DeshadowGAN: A Deep Learning Approach to Remove Shadows from Optical Coherence Tomography Images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical coherence tomography (OCT) is crucial for diagnosing optic nerve head (ONH) pathologies.
• Blood vessel shadows in OCT images can obscure important diagnostic features.
• Accurate shadow removal is essential for reliable OCT image analysis.

Purpose of the Study:

• To develop and validate a novel deep learning algorithm for removing blood vessel shadows from OCT images of the ONH.
• To quantitatively assess the effectiveness of the proposed algorithm in reducing shadow visibility across different retinal layers.

Main Methods:

• A custom generative adversarial network, DeshadowGAN, was designed and trained using a large dataset of OCT B-scans.
• OCT volume scans of the ONH were acquired from 13 subjects.
• Image quality was evaluated using intralayer contrast measurements in the RNFL, IPL, PR, and RPE layers, and compared to standard compensation methods.

Main Results:

• DeshadowGAN significantly reduced intralayer contrast across all analyzed retinal layers, indicating effective shadow removal.
• Average contrast reduction ranged from 28.8% to 43.0% across different layers.
• The algorithm produced artifact-free images, outperforming traditional compensation techniques.

Conclusions:

• DeshadowGAN effectively corrects blood vessel shadows in OCT images of the ONH.
• The algorithm serves as a valuable preprocessing step to enhance OCT segmentation, denoising, and classification algorithms.
• Integration of DeshadowGAN into OCT devices can improve the diagnosis and prognosis of ocular diseases.