Table 1.
Characteristics of the included studies. Abbreviations: P/R = Prospective/Retrospective; DWI = Diffusion Weighted Imaging; DTI = Diffusion Tensor Imaging; DKI = Diffusion Kurtosis Imaging; IVIM = Intravoxel Incoherent Motion; SE = Stretched Exponential; RSI = Restriction Spectrum Imaging; SO = Survival Outcome; ROI = Region of Interest; OS = Overall Survival; PFS = Progression-Free Survival; NAWM = Normal Appearing White Matter; CEL = Contrast Enhancing Lesion; NEL = Non-Enhancing Lesion; EOR = Extent of Resection; KPS = Karnofsky Performance Score; 2-GMM = 2 Gaussian Mixture Model; 4-GMM = 4 Gaussian Mixture Model; ND/R = Newly Diagnosed/Recurrent; RT = Radio Therapy; TMZ = Temozolomide; BV = Bevacizumab; FS = Field Strength; EPI = Echo Planar Imaging; FOV = Field of View; TR = Relaxation Time; TE = Echi Time; Seq = Sequence; ST = Slice Thickness.
| Author | Year | Np | Study Design (P/R) | Diagnosis (ND/R) |
Treatment | MRI Timepoints | Diffusion Model/s | Diffusion Acquisition Details | Diffusion Metrics Investigated | SO | ROI Info | Survival Analysis | Main Findings |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Saksena et al. [11] | 2010 | 34 | R | ND | Surgery/biopsy; RT; Chemotherapy | Preoperative | DTI | FS = 3 T, Seq.= EPI, FOV = 128 × 128 mm, TR = 17 ms, TE = 84.3 ms, ST = 5 mm, no gape, NEX = 1, gradients applied in 25 non collinear directions, b = 0, 1000 s/mm2 | FA, ADC, CL, CP, and CS (mean and min) | PFS | ROIs on FLAIR signal abnormality in CEL and NEL; manual segmentation. | Kaplan–Meier survival curves; univariate and multivariate Cox proportional hazards models adjusted for age, KPS, EOR. | Univariate analysis showed that min values of FA, MD, CP, CS were associated with PFS rate. The multivariate analysis demonstrated that only min CP was a PFS predictor. |
| Li et al. [41] | 2011 | 64 | P | ND | Surgery/biopsy; RT; Chemotherapy: 23 TMZ, 29 TMZ with tarceva, 10 poly ICLC, 2 R115777 | Postoperative: pre-treatment, post-treatment | DWI | FS = 3 T or 1.5T, Directions: 3–6, Seq. = EPI, TR/TE = 5000–10,000/63–110 ms, matrix 128 × 128 or 256 × 256, ST 3–5 mm, 21–40 slices, b 0–1000 s/mm2. | nADC median and percentiles (pre-RT, changes between preRT and post-RT) | OS, PFS | ROIs on NAWM, CEL, T2 hyperintensity lesions, and the NE lesion; automatic segmentation | Kaplan–Meier survival curves; Univariate and multivariate Cox proportional hazards models adjusted for age and field strength. | No diffusion parameters associated with OS or PFS in univariate analysis; lower nADC in CEL and higher nADC in T2 hyperintensity lesion and NEL associated with worse OS in multivariate analysis. |
| Pope et al. [42] | 2011 | 121 | R | ND | Surgery/biopsy; RT; Chemotherapy post resection: 59 TMZ and BV, 62 TMZ. Chemotherapy on recurrence: 34 add BV. | Preoperative | DWI | FS = 1.5 T, Seq. = EPI, section thickness of 3–5 mm, FOV of 24 cm, matrix size: 256 × 256 for most patients. b = 0, 1000 s/mm2. | ADC_L (2-GMM histogram) | OS, PFS | ROIs on T1CEL; semiautomatic segmentation. | Kaplan–Meier with log-rank and Wilcoxon test; uni- and multivariate Cox regression models with RPA class and MGMT methylation status | ADC values did not stratify OS and PFS in the control group; pretreatment ADC histogram analysis can stratify PFS in BV-treated patients with newly diagnosed GBM. |
| Ellingson et al. [43] | 2012 | 143 | R | ND | Surgery/biopsy; RT; Chemotherapy: TMZ | Postoperative: pretreatment, posttreatment | DWI | FS = 1.5 T, Seq. = EPI, TE/TR = 102.2/8000 ms, NEX = 1, ST = 5 mm, gap = 1 mm, matrix size = 128 × 128, FOV = 24 cm using a twice-refocused epi, b = 0 s/mm2 and b = 1000 s/mm2 | fDM metrics | OS, PFS | ROIs on regions of FLAIR signal abnormality and T1 CEL; segmentation method not specified | Log-rank analysis of Kaplan–Meier curves; Cox-regression analysis adjusted for age and KPS | Patients exhibiting a large volume of tissue with decreased ADC are statistically more likely to have a short PFS and OS. |
