Table 1 Summary of representative AI studies in lung cancer
From: Progress and challenges of artificial intelligence in lung cancer clinical translation
Reference | Year | Steps | Tasks | Number of cases | Data modality | Algorithm | Performance |
|---|---|---|---|---|---|---|---|
Engelhard et al.20 | 2021 | Prevention | Predict environment-associated risk of smoking craving | 48 | Environment images | CNN | AUC > 0.8 |
Senyurek et al.22 | 2020 | Prevention | Smoking detection | 40 | Puff topography | CNN | F1-score 78% |
Gould et al.28 | 2021 | Screening | Predict lung cancer risk | 196,102 | Clinical and laboratory data | XGBoost | AUC 0.86 |
Lu et al.29 | 2020 | Screening | Predict lung cancer risk | 41,856 | Medical records and chest X-ray | CNN | AUC 0.755 |
White et al.30 | 2017 | Screening | Predict lung cancer risk | 4,813,985 | Web Search Logs | Statistical classifier | True-positive rate ranging from 3% to 57%; False-positive rates ranging from 0.00001 to 0.001 |
Ardila et al.33 | 2019 | Screening | Malignancy risk estimation of lung nodules | 6,716 | Low-dose CT | CNN | AUC 0.944 |
Mikhael et al.34 | 2023 | Screening | Malignancy risk estimation of lung nodules | 15,000 | Low-dose CT | CNN | AUC 0.86–0.94 |
Kirienko et al.46 | 2018 | Diagnosis | Differentiation between primary and metastatic lung cancer | 534 | PET | Linear discriminant analysis (radiomics) | AUC 0.92 |
Wang et al.47 | 2023 | Diagnosis | Diagnosis of malignant pleural effusion | 918 | CT | CNN | AUC 0.842 |
Perez-Johnston et al.49 | 2022 | Diagnosis | Predict pathologic subtypes | 219 | CT | K-means clustering (radiomics) | N/A |
Coudray et al.55 | 2018 | Diagnosis | Subtype classification of NSCLC, driver mutation prediction | 1,634 | Whole-slide images | CNN | AUC 0.970 |
Rossi et al.51 | 2021 | Diagnosis | EGFR mutation prediction | 109 | CT | Support-vector machine (radiomics) | AUC 0.850 |
Yamamoto et al.52 | 2014 | Diagnosis | ALK mutation prediction | 172 | CT | Modified Random Forest (radiomics) | Accuracy 78.8% |
Wu et al.60 | 2022 | Diagnosis | PD-L1 expression assessment | 173 | Whole-slide images | CNN | Strong consistency between AI and pathologists (R = 0.9429–0.9458). |
Chen et al.53 | 2023 | Diagnosis | PD-L1 expression assessment | 194 | Contrast-enhanced CT | Linear Regression with FDR control; Elastic Net Regularization (radiomics) | AUC 0.70–0.72 |
Sun et al.54 | 2018 | Diagnosis | CD8+ T cell assessment | 135 | CT | Elastic net regression (radiomics) | AUC 0.67 |
Park et al.61 | 2022 | Diagnosis | Tumor-infiltrating lymphocytes assessment | 923 | Whole-slide images | Lunit SCOPE IO (self-developed analytical tool) | AUC 0.9715 |
Zhao et al.68 | 2018 | Prognosis | Predict invasiveness | 523 | CT | CNN | AUC 0.880 |
Hosny et al.73 | 2018 | Prognosis | Prognostic prediction | 1,194 | CT | CNN | AUC 0.70 |
Lu et al.74 | 2020 | Prognosis | Prognostic prediction | 1,057 | Whole-slide images | CNN | AUC 0.63 |
Zhong et al.75 | 2022 | Treatment | Predict lymph node metastasis | 3,096 | CT | CNN | AUC 0.82 |
Das et al.76 | 2023 | Treatment | Interpretation of pulmonary function tests | 24 | Pulmonary function test | Self-developed AI software | Accuracy increased by 10.4% |
Kadomatsu et al.79 | 2022 | Treatment | Intraoperative air leak sites detection | 110 | Endoscopic images | CNN | sensitivity and specificity of 81.3% and 68.9% |
Choi et al.81 | 2024 | Treatment | Predict cardiac toxicity after radiotherapy | 209 | PET/CT | Tree-based Pipeline Optimization Tool (radiomics) | 93% predictive accuracy |
Dercle et al.84 | 2023 | Treatment | Predict nivolumab response | 758 | CT | Random Forest (radiomics) | Sensitivity and specificity for predicting 3-month OS were 86% and 77.8%, |
Saad et al.88 | 2023 | Treatment | Predicting response to immune checkpoint inhibitors | 976 | CT | CNN | C-index 0.75 |
Arbour et al.102 | 2021 | Monitoring | RECIST evaluation | 453 | Text reports | Fully Connected Deep Learning Model | AUC 0.9 |
Tran et al.103 | 2024 | Monitoring | Identify patients with increased risk of relapse | 85 | CT and ctDNA | Cox regression model | AUROC 0.682 |
