Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SDPA++: A General Framework for Self-Supervised Denoising with Patch Aggregation

Huy Minh Nhat Nguyen, Triet Hoang Minh Dao, Chau Vinh Hoang Truong, Cuong Tuan Nguyen

TL;DR

SDPA++ tackles OCT denoising in the absence of ground truth by first creating pseudo-clean references through self-fusion and neighbor-based self-supervision, then training an ensemble of denoisers whose outputs are fused via a patch-wise strategy. The two-phase pipeline leverages both self-supervised targets and supervised models (NAFNet, CGNet) with a patch-based selection to improve noise suppression while preserving textures and edges. Experimental results on the VIP Cup dataset show improvements in metrics such as MSR, TP, and EP, with competitive CNR, suggesting more reliable retinal image quality for diagnosis. The approach is practical for clinical deployment since it operates on real noisy OCT data and culminates in a single deployable model for inference, potentially benefiting OCT-based diagnostics.

Abstract

Optical Coherence Tomography (OCT) is a widely used non-invasive imaging technique that provides detailed three-dimensional views of the retina, which are essential for the early and accurate diagnosis of ocular diseases. Consequently, OCT image analysis and processing have emerged as key research areas in biomedical imaging. However, acquiring paired datasets of clean and real-world noisy OCT images for supervised denoising models remains a formidable challenge due to intrinsic speckle noise and practical constraints in clinical imaging environments. To address these issues, we propose SDPA++: A General Framework for Self-Supervised Denoising with Patch Aggregation. Our novel approach leverages only noisy OCT images by first generating pseudo-ground-truth images through self-fusion and self-supervised denoising. These refined images then serve as targets to train an ensemble of denoising models using a patch-based strategy that effectively enhances image clarity. Performance improvements are validated via metrics such as Contrast-to-Noise Ratio (CNR), Mean Square Ratio (MSR), Texture Preservation (TP), and Edge Preservation (EP) on the real-world dataset from the IEEE SPS Video and Image Processing Cup. Notably, the VIP Cup dataset contains only real-world noisy OCT images without clean references, highlighting our method's potential for improving image quality and diagnostic outcomes in clinical practice.

SDPA++: A General Framework for Self-Supervised Denoising with Patch Aggregation

TL;DR

SDPA++ tackles OCT denoising in the absence of ground truth by first creating pseudo-clean references through self-fusion and neighbor-based self-supervision, then training an ensemble of denoisers whose outputs are fused via a patch-wise strategy. The two-phase pipeline leverages both self-supervised targets and supervised models (NAFNet, CGNet) with a patch-based selection to improve noise suppression while preserving textures and edges. Experimental results on the VIP Cup dataset show improvements in metrics such as MSR, TP, and EP, with competitive CNR, suggesting more reliable retinal image quality for diagnosis. The approach is practical for clinical deployment since it operates on real noisy OCT data and culminates in a single deployable model for inference, potentially benefiting OCT-based diagnostics.

Abstract

Optical Coherence Tomography (OCT) is a widely used non-invasive imaging technique that provides detailed three-dimensional views of the retina, which are essential for the early and accurate diagnosis of ocular diseases. Consequently, OCT image analysis and processing have emerged as key research areas in biomedical imaging. However, acquiring paired datasets of clean and real-world noisy OCT images for supervised denoising models remains a formidable challenge due to intrinsic speckle noise and practical constraints in clinical imaging environments. To address these issues, we propose SDPA++: A General Framework for Self-Supervised Denoising with Patch Aggregation. Our novel approach leverages only noisy OCT images by first generating pseudo-ground-truth images through self-fusion and self-supervised denoising. These refined images then serve as targets to train an ensemble of denoising models using a patch-based strategy that effectively enhances image clarity. Performance improvements are validated via metrics such as Contrast-to-Noise Ratio (CNR), Mean Square Ratio (MSR), Texture Preservation (TP), and Edge Preservation (EP) on the real-world dataset from the IEEE SPS Video and Image Processing Cup. Notably, the VIP Cup dataset contains only real-world noisy OCT images without clean references, highlighting our method's potential for improving image quality and diagnostic outcomes in clinical practice.
Paper Structure (20 sections, 7 equations, 5 figures, 5 tables)

This paper contains 20 sections, 7 equations, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Dataset Description
  4. Motivation
  5. Contribution
  6. Related Works
  7. Traditional Denoising Methods
  8. Supervised Denoising Methods
  9. Self-Supervised and Unsupervised Denoising Methods
  10. Methodology
  11. Phase 1: Self-Fusion and Self-Supervised Refinement
  12. Self-Fusion Preprocessing.
  13. Neighbor2Neighbor.
  14. Phase 2: Supervised Denoising and Patch Ensemble
  15. Color Scaling Post-Processing
  16. ...and 5 more sections

Figures (5)

  • Figure 1: Overview of proposed denoising method
  • Figure 2: Patch Ensemble Strategy
  • Figure 3: The effect of applying the proposed contrast enhancement function on the image.
  • Figure 4: Visual results of used methods on a B-scan
  • Figure 5: Visual of color scaling method