Multi-task generative adversarial network for retinal optical coherence tomography image denoising.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Speckle noise in Optical Coherence Tomography (OCT) images degrades image quality and obscures critical details.
• Effective denoising is crucial for accurate ophthalmic disease assessment and clinical diagnosis.

Purpose of the Study:

• To introduce a novel Multi-task Generative Adversarial Network (MGAN) for denoising retinal OCT images.
• To enhance the preservation of retinal structural information during the denoising process.

Main Methods:

• The proposed MGAN integrates adversarial and multi-task learning for OCT image denoising.
• The generator performs both denoising and segmentation tasks simultaneously.
• A retina-attention module guides the denoising task to focus on retinal regions, preserving structural details using structural similarity index measure (SSIM) loss.

Main Results:

• MGAN demonstrated superior performance in speckle noise reduction and structural information preservation compared to existing methods.
• Evaluations on three public OCT datasets confirmed MGAN's effectiveness in improving image quality.
• Qualitative and quantitative analyses validated the method's advantages.

Conclusions:

• MGAN offers an effective solution for denoising retinal OCT images by preserving structural information.
• The proposed method has the potential to advance clinical applications of OCT in retinopathy observation and diagnosis.