Table 2.
Summary of studies on treatment of prostate cancer using deep learning models.
| Author | Year | Objective | Sample Size | Study Design | Model | AUC | DSC | SDI | MAE | Sn | Sp |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Sumitomo et al. [20] | 2020 | To predict risk of urinary incontinence following RARP using DL model based on MRI images | 400 patients | Retrospective | CNN model | 0.775 | N/A | N/A | N/A | N/A | N/A |
| Lai et al. [21] | 2021 | To apply DL methods for auto-segmentation of biparametric images into prostate zones and cancer regions. | 204 patients; T2W, DWI, ADC images used. |
Retrospective | Segnet | 0.958 | N/A | N/A | N/A | N/A | N/A |
| Sloun et al. [22] | 2020 | To use DL for automated real-time prostate segmentation on TRUS pictures. | 436 images 181 patients |
Prospective | U-Net | 0.98 | N/A | N/A | N/A | N/A | N/A |
| Schelb et al. [23] | 2020 | To compare DL system and multiple radiologists agreement on prostate MRI lesion segmentation | 165 patients; T2W and DWI used |
Retrospective | U-Net | N/A | 0.22 | N/A | N/A | N/A | N/A |
| Soerensen et al. [24] | 2021 | To develop a DL model for segmenting the prostate on MRI, and apply it in clinics as part of regular MR-US fusion biopsy. | 905 patients; T2W images |
Prospective | ProGNet and U-Net | N/A | 0.92 | N/A | N/A | N/A | N/A |
| Nils et al. [25] | 2021 | To assess the effects of diverse training data on DL performance in detecting and segmenting csPCa. | 1488 images; T2W and DWI images |
Retrospective | U-Net | N/A | 0.90 | N/A | N/A | 97% | 90% |
| Polymeri et al. [26] | 2019 | To validate DL model for automated PCa assessment on PET/CT and evaluation of PET/CT measurements as prognostic indicators | 100 patients | Retrospective | Fully CNN | N/A | N/A | 0.78 | N/A | N/A | N/A |
| Gentile et al. [27] | 2021 | To identify high grade PCa using a combination of different PSA molecular forms and PSA density in a DL model. | 222 patients | Prospective | 7-hidden-layer CNN | N/A | N/A | N/A | N/A | 86% | 89% |
| Ma et al. [28] | 2017 | To autonomously segment CT images using DL and multi-atlas fusion. | 92 patients | NA | CNN model | N/A | 0.86 | N/A | N/A | N/A | N/A |
| Hung et al. [29] | 2019 | To develop a DL model to predict urinary continence recovery following RARP and then use it to evaluate the surgeon’s past medical results. | 79 patients | Prospective | DeepSurv | N/A | N/A | N/A | 85.9 | N/A | N/A |