Table 1.
Participant demographics, study characteristics and outcomes of the included studies.
| Training Set | Validation Set | Performance * | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| First Author (Year of Publication) (Reference) |
N | Age (Mean ± SD) |
Gender (Male-Female) |
N | Validated on an Independent External Dataset? |
Input Imaging Data |
MLA Method | Target Condition |
Sensitivity | Specificity | AUC (±SD) | Accuracy |
| Ahammed Muneer (2019) [46] | 389 | NR | NR | 168 | No | T2w images; tumor segmentation | Deep CNN | Glioma grade | 92.72 | 98.13 | NR | 94.64 |
| Arita (2018) [27] | 111 | NR | NR | 58 | No | T2w-based VOI segmentation and T1w, T2w, FLAIR, and T1w +c images | Lasso and Elastic-Net Regularized Generalized Linear Model | IDH genotype | NR | NR | NR | 87 |
| Bakas (2018) [47] | 86 | NR | NR | NR | No | T1w, T2w, FLAIR, T1w +c images; DTI series and DSC-PWI series | Multivariate machine learning model with a Random Forests algorithm | IDH genotype | 66.7 | 92.9 | NR | 88.4 |
| Bangalore Yogananda (2020) [48] | 214 | NR | NR | 214 | No | T2w, FLAIR, and T1w +c images | 3D Dense-UNet: T2-Net | IDH genotype | 97 | 98 | 0.98 ± 0.146 | 97.14 |
| 3D Dense-UNet: TS-Net | IDH genotype | 98 | 97 | 0.99 ± 0.146 | 97.12 | |||||||
| Batchala (2019) [84] | 102 | NR | 50–52 | 106 | No | T1w, T2w, FLAIR, and T1w +c images; DSC-PWI series | Multivariate model | 1p/19q integrity | NR | NR | NR | 81.1 |
| Bonte (2016) [68] | 274 | NR | NR | NR | No | BraTS-data (T1w, T2w, FLAIR, and T1w +c images) | Random Forests algorithm | LGG/HGG | 95.5 | 79.5 | NR | 92.3 |
| Cao (2020) [28] | 141 | NR | 74–67 | 88 | No | T1w, T2w, FLAIR, and T1w +c images | Lasso and Elastic-Net Regularized Generalized Linear Model with Support vector machine classifier | LGG/HGG | NR | NR | 0.915 ± 0.356 | NR |
| Carver (2019) [29] | 78 | NR | NR | 50 | Yes | T1w, T2w, FLAIR, and T1w +c images | Lasso and Elastic-Net Regularized Generalized Linear Model | IDH genotype | NR | NR | NR | 74 |
| Chang (2018) [77] | 1188 | NR | NR | 153 | No | T1w, T2w, FLAIR, and T1w +c images | Residual CNN model | IDH genotype | NR | NR | 0.93 | 83.0 |
| Citak-Er (2018) [49] | 43 | 49.5 ± 12.8 | 25–18 | NR | No | T1w, T2w, DW images; DTI series, DSC-PWI series, and MRS | Support vector machine classifier with linear kernel and logistic regression with a Random Forests algorithm | LGG/HGG | 86.7 | 96.4 | NR | 93.0 |
| Cui (2018) [30] | 40 | NR | NR | NR | No | T1w, T2w, FLAIR, and T1w +c images; tumor segmentation | Lasso and Elastic-Net Regularized Generalized Linear Model | LGG/HGG | NR | NR | 0.84 | NR |
| De Looze (2018) [50] | 381 | NR | 251–130 | NR | No | Three VASARI criteria as assessed on T1w, T2w, FLAIR, and DW images | Random Forests model | IDH genotype | 81 | 77 | 0.88 | NR |
| Glioma grade II/III | 82 | 94 | 0.98 | NR | ||||||||
| Glioma grade II/IV | 100 | 100 | 1.0 | NR | ||||||||
| Glioma grade III/IV | 83 | 97 | 0.97 | NR | ||||||||
| Fan (2019) [45] | 126 | 46.8 | NR | NR | No | T1w +c images | Lasso and Elastic-Net Regularized Generalized Linear Model adopted into linear discriminant analysis and Support vector machine classifier | glioblastoma/anaplastic oligodendro-glioma | 100.0 | 91.0 | 0.923 | 93.8 |
| Gates (2020) [51] | 23 | NR | NR | NR | No | T2, ADC, CBV, and Ktrans | Random Forests algorithm | Glioma grade | NR | NR | NR | 96 |