| Romano et al. [44] | 2012 | 47 | R | ND | Surgery; RT; Chemotherapy: TMZ following by adjuvant tmz therapy. | Preoperative | DWI | FS = 1.5 T, Seq. = EPI, b-values 0, 500, and 1000 s/mm2, ST = 5 mm; TR= 3000 ms; TE = 84 ms; gap, = 0.3mm; matrix = 256 × 256 mm, acquisition time = 1.40 min. | ADC min | OS, PFS | ROIs on T1CEL; semiautomatic segmentation. | Kaplan–Meier, log rank, uni- and multivariate Cox regression models with MGMT methylation status | patients with high ADCmin values have higher OS and PFS than patients with low ADCmin values. |
| Pope et al. [45] | 2012 | 97 | R | R | Treatments pre-recurrence: RT and TMZ; Chemotherapy on recurrence: BV or BV and CPT-11 (Irinotecan) | Pretreatment | DWI | NR | ADC_L and LCP (2-GMM histogram) | 6-PFS, overall PFS, OS | ROIs on T1CEL; semiautomatic segmentation. | The Kaplan–Meier method with log-rank test, uni- and multivariate Cox models adjusted for age and enhancing tumour volume at recurrence | ADC-L was predictive for 6-PFS, OS and PFS. LCP alone was only predictive of 6-PFS. |
| Paldino et al. [46] | 2012 | 15 | P | R | Treatments pre-recurrence: RT and TMZ; Chemotherapy on recurrence: BV and Irinotecan | Pretreatment, posttreatment | DTI | FS = 1.5 T, Seq. = EPI, TR/TE = 6000/100 ms; flip angle, 90 degrees; 4 NEX; matrix = 128 × 128; voxel size 1.72 × 1.72 × 5 mm | Changes in MD and FA mean | OS, PFS | ROIs on T1 CEL and abnormalities on FLAIR images; semiautomatic segmentation | Cox proportional hazard model. | Patients with a change in MD within FLAIR signal abnormality region had significantly shorter OS and PFS than those with no change. |
| Zikou et al. [47] | 2012 | 17 | P | ND | Surgery; RT; Chemotherapy: TMZ. | Preoperative | DTI | FS = 1.5 T, Seq. = EPI, TR:9807 ms, TE:131 ms, FOV:230 mm, acquisition matrix: 128 × 128, slice thickness: 3 mm, max b-value: 700 s/mm2, 16 non-collinear diffusion directions | normalized MD and FA | OS | ROIs on T1CEL; manual segmentation | Log-rank analysis of Kaplan–Meier curves; Multivariate Cox regression analysis not performed due to statistical non-significance. | No significant correlation was found between MD, FA and OS. |
| Sunwoo et al. [48] | 2013 | 26 | R | ND | Surgery/biopsy; RT; Chemotherapy: TMZ. | Preoperative | DWI | FS = 1.5 T, TR/TE = 6000/63 (at b 0 and 1000 s/mm2, 25 sections, bandwidth of 1953Hz/voxel, ST 5 mm, gap 1, FOV = 240 × 240 mm, a matrix = 160 × 160, voxel resolution 1.5 × 1.5 × 5.0 mm, directions 3. | ADC mean, ADC 5th percentile (histogram) | PFS | ROIs on T1CEL; manual segmentation | Kaplan–Meier | A positive significant relationship was demonstrated between PFS and the mean ADC. 5th percentile was not significantly associated with PFS. |
| Ellingson et al. [49] | 2013 | 143 | R | ND | Surgery/biopsy; RT; Chemotherapy: TMZ | Postoperative: pretreatment, posttreatment | DWI | FS = 1.5 T, TE/TR = 102.2 ms/8000 ms, NEX = 1, ST = 5 mm, gap = 1, matrix size = 128 × 128, FOV = 24 cm using a twice-refocused EPI, b 1000 s/mm2 and b 0 s/mm2. | prob-fDM metrics | OS, PFS | ROIs on regions of FLAIR signal abnormality; semiautomatic segmentation | Log-rank analysis of Kaplan–Meier curves | Patients with a large volume fraction of tumour showing a decrease in ADC through prob-fDM had a significantly shorter PFS and OS. |
| Nakamura et al. [50] | 2013 | 138 | R | ND | Surgery/biopsy; RT; Chemotherapy | Preoperative | DWI | FS = 1.5 T, Seq. = EPI, TE/TR = 3600/81 ms, ST = 5, gap = 1 mm, 128 × 128 matrix, 230 mm FOV, one acquisition, b = 1000 s/mm2 | ADC min | OS, PFS | ROIs on T1CEL; manual segmentation | Log-rank analysis of Kaplan–Meier curves; multivariate Cox regression analysis with age, KPS and surgery/biopsy | Tumours with low ADC min are associated with low PFS and OS. |
