Table 1.
Patient characteristics and pretreatment clinical data.
Fig 1.
Input PU-MRI images (ILR) were 4 times upscaled by each SR model. (a) EDSR, (b) WDSR, (c) SRGAN, and (d) RDN. Abbreviations: AF, activation function; BN, batch normalization; EDSR, enhanced deep super resolution network; PU-MRI, post urination magnetic resonance imaging; SR, super resolution; SRGAN, photo-realistic single image super resolution using a generative adversarial network; RDB, residual dense block; RDN, residual dense network; WDSR, wide activation for efficient and accurate image super resolution network.
Fig 2.
Representative images identifying the prostatic urinary tract in the axial direction for each SR image.
The above representative images show the best case (No. 15), while the bottom images show the worst case (No. 16) in the subjective evaluation. (A) On the left side of the PU-MRI image is the input image for each SR deep learning model, while on the right side of the PU-MRI image is the enlarged image of the prostate gland and urethra area. The output SR images (B–E) have the same matrix as the magnified PU-MRI image. Yellow arrow shows the prostatic urinary tract. Green arrow shows the gel spacer. Abbreviations: EDSR, enhanced deep super resolution network; PU-MRI, post urination magnetic resonance imaging; RDN, residual dense network; SRGAN, photo-realistic single image super resolution using a generative adversarial network; WDSR, wide activation for efficient and accurate image super resolution network.
Fig 3.
Bland-Altman plot showing the inter-operator difference in the urethra’s visibility score for each image set.
The red line denotes the mean of the difference; the orange dashed line denotes the 95% limits of agreement. Abbreviations: EDSR, enhanced deep super resolution network; PU-MRI, post urination magnetic resonance imaging; RDN, residual dense network; SRGAN, photo-realistic single image super resolution using a generative adversarial network; WDSR, wide activation for efficient and accurate image super resolution network.