Table 1.
Specifications of radiomics studies included in this narrative review.
| Reference | Study design |
Patients (No) |
Diagnostic modality | Radiomics imaging features selected (No) | Prediction |
Sensitivity (%) |
Specificity (%) |
Accurancy (%) |
AUC |
|---|---|---|---|---|---|---|---|---|---|
| Parekh et al. (2017) [13] | Retrospective | 124 | MRI (3T) | 690 (RFMs) | Malignancy | 93 | 85 | ||
|
| |||||||||
| Whitney et al. (2018) [14] | Retrospective | 508 | MRI (1.5 and 3 T) | 38 | Malignancy | 0.846 (including size features) 0.848 (excluding size features) |
|||
|
| |||||||||
| Bickelhaupt et al. (2017) [15] | Retrospective | 50 | MRI (1.5 T) | 188 | Malignancy | 0.842-0.851 | |||
|
| |||||||||
| Bickelhaupt et al. (2018) [16] | Retrospective | 222 | MRI (1.5 T) | 359 | Malignancy | 98.4 | 69.7 | ||
|
| |||||||||
| Zhang et al. (2017)[17] | Retrospective | 117 | US | 364 | Malignancy | 85.7 | 89.3 | ||
|
| |||||||||
| Tagliafico et al. (2018)[18] | Prospective | 20 | Mammography (DBT) | 104 | Malignancy | 0.567 | |||
|
| |||||||||
| Braman et al. (2017)[19] | Retrospective | 117 | MRI (1.5 and 3 T) | 99 | NAC | 0.78 (training dataset) 0.74(independent testing set) 0.83 (HR+, HER2−) 0.93 (TN/HER2+) |
|||
|
| |||||||||
| Dong et al. (2017)[20] | Retrospective | 146 | MRI (1.5 T) | 25 | Prognostic factors | 0.847 (training set; model 10 T2-fat suppression) 0.770 (validation set; model 10 T2-fat suppression) 0.847 (training set; model 8 DWI) 0.787 (validation set; model 8) 0.863 (training set; model 10 joint T2-fat suppression/DWI) 0.805 (validation set; model 10 joint T2-fat suppression/DWI) |
|||
|
| |||||||||
| Obeid et al. (2016) [21] | Retrospective | 63 | MRI (1.5 and 3 T) | 13 | Prognostic factors | - | - | - | - |
|
| |||||||||
| Ma et al. (2018)[22] | Retrospective | 377 | MRI (3 T) | 56 | Prognostic factors | 77.7 | 76.9 | 0.757 | 0.773 |
|
| |||||||||
| Liang et al. (2018)[23] | Retrospective | 318 | MRI (1.5 T) | 30 | Prognostic factors | 0.762 (training dataset) 0.740 (validation dataset) |
|||
|
| |||||||||
| Guo et al. (2015)[24] | Retrospective | 91 | MRI (1.5 T) | 38 | Molecular subtypes | 0.877 (stage) ∗ 0.693 (lymph node) ∗ 0.789 (ER) ∗ 0.689 (PR) ∗ 0.641 (HER2) ∗ |
|||
|
| |||||||||
| Li et al. (2016) [4] | Retrospective | 91 | MRI (1.5 T) | 38 | Molecular subtypes | 0.89 (ER+ vs ER−) 0.69 (PR+ vs PR-) 0.65 (HER”+ vs HER2-) 0.67 TN vs others) |
|||
|
| |||||||||
| Wang et al. (2015) [25] | Retrospective | 84 | MRI (3 T) | 85 | Molecular subtypes | 57.0 (TN vs others) ∗∗ 62.0 (TN vs ER+)∗∗ 53.0 (TN vs PR+)∗∗ 49.5 (TN vs LumA) ∗∗ 69.5 (TN vs LumB) ∗∗ |
94.7(TN vs others) ∗∗ 93.6 (TN vs ER+)∗∗ 94.1 (TN vs PR+)∗∗ 89.8(TN vs LumA) ∗∗ 90.0 (TN vs LumB) ∗∗ |
90.0 (TN vs others) ∗∗ 89.4 (TN vs ER+)∗∗ 87.8 (TN vs PR+)∗∗ 81.8 (TN vs LumA) ∗∗ 84.3 (TN vs LumB) ∗∗ |
0.878 (TN vs others) ∗∗ 0.883 (TN vs ER+)∗∗ 0.859 (TN vs PR+)∗∗ 0.814 (TN vs LumA) ∗∗ 0.789 (TN vs LumB) ∗∗ |
|
| |||||||||
| Fan et al. (2017)[26] | Retrospective | 60 | MRI (1.5 T) | 88 | Molecular subtypes | 88. 2 (LumA) 86.5 (LumB) 81.1 (HER2) 81.1 (basal-like) |
76.9 (LumA) 62.5 (LumB) 100 (HER2) 100 (basal-like) |
0.867 (LumA) 0.786 (LumB) 0.888 (HER2) 0.923 (basal-like) |
|
|
| |||||||||
| Guo et al. (2017)[27] | Retrospective | 215 | US | 463 | Molecular subtypes | 0.760 | |||
|
| |||||||||
| Ma et al. (2018)[28] | Retrospective | 331 | Mammography | 39 | Molecular subtypes | 0.865 (TN vs non TN) 0.784 (HER2 vs non HER2) 0.752 (Lum vs non-Lum) |
|||
|
| |||||||||
| Li et al. (2016) [29] | Retrospective | 84 | MRI (1.5 and 3T) | 38 | Recurrence | 0.88 (MammaPrint) 0.76 (Oncotype DX) 0.68 (PAM50 risk of relapse based on subtype) 0.55 (PAM50 risk of relapse based on subtype and proliferation) |
|||
|
| |||||||||
| Park et al. (2018)[30] | Retrospective | 294 | MRI (1.5 T) | 156 | Recurrence | - | - | - | - |
|
| |||||||||
| Drukker et al. (2018)[31] | Retrospective | 162 | MRI (1.5 T) | 1 | Recurrence | - | - | - | - |
(i) ∗AUC considering only radiomics models.
(ii) ∗∗Considering both tumor and BPE features.