.
      ). This same conclusion is maintained under the new classification and may be related to the epigenetics, metabolism, and redox homeostasis of tumor cells, which change the tumor extracellular volume (
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ,
      • Sun Y
      • Su C
      • Deng K
      • et al.
      Mean apparent propagator-MRI in evaluation of glioma grade, cellular proliferation, and IDH-1 gene mutation status.
      ). IDHwt glioma has the worst prognosis (
      • Ceccarelli M
      • Barthel FP
      • Malta TM
      • et al.
      Molecular profiling reveals biologically discrete subsets and pathways of progression in diffuse glioma.
      ,
      • Yan H
      • Parsons DW
      • Jin G
      • et al.
      IDH1 and IDH2 mutations in gliomas.
      ,
      • van den Bent MJ
      • Brandes AA
      • Taphoorn MJ
      • et al.
      Adjuvant procarbazine, lomustine, and vincristine chemotherapy in newly diagnosed anaplastic oligodendroglioma: long-term follow-up of EORTC brain tumor group study 26951.
      ), which indirectly shows the reliability of our conjecture. We propose that this finding might represent a breakthrough, as previous diffusion parameters have been influenced by the glioma grade (
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ,
      • Hempel JM
      • Schittenhelm J
      • Brendle C
      • et al.
      Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
      ,
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ,
      • Wang P
      • Weng L
      • Xie S
      • et al.
      Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma.
      ), resulting in conflicting clinical applications.
      Our study did not find that MD is useful to identify WHO grade 2/3 tumors, possibly because typically IDH mutant gliomas show a larger extracellular volume as a percentage of the tumor volume than IDH wild-type gliomas (
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ,
      • Sun Y
      • Su C
      • Deng K
      • et al.
      Mean apparent propagator-MRI in evaluation of glioma grade, cellular proliferation, and IDH-1 gene mutation status.
      ), thereby reducing the extent to which MD captures this part of the microstructure as a quantitative parameter in the Gaussian model. This characteristic allows the remaining tumor biomorphology within the group to be classified as relatively uniform after the exclusion of IDH wild-type gliomas. Numerical overlap has been described in previous studies; thus, complete reliance on diffusion parameters for classification or molecular type identification is impossible (
      • Maynard J
      • Okuchi S
      • Wastling S
      • et al.
      World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
      ,
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ,
      • Hempel JM
      • Schittenhelm J
      • Brendle C
      • et al.
      Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
      ,
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ,
      • Wang P
      • Weng L
      • Xie S
      • et al.
      Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma.
      ). We also encountered this problem in our study. However, we found that QIV did not show obvious numerical overlap in some of the groups. As a surrogate indicator of MSD, QIV has a similar contrast to MSD in different tissue types (
      • Ning L
      • Westin CF
      • Rathi Y.
      Estimating diffusion propagator and its moments using directional radial basis functions.
      ), and QIV is significant and stable in predicting the 1p/19q combined deletion. Reductions in RTOP are associated with axonal damage to fiber bundles, accompanied by an increase in isotropic tissue (
      • Avram AV
      • Sarlls JE
      • Barnett AS
      • et al.
      Clinical feasibility of using mean apparent propagator (MAP) MRI to characterize brain tissue microstructure.
      ). After comparing the white matter of the contralateral side (
      • Wang P
      • Weng L
      • Xie S
      • et al.
      Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma.
      ), we surmise that this change reflects diffuse tumor infiltration with an increased cell density in the extracellular matrix, which has a certain correlation with tumor heterogeneity. After eliminating the effect of the grade, RTOP also identifies the 1p/19q combined deletion, probably because RTOP appears to reflect cellularity and restrictions better than MD (
      • Avram AV
      • Sarlls JE
      • Barnett AS
      • et al.
      Clinical feasibility of using mean apparent propagator (MAP) MRI to characterize brain tissue microstructure.
      ). The accuracy of the minimum QIV or maximum RTOP alone for predicting the IDH mutation status of adult-type diffuse gliomas (AUC = 0.97) surpassed that of ADC (DWI; AUC maximum, 0.83) (
      • Maynard J
      • Okuchi S
      • Wastling S
      • et al.
      World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
      ), MD (DTI; AUC maximum, 0.93) (
      • Tan WL
      • Huang WY
      • Yin B
      • et al.
      Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
      ) and KA (neurite orientation dispersion and density imaging; AUC maximum, 0.76) (
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ). Based on these results, QIV and RTOP may strongly correlate with the IDH1/2 mutation and 1p/19q combined deletion.
      Oligodendroglial tumors with the 1p/19q combined deletion often show calcification on conventional images, but we found that MK does not identify the 1p/19q combined deletion, similar to the conclusion reported by Figini (
      • Figini M
      • Riva M
      • Graham M
      • et al.
      Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
      ). However, Hempel et al. (
      • Hempel JM
      • Schittenhelm J
      • Brendle C
      • et al.
      Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
      ), Chu et al. (
      • Chu JP
      • Song YK
      • Tian YS
      • et al.
      Diffusion kurtosis imaging in evaluating gliomas: different region of interest selection methods on time efficiency, measurement repeatability, and diagnostic ability.
      ), and Gao et al. (
      • Gao A
      • Zhang H
      • Yan X
      • et al.
      Whole-Tumor Histogram Analysis of Multiple Diffusion Metrics for Glioma Genotyping.
      ) documented the opposite results, which may be caused by the difference in b value and ROI selection. The scan time is usually also reduced because the NG may be more sensitive to local reference frame estimates, which may be influenced by differences in the subject's head orientation and motion (
      • Avram AV
      • Sarlls JE
      • Barnett AS
      • et al.
      Clinical feasibility of using mean apparent propagator (MAP) MRI to characterize brain tissue microstructure.
      ). However, in a similar study where the scan time was reduced, NG still did not seem to perform very well (
      • Gao A
      • Zhang H
      • Yan X
      • et al.
      Whole-Tumor Histogram Analysis of Multiple Diffusion Metrics for Glioma Genotyping.
      ).
      Our research has some limitations. First, participants with grade 4 IDHmut glioma were not included in this study due to the insufficient sample size, which may have resulted in some erroneous results. However, the proportion of all our participants with glioma was approximately the same as their natural epidemiological rate. Therefore, we will continue to collect relevant samples or overcome this limitation by conducting multicenter research. In addition, in future studies, we will evaluate the direct acquisition of DTI and DKI parameter maps through the postprocessing and analysis of DSI. When the scan parameters are the same, the differences between DTI, DKI and MAP-MRI can be evaluated. This approach may effectively avoid the difference in results caused by the use of different scanning parameters. Finally, an integrated diagnostic model should be built in the future that includes information such as conventional clinical information and conventional image features.
      In conclusion, mean apparent propagation diffusion-MRI is useful to identify adult-type diffuse gliomas with WHO grade 2/3, IDH 1/2 mutation, and 1p/19q combined deletion genotypes with higher diagnostic efficacy and robustness than DTI and DKI. Although it may not currently exhibit a detection performance that exceeds that of pathology, the noninvasiveness of imaging-based detection is still beneficial for the diagnosis and treatment of patients with glioma.

