Retinal optical coherence tomography image enhancement via deep learning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical coherence tomography (OCT) is crucial for diagnosing retinal diseases.
• Speckle noise in OCT images degrades quality and diagnostic reliability.
• Deep learning offers potential for advanced image processing in medical fields.

Purpose of the Study:

• To develop and evaluate novel deep learning-based speckle reduction methods for OCT images.
• To compare the performance of two distinct deep learning approaches for OCT denoising.
• To assess the impact of speckle reduction on image quality and diagnostic utility.

Main Methods:

• Training convolutional neural networks (CNNs) on retinal OCT cross-sections.
• Implementing two denoising strategies: mean-squared error (MSE) and generative adversarial network (GAN) with perceptual loss.
• Evaluating denoising performance using quantitative metrics (PSNR, SSIM) and qualitative assessments.

Main Results:

• The MSE-based method achieved state-of-the-art quantitative improvements and aided layer segmentation.
• The GAN-based method, prioritizing visual perception, showed superior qualitative results in accuracy, clarity, and user preference.
• Both deep learning approaches demonstrated effectiveness in reducing speckle noise in OCT images.

Conclusions:

• Deep learning provides an effective and efficient solution for speckle noise reduction in retinal OCT imaging.
• The choice between MSE and GAN-based denoising depends on whether quantitative metrics or visual perception is prioritized.
• These advanced denoising techniques can enhance the reliability of OCT-based eye disease diagnosis and monitoring.