Generative Adversarial Networks for Enhancing Image Resolution and Quality in Medical Imaging
Abstract
To obtain high-resolution (HR) Magnetic Resonance (MR) images, patients must remain stationary for extended periods, which can
cause discomfort and increase the risk of motion artifacts in the images. One approach to address this issue is to capture low-resolution (LR)
images and then enhance their quality using a Super Resolution Generative Adversarial Network (SRGAN) to produce an HR version. Capturing LR images is quicker and less uncomfortable for the patient, and it improves scanner efficiency. In this study, SRGAN is applied to MR
images of the prostate to enhance in-plane resolution by factors of 4 and 8. Here, super resolution refers to the post-processing enhancement
of medical images, whereas high resolution denotes the native resolution obtained during the MR scan. We also evaluated SRGAN against
three other models: SRCNN, SRResNet, and Sparse Representation. Although SRGAN may not achieve the highest Peak Signal-to-Noise
Ratio (PSNR) or Structural Similarity Index (SSIM) scores, it provides the closest visual resemblance to the original HR images, as evidenced
by the Mean Opinion Score (MOS).
cause discomfort and increase the risk of motion artifacts in the images. One approach to address this issue is to capture low-resolution (LR)
images and then enhance their quality using a Super Resolution Generative Adversarial Network (SRGAN) to produce an HR version. Capturing LR images is quicker and less uncomfortable for the patient, and it improves scanner efficiency. In this study, SRGAN is applied to MR
images of the prostate to enhance in-plane resolution by factors of 4 and 8. Here, super resolution refers to the post-processing enhancement
of medical images, whereas high resolution denotes the native resolution obtained during the MR scan. We also evaluated SRGAN against
three other models: SRCNN, SRResNet, and Sparse Representation. Although SRGAN may not achieve the highest Peak Signal-to-Noise
Ratio (PSNR) or Structural Similarity Index (SSIM) scores, it provides the closest visual resemblance to the original HR images, as evidenced
by the Mean Opinion Score (MOS).
Keywords
Generative Adversarial Networks; Image Resolution; Medical Imaging; Image Quality; Enhancement Techniques
Full Text:
PDFReferences
[1] Koshino, K., Werner, R.A., Pomper, M.G. Narrative review of generative adversarial networks in medical and molecular imaging // Nuclear Medicine. 2021.
[2] Singh, N.K., Raza, K. Medical image generation using generative adversarial networks: A review // Health informatics: A computational
perspective in healthcare.
[3] Gong, M., Chen, S., Chen, Q., Zeng, Y. Generative adversarial networks in medical image processing // Current Pharmaceutical Design.
DOI: http://dx.doi.org/10.70711/aitr.v2i4.4871
Refbacks
- There are currently no refbacks.
