Table 1.
Characteristics of the selected radiomics studies.
| Authors | Publication Year | Objective | Journal | Number of Patients | Imaging Modality | Segmentation | Technique Used for Feature Selection | Validation | Classification | Features | Best Results | Calibration Statistics | Decision Curve Analysis | RQS |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Jiang X et al. [12] | 2021 | Detect EGFR mutation in spinal metastasis in patients with primary lung adenocarcinoma |
Journal of Magnetic Resonance Imaging | 97 | 3T MRI T1W, T2W, T2W-FS | Manual, ITK-SNAP | Mann–Whitney U-test, LASSO, 10-fold cross-validation | Y | Logistic regression models | Handcrafted features: first-order, shape- and size- based, texture, filtered features | Fusion features: AUC = 0.771, ACC = 0.550, SEN = 0.750, SPE = 0.875 | Y | Y | 10/36 = 27.7% |
| Ren M et al. [13] | 2021 | Detect EGFR mutation in spinal metastasis in patients with primary lung adenocarcinoma |
Medical Physics | 110 | 3T MRI T1W, T2W, T2W-FS | Manual, ITK-SNAP | Intraclass correlation coefficient (ICC) analysis, Mann–Whitney U, LASSO, 10-fold cross-validation |
Y | Logistic regression, random forest, neural network, and support vector machine | First-order, shape-based, and texture (1967) |
Fusion features: AUC = 0.803 (0.682–0.924), SEN = 0.700, SPE = 0.818; nomogram, AUC = 0.882 (0.695–0.974), ACC = 0.808, SEN = 0.846, SPE = 0.846 | Y | Y | 11/36 = 30.5% |
| Fan Y et al. [14] | 2021 | Detect EGFR mutation in spinal metastasis in patients with primary lung adenocarcinoma |
Physics in Medicine & Biology | 94 | 3T MRI, T1W, T2W-FS | Manual, ITK-SNAP | Mann–Whitney U-test, LASSO, 10-fold cross-validation | Y | Logistic regression models | First-order, shape- and size- based, texture, high-dimensional features (1595) | Multiregional radiomics signature: AUC = 0.777 (0.612–0.967), ACC = 0.688, SEN = 0.615, SPE = 0.947 | N | N | 8/36 = 22.2% |
| Ran C et al. [15] | 2020 | Detect EGFR mutation subtypes in exons 19 and 21 in spinal metastasis in patients with primary lung adenocarcinoma |
Academic Radiology | 76 | 3T MRI, T1W, T2W-FS | Manual | Mann–Whitney U, LASSO, 10-fold cross-validation |
Y | Logistic regression models | First-order, shape-based, and texture (1967) | T1W: AUC = 0.728 (0.526 0.903), ACC = 0.692, SEN = 0.692, SPE = 0.769; T2W-FS: AUC = 0.852 (0.706 0.998), ACC = 0.731, SEN = 0.846, SPE = 0.769; nomogram, AUC = 0.821 (0.692–0.929), SEN = 0.667, SPE = 0.909 | Y | Y | 12/36 = 33.3% |
| Wang Y et al. [16] | 2019 | Pretreatment prediction of bone metastasis in patients with prostate cancer | Magnetic Resonance Imaging | 176 | 3T MRI T2W, T1W DCE | Manual, IBEX | Linear regression, ridge regression, logistic regression models | Y | Linear regression, ridge regression, logistics regression models | Shape, intensity, intensity histogram, GLCM, gray-level run (976) | Combined T2W and DCE: AUC = 0.898 (0.833–0.937), ACC = 0.821, SEN = 0.647, SPE = 0.782 | Y | N | 8/36 = 22.2% |
| Hayakawa T et al. [17] | 2020 | Investigate the potential prognostic value of clinical risk factors, image features, and radiomics of pelvic bone metastasis in newly diagnosed prostate cancer patients |
Japanese Journal of Radiology | 69 | CT | Manual, 3D Slicer | N | N | Not available | Shape-based, first-order statistics, texture (105) | Maximum 2D diameter and least axis were detected as risk factors for OS (HR 1.007 and 1.013, respectively) | N | N | 0/36 = 0% |