| Mohsen et al. [51] | 2013 | 25 | R | ND | Surgery; RT; Chemotherapy: TMZ. | Preoperative or immediately postoperative | DTI | FS = 1.5 T, Seq. = EPI, TR/TE: 12 k/95 ms, ST/inter-slice gap: 4/4 mm, resolution 256 × 256, 25 directions, two b = 0, 1000 s/mm2 FOV: 24 × 24 cm. | p and q maps pattern (diffuse, localised or minimally invasive | PFS | ROIs on the visible abnormality on p and q maps; manual segmentation | Log-rank analysis of Kaplan–Meier curves; univariate Cox regression analysis | Invasiveness of DTI pattern was associated with PFS. A minimal invasive pattern predicts a higher PFS. |
| Ellingson et al. [52] | 2013 | 132 | R | R | Chemotherapy: 89 patients BV; 43 variety of chemotherapies but never exposed to BV | Pretreatment | DWI | FS = 1.5 or 3 T, Seq. = EPI, TE/TR = 80–110 ms/4–10 s, 1 average, section thickness = 5 mm with gap = 1 mm, matrix size = 128 × 128, and FOV = 22–25 cm, b = 1000 and b 0 s/mm2. | ADC_L (2-GMM histogram) | OS, PFS | ROIs on T1CEL; semiautomatic segmentation | Log-rank analysis of Kaplan–Meier curves; univariate and multivariate Cox regression analysis adjusted for age, treatment cohort | Patients with lower ADC_L had a significantly longer PFS and OS compared with those having higher ADC_L. |
| Omuro et al. [53] | 2014 | 40 | P | ND | Surgery/biopsy; RT; Chemotherapy: TMZ and BV | Postoperative (pretreatment) | DWI | NR | nADC mean | PFS, 1y-OS | ROIs on T1 CEL; manual segmentation | Log-rank analysis of Kaplan–Meier curves | Lower baseline ADC was associated with prolonged OS, but not PFS. |
| Rahman et al. [54] | 2014 | 91 | R | R | Treatments pre-recurrence: standard radiation and TMZ therapy; Chemotherapy on recurrence: BV | Pretreatment | DWI | FS = 1.5, Seq. = monopolar EPI, TE/TR = 80–110 ms/4–10 ms, ST = 5 mm, gap = 1 mm, matrix size = 128 × 128 mm, FOV = 22–25 cm, b value 1000 and 0 s/mm2 | %ADC_L, %ADC_H, and ADC_L/ADC_M (4-GMM histogram) | OS, PFS | ROIs on T1 CEL NE T2/FLAIR abnormality; automatic segmentation | Kaplan–Meier curves; uni- and multivariable analysis with Cox proportional hazards model adjusted for clinical variables | Baseline ADC_L/ADC_M within NE T2/FLAIR volume and ADC_H within T1 CEL can stratify OS and PFS. |
| Wen at al. [55] | 2015 | 36 | R | ND | Surgery/biopsy; RT; Chemotherapy: TMZ, erlotinib and BV. | Postoperative: pretreatment and posttreatment at 1 month, 2 months and every 2 months (up to a maximum of 14 months) | DWI | FS = 3 T, b = 1000 (dir = 6, NEX = 4) and ADC maps were calculated using in-house developed software. | ADC percentiles (histogram); 2-GMM histogram metrics; fDM metrics | OS, PFS | ROIs on T1 CEL and T2/FLAIR hyperintensity; semiautomatic segmentation | Kaplan–Meier curves; Univariate and multivariate Cox regression analysis adjusted for age, KPS, EOR | ADC10% within the T2L at 2 months was strongly associated with OS and PFS. fDM metrics showed an association with OS and PFS within the CEL when considered by univariate analysis, but not in the T2L. |
| Coban et al. [12] | 2015 | 58 | R | ND | Surgery; RT; Chemotherapy | Preoperative | DWI | FS = 3 T, Seq. = EPI, acceleration factor of 2, FOV = 22 × 22 cm2; b 0, 1000 s/mm2, section thickness = 3 mm; number of sections = 40; acquisition time = 8 min. | ADC min | 15 months OS | ROIs on T1 CEL and visually low ADC; manual segmentation | ROC analysis, Kaplan–Meier curves | ADC min was not useful for differentiating patients having short or long survival. |
| Elson et al. [56] | 2015 | 52 | R | ND | Surgery; RT; Chemotherapy | Postoperative | DWI | NR | ADCmean, ADCmin, nADCmean, nADCmin | OS, PFS | ROIs on hyperintense T2/FLAIR; manual segmentation | Log-rank analysis on Kaplan–Meier data; multivariate Cox regression analysis adjusted for age, EOR, KPS | Regression analysis indicated that normalized ADC values provide the strongest association with PFS and OS. |