      COMPLIANCE WITH ETHICAL STANDARDS

      Guarantor

      The scientific guarantor of this publication is Yang Gao.

      Statistics and Biometry

      One of the authors has significant statistical expertise.
      No complex statistical methods were necessary for this paper.

      Informed Consent

      Only if the study is on human subjects:
      Written informed consent was obtained from all subjects (participants) in this study.

      Ethical Approval

      Institutional Review Board approval was obtained.

      Study Subjects or Cohorts Overlap

      Some study subjects or cohorts have been previously reported in “Wang P, Weng L, Xie S, et al. Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma. Eur J Radiol 2021;138:109622.”

      Methodology

      Methodology:
      • prospective
      • diagnostic or prognostic study
      • performed at one institution

      ACKNOWLEDGMENTS

      The authors thank Fei Wang, Haitao Ju, Xuefei Bai, Ru Wang, and Bo Li, who provided valuable assistance with the clinical studies.

      FUNDING

      This work was supported by the Inner Mongolia Autonomous Region Science and Technology Plan Project [grant number 2019GG047 ], Natural Science Foundation of Inner Mongolia Autonomous Region [grant number 2018MS08078 ], Doctor Starts Funds in Affiliated Hospital of Inner Mongolia Medical University [grant number NYFY BS 202108 ], and Natural Science Foundation of Inner Mongolia Autonomous Region [grant number 2020MS08051 ].

