Table 1.
Summary of reported diagnostic performance of US radiomics in head and neck oncology (studies have been summarized in alphabetical order).
| Authors | Radiomics Platform | Number of Features | Statistical Analysis/Model Performance |
|---|---|---|---|
| Acharya et al., 2012 [13] | Not reported | 5 | Malignant vs. benign thyroid nodules Training AUC: 0.99 |
| Ardakani et al., 2018 (Eur J Radiol) [14] | Not reported | 40 | Hot (hyperfunctioning) vs. cold (hypofunctioning) thyroid nodules Training AUC: 0.99 (95% CI: 0.978, 1.000) Validation AUC: 0.95 (95% CI: 0.874, 1.000) |
| Ardakani et al., 2018 (J Ultrasound Med) [15] | Not reported | 4 | LNM vs. no LNM Radiologic + textural features Training AUC: 0.99 (95% CI: 0.998, 0.999) Validation AUC: 0.95 (95% CI: 0.911, 0.993) |
| Bhatia et al., 2016 [16] | MATLAB | 15 | Malignant vs. benign thyroid nodules Training AUC: 0.97 (p < 0.0001) |
| Chang et al., 2016 [17] | Not reported | 74 | Malignant vs. benign thyroid nodules Adaboost CAD AUC: 0.98 RAD AUC: 0.98 |
| Chen et al., 2020 [18] | MATLAB | 23 | Benign vs. lymphomatous AUC: 0.95 (p < 0.001) Lymphomatous vs. metastatic AUC: 0.93 (p < 0.001) Benign vs. malignant AUC: 0.84 (p < 0.001) Benign vs. metastatic AUC: 0.72 (p < 0.001) |
| Ding et al., 2012 [19] | Not reported | Not reported | Malignant vs. benign thyroid nodules Training accuracy: 95% |
| Galimzianova et al., 2020 [20] | Not reported | 219 | Malignant vs. benign thyroid nodules Training AUC: 0.83 (95% CI: 0.715, 0.942) |
| Jiang et al., 2020 [21] | PyRadiomics | 6 | LNM vs. no LNM Training AUC: 0.85 (95% CI: 0.791, 0.912) Validation AUC: 0.83 (95% CI: 0.749, 0.916) |
| Kim et al., 2015 [22] | MATLAB | 10 | Malignant vs. benign thyroid nodules Gray-scale AUC: 0.80 Elastography AUC: 0.68 |
| Kim et al., 2017 [23] | MATLAB | 5 | LNM vs. no LNM OR: 0.98; 95% CI: 0.48-1.99, p > 0.05 |
| Kwon et al., 2020 [24] | PyRadiomics | 6 | LNM vs. no LNM Training AUC: 0.93 Validation AUC: 0.90 |
| Li et al., 2020 [25] | Ultrosomics | 690 | LNM vs. no LNM Training AUC: 0.76 Validation AUC: 0.80 |
| Liang et al., 2018 [26] | AI Kit | 19 | Malignant vs. benign thyroid nodules Training AUC: 0.92 (95% CI: 0.877, 0.965) Validation AUC: 0.93 (95% CI: 0.884, 0.977) |
| Liu et al., 2018 [27] | MATLAB | 25 | LNM vs. no LNM (B-mode + SE-US) Training AUC: 0.90 |
| Liu et al., 2019 [28] | MATLAB | 50 | LNM vs. no LNM Training AUC: 0.78 (95% CI: 0.731, 0.833) Validation AUC: 0.73 (95% CI: 0.653, 0.801) |
| Nam et al., 2016 [29] | MATLAB | 5 | Malignant vs. benign thyroid nodules Skewness AUC: 0.61 (95% CI: 0.563, 0.647) Kurtosis AUC: 0.65 (95% CI: 0.607, 0.691) Entropy AUC: 0.64 (95% CI: 0.596, 0.681) |
| Park et al., 2020 [30] | MATLAB | 14 | LNM vs. no LNM Training AUC: 0.71 (95% CI: 0.649, 0.770) Validation AUC: 0.62 (95% CI: 0.560, 0.682) |
| Park et al., 2021 [31] | MATLAB | 66 | Malignant vs. benign thyroid nodules Training AUC: 0.85 (95% CI: 0.830, 0.870) Validation AUC: 0.75 (95% CI: 0.690, 0.810) |
| Prochazka et al., 2019 [32] | MATLAB | Not reported | Malignant vs. benign thyroid nodules Accuracy: 94% |
| Raghavendra et al., 2017 [33] | Not reported | Not reported | Malignant vs. benign thyroid nodules Training AUC: 0.94 |
| Tong et al., 2020 [34] | MATLAB | 21 | LNM vs. no LNM US radiomics nomogram Training AUC: 0.94 (95% CI: 0.911, 0.982) Validation AUC: 0.91 (95% CI: 0.842, 0.987) |
| Yoon et al., 2021 [35] | MATLAB | 15 | Malignant vs. benign thyroid nodules Radiomics score + clinical variables AUC: 0.84 (95% CI: 0.775, 0.897) Clinical variables alone AUC: 0.58 (95% CI: 0.435, 0.693) |
| Zhao et al., 2021 [36] | Intelligence Foundry | 6 | Malignant vs. benign thyroid nodules ML-assisted US visual approach Training AUC: Not reported Validation AUC: 0.90 Test AUC: 0.92 |
| Zhou et al., 2020 [37] | MATLAB | 23 | LNM vs. no LNM Training AUC: 0.87 (95% CI: 0.802, 0.938) Validation AUC: 0.86 (95% CI: 0.785, 0.930) |
AUC, area under the curve; LNM, lymph node metastasis; CAD AUC, computer-aided diagnosis area under the curve; RAD AUC, radiologist area under the curve.