| Zhang W et al. [18] | 2020 | Pretreatment prediction of bone metastasis in patients with prostate cancer | BMC Medical Imaging | 116 | 3T MRI, T2W-FS, DWI, DCE T1W | Manual, AK software | ANOVA | Y | Logistic regression models | Not available (204) | AUC = 0.84 | Y | Y | 14/36 = 38.8% |
| Sun W et al. [19] | 2021 | Distinguish between benign and malignant bone tumors | Cancer Imaging | 206 | CT | Manual, ITK-SNAP | LASSO | Y | Logistic regression models | Shape, statistical, texture, wavelet (1130) | Radiomic model, AUC = 0.781 (0.643–0.918); nomogram, AUC = 0.917 | Y | Y | 12/36 = 33.3% |
| Xiong X et al. [20] | 2021 | Differentiating between multiple myeloma and different tumor metastasis lesions of the lumbar vertebra |
Frontiers in Oncology | 107 | 3T MRI, T1W, T2W-FS | Manual | LASSO, 10-fold cross-validation | Y | Support vector machine, k-nearest neighbor, random forest, artificial neural networks (ANNs), and naïve Bayes |
Histogram features, GLCM, GRLM, and an autoregressive model (282) |
Differentiating myeloma and metastasis, ANN T2W-FS: AUC = 0.815, SEN = 0.879, SPE = 0.790; differentiating myeloma and metastasis subtypes, ANN T2W-FS: AUC = 0.648, SEN = 0.714, SPE = 0.775 | N | N | 8/36 = 22.2% |
| Yin P et al. [21] | 2018 | Differentiation between primary sacral chordoma, sacral giant cell tumor, and sacral metastatic tumor |
Journal of Magnetic Resonance Imaging | 167 | 3T MRI, T2W-FS, T1W CE | Manual, ITK-SNAP | ANOVA, LASSO, Pearson correlation, random forest | Y | Random forest | Histogram features, form factor features, Haralick, GLCM features, RLM (385). |
Combined T2W and T1W CE: AUC = 0.773, ACC = 0.711; T2W, AUC = 0.678, ACC = 0.541; T1W CE, AUC = 0.592, ACC = 0.568 | Y | N | 9/36 = 25% |
| Zhong X et al. [22] | 2020 | Differentiating of cervical spine osteoradionecrosis from metastasis after radiotherapy in nasopharyngeal carcinoma |
BMC Medical Imaging | 123 | 1.5 MRI, T1W CE | Manual, MaZda | Intraclass correlation coefficient (ICC) analysis, combination feature selection algorithm (combination of Fisher coefficient, classification error probability combined with average correlation coefficients, and mutual information), LASSO, 10-fold cross-validation |
Y | Logistic regression models | Histogram, gray-level co-occurrence matrix, run-length matrix, absolute gradient, autoregressive model, and wavelet (279) |
Nomogram: AUC = 0.720 (0.573–0.867), ACC = 0.723, SEN = 0.800, SPE = 0.640 | Y | Y | 11/36 = 30.5% |
| Filograna L et al. [23] | 2019 | Differentiate between metastatic and nonmetastatic vertebral bodies in patients with bone marrow metastatic disease |
La Radiologia Medica | 8 | 1.5 MRI, T1W, T2W-FS | Not available | Wilcoxon test | N | Logistic regression models | Statistical/ histogram, morphological, and textural features (89) |
T1W: AUC = 0.814 (0.685–0.942); T2W: AUC = 0.911 (0.829–0.993) | N | N | 2/36 = 5.5% |
| Lang N et al. [24] | 2019 | Differentiate metastatic cancer in the spine originated from lung cancer and other nonlung tumors | Magnetic Resonance Imaging | 61 | 3T MRI DCE | Manual, automatic | Random forest algorithm | N | Logistic regression models | Texture, histogram (33 × 3 maps) | 3 features, histogram + texture: ACC = 0.68; 5 features, histogram + texture: ACC = 0.71; | N | N | 1/36 = 2.7% |