      Appendix. Supplementary materials

      References

        • GBD 2016 Brain and Other CNS Cancer Collaborators
        Global, regional, and national burden of brain and other CNS cancer, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016.
        Lancet Neurol. 2019; 18: 376-393
        • Reni M
        • Mazza E
        • Zanon S
        • et al.
        Central nervous system gliomas.
        Crit Rev Oncol Hematol. 2017; 113: 213-234
        • Ceccarelli M
        • Barthel FP
        • Malta TM
        • et al.
        Molecular profiling reveals biologically discrete subsets and pathways of progression in diffuse glioma.
        Cell. 2016; 164: 550-563
        • Yan H
        • Parsons DW
        • Jin G
        • et al.
        IDH1 and IDH2 mutations in gliomas.
        N Engl J Med. 2009; 360: 765-773
        • van den Bent MJ
        • Brandes AA
        • Taphoorn MJ
        • et al.
        Adjuvant procarbazine, lomustine, and vincristine chemotherapy in newly diagnosed anaplastic oligodendroglioma: long-term follow-up of EORTC brain tumor group study 26951.
        J Clin Oncol. 2013; 31: 344-350
        • Louis DN
        • Perry A
        • Wesseling P
        • et al.
        The 2021 WHO classification of tumors of the central nervous system: a summary.
        Neuro Oncol. 2021; 23: 1231-1251
        • Jiang T
        • Nam DH
        • Ram Z
        • et al.
        Clinical practice guidelines for the management of adult diffuse gliomas.
        Cancer Lett. 2021; 499: 60-72
        • McGirt MJ
        • Woodworth GF
        • Coon AL
        • et al.
        Independent predictors of morbidity after image-guided stereotactic brain biopsy: a risk assessment of 270 cases.
        J Neurosurg. 2005; 102: 897-901
        • Woehrer A
        • Hainfellner JA.
        Molecular diagnostics: techniques and recommendations for 1p/19q assessment.
        CNS Oncol. 2015; 4: 295-306
        • Smits M.
        MRI biomarkers in neuro-oncology.
        Nat Rev Neurol. 2021; 17: 486-500
        • Maynard J
        • Okuchi S
        • Wastling S
        • et al.
        World Health Organization grade II/III glioma molecular status: prediction by MRI morphologic features and apparent diffusion coefficient.
        Radiology. 2020; 296: 111-121
        • Tan WL
        • Huang WY
        • Yin B
        • et al.
        Can diffusion tensor imaging noninvasively detect IDH1 gene mutations in astrogliomas? A retrospective study of 112 cases.
        AJNR Am J Neuroradiol. 2014; 35: 920-927
        • Hempel JM
        • Schittenhelm J
        • Brendle C
        • et al.
        Effect of perfusion on diffusion kurtosis imaging estimates for in vivo assessment of integrated 2016 WHO glioma grades: a cross-sectional observational study.
        Clin Neuroradiol. 2018; 28: 481-491
        • Figini M
        • Riva M
        • Graham M
        • et al.
        Prediction of isocitrate dehydrogenase genotype in brain gliomas with MRI: single-shell versus multishell diffusion models.
        Radiology. 2018; 289: 788-796
        • Chu JP
        • Song YK
        • Tian YS
        • et al.
        Diffusion kurtosis imaging in evaluating gliomas: different region of interest selection methods on time efficiency, measurement repeatability, and diagnostic ability.
        Eur Radiol. 2021; 31: 729-739
        • Jensen JH
        • Helpern JA
        • Ramani A
        • et al.
        Diffusional kurtosis imaging: the quantification of non-Gaussian water diffusion by means of magnetic resonance imaging.
        Magn Reson Med. 2005; 53: 1432-1440
        • Özarslan E
        • Koay CG
        • Shepherd TM
        • et al.
        Mean apparent propagator (MAP) MRI: a novel diffusion imaging method for mapping tissue microstructure.
        Neuroimage. 2013; 78: 16-32
        • Avram AV
        • Sarlls JE
        • Barnett AS
        • et al.
        Clinical feasibility of using mean apparent propagator (MAP) MRI to characterize brain tissue microstructure.
        Neuroimage. 2016; 127: 422-434
        • Fick RHJ
