Table 6.
Overview of published studies regarding predictive models for gene mutation status based on nodule features (2017–2021).
| Authors | Year | Dataset | Methods | Performance Results (%) |
|---|---|---|---|---|
| Zou et al. [177] | 2017 | Private (209 patients) |
Multivariable analyses | EGFR: AUC = 73.7 |
| Cheng et al. [176] | 2017 | Private (2146 patients) |
Weighted mean difference, inverse variance | EGFR: OR = 49.0 |
| Li et al. [179] | 2018 | Private (1010 patients) |
Random forest/CNNs | EGFR: AUC = 83.4 |
| Koyasu et al. [178] | 2019 | NSCLC-radiogenomics | XGBoost/random forest | EGFR: AUC = 65.9 |
| Wang et al. [180] | 2019 | Private (844 patients) |
CNNs | EGFR: AUC = 85.0 |
| Zhao et al. [181] | 2019 | TCIA and private (879 patients) |
3D DenseNet | EGFR: AUC = 75.8 |
| Moreno et al. [183] | 2021 | NSCLC-radiogenomics | SCAV with ML/CNN |
EGFR: AUC = 82.0 (CNN) KRAS: AUC = 73.9 (CNN) |
| Zhang et al. [182] | 2021 | Private (914 patients) |
Machine learning (SVM/RF/MLP) Deep learning (SE-CNN/CNN/1D-CNN/AlexNet/Fine-tuned VG16/Fine-tuned VGG19) |
EGFR: AUC = 91.0 (SE-CNN) AUC = 83.6 (SVM) |
| Le et al. [184] | 2021 | NSCLC-radiogenomics | LR / KNN / RF / XGBoost |
EGFR: ACC = 77.8 KRAS: ACC = 83.3 |
| Cheng et al. [187] | 2021 | Private (670 patients) |
Pre-trained 3D DenseNet |
EGFR: AUC = 76.0 ACC = 72.5 F-score = 71.3 |
| Zhang et al. [186] | 2021 | Private (134 patients) |
Logistic regression |
EGFR: AUC = 78.0 KRAS: AUC = 81.0 ERBB2: AUC = 87.0 TP53: AUC = 84.0 |
| Han et al. [185] | 2021 | Private (827 patients) |
Logistic Regression |
EGFR: AUC = 75.8 ALK: AUC = 73.9 |
ACC: Accuracy; AUC: area under the ROC curve; KNN: K-nearest neighbors; LR: logistic regression; MLP: multilayer perceptron; OR: odds ratio; RF: random forest; SCAV: selective class average voting; SE-CNN: squeezeand-excitation convolutional neural network; SVM: support vector machine; XGBoost: extreme gradient boosting.