Table 2.

Summary of radiomics literature in prostate cancer using MR.

Author	Study Type	Application	Number of Patients	Results
Fehr et al., 2015 [90]	Retrospective	Cancer risk prediction	217	Textural features from T2WI and ADC could distinguish between different Gleason scores. Accuracy of 93% after cross-validation for discrimination of Gleason 6 (3 + 3) vs. Gleason ≥ 7, and 92% for discrimination of Gleason 3 + 4 = 7a vs. 4 + 3 = 7b.
Woźnicki et al., 2020 [91]	Retrospective	Cancer risk prediction	191	Radiomics characterizes prostatic index lesions accurate and perform comparable to radiologists for prostate cancer characterization. Prognostic machine learning models could help in detection of clinically significant prostate cancer and patient selection for MRI-guided fusion biopsy.
Li et al., 2020 [92]	Retrospective	Cancer risk prediction	381	3 models were developed: a clinical model, a radiomics model (T2WI and ADC), and a clinical-radiomics combined model. Radiomic (AUC 0.98) and combined model (AUC 0.98) perform better in prediction of clinically significant cancer than clinical model (AUC 0.79)
Xu et al., 2019 [93]	Retrospective	Cancer risk prediction	331	6 selected radiomics features of MRI (T2WI and ADC) performed better (AUC 0.92) than each alone (T2WI: AUC 0.81, ADC: AUC 0.89). Individual preoperative prediction model performs better when including clinical factors and radiomic features (clinical model: AUC 0.73; combined model: AUC 0.93).
Ma et al., 2020 [94]	Retrospective	Staging	119	Radiomics signature based on 17 features on T2WIs has the potential to predict preoperative risk of extracapsular extension, good performance in the validation set (AUC 0.821).
Zhang et al., 2020 [95]	Retrospective	Tumor grading	166	Radiomics model with signatures from T2WI, ADC and DCE perform better than any single sequence (AUC: radiomics model 0.87; AUC T2WI/ADC/DCE: 0.70/0.76/0.73). Combined model with radiomics signature, clinical stage, and time from biopsy to RP outperformed the clinical model and radiomics model (AUC: combined model 0.91, clinical model 0.65, radiomics model 0.87). MpMRI had the potential to predict tumor upgrade from biopsy to RP.
Gnep et al., 2017 [96]	Retrospective	Therapy response Biochemical recurrence	74	T2WI Haralick textural features appear be strongly correlate with biochemical recurrence after radiotherapy.
Shiradkar et al., 2018 [97]	Retrospective	Therapy response Biochemical recurrence	120	10 extracted radiomic features from pretreatment T2WI and ADC are significantly correlated with BCR and could be used for BCR prediction; after radiotherapy?
Stoyanova et al., 2016 [98]	Retrospective	Radiogenomics	17	Radiomic features extracted from biopsy regions of primary tumors (?) and normal tissues correlate significant with gene signatures associated with adverse outcome.
Fischer et al., 2019 [99]	Retrospective	Radiogenomics	298	Biomarkers that play critical roles in PCa showed high correlation with aggressiveness-related imaging features extracted from mp-MRI images. The use of multi-omics data has the potential of significantly improving prediction of prostate cancer aggressiveness.