| Lee et al. [57] | 2015 | 24 | R | ND | Surgery; RT; Chemotherapy: TMZ | Postoperative | DWI | FS = 3 T, Seq. = EPI, b-values of 0 and 1000 s/mm2, three orthogonal directions. | nADC (histogram metrics) | PFS | ROIs on T1 CEL; manual segmentation | Log-rank analysis on Kaplan–Meier data | nADC not associated with PFS |
| Zhang et al. [58] | 2015 | 52 | R | R | Treatments pre-recurrence: surgical resection + RT + TMZ; Chemotherapy on recurrence: BV | Pretreatment; 2 posttreatment scans | DWI | FS = 1.5 or 3 T, Seq. = EPI, b = 0 and 3 diffusion-weighted acquisitions with b 1000 s/mm2 | low-ADC volume and percent change, normalized 5th percentile low ADC values and percent changes | OS | ROIs on T1 CEL and FLAIR hyperintense abnormality corresponding to low ADC signal; manual segmentation | Kaplan–Meier curves; uni- and multivariate Cox regression analysis with clinical and imaging metrics | At the second post-BV scan, the volume of the low-ADC lesion was inversely associated with OS. Normalized 5th percentile low-ADC value and its percent change were not associated with OS. |
| Jamjoom et al. [59] | 2015 | 46 | R | ND | No surgery; 4 treatment groups | Preoperative | DTI | FS = 3 T, Seq. = EPI, Acceleration factor of 2, b = 0 and b = 1000, six directions, TR = 2435–4813 ms, TE = 48–62 ms, voxel size 1.6 × 1.6 × 5 mm, FOV = 230 × 180 × 159 mm. 15 directions, TR = 3175–8000 ms, TE = 57–90 ms, voxel size 2 × 2 × 3.3 mm, FOV 224 × 224 × 105 mm | MDmin (from MD map); histogram metrics (from MD gradient maps) | OS | ROIs on T1 CEL that visually appeared dark on the MD maps; semiautomatic segmentation | Univariate and multivariate Cox regression analysis adjusted for treatment protocol and gender | Lower minMD and higher MD gradient values for the 10th and 75th percentile of the tumour boundary predict short OS. |
| Wen at al. [60] | 2015 | 75 | R | ND | Surgery/biopsy; 44 RT; Chemotherapy: TMZ and enzastaur, 31 TMZ + erlotinib and BV | Postoperative: pretreatment, posttreatment (after 1, 2 and 4 months) | DTI | FS = 3 T, six-directional, Seq. = DWI, b = 1000 s/mm2, number of excitations = 4. | MD, FA and longitudinal and radial eigenvalues (histogram metrics) | OS, PFS | ROIs on T1 CEL and T2 hyperintense lesions; manual segmentation | Log-rank analysis on Kaplan–Meier data; multivariate Cox regression analysis adjusted for age, EOR, KPS | For the TMZ + enza cohort: volumes of regions with low MD values at 1-month scan associated with OS and at 2-month scan associated with PFS. For the TMZ + erl + bev cohort, volumetric diffusion parameters and MD and EVrad were associated with OS and PFS at different timepoints. |
| Chang et al. [61] | 2015 | 120 | R | ND | Surgery/biopsy; RT; Chemotherapy: TMZ following by adjuvant tmz therapy. | Postoperative | DWI | FS = 1.5 T, TE/TR = 80–120 ms/5000 ms, matrix size = 128 128, ST = 3 mm with no interslice gap, and b-values of 0 and 1000 s/mm2 in three orthogonal directions. | ADC_L, ADC_H (2-GMM histogram) | OS, PFS | ROIs on CEL on T1 subtraction images; segmentation method not specified | Log-rank analysis on Kaplan–Meier data and multivariate Cox regression analysis adjusted for age | Patients with lower ADC_L have shorter OS and PFS. ADC_H was not predictive. |
| Burth et al. [62] | 2016 | 125 | R | ND | Surgery/biopsy; Radiotherapy and Chemotherapy: 5 different treatment regimens | Preoperative | DWI | FS = 3 T, TR = 5300 ms, TE = 90 ms, b 0 and b 1200, pixel size 1.769 mm/1.769 mm, image matrix 130 × 130, ST 5 mm, flip angle 908, FoV = 229 × 229 mm. | ADC histogram metrics | OS, PFS | ROIs on T1 CEL and T2/FLAIR hyperintensity; semiautomatic segmentation | Univariate and multivariable Cox regression analyses including age, sex, EOR, KPS, rCBV | Univariate analysis showed that 10th percentile ADC in CEL and T2/FLAIR were significantly associated with OS, but not with PFS. In multivariable analysis diffusion-derived MRI parameters did not predict survival. |
