DeSpecNet: a CNN-based method for speckle reduction in retinal optical coherence tomography images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Speckle noise significantly degrades the quality of optical coherence tomography (OCT) images.
• Accurate analysis of retinal structures in OCT scans is crucial for diagnosing eye conditions.
• Existing speckle reduction methods often require manual parameter tuning, limiting their adaptability.

Purpose of the Study:

• To develop and evaluate a novel deep learning network, DeSpecNet, for automated speckle reduction in retinal OCT images.
• To improve the visual quality and quantitative metrics of OCT images through advanced noise suppression.
• To demonstrate the generalizability of the proposed method across various retinal OCT image types.

Main Methods:

• A deep convolutional neural network (CNN) architecture named DeSpecNet was designed.
• The network incorporates residual learning, shortcut connections, batch normalization, and leaky rectified linear units.
• The model learns speckle reduction directly from training data, eliminating the need for manual parameter selection.

Main Results:

• DeSpecNet demonstrated significant improvements in both visual quality and quantitative indices of retinal OCT images.
• The network effectively suppressed speckle noise while preserving crucial image details and edges.
• The proposed method outperformed existing state-of-the-art techniques in speckle reduction and feature enhancement.

Conclusions:

• DeSpecNet offers a powerful and automated solution for speckle reduction in OCT imaging.
• The network's ability to learn from data enhances its adaptability and performance compared to traditional algorithms.
• This deep learning approach holds significant potential for improving the diagnostic accuracy of OCT-based ophthalmology.