Table 1.
Summary of literature for magnetic resonance imaging radiomics feature robustness
| Reference | Disease / phantom | MR sequences | # features | Feature classes | Parameters evaluated | Statistical analysis | Robustness evaluation |
|---|---|---|---|---|---|---|---|
| Baessler et al. [26], 2019 | Vegetable/fruit phantom | FLAIR, T1w, T2w | 45 | Intensity, shape, texture | MR sequence, resolution | CCC, DR, Bland-Altman analyses, ICC | Test-retest robustness, intraobserver and interobserver reproducibility |
| Traverso et al. [50], 2019 | Locally advanced rectal cancer | DWI (ADC map) | 70 | Intensity, shape, texture | Pre-processing filter, re-binning and resampling | CCC, ICC, Spearman correlation | Inter-observer dependence |
| Duron et al. [39], 2019 | Lacrymal gland tumor and breast lesion | T1w, DWI (ADC map), DIXON, DISCO | 69/57 (2 softwares) | Texture | Discretization method, bin width and bin number | CCC, ICC(2,1) | Intra- and inter-observer reproducibility |
| Lecler et al. [37], 2019 | Lacrimal gland tumor | T1w, DWI (ADC map), DIXON | 85 | Intensity, shape, texture | MR sequence, metric threshold | CCC, ICC(2,1), Spearman correlation | Intra- and inter-observer reproducibility, non-redundancy |
| Um et al. [51], 2019 | Glioblastoma multiforme | FLAIR, T1w, post-contrast T1w | 420 | Intensity, shape, texture, filter-based | Preprocessing technique on multi-scanner datasets, bin number | Two-sided Wilcoxon tests | Feature variability |
| Schwier et al. [24], 2019 | Prostate cancer | T2w, DWI (ADC map) | NA | Intensity, shape, texture, filter-based | Image normalization, 2D/3D texture computation, bin widths, and image pre-filtering | ICC(1,1) | Test-retest repeatability |
| Fiset et al. [38], 2019 | Cervical cancer | T2w | 1761 | Intensity, shape, texture, filter-based | Quantization method, LoG kernel sizes, | ICC(1,1), ICC(2,1), Pearson correlation, Krippendorff’s alpha | Test-retest repeatability, cross-scanner reproducibility, inter-observer reproducibility |
| Peerlings et al. [33], 2019 | Ovarian, lung and colorectal liver metastasis cancer | DWI (ADC map) | 1322 | Intensity, shape, texture, filter-based | Center and vendor | CCC | Feature stability |
| Buch et al. [52], 2018 | Nonanatomic Gd-DTPA phantom | T1w | 41 | Intensity, texture, filter-based (Laws) | Magnet strength, flip-angle, number of excitations, scanner platform | Q values | Feature variability |
| Yang et al. [53], 2018 | Simulated data from digital phantom and glioma | T1w, T2w | 26 | Texture | Noise level, acceleration factor, and image reconstruction algorithm | Student’s t-test, CV | Feature variance |
| Bologna et al. [32], 2018 | Soft tissue sarcoma and oropharyngeal cancer | DWI (ADC map) | 69 | Intensity, texture | ROI transformation and bin number | Absolute percentage variation, two-way mixed effect ICC | Feature stability and discrimination |
| Chirra et al. [40], 2018 | Prostate cancer | T2w | 406 | Intensity, texture, filter-based | Different sites | Multivariate CV and Instability Score | Cross-site reproducibility |
| Saha et al. [31], 2018 | Breast cancer | DCE-MRI (first postcontrast, PE, SER, washing rate maps) | 529 | Intensity, shape, texture | Scanner, contrast agent | ICC(3,1), Pearson correlation, average DSC | Inter-reader stability, inter-relations within feature groups, pairwise reader variability |
| Molina et al. [27], 2017 | Glioblastoma | T1w | 16 | Texture | Spatial resolution and bin number | CV | Feature variation |
| Brynolfsson et al. [54], 2017 | Glioma and prostate cancer | DWI (ADC map) | 19 | Texture | noise level, resolution, ADC map construction, quantization method, and bin number | Two-sample Kolmogorov-Smirnov tests | Feature distribution variation |
| Gourtsoyianni et al. [41], 2017 | Primary rectal cancer | T2w | 46 | Intensity, texture, filter-based | 2 baseline examinations | wCV | Test-retest repeatability |
| Guan et al. [55], 2016 | Cervical cancer | DWI (ADC map) | 8 | Intensity, texture | GLCM direction | ICC, Wilcoxon test, Kruskal-Wallis test, and ROC curve | Inter- and intra-observer agreement |
| Molina et al. [56], 2016 | Glioblastoma | T1w | 16 | Texture | Matrix size and bin number | CV | Feature variation |
| Savio et al. [57], 2010 | Multiple sclerosis | T1w | 264 | Intensity, texture, filter-based | Global, regional and local features | Wilcoxon’s signed ranks test | Feature variation |
| Mayerhoefer et al. [58], 2009 | PSAG phantom | T2w | NA | Texture, filter-based | Spatial resolution, NAs, TR, TE, and SBW | LDA and k-NN classifier | Ability to distinguish between different patterns |
| Collewet et al. [59], 2004 | Cheese phantom | T2w, PDW | 90 | Texture, filter-based | MRI acquisition protocol and quantization method | POE, ACC, 1-NN classifier | Classification |
MR Magnetic resonance, FLAIR Fluid-attenuated inversion recovery, DWI Diffusion-weighted imaging, ADC Apparent diffusion coefficient, DISCO Differential subsampling with cartesian ordering, DCE-MRI Dynamic contrast-enhanced magnetic resonance imaging, PE Peak enhancement, SER Signal enhancement ratio, PDW Proton density weighted, LoG Laplacian of Gaussian, NAs Number of acquisitions, TR Repetition time, TE Echo time, SBW Sampling bandwidth, CCC Concordance correlation coefficient, DR Dynamic range, ICC Intraclass correlation coefficient, wCV Within-subject coefficient of variation, ROC Receiver operating characteristic, CV Coefficient of variation, DSC Dice similarity coefficients, LDA Linear discriminant analysis, k-NN k nearest neighbor, POE Probability of error, ACC Average correlation coefficient, 1-NN 1-nearest neighbor