| Shankar et al. [63] | 2016 | 84 | R | ND | Surgery/biopsy; RT; Chemotherapy: TMZ | Preoperative | DWI | FS = 1.5 T, Seq. = EPI. TR = 8000 ms, TE = 73.6 ms, FOV = 260 mm, matrix size = 160 × 192, section ST = 5 mm, gap = 1.5 mm, b = 0 and b = 1000 in three orthogonal directions. | nADC min | OS | Whole tumour volume identified on T1 CEL and FLAIR; restricted diffusion ROIs identified on ADC map; manual segmentation | Log-rank analysis on Kaplan–Meier data and multivariate Cox regression analysis | Positive association between nADC min and OS. |
| Van der Hoorn et al. [64] | 2016 | 14 | R | ND | Surgery; RT; Chemotherapy: TMZ and adjuvant TMZ | Postoperative: preradiotherapy, postradiotherapy | DWI | FS = 1.5 T, Seq. = EPI, TR/TE = 6000–12,500/64–108 ms; flip angle 90°; FOV 220–300 × 220–300 mm; 52–66 slices; 0–4 mm slice gap; voxel size 0.86–1.2 × 0.86–1.2 × 4–5 mm, b-value of 0 and 1000 s/mm2, scanned in 3–25 directions. | nADC histogram metrics | OS, PFS | ROIs automatically segmented in periresectional area and manually adjusted. | Univariate and multivariate Cox regression analysis adjusted for age and MGMT methylation status | The increase in ADC value postradiotherapy in comparison to preradiotherapy did not predict an increase in PFS or OS neither in univariate nor multivariate analysis. |
| Chang et al. [65] | 2016 | 126 | R | R | Treatments pre-recurrence: surgical resection + RT + TMZ; Chemotherapy on recurrence: BV | Pretreatment, posttreatment | DWI | Seq. = Monopolar EPI, TE/TR = 80–110 ms/4–10 s, section thickness = 5 mm, gap = 1 mm, matrix size = 128 × 128 mm, FOV= 22–25 cm, b-value 1000 and 0 s/mm2. | ADC (texture, 4-GMM histogram metrics) | OS, PFS | ROIs on T1 CEL, T2/FLAIR; semiautomatic segmentation | Machine-learning predictive model based on random-forest and including conventional MRI and DWI metrics | Model based on multiparametric MRI imaging metrics (of which DWI) was able to predict OS |
| Zolal et al. [39] | 2016 | 31 | R | R | Surgery; RT; Chemotherapy: TMZ | Preoperative (prior to second surgery) | DWI | FS = 1.5 T, b = 0 and 1000 s/mm2, ST of 5 mm, and voxel sizes between 0.9 and 2 mm. | ADC histogram metrics | OS, PFS, Survival after 2nd surgery | ROIs in T1 CEL (manual selection or semi-automated adaptive thresholding) | Log-rank analysis on Kaplan–Meier data and multivariate Cox regression analysis including also age, EOR, tumour size | ADC histogram skewness associated with OS and PFS in univariate analysis and with survival after 2nd surgery in multivariate analysis. |
| Choi et al. [66] | 2016 | 112 | R | ND | Surgery; RT; Chemotherapy: TMZ. | Preoperative | DTI | FS = 3 T, b values of 600 s/mm2 and 0 s/mm2, 32 directions, FOV = 8413.4/77; 220 mm; section thickness = 2 mm, matrix = 112 × 3 × 112. | MD histogram metrics | 12-OS, 16-OS, 12-PFS | ROIs on T1 CEL; semiautomatic segmentation | Log-rank analysis on Kaplan–Meier data and multivariate Cox regression analysis with MGMT methylation status, age, KPS, EOR | At univariate analysis, lower MD histogram parameters were significant predictors of poor OS and PFS; Multivariable models with MD parameters had significantly higher performances that those without MD parameters for OS and PFS prediction. |
| Huber et al. [67] | 2016 | 122 | R | ND | Surgery; RT; Chemotherapy: TMZ. | Preoperative | DTI | FS = 3 T, DTI direction 15 or 6 directions | mean ADC, FA | OS | ROIs in the CEL, central region (CR), and the FLAIR-hyperintense NE peritumoral region | Kaplan–Meier curves; multivariate Cox regression analysis with age, KPS, tumour volume, infiltration | Patients with low FA values in CEL showed a significantly improved OS in univariate analysis. In multivariate analysis FA values could not be identified as independent prognostic parameters besides clinical factors. |