| Han (2018) [76] | 117 | NR | NR | 21 | No | T1w, T2w, and FLAIR images | Recurrent CNN model | MGMT promoter methylation status | NR | NR | 0.54 | 53 |
| Han (2018) [71] | 184 | 41.67 | 120–64 | 93 | No | T2w images and T2w-based segmentation |
Random Forests algorithm | 1p/19q integrity | 68.3 | 71.2 | 0.760 ± 0.477 | 70.0 |
| Hwan-Ho (2017) [33] | 108 | NR | NR | NR | No | BraTS-data (T1w, T2w, FLAIR, and T1w +c images) and BraTS-segmentation | Lasso and Elastic-Net Regularized Generalized Linear Model and logistic regression | Glioma grade | 88.89 | 90.74 | 0.8870 | 89.81 |
| Inano (2014) [52] | 33 | NR | 22–11 | 33 | No | DW images, FA-maps, first eigenvalue, second eigenvalue, third eigenvalue, MD-maps, and raw T2 signal with no diffusion-weighting |
Support vector machine classifiers | Glioma grade | 84.8 | 74.5 | 0.912 ± 0.028 | 80.4 |
| Jiang (2019) [34] | 87 | 45.4 ± 13.1 | 43–44 | 35 | Yes | T2w and T1w +c images | Lasso regression model with fusion Radiomics model and Support vector machine classifier | MGMT promoter methylation status | 82.1 | 85.7 | 0.898 ± 0.323 | 88.6 |
| Jiang (2020) [35] | 83 | 45.5 ± 12.3 | 50–33 | 33 | Yes | T2w and T1w +c images | Lasso regression model with radiomics signature model and Support vector machine classifier | TERT promoter mutation status | 71.4 | 89.5 | 0.827 ± 0.470 | 84.8 |
| Kim (2020) [53] | 127 | NR | 68–59 | 28 | No | T1w, T2w, FLAIR, T1w +c, DW images; DSC-PWI series | Recursive feature elimination with Support vector machine, completed with a Random Forests algorithm and a logistic regression classifier | IDH genotype | 53.6 | 86.7 | 0.747 ± 0.228 | NR |
| Kinoshita (2018) [70] | 199 | NR | NR | NR | No | Conventional MR sequences (NOS) | Random Forests algorithm | Glioma grade | NR | NR | 0.711 | 64.5 |
| Lee (2019) [54] | 88 | NR | 47–41 | 35 | Yes | T1w, T2w, FLAIR, DW images; DSC-PWI series | Eight machine learning classifiers: K-Nearest Neighbors, Support vector classification, Decision Tree, Random Forest, AdaBoost, Naive Bayes, Linear Discriminant Analysis, and Gradient Boosting |
IDH genotype | NR | NR | NR | 83.4 |
| Li (2019) [55] | 69 | 60.0 | 37–32 | 40 | Yes | T2w and T1w +c images | Support vector machine classifier with Support vector machine classifier | PTEN genotype | 86.7 | 70.0 | 0.787 | 82.5 |
| Li (2018) [32] | 63 | 43.6 | 25–38 | 91 | Yes | T2w images | Lasso regression model with Support vector machine classifier | ATRX genotype | 57.1 | 85.7 | 0.725 | 76.9 |
| Li (2018) [33] | 180 | 39.2 | 111–69 | 92 | No | T2w images | Lasso regression model with Support vector machine classifier | P53 status | 62.2 | 85.1 | 0.763 | 70.7 |
| Li (2017) [56] | 151 | 40.7 ± 10.8 | 81–70 | 151 | No | T1w and FLAIR images | CNN for segmentation followed by DLR model with Support vector machine classifier | IDH genotype | 94.38 | 86.67 | 0.9521 | 92.44 |
| Li (2018) [77] | 133 | 54.2 | 79–54 | 60 | No | T1w, T2w, FLAIR, and T1w +c images | Multiregional Radiomics model | MGMT-methylation | NR | NR | 0.88 | 80 |
| Li (2018) [78] | 118 | 53.6 | 70–48 | 107 | No | T1w, T2w, FLAIR, and T1w +c images | Multiregional Radiomics models | IDH genotype | 80 | 99 | 0.96 | 97 |
| Liang (2018) [79] | 167 | 52.4 ± 15.5 | NR | NR | No | BraTS-data (T1w, T2w, FLAIR, and T1w +c images) | Multimodal Three-Dimensional DenseNet | IDH genotype | 78.5 | 88.0 | 0.857 | 84.6 |
| Lo (2020) [57] | 39 | NR | 28–11 | NR | No | T1w +c images; processed by transformed ranklet images. | Logistic regression classifier |
