Table 2 Main studies on radiomics/AI imaging applications for primary treatments with curative intent in the setting of prostate cancer management (one example for each main potential application is proposed)
From: Beyond diagnosis: is there a role for radiomics in prostate cancer management?
First author [reference number] | Publication year | Country | Treatment | Aim | Design | Sample size | Imaging modality | Main outcome | Potential impact |
|---|---|---|---|---|---|---|---|---|---|
Künzel [32] | 2021 | Germany | RT | Autonomous treatment planning | Prospective, single-centre | 1 | MRI (AI) | Feasibility of autonomous treatment planning for adaptive MRIgRT proven | Reduction of time to treatment and RO workload. Potential first step toward real-time RT |
Shiradkar [35] | 2016 | USA | RT | Targeted focal treatment planning | Retrospective, multicentre | 23 | MRI (radiomics) | Radiomics framework showed reduction in dosage to OARs and boosted dose to index lesions | Decision support framework for RO to elaborate more effective and targeted treatments |
Rossi [48] | 2018 | The Netherlands | RT | Toxicity prediction | Retrospective, multicentre | 351 | Dose distributions (dosiomics) | Dosiomics showed an added value for the prediction of RT toxicity, statistically significant for GI toxicity compared to NCTP | More reliable prediction of adverse events might aid RO in treatment planning |
Abdollahi [38] | 2019 | Iran | RT | Prediction of treatment response | Retrospective, single-centre | 33 | MRI (radiomics) | Pre-treatment MRI radiomics might identify non-responders to IMRT (AUC 0.78) | Different treatment options could be preferred for potential non-responders |
Solari [52] | 2021 | Germany | RP | Prediction of GS upgrade from Bx | Retrospective, single-centre | 101 | [68Ga] Ga-PSMA-11 PET/MRI (radiomics) | Combined PET+ADC radiomics model outperformed Bx in the prediction of GS at RP (accuracy of 82.5% versus 72.4%) | More reliable patient risk stratification could be achieved to guide management |
Cuocolo [59] | 2021 | Italy | RP | Detection of EPE | Retrospective, multi-centre | 193 | MRI (radiomics) | Radiomics model for EPE detection showed good generalizability in a multicentre setting and might aid radiologists in PCa staging | Low EPE probability at pre-treatment MRI might help select suitable candidates for nerve-sparing surgery |
Zheng [67] | 2022 | USA | RP | Evaluation of lymph node involvement | Retrospective, single-centre | 244 | MRI (radiomics) | A combined model (clinical and radiomics features) outperformed pre-existing nomograms for the prediction of lymph node status (AUC 0.915 versus 0.724) | Reducing the number of unnecessary ePLND identifying patients with low probability of lymph node involvement |