| Yan et al. [68] | 2016 | 31 | R | ND | Surgery; Chemotherapy: TMZ. | Preoperative, postoperative | DTI | FS = 3 T, Seq. = EPI, TR/TE = 8300/98 ms; flip angle 90°; FOV 192 × 192 mm; 63 slices; no slice gap; and voxel size 2 × 2 × 2 mm, b-values = 0, 350, 650, 1000, 1300, and 1600 s/mm2, 13 directions. | EOR (extent of resection) based on p and q maps | OS, PFS | ROIs representing EOR manually placed on T1 | Univariate and multivariate Cox regression analysis including age, MGMT methylation status, IDH-1 mutation, tumour volume and location | larger residual abnormal q volume predicted significantly shorter PFS; larger resection of abnormal q area improved OS. |
| Puig et al. [25] | 2016 | 15 | P | ND | Surgery; RT; Chemotherapy: TMZ | Preoperative | IVIM MRI | FS = 1.5, Seq. = EPI, slice = 24, TR = 3000 ms, TE = 76 ms, EPI factor was 41, FOV = 200 mm, section thickness = 5 mm, matrix = 96 × 77 mm, pixel size = 2.4 × 2.9 × 5 mm.13 b-values: 0, 10, 20, 30, 50, 100, 150, 200, 350, 500, 650, 800, and 1000 s/mm2, acquisition time was 3 min 48 s per patient. | D, D*, f | OS | ROIs in T1 CEL and NEL; manual segmentation | Kaplan–Meier curves; multivariate Cox regression analysis with clinical and DSC metrics | f and D* in CEL are associated with 6 months survival |
| Kondo et al. [69] | 2017 | 76 | R | ND | NS | Preoperative | DWI | NR | L-ADC_L, B-ADC_L, B&L-ADC_L (2-GMM histogram) | OS, PFS | ROIs in T1 CEL manual segmentation | Kaplan–Meier curves; univariate Cox regression analysis | B&L-ADCL was strongly associated with poor PFS and OS |
| Krishnan et al. [28] | 2017 | 45 | R | ND | Surgery | Postoperative (pretreatment) | DWI, RSI | FS = 3 T, Seq. = EPI, TE/TR = 96 ms/17 ms, FOV = 24 cm, matrix = 96 × 966 × 48, voxel size = 2.5 mm, 4 b-values (b 0, 500, 1500, and 4000 s/mm2, 6 and 15 unique diffusion directions for each nonzero b-value, respectively 8 min scan time. | ADC and RSI volume fraction, 10th and 90th percentile | OS, PFS | 3D ROIs on T1 CEL and FLAIR hyperintensity; semiautomatic segmentation | Univariate and multivariate Cox regression analysis combined with age, gender and resection type | No ADC metrics were associated with PFS and OS. RSI volume fraction was associated with PFS and OS, RSI 90th percentile associated with OS. |
| Ellingson et al. [70] | 2017 | 258 | R | R | Chemotherapy: 5 different regimens | Pretreatment | DWI | FS = 1.5 or 3 T, Seq. = monopolar EPI, TE/TR = 80–110 ms/4–10 s, NEX = 1, ST = 5 with 0–1 mm interslice gap, matrix size = 128 × 128, FOV = 220–256 mm. b = 0 and b = 1000 s/mm2. | ADC_L (2-GMM histogram) | OS | 3D ROIs on T1 subtraction maps; semiautomatic segmentation | Log-rank analysis and multivariate Cox regression analysis including age, enhancing tumour volume | Pretreatment ADC_L was an independent predictive biomarker for OS in anti-VEGF therapies, but not in lomustine. |
| Galla et al. [71] | 2017 | 65 | R | R | Chemotherapy: BV | Pretreatment, posttreatment | DWI | FS = 3 T, Seq. = EPI, FOV= 24 × 24 cm2, b= 0, 1000 s/mm2, ST = 5 mm. | changes in nADC mean and min | OS | ROIs on ADC maps corresponding to the T1 CEL and NEL; manual segmentation | Cox regression analysis adjusted for age, tumour size, prior treatments | The change in mean ADC was significantly associated with OS. The change in min ADC in the NEL was not associated with OS |
| Boonzaier et al. [72] | 2017 | 43 | R | ND | Surgery; RT; Chemotherapy | Preoperative | DTI | FS = 3 T, Seq. EPI, TR/TE = 8300/98; flip angle = 90°; FOV = 192 × 3 × 192 mm; 63 sections; no section gap; voxel size = 2.0 × 3.2 × 3.2 mm; b values 0–1000 s/mm2. | volumetric analysis: ADC low volume | OS, PFS | ROIs in CEL and NEL regions based on ADC-rCBV maps; manual segmentation | Log-rank analysis and multivariate Cox regression accounting for age, CE volume, IDH-1 mutation, MGMT methylation status, EOR | Volumetric analysis of ADC-rCBV ROIs in NEL helps in stratifying PFS and OS |