IDH genotype | 57 | 97 | NR | 90 |
| Lu (2018) [58] | 214 | NR | NR | 70 | Yes | T1w, T2w, FLAIR, T1w +c, and DW images (T2w and DW images were optional) |
Three-level machine learning model | LGG/HGG | 82.5 | 90.5 | NR | 87.7 |
| Matsui (2020) [36] | 217 | 42 | 131–86 | NR | No | T1w, T2w, and FLAIR images | Lasso regression model with DLR model | Grading LGG | NR | NR | NR | 58.5 |
| Mzoughi (2020) [37] | 284 | NR | NR | 67 | Yes | T1w +c images | Lasso regression model with 3D CNN model with Support vector machine classifier | Glioma grade | NR | NR | NR | 96.4 |
| Park (2020) [71] | 168 | NR | NR | 168 | No | T2w, FLAIR, and T1w +c images | Random Forests algorithm | IDH genotype | NR | NR | 0.900 ± 0.298 | NR |
| Park (2019) [72] | 136 | 44.99 ± 12.94 | 65–71 | 99 | Yes | T2w, FLAIR, and T1w +c images; DTI series | Random Forests algorithm | Glioma grade | 72.6 | 60.4 | 0.72 ± 0.51 | 66.7 |
| Rathore (2019) [59] | 202 | NR | NR | NR | No | T1w, T2w, FLAIR, and T1w +c images. Data were sometimes complemented with DTI and DSC-PWI series |
CNN adjusted with a Support vector machine classifier | IDH genotype | 83 | 86 | 0.85 | 85 |
| MGMT | 83 | 85 | 0.84 | 83 | ||||||||
| Rathore (2018) [67] | 111 | NR | NR | NR | No | T1w, T2w, FLAIR, and T1w +c images | Support Vector Machine model with a Random Forests algorithm | MGMT-methylation | 75.0 | 97.0 | 0.80 | 88.28 |
| Rathore (2019) [59] | 270 | NR | NR | NR | No | T1w, T2w, FLAIR, and T1w +c images; DTI and DSC-PWI series | Cross-validated sequential feature selection | MGMT-methylation | NR | NR | NR | 86.95 |
| Sasaki (2018) [39] | 207 | NR | NR | NR |
No | T1w, T2w, FLAIR, and T1w +c images | Lasso regression model with supervised component principal analysis | MGMT-methylation | NR | NR | NR | 68 |
| Sasaki (2019) [38] | 201 | NR | NR | NR | No | T1w, T2w, and T1w +c images | Lasso regression model with supervised component principal analysis | MGMT-methylation | 67 | 66 | NR | 67 |
| Shboul (2020) [40] | 81 | NR | NR | 27 | No | T1w, T2w, FLAIR, and T1w +c images | Lasso regression model with supervised component principal analysis and multi-resolution fractal modeling | IDH genotype | 90 | 79 | 0.84 ± 0.156 | NR |
| 1p/19q integrity | 75 | 85 | 0.80 ± 0.208 | NR | ||||||||
| MGMT-methylation | 93 | 73 | 0.83 ± 0.208 | NR | ||||||||
| ATRX genotype | 69 | 83 | 0.70 ± 0.468 | NR | ||||||||
| TERT promoter mutation status | 77 | 86 | 0.82 ± 0.208 | NR | ||||||||
| Shofty (2018) [60] | 47 | 37.7 ± 10.6 | 27–20 | NR | No | T2w, FLAIR, and T1w +c images | Ensemble Radiomic Classifier model with a Support vector machine classifier | 1p/19q integrity | 92 | 83 | 0.87 | 87 |
| Sun (2020) [41] | 92 | NR | NR | NR | No | T1w, T2w images | Lasso regression model with logistic regression models | P53 status | 100 | 40 | 0.709 | 81.3 |
| Takahashi (2019) [80] | 44 | NR | NR | 11 | No | DW (b1000 and b2000) images, ADC-maps, FA-maps, and MK-maps | Deep CNN model | Glioma grade | NR | NR | NR | 82 |
| Takahashi (2019) [82] | 38 | NR | NR | NR | No | T2w-based VOI segmentation | Logistic regression models | 1p/19q integrity | 69.7 | 73.3 | 0.736 | 71.1 |
| Tan (2019) [42] | 74 | 47.93 ± 13.28 | 45–29 | 31 | No | FLAIR and T1w +c images; ADC-maps | Radiomics Nomogram model | IDH genotype | 86.7 | 87.5 | 0.900 ± 0.116 | 87.1 |
