RBAD: A Dataset and Benchmark for Retinal Vessels Branching Angle Detection

TL;DR

This work introduces RBAD, a self-configured approach for detecting retinal branching angles, accompanied by an open-source annotation tool and a DRIVE-derived benchmark dataset of 40 images annotated for branching angles. The proposed method skeletonizes retinal segmentation, identifies bifurcations via a fast key-points strategy, aggregates a Gaussian heatmap to locate the root, and computes angles with a vector-based formula, yielding an angle map for analysis. Quantitative benchmarking shows the method outperforms previous approaches in MAE and MSE while maintaining robust and consistent angle estimates across images. The dataset and tools aim to accelerate interpretable retinal analysis and facilitate large-scale clinical research for early disease detection.

Abstract

Detecting retinal image analysis, particularly the geometrical features of branching points, plays an essential role in diagnosing eye diseases. However, existing methods used for this purpose often are coarse-level and lack fine-grained analysis for efficient annotation. To mitigate these issues, this paper proposes a novel method for detecting retinal branching angles using a self-configured image processing technique. Additionally, we offer an open-source annotation tool and a benchmark dataset comprising 40 images annotated with retinal branching angles. Our methodology for retinal branching angle detection and calculation is detailed, followed by a benchmark analysis comparing our method with previous approaches. The results indicate that our method is robust under various conditions with high accuracy and efficiency, which offers a valuable instrument for ophthalmic research and clinical applications.

Figures (6)

• Figure 1: Framework. (I) shows the annotation process, (b) is the proposed annotation tool, and (c) is the proposed angle detection method. Specifically, (a) is the original RGB image, (b) is the annotated image, (c) is the green channel, (d) is the edge-enhanced image with Laplacian filter, (e) is the image with high-pass-filter, (e) is the segmentation map of RGB image, and (f) is the key points map.
• Figure 2: Interface of Annotation Tool. (a) is the visually enhanced image with angle annotations overlaid. (b) is the annotation process, where an angle annotation is marked with three consecutive clicks, and the second point becomes the bifurcation. (c) shows the annotation editing function, angles can be deleted by alter-clicking (right-click) the bifurcation point.
• Figure 3: Some of the problems that tend to occur when labeling are listed, including blurred, overlapping vessels, and false angles due to overlap.
• Figure 4: The process of retina branching angle detection. (a) is the segmentation map, (b) is the Fast Key Points searching process in the selected region of (a), (c) is the key points map, where the red circle indicates an area that has the most bifurcations, (d) is the heatmap of bifurcation points, where the red mark is the location of highest pixel intensity which indicates the root, and (e) is the branching angle map.
• Figure 5: Retinal angle detection and calculation methods. (a) is the Line Detection-based method, (b) is the ROI-Window method, (c) is the rule-based method, and (d) is our proposed method.