| Heiland et al. [73] | 2017 | 21 | P | ND | Surgery | Preoperative | DTI | NR | AD, RD, mean MD and FA | OS, PFS | ROIs on the whole CEL; manual segmentation | The Kaplan–Meier, univariate Cox-Regression tests | Patients with high MD in the contrast-enhancing region had a significantly better OS. Patients with low FA in the contrast-enhancing region had a significantly better OS. |
| Chakhoyan et al. [26] | 2018 | 23 | P | ND | Surgery; RT; Chemotherapy: TMZ | Postoperative: pretreatment, posttreatment | DWI, DKI, SE | FS = 3 T, Seq. EPI, TR/TE = 13,400/103 ms, a flip angle of 90°, 52 contiguous slices, ST = 3 mm, no inter-slice gap, an in-plane resolution of 2 mm × 2 mm, matrix = 128 × 128 mm, b values: 0, 50, 100, 250, 500, 750, 1000, 2500, 3500 and 5000 s/mm2. The total acquisition time for the DWI scan was 6 min. | ADC, K, D, DDC, alpha | OS | ROIs placed in NAWM and CEL; semiautomatic segmentation | Log-rank analysis on Kaplan–Meier curves and multivariate Cox regression analysis including age, MGMT status, tumour volume at baseline | ADC show significant value in predicting OS. DKI and SE metrics did not show significant value in OS prediction. |
| Li et al. [24] | 2018 | 115 | P | ND | Surgery; Adjuvant therapy postoperative | Preoperative | DTI | NR | Joint histogram features from Normalized DTI-p and q maps | OS, PFS | ROIs in CEL and NEL; manual segmentation | Kaplan–Meier, Cox regression accounting for IDH-1 mutation, MGMT methylation status, sex, age, EOR | joint histogram features were associated with OS and PFS and improved survival model performance. |
| Bae et al. [74] | 2018 | 217 | R | ND | Surgery; RT; Chemotherapy: TMZ | Preoperative | DTI | FS = 3 T, b values 600 and 0 s/mm2, 32 directions, FOV = 8413.4/77; 220 mm; section thickness = 2 mm; matrix 112 × 3 × 112. | Radiomic features | OS, PFS | ROIs in necrosis, CEL, NEL on T2/FLAIR; semiautomatic segmentation | Random Survival Forest prediction model from multiparametric MRI | Radiomic prediction model including clinical and multiparametric MRI metrics (of which DTI metrics) was able to predict OS and PFS |
| Rulseh et al. [75] | 2019 | 43 | R | ND | Surgery; RT; Chemotherapy: TMZ | Preoperative | DWI | NR | ADC histogram metrics | OS, PFS | Whole-brain masks generated using FSL | The Kaplan–Meier, univariate Cox-Regression analysis | Median ADC was the best variablefor PFS prediction, while p85 was the best variable for OS prediction. |
| Kim et al. [76] | 2019 | 93 | R | ND | Surgery/biopsy; RT; Chemotherapy: TMZ | Preoperative | DWI | FS = 1.5 T, Seq. = EPI, TR/TE 3000 ms/80 ms; FOV = 240 × 240 mm; matrix = 164 × 162; ST = 5 mm; gap = 1 mm, b value 0 and 1000 s/mm2. | ADC histogram metrics | OS | ROIs on T1CE and FLAIR; manual segmentation | Log-rank test on Kaplan–Meier curves, unsupervised K-means clustering | ADC histogram parameters demonstrated a significant association with OS. |
| Buemi et al. [77] | 2019 | 17 | R | R | Chemotherapy: 13 BV, 4 fotemustine and BV | Pretreatment, posttreatment | DWI | FS = 1.5 T, Seq = EPI, TE/TR = 90 ms/1000 ms, NEX 2, slice thickness 5 mm with 1 mm interslice distance, matrix size = 320 × 320 mm, and FOV = RL 240 mm, AP 282 mm, FH 131 mm, b = 0 and b = 1000 s/mm2. | ADC_L, ADC_H (2-GMM histogram) | OS, PFS | ROIs on T1 CEL and T2/FLAIR abnormalities; semiautomatic segmentation | Kaplan–Meier with log-rank test, multivariate Cox regression adjusted for clinical variables | In univariate analysis, ADC_L in CEL was significantly predictive of PFS and OS. In multivariate analysis, the ADC_L was predictive for PSF but not OS. |
| Petrova et al. [78] | 2019 | 54 | R | R | Chemotherapy: RT and TMZ Chemotherapy on recurrence: BV | Pretreatment | DWI | FS = 1.5 T or 3 T, b = 0 and b = 1000, matrix = 128 × 128, FOV = 22–24 cm with a ST of 5 mm. | ADC histogram metrics | OS, 6PFS | ROIs on T1 CE; manual segmentation | 6 machine learning classifiers | Diffusion and perfusion imaging using an SVM was able to predict 6PFS. Less power was shown to predict OS. |