        • Wassermann D
        • Caruyer E
        • et al.
        MAPL: tissue microstructure estimation using Laplacian-regularized MAP-MRI and its application to HCP data.
        Neuroimage. 2016; 134: 365-385
        • Le H
        • Zeng W
        • Zhang H
        • et al.
        Mean apparent propagator MRI is better than conventional diffusion tensor imaging for the evaluation of Parkinson's disease: a prospective pilot study.
        Front Aging Neurosci. 2020; 12563595
        • Wang P
        • Weng L
        • Xie S
        • et al.
        Primary application of mean apparent propagator-MRI diffusion model in the grading of diffuse glioma.
        Eur J Radiol. 2021; 138109622
        • Sun Y
        • Su C
        • Deng K
        • et al.
        Mean apparent propagator-MRI in evaluation of glioma grade, cellular proliferation, and IDH-1 gene mutation status.
        Eur Radiol. 2022; 32: 3744-3754
        • Xie S
        • Chen L
        • Zuo N
        • et al.
        DiffusionKit: a light one-stop solution for diffusion MRI data analysis.
        J Neurosci Methods. 2016; 273: 107-119
        • Tabesh A
        • Jensen JH
        • Ardekani BA
        • et al.
        Estimation of tensors and tensor-derived measures in diffusional kurtosis imaging.
        Magn Reson Med. 2011; 65: 823-836
        • Garyfallidis E
        • Brett M
        • Amirbekian B
        • et al.
        Dipy, a library for the analysis of diffusion MRI data.
        Front Neuroinform. 2014; 8: 8
        • Yushkevich PA
        • Piven J
        • Hazlett HC
        • et al.
        User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability.
        Neuroimage. 2006; 31: 1116-1128
        • Higa N
        • Akahane T
        • Yokoyama S
        • et al.
        A tailored next-generation sequencing panel identified distinct subtypes of wildtype IDH and TERT promoter glioblastomas.
        Cancer Sci. 2020; 111: 3902-3911
        • Zacher A
        • Kaulich K
        • Stepanow S
        • et al.
        Molecular diagnostics of gliomas using next generation sequencing of a glioma-tailored gene panel.
        Brain Pathol. 2017; 27: 146-159
        • Li J
        • Jiang B
        • Fine JP.
        Multicategory reclassification statistics for assessing improvements in diagnostic accuracy.
        Biostatistics. 2013; 14: 382-394
        • Wu YC
        • Field AS
        • Alexander AL.
        Computation of diffusion function measures in q-space using magnetic resonance hybrid diffusion imaging.
        IEEE Trans Med Imaging. 2008; 27: 858-865
        • Ning L
        • Westin CF
        • Rathi Y.
        Estimating diffusion propagator and its moments using directional radial basis functions.
        IEEE Trans Med Imaging. 2015; 34: 2058-2078
        • Seow P
        • Hernowo AT
        • Narayanan V
        • et al.
        Neural Fiber Integrity in High- Versus Low-Grade Glioma using Probabilistic Fiber Tracking.
        Acad Radiol. 2021; 28: 1721-1732
        • Price SJ
        • Jena R
        • Burnet NG
        • et al.
        Improved delineation of glioma margins and regions of infiltration with the use of diffusion tensor imaging: an image-guided biopsy study.
        AJNR Am J Neuroradiol. 2006; 27: 1969-1974
        • Pallud J
        • Varlet P
        • Devaux B
        • et al.
        Diffuse low-grade oligodendrogliomas extend beyond MRI-defined abnormalities.
        Neurology. 2010; 74: 1724-1731
        • Zhang J
        • van Zijl PC
        • Laterra B
        Unique patterns of diffusion directionality in rat brain tumors revealed by high-resolution diffusion tensor MRI.
        Magn Reson Imaging. 2007; 58: 454-462
        • Muftuler LT
        • Nencka AS
        • Koch KM
        Diffusion propagator metrics are biased when simultaneous multi-slice acceleration is used.
        Magn Reson Imaging. 2022; 86: 46-54
        • Gao A
        • Zhang H
        • Yan X
        • et al.
        Whole-Tumor Histogram Analysis of Multiple Diffusion Metrics for Glioma Genotyping.
        Radiology. 2022 Mar; 302: 652-661