Table 6.

Performance of machine learning

Authors	Study task	Method	Result
Park et al. [21]	To investigate whether CT slice thickness influences the performance of radiomics prognostic models	Patients with NSCLC with development set (n=185) and validation set (n=126), using radiomic prediction with CT slices thickness 1, 3, 5 mm	AUC with development set=0.68 to 0.7, AUC with validation set=0.73 to 0.76
Lao et al. [22]	To investigate if deep features extracted via transfer learning can generate radiomics signatures for prediction of overall survival in patients with Glioblastoma Multiforme	LASSO Cox regression model was applied for 75 patient set for training and 37 patient set for validation	AUC =0.71 to 0.739
Liu et al. [23]	Prediction of distant metastasis through deep learning radiomics	235 patients with neoadjuvant chemotherapy using deep learning radiomics signature	AUC =0.747 to 0.775
Vils et al. [24]	Evaluation of radiomics in recurrent glioblastoma	Multivariable models using radiomic feature selection for patients (n=69 training, n=49 validation)	AUC=0.67 to 0..673
Our study	Different machine learning algorithms to evaluate radiomics prediction of NSCLC	6 machine algorithms used for 422 patients with 70:30 cross-validation	Average AUC=0.67 to 0.91