ROSE: A Retinal OCT-Angiography Vessel Segmentation Dataset and New Model (2007.05201v2)

Abstract: Optical Coherence Tomography Angiography (OCT-A) is a non-invasive imaging technique, and has been increasingly used to image the retinal vasculature at capillary level resolution. However, automated segmentation of retinal vessels in OCT-A has been under-studied due to various challenges such as low capillary visibility and high vessel complexity, despite its significance in understanding many eye-related diseases. In addition, there is no publicly available OCT-A dataset with manually graded vessels for training and validation. To address these issues, for the first time in the field of retinal image analysis we construct a dedicated Retinal OCT-A SEgmentation dataset (ROSE), which consists of 229 OCT-A images with vessel annotations at either centerline-level or pixel level. This dataset has been released for public access to assist researchers in the community in undertaking research in related topics. Secondly, we propose a novel Split-based Coarse-to-Fine vessel segmentation network (SCF-Net), with the ability to detect thick and thin vessels separately. In the SCF-Net, a split-based coarse segmentation (SCS) module is first introduced to produce a preliminary confidence map of vessels, and a split-based refinement (SRN) module is then used to optimize the shape/contour of the retinal microvasculature. Thirdly, we perform a thorough evaluation of the state-of-the-art vessel segmentation models and our SCF-Net on the proposed ROSE dataset. The experimental results demonstrate that our SCF-Net yields better vessel segmentation performance in OCT-A than both traditional methods and other deep learning methods.

• The paper presents ROSE, the first annotated OCTA dataset with centerline and pixel-level vessel annotations for 229 images.
• The paper proposes OCTA-Net, a dual-stage segmentation model that uses a split-based coarse-to-fine approach to accurately detect both thick and thin vessels.
• Performance tests demonstrate that OCTA-Net outperforms state-of-the-art models, highlighting its potential in neurodegenerative disease research.

An Analysis of "ROSE: A Retinal OCT-Angiography Vessel Segmentation Dataset and New Model"

The paper "ROSE: A Retinal OCT-Angiography Vessel Segmentation Dataset and New Model," authored by Ma et al., presents an intriguing advancement in the field of retinal image analysis, specifically addressing the challenge of vessel segmentation in Optical Coherence Tomography Angiography (OCTA) images. The significance of automated vessel segmentation in OCTA stems from its potential utility in diagnosing and managing vision-related and neurodegenerative diseases. The authors introduce the ROSE dataset, a novel annotated OCTA dataset, and propose a specialized segmentation model named OCTA-Net.

Contributions and Dataset

The ROSE dataset constitutes a pivotal aspect of this work. As the first publicly available dataset with precise vessel annotations for OCTA images, ROSE is a significant contribution to the field. It includes 229 OCTA images, meticulously annotated at both centerline and pixel levels, facilitating the training and evaluation of segmentation algorithms. This dataset addresses the notable gap in resources available for OCTA-specific algorithm development, offering a benchmark for researchers to validate and compare segmentation techniques.

OCTA-Net Architecture

The proposed OCTA-Net introduces a dual-stage segmentation network specifically designed for OCTA images. The architecture incorporates a split-based coarse-to-fine approach, enabling the differentiation and precise detection of both thick and thin vessels. The initial stage employs a split-based coarse segmentation module, generating a preliminary confidence map of vessel structures. The subsequent refined segmentation module then enhances the shape and contour precision of the retinal microvasculature. This dual-stage approach facilitates improved segmentation accuracy, particularly for the densely connected capillaries that often challenge existing methods.

Performance Evaluation

The experimental results underscore the efficacy of OCTA-Net when evaluated against state-of-the-art segmentation models on the ROSE dataset. OCTA-Net demonstrates superior performance metrics, including Area Under the ROC Curve (AUC), Dice coefficient, and accuracy in both pixel-level and centerline-level segmentation tasks. Furthermore, a detailed fractal dimension analysis of the segmented microvasculature reveals significant differences between healthy control groups and those with Alzheimer's Disease, suggesting the potential application of OCTA-Net in neurodegenerative disease research.

Methodological Implications

The application of ResNeSt blocks within the OCTA-Net architecture is a noteworthy methodological choice, leveraging split attention mechanisms for improved feature extraction. The inclusion of a fine segmentation stage is particularly beneficial for optimizing the segmentation of small capillaries, which are crucial for detecting subtle pathological changes in the retina.

Future Directions

The authors posit that ROSE and OCTA-Net pave the way for further exploration of retinal microvascular changes as biomarkers for systemic and neurodegenerative diseases. Future research could expand ROSE to include images from diverse pathological conditions, enhancing its applicability as a robust benchmarking tool. Additionally, improvements in the OCTA imaging process, particularly in the removal of projection artifacts, could further refine segmentation performance.

In conclusion, this paper presents a comprehensive analysis of OCTA image segmentation, leveraging both a novel dataset and an innovative segmentation model to advance the capabilities of automated retinal imaging. The implications of this work are far-reaching, offering potential insights into the intersection of retinal image analysis and broader medical diagnostics.