| Flores-Alvarez et al. [79] | 2019 | 36 | R | ND | Surgery; RT; Chemotherapy | Preoperative | DTI | FS = 3 T, FOV of 22 × 22 mm2, b-value of 1000 s/mm2, 25 directions, TR = 10.000 ms, TE = 101.8 ms, ST of 3 mm and a Matrix array of 112 × 112. | FA | OS | ROIs in necrosis, CEL, oedema, normal controlateral, peritumoral oedema; manual segmentation | Log-rank test on Kaplan–Meier curves | Significant association between the peritumoral oedema measurement of FA with intervals of OS. |
| Li et al. [80] | 2019 | 80 | P | ND | Surgery/biopsy; RT (17.4%, 20/115); Chemotherapy: TMZ (73.0%, 84); Best supportive care (9.6%, 11/115); | Preoperative | DTI | NR | histogram analysis of normalized DTI-p and q maps | OS, PFS | ROIs in CEL and NEL; manual segmentation | Kaplan–Meier and Cox regression accounting for IDH-1, mutation, MGMT methylation status, sex, age, EOR, tumour volume | A higher mean value of anisotropic diffusion (q) in NE region was associated with worse OS and PFS. 5 p and q histogram metrics showed significantly incremental value in predicting 12-month OS and PFS. |
| Mohan et al. [81] | 2019 | 48 | R | ND | Surgery; RT; Chemotherapy: TMZ. Other therapies: BV (8 patients) | Preoperative | DTI | FS = 3 T, Seq. = Epi, parallel acquisition (GRAPPA), acceleration factor of 2.30 Directions, a b-value of 1000 s/mm2, a total acquisition time of 8 min, TR/TE 5000/86 ms, NEX = 3, FOV = 22 × 22 cm2, slice thickness = 3 mm, number of sections = 40. | mean FA, MD | OS | ROIs in CC if there were CE lesions on T1 or signal abnormality on T2 and FLAIR; manual segmentation | Kaplan–Meier with log-rank test, multivariate Cox regression adjusted for clinical variables | FA from the invaded CC was positively correlated with OS. |
| Yan et al. [82] | 2019 | 51 | R | ND | Surgery; RT; Chemotherapy: TMZ. | Preoperative, postoperative | DTI | FS = 3 T, Seq. = EPI, TR/TE = 8300/98 ms; flip angle 90°; FOV = 192 mm2; 63 slices; no slice gap; voxel size 2 mm3, b values (0, 350, 650, 1000, 1300 and 1600 s/mm2), scanned in 12 directions. | EOR (extent of resection) based on p and q maps | OS, PFS | ROIs representing EOR on pre-MR; manual segmentation | Kaplan–Meier with log-rank test | Larger abnormal q regions showed better PFS and OS |
| Park et al. [83] | 2020 | 248 | R | ND | Surgery; RT; 60 Gy Chemotherapy: TMZ | Preoperative | DWI | FS = 3 T, Seq. = EPI, TR/TE = 3000/56 ms; b = 0 and b = 1000 s/mm2, FOV= 25 cm; ST/gap = 5 mm/2 mm; matrix = 256 × 256; acquisition time, 39 s. | radiomic features | OS | ROIs on T1 CEL; semiautomatic segmentation | Log-rank test on Kaplan–Meier curves; radiomics predictive models | Multiparametric MR model (incorporating also ADC features) was able to predict OS |
| Song et al. [84] | 2020 | 19 | R | R | Before recurrence: RT + TMZ Chemotherapy on recurrence: immune checkpoint inhibitors (ICIs) | Pretreatment, posttreatment | DWI | FS = 3 T, Seq. = EPI, TR/TE = 4025/82 ms, b value 0 and 1000 s/mm2. | nADC, changes between pre and post treatment | PFS6 | ROIs on T1 CEL; semiautomatic segmentation | Univariate analysis | nADC was able to assess PFS6 |
| Patel et al. [40] | 2020 | 67 | R | R | 35 treated with BV; 35 repeated surgery Chemotherapy: 19% BV monotherapy; 81% BV + TMZ and small molecular inhibitors. | Pretreatment | DWI | FS = 1.5 or 3 T, Seq. = EPI, ST = 3 mm with no interslice gap, b-values of 0, 500 and 1000 s/mm2) or diffusion tensor imaging with 64 directions. | ADC_L (2-GMM histogram) | OS | ROIs on T1 CEL; manual segmentation | Log-rank analyses on Kaplan–Meier data and Cox proportional hazard models adjusted for age, tumour volume | ADCL was an independent predictor of OS in the BV cohort, but not the surgical cohort. |