Table 2.
Clinical application of radiomics in molecular and pathological analysis.
| Study | Number of patients | Tumor characteristic | Imaging modality | Function and Prediction results | Segmentation and feature selection method/model | Machine learning algorithm |
|---|---|---|---|---|---|---|
| Yang et al.39 | 467 | lung adenocarcinoma (LADC) | CT | Predicting EGFR mutationAUC 0.789 | Nodule segmentation 3D U-net model and pyradiomics | RF |
| Feng et al.40 | 300 | Breast cancer | CT | Predicting triple negative breast cancerAUC 0.851 | Manual segmentation and LASSO logistic method | Statistics |
| Ma et al.41 | 140 | Solid Lung Adenocarcinoma | CT | Predicting AnaplasticLymphoma Kinase Gene RearrangementAUC 0.801 | Pearson correlation coefficient and ANOVA or RFE | SVM |
| Marentakis et al.22 | 102 | Lung cancer | CT | Histological classificationAUC 0.78 | Joint FDG-PET and MRI prediction of lung metastases | LSTM + Inception |
| Zhang et al.42 | 420 | lung adenocarcinoma | CT | Predicting EGFR mutation statusAUC 0.835 | LASSO and Wilcoxon test, DT, logistic regression | SVM |
| Wu et al.43 | 74 | hepatocellular carcinoma | CT | Predicting the Ki-67 marker index | Statistics | Logistic regression |
| Zhao et al.27 | 579 | pulmonary adenocarcinoma | CT | Predicting EGFR mutation statusAUC 0.75 | Manual delineate | 3D DenseNets |
| Li et al.44 | 207 | colon cancer | CT | Predicting perineural invasion and KRAS mutationAUC 0.793 and 0.862 | Manual delineate | SVM |
| Wang et al.28 | 844 | lung adenocarcinoma | CT | Predicting EGFR mutation statusAUC 0.81 | A cubic ROI containing the entire tumour manual select | Deep learning model |
| Sutton et al.45 | 273 | breast cancer | MRI | Classifying pathologic response post-neoadjuvant chemotherapyAUC 0.83 | GMMGLMNet-RF-RFE | Statistics |
| Fan et al.46 | 144 | Breast Cancer | MRI | Predicting histological grade and Ki-67 expression levelAUC 0.814 and 0.810 | Spatial fuzzy C-means algorithm refined by a Markov random field | Multitask learning method |
| Shofty et al.47 | 47 | low-grade gliomas | MRI | 1p/19q codeletion status predictionAUC 0.87 | AnalyzeDirect software segmentation | Ensemble Bagged Trees |
| Park et al.48 | 121 | low-grade gliomas | MRI | Predicting molecular features of glioblastoma in Isocitrate Dehydrogenase Wild-TypeAUC 0.863 | Clinical feature + VASARI + radiomics feature | RFESVM |
| Yan et al.29 | 357 | glioma | MRI | Predicting IDH and TERT statusAUC 0.884 and 0.669 | WaveletLASSO | Bayesian-regularization neural networks |
| Wu et al.49 | 126 | diffuse gliomas | MRI | Predicting isocitrate dehydrogenase genotypeAUC 0.931 | Automated segmentation | RF |
| Braman et al.50 | 117 | Breast cancer | MRI | Predicting pathological complete response to neoadjuvant chemotherapyAUC 0.74 | A combined intratumoral and peritumoral radiomics approach | Cluster |
| Niu et al.51 | 182 | High-Grade Gliomas | MRI | Estimating the IDH1 GenotypeAUC 0.86 | Statistics | LASSO |
| Umutlu et al.52 | 124 | Breast cancer | PET/MRI | Breast Cancer Phenotyping and Tumor Decoding | Statistics | LASSO |
| Zheng et al.26 | 584 | Breast cancer | US | Predicting axillary lymphnode statusAUC 0.902 | Deep learning radiomics model | Deep learning radiomics model |
ANOVA, analysis of variance; DT, decision tree; RF, random forest; LASSO, least absolute shrinkage and selection operator; RFE, recursive feature elimination; SVM, support vector machine.