Eliminating Shadow Artifacts via Generative Inpainting Networks to Quantify Vascular Changes of the Choroid

Huihong Zhang, Jianlong Yang, Kang Zhou, Fei Li, Yan Hu, Shenghua Gao, Jun Cheng, Yanwu XuarXiv 2019 · DOI · arXiv

Abstract

Shadow artifacts from retinal vessels hinder the development of quantitative choroidal biomarkers in Optical Coherence Tomography (OCT), which limits the clinical applications of the choroidal OCT to the measures of layer thickness. In this paper, we present a novel framework that could eliminate the retinal shadows and realize the quantification of choroidal vasculature. Different from existing methods, we convert the shadow elimination into an object removal task, which can be handled by image inpainting techniques. We adopt and finetune a two-stage generative inpainting network which connects the edges of the vessels ahead of refilling the shadow-contaminated areas. It shows surpassing performance in repairing the choroidal vasculature compared with traditional inpainting algorithms. We further verify the feasibility of the proposed frame using a prospective observational study, which detects the choroidal vascular changes related to the Intra-Ocular Pressure (IOP) elevation of 34 healthy volunteers. As the average IOP increases from $15.84\pm1.99$ mmHg to $34.48\pm5.35$ mmHg, we achieve the decreases of the choroidal vessel density and flow index from $0.491\pm0.020$ to $0.463\pm0.019$ and from $0.336\pm0.025$ to $0.300\pm0.019$, respectively.

Introduction

Optical Coherence Tomography (OCT) is a cornerstone of modern ophthalmology, providing detailed 3D visualizations of ocular layers. However, a persistent challenge in OCT imaging is the presence of shadow artifacts. Superficial retinal blood vessels scatter and absorb light, casting dark, tail-like shadows onto deeper structures like the choroid—the vital vascular layer situated between the retina and the sclera.

Because these shadows obscure underlying tissue, clinical assessment of the choroid has been largely limited to simple thickness measurements. While Attenuation Compensation (AC) algorithms can improve contrast in dim areas, they often fail to recover information lost in the dense shadows of large vessels. This research proposes a paradigm shift: treating these shadows not as areas to be brightened, but as missing data to be reconstructed through advanced image inpainting.

Methodology

The proposed framework operates in three distinct stages: Preprocessing, Generative Inpainting, and Quantification.

The three-stage framework: Preprocessing, Generative Inpainting Network (GIN), and Quantification.
  1. Preprocessing: An AC algorithm enhances image contrast, followed by the segmentation of the Retinal Pigment Epithelium (RPE) and choroid layers. The RPE, which is naturally avascular and uniform, serves as a "shadow mask." Any dark regions detected in the RPE are identified as shadows cast by overlying retinal vessels.
  2. Generative Inpainting Network (GIN): The system utilizes a two-stage Generative Adversarial Network (GAN) architecture. The first stage, the Edge Generator, predicts the missing edges of choroidal vessels within shadowed regions. The second stage, Image Completion, uses these predicted edges as a structural guide to fill in textures and intensities, ensuring morphological consistency.
  3. Quantification: Following shadow removal, the system calculates two critical biomarkers: Vessel Density and a Relative Flow Index.

Results and Clinical Validation

The framework was evaluated using a dataset of 68 eyes from 34 healthy volunteers. To simulate clinical utility, scans were taken at baseline and in an upside-down position to naturally induce high Intraocular Pressure (IOP).

Quantitative and Qualitative Performance

The fine-tuned GIN significantly outperformed traditional inpainting methods, such as Exemplar-Based Inpainting (EBI) and Coherence Transport Inpainting (CTI). Quantitatively, the model achieved a Structural Similarity Index (SSIM) of 0.946 and a Peak Signal-to-Noise Ratio (PSNR) of 30.615. Qualitatively, the GIN produced choroidal maps that were virtually indistinguishable from unshadowed tissue, avoiding the unnatural artifacts seen in traditional methods.

Qualitative comparison of the proposed method against original images, shadow masks, and traditional inpainting techniques.

Clinical Findings

The framework successfully detected vascular changes associated with IOP elevation. When volunteers moved to the upside-down position, the average IOP rose from 15.84 mmHg to 34.48 mmHg. Correspondingly, the system measured a decrease in choroidal vessel density (from 0.491 to 0.463) and a decrease in the flow index (from 0.336 to 0.300), demonstrating the sensitivity of these new biomarkers.

Discussion and Conclusion

By reframing shadow removal as an inpainting task, this framework enables the first accurate quantification of choroidal vasculature in the presence of retinal shadows. However, the method currently relies on a healthy RPE for mask generation, which may limit its application in patients with diseases like Age-related Macular Degeneration (AMD) that distort the RPE. Additionally, while effective for larger choroidal vessels, care must be taken when applying such generative models to microvasculature to avoid "hallucinating" connections that do not exist. Future research will focus on RPE-independent shadow detection and broader clinical validation.

Summary updated

Abstract

Shadow artifacts from retinal vessels hinder the development of quantitative choroidal biomarkers in Optical Coherence Tomography (OCT), which limits the clinical applications of the choroidal OCT to the measures of layer thickness. In this paper, we present a novel framework that could eliminate the retinal shadows and realize the quantification of choroidal vasculature. Different from existing methods, we convert the shadow elimination into an object removal task, which can be handled by image inpainting techniques. We adopt and finetune a two-stage generative inpainting network which connects the edges of the vessels ahead of refilling the shadow-contaminated areas. It shows surpassing performance in repairing the choroidal vasculature compared with traditional inpainting algorithms. We further verify the feasibility of the proposed frame using a prospective observational study, which detects the choroidal vascular changes related to the Intra-Ocular Pressure (IOP) elevation of 34 healthy volunteers. As the average IOP increases from $15.84\pm1.99$ mmHg to $34.48\pm5.35$ mmHg, we achieve the decreases of the choroidal vessel density and flow index from $0.491\pm0.020$ to $0.463\pm0.019$ and from $0.336\pm0.025$ to $0.300\pm0.019$, respectively.
BibTeX
@misc{zhang2019eliminating,
  author = {Huihong Zhang and Jianlong Yang and Kang Zhou and Fei Li and Yan Hu and Shenghua Gao and Jun Cheng and Yanwu Xu},
  title = {{Eliminating Shadow Artifacts via Generative Inpainting Networks to Quantify Vascular Changes of the Choroid}},
  year = {2019},
  eprint = {1907.01271},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1907.01271},
}
Figure