| Tian (2020) [43] | 88 | NR | 53–35 | 38 | No | T1w, T2w, FLAIR, and T1w +c images; MRS | Lasso regression model with Radiomics Nomogram model | TERT promoter mutation status | 75.0 | 90.9 | 0.889 ± 0.335 | 84.2 |
| Tongtong (2017) [61] | 110 | NR | NR | NR | No | 3D FLAIR images | Support vector machine classifier with minimum redundancy, maximum relevance, and maximum sparse representation coefficient | IDH genotype | 88 | 79 | 0.90 | 85 |
| van der Voort (2019) [62] | 284 | NR | 161–123 | 129 | Yes | T2w and T1w +c images. Data were sometimes complemented with FLAIR images | Support vector machine classifier | 1p/19q integrity | 73.2 | 61.7 | 0.723 ± 0.084 | 69.3 |
| Wei (2019) [83] | 74 | NR | 42–32 | 31 | No | FLAIR and T1w +c images; ADC-maps | Fusion Radiomics model by logistic regression modelling | MGMT promoter methylation | 94.4 | 53.9 | 0.902 ± 0.305 | 77.4 |
| Wu (2019) [73] | 84 | 53.5 ± 15.0 | 67–59 | 42 | No | T1w, T2w, FLAIR, and T1w +c images | Random Forests algorithm | IDH genotype | NR | NR | 0.931 ± 0.233 | 89.5 |
| Xi (2018) [44] | 98 | NR | 55–43 | 20 | Yes | T1w, T2w, and T1w +c images | Lasso regression model with Support vector machine model | MGMT promoter methylation | 87.5 | 75.0 | NR | 80.0 |
| Yang (2018) [81] | 113 | NR | NR | NR | No | T1w, T2w, FLAIR, and T1w +c images | CNN model | LGG/HGG | NR | NR | NR | 86.7 |
| Yu (2017) [85] | 110 | 40.3 ± 11.3 | 54–56 | 30 | No | FLAIR images | Radiomics model | IDH genotype | 88 | 67 | 0.79 | 83 |
| Zhang (2017) [63] | 90 | 51.4 | 52–38 | 30 | No | T1w, T2w, FLAIR, T1w +c, and DW images | Random Forests algorithm | IDH genotype | NR | NR | 0.9231 | 89 |
| Zhang (2018) [64] | 73 | NR | NR | 30 | No | T1w, T2w, FLAIR, and T1w +c images | Support vector machine-based recursive feature elimination | IDH genotype | 85.0 | 70.0 | 0.792 | 80.0 |
| P53 status | 84.6 | 85.7 | 0.869 | 85.0 | ||||||||
| Zhang (2020) [65] | 108 | NR | 61–47 | NR | No | DTI series | CNN model with a Support vector machine classifier | LGG/HGG | 98 | 86 | 0.93 | 94 |
| Glioma grade III/IV | 98 | 100 | 0.99 | 98 | ||||||||
| Zhao (2020) [74] | 36 | 45.0 ± 14.4 | 19–17 | 36 | No | FLAIR and T1w +c images | Random Forests algorithm | Glioma grade II/III | 77.8 | 78.3 | 0.861 ± 0.240 | 78.1 |
| Zhou (2019) [66] | 538 | NR | 303–235 | 206 | Yes | FLAIR and T1w +c images | Random Forests algorithm with a Support vector machine classifier | IDH genotype | NR | NR | 0.919 ± 0.286 | NR |
Legend: ADC: Apparent diffusion coefficient; ARTX: Alpha thalassemia/mental retardation syndrome X linked gene; BraTS: Brain Tumor Segmentation Challenge; CNN: Convolutional neural network; DW: Diffusion-weighted images; DTI: Diffusion tensor imaging; DSC-PWI: Dynamic susceptibility contrast perfusions weighted imaging; FA: Fractional anisotropy imaging; FLAIR: fluid attenuated inversion recovery; HGG: High grade glioma; IDH: Isocitrate dehydrogenase gene; LGG: Low-grade glioma; MGMT: O6-Methylguanine-DNA Methyltransferase; MRS: Magnetic resonance spectroscopy imaging; TERT: Telomerase reverse transcriptase gene; T1w: T1-weighted images; T1w +c: T1-weighted post-contrast images; T2w: T2-weighted images; VASARI: Visually AcceSAble Rembrandt Images. * If cross-validation was used, the Performance values of the cross-validation set were provided here. When the dataset was split into training/validation/test sets, the Performance evaluation values with regard to the investigated outcome (e.g., IDH genotype) of the Validation set were provided here.