Automated Detection of Leakage in Fluorescein Angiography Images with Application to Malarial Retinopathy

Yitian Zhao, Ian J. C. MacCormick, David Parry, Sophie Leach, Nicholas A. V. Beare, Simon Harding, Yalin ZhengScientific 2015 · DOI

Abstract

The detection and assessment of leakage in retinal fluorescein angiogram images is important for the management of a wide range of retinal diseases. We have developed a framework that can automatically detect three types of leakage (large focal, punctate focal, and vessel segment leakage) and validated it on images from patients with malarial retinopathy. This framework comprises three steps: vessel segmentation, saliency feature generation and leakage detection. We tested the effectiveness of this framework by applying it to images from 20 patients with large focal leak, 10 patients with punctate focal leak, and 5,846 vessel segments from 10 patients with vessel leakage. The sensitivity in detecting large focal, punctate focal and vessel segment leakage are 95%, 82% and 81%, respectively, when compared to manual annotation by expert human observers. Our framework has the potential to become a powerful new tool for studying malarial retinopathy, and other conditions involving retinal leakage.

Introduction

Cerebral malaria (CM) remains a devastating condition, particularly among pediatric populations, often leading to severe brain swelling and mortality. Because the retina is embryologically an extension of the central nervous system, "malarial retinopathy" (MR) provides a unique, non-invasive window into the disease's progression in the brain. A hallmark of MR is the leakage of fluid from retinal vessels, which clinicians typically assess using Fluorescein Angiography (FA). In FA, a fluorescent dye highlights vascular abnormalities, but traditional analysis is qualitative, subjective, and time-consuming. There is a critical need for automated, quantitative tools capable of processing clinical images from critically ill patients.

Methodology

The researchers developed a three-stage automated pipeline designed to identify and quantify leakage in FA images:

  1. Vessel Segmentation: The system first maps the vascular tree using local phase-based enhancement—which remains invariant to uneven lighting—followed by a Graph Cut-based Active Contour model.
  2. Saliency Feature Generation: Drawing from computer vision concepts, "saliency" identifies objects that stand out from their background. Leaking regions are modeled as salient features based on intensity differences and spatial weighting, appearing as bright, blurred "clouds" against the darker retinal background.
  3. Leakage Detection: The framework categorizes leaks into three distinct types based on morphology and location:
  4. Large Focal Leakage: Circular areas with a diameter greater than 125μm.
  5. Punctate Focal Leakage: Small, intense spots less than 125μm.
  6. Vessel Segment Leakage: Increased brightness and blurring specifically along vessel walls.
Examples of leakage types: (a) large focal leakage, (b) punctate focal leakage, and (c) vessel segment leakage.

To refine detection, the system utilizes Retinex theory to correct for heterogeneous brightness and automatically filters out the optic disc to prevent false positives.

Results and Discussion

The automated framework was evaluated against a ground truth established by expert human graders using a dataset of FA images from children with cerebral malaria in Malawi. The system demonstrated high sensitivity across all categories:

  • Large Focal Leakage: 95% sensitivity with zero false positives.
  • Punctate Focal Leakage: 82% sensitivity. Discrepancies often occurred in areas where human experts also showed low inter-observer agreement.
  • Vessel Segment Leakage: 81% sensitivity and 82% specificity. Notably, the system achieved an Area Under the Curve (AUC) of 0.815, outperforming one of the individual human experts (AUC 0.765).

While the system is highly effective, limitations include "crowding effects" where close punctate leaks are merged, and occasional misclassification of leaks near vessels. Artifacts from involuntary eye movements (nystagmus) in comatose patients also present challenges for automated sensors.

Conclusion

This framework represents the first automated tool for quantifying multiple types of retinal leakage in FA images. By turning qualitative observations into repeatable metrics, such as the 125μm threshold for focal leaks, this technology provides a vital tool for studying malarial retinopathy. Future work aims to extend this application to other major retinal diseases, including diabetic retinopathy and age-related macular degeneration.

Summary updated

Abstract

The detection and assessment of leakage in retinal fluorescein angiogram images is important for the management of a wide range of retinal diseases. We have developed a framework that can automatically detect three types of leakage (large focal, punctate focal, and vessel segment leakage) and validated it on images from patients with malarial retinopathy. This framework comprises three steps: vessel segmentation, saliency feature generation and leakage detection. We tested the effectiveness of this framework by applying it to images from 20 patients with large focal leak, 10 patients with punctate focal leak, and 5,846 vessel segments from 10 patients with vessel leakage. The sensitivity in detecting large focal, punctate focal and vessel segment leakage are 95%, 82% and 81%, respectively, when compared to manual annotation by expert human observers. Our framework has the potential to become a powerful new tool for studying malarial retinopathy, and other conditions involving retinal leakage.
BibTeX
@inproceedings{zhao2015automated,
  author = {Yitian Zhao and Ian J. C. MacCormick and David Parry and Sophie Leach and Nicholas A. V. Beare and Simon Harding and Yalin Zheng},
  title = {{Automated Detection of Leakage in Fluorescein Angiography Images with Application to Malarial Retinopathy}},
  booktitle = {Scientific},
  year = {2015},
  doi = {10.1038/srep10425},
}
Figure