Table 8.
Overview of published works regarding the prediction of PD-L1 expression status in lung cancer CT images (2017–2021).
| Authors | Year | Dataset | Methods | Performance Results (%) |
|---|---|---|---|---|
| Toyokawa et al. [193] | 2017 | Private (394 patients) |
Fisher’s exact test Univariate/multivariate LR (CT features) |
PD-L1+ statistical association: (p < 0.01)—convergence, notching, spiculation, cavitation |
| Wu et al. [194] | 2019 | Private (350 patients) |
Univariate/multivariate LR Fisher’s exact test Mann–Whitney U test |
AUC = 78.3 TPR = 81.1 TNR = 64.1 |
| Zhu et al. [195] | 2020 | Private (127 patients) |
Univariate/multivariate LR 3D DenseNet |
AUC = 78.0 ACC = 77.8 TPR = 77.8 TNR = 77.4 |
| Jiang et al. [197] | 2020 | Private (399 patients) |
Random forest Logistic regression |
AUC = 97.0 (≥1%) AUC = 80.0 (≥50%) |
| Tian et al. [196] | 2021 | Private (939 patients) |
Fully connected classifier | AUC = 76.0 |
| Yang et al. [199] | 2021 | Private (200 patients) |
Simple temporal attention (SimTA) module | AUC = 77.0 (SimTA60) AUC = 80.0 (SimTA90) AUC = 69.0 (RNN) AUC = 64.0 (Radiomics) |
| Jiang et al. [198] | 2021 | Private (125 patients) |
Random forest Decision tree Logistic regression AdaBoost Support vector machine |
(Internal validation) AUC = 96.0 TNR = 80.0 TPR = 98.5 (External validation) AUC = 85.0 TNR = 63.6 TPR = 91.3 |
ACC: accuracy; AUC: area under the ROC curve; LR: logistic regression; RNN: recurrent neural network; SimTAx: response prediction x days post immunotherapy; TNR: true negative rate; TPR: true positive rate.