Table 1.
Description of pan-cancer radiomic studies focused on genomics.
| Cancer type | Genomic feature | Imaging modality | Statistics and modelling | Patients | Result | References |
|---|---|---|---|---|---|---|
| Gliomas | ||||||
| IDH1 mutation | MRI | Multivariate RF classifier model |
Training, n = 100 Validation, n = 100 |
The combined model produced a maximum AUC of 78.24%. | [3] | |
| IDH1 mutation | MRI | the SVM-based recursive feature elimination (SVM-RFE) algorithm | The AUC of the predicted IDH1( + ) STATUS is 94.74%. | [4] | ||
| IDH1 mutation | MRI |
Training, n = 60 Validation, n = 20 |
The DWI-trained XGBoost model performed best, which achieved ROC on the test set with an area under the curve (AUC) of 0.97. | [5] | ||
| IDH mutation | MRI | Convolutional Neural Networks | n = 214 | T2-net demonstrated a mean cross-validation accuracy of 97.14% ± 0.04 in predicting IDH mutation status. | [6] | |
| IDH mutation | MRI | Convolutional Neural Networks |
Training, n = 727 Validation, n = 129 |
The hybrid model achieved accuracies of 93.8%, 87.9% and 78.8%. | [7] | |
| ATRX mutation | MRI | Linear SVM model and random forest model | 95 | 94.0% | [8] | |
| ATRX mutation | 18F-FET/PET and MRI | Support vector machines and random forest models | 42 | The five fold cross-validated area under the curve in predicting the ATRX mutation was 85.1%. | [9] | |
| ATRX mutation | MRI | the Elastic Net regression model | 111 | The radiomics nomogram identified LrGG patients for ATRX loss (C-index: training sets = 0.863, validation sets = 0.840). | [10] | |
| MGMT methylation | MRI | Multiple logistic regression model |
Training, n = 105 Validation, n = 31 |
The fusion radiomics signature exhibited supreme power for predicting MGMT promoter methylation, with an AUC of 0.925 in the training cohort and 0.902 in the validation cohort. | [11] | |
| MGMT methylation | 18F-DOPA-PET | Random forest models |
Training, n = 59 Validation, n = 10 |
Achieved 80% ± 10% accuracy for a 95% confidence level in predicting MGMT status. | [12] | |
| Glioblastomas | ||||||
| MGMT methylation | MRI | random forest models |
Training, n = 130 Validation, n = 60 |
Radiomics model built from multiregional and multiparameter MRI may serve as a potential imaging biomarker for pre-treatment prediction of MGMT methylation in GBM. | [13] | |
| MGMT methylation | MRI | Multivariate Cox model |
Training, n = 120 Validation, n = 61 |
Radiological characteristics together with MGMT status were the only parameters with independent significance in the multivariate analysis (P ≤ 0.01). | [14] | |
| MGMT methylation | MRI | multivariable Cox-regression model |
Training, n = 142 Validation, n = 46 |
The predictive model performed significantly in the external validation of MGMT methylation (AUC 0.667, 95% CI 0.522–0.82). | [15] | |
| NSCLC | ||||||
| EGFR mutation | 18F-FDG-PET and CT | CS model/ multivariable logistic regression analyses |
Training, n = 429 Validation, n = 187 |
Deep-learning score (EGFR-DLS) is significantly and positively associated with longer progression-free survival (PFS). | [16] | |
| EGFR mutation | MRI | Gradient boosting classifier model | Patients=110 | Data support the use of radiological scores based on MR imaging of NSCLC brain metastases as a non-invasive biomarker of survival. | [17] | |
| KRAS mutation | CT | LASSO regression model |
Training, n = 145 Validation, n = 101 |
This diagnostic/prognostic study examined a CT-based DL approach to predict the efficacy of EGFR-TKI therapy in patients. | [18] | |
| KRAS mutation | CT | LASSO regression model | 134 | The AUCs for the combined models used to identify KRAS and TP53 mutations were 0.81, and 0.84, respectively. | [19] | |
| KRAS mutation | PET and CT | Radiomics score (RS) models |
Training, n = 180 Validation, n = 78 |
the PET/CT radiomics score model exhibited a higher AUC for predicting KRAS mutations (0.83). | [20] | |
| ALK rearrangements | CT | Generalised boosted regression model (GBM) | Training, n = 84 | The average accuracy of the model calculated on the independent nested validation set was 0.81. | [21] | |
| ALK rearrangements | CT | LASSO regression model |
Training, n = 268 Validation, n = 67 |
The addition of conventional CT features enhanced the validation performance of the radiomic model in the primary cohort (AUC = 0.83–0.88). | [22] | |
| ALK rearrangements | PET and CT | LASSO logistic regression |
Training, n = 368 Validation, n = 158 |
This combined model PET/CT clinical model has a significant advantage to predict the ALK mutation status in the training group (AUC = 0.87). | [23] | |
| Colorectal cancer | ||||||
| KRAS mutation | CT | RELIEF and support vector machine methods |
Training, n = 61 Validation, n = 56 |
The AUC, sensitivity, and specificity for predicting KRAS/NRAS/BRAF mutations were 0.86. | [24] | |
| KRAS mutation | MRI | SVM classifiers |
Training, n = 213 Validation, n = 177 |
The proposed T2WI-based radiomics signature has a moderate performance to predict KRAS status. | [25] | |
| KRAS mutation | CECT | Artificial neural network method (ANN) |
Training, n = 93 Validation, n = 66 |
The combined score could distinguish between wild-type and mutant patients with an AUC of 0.95 in the primary cohort. | [116] | |
| Clear cell renal cell carcinoma | ||||||
| BAPI mutation | CT | Random forest model | Patients=54 | The AUC of the random forest model for predicting the mutation status of BAP1 was 0.77. | [117] | |
| VHL mutation | CT | Random forest model |
Training, n = 170 Validation, n = 85 |
The model with eight all-relevant features achieved an AUC of 0.949 in the validation cohort. | [118] | |
| VHL mutation | CT | Random forest model |
Training, n = 207 Validation, n = 175 |
Using radiomics features, the random forest algorithm showed a good capacity to identify the mutations VHL (AUC = 0.971). | [119] | |