Detection and quantification of fluorescein angiography leakage: From manual grading to advances in machine learning.

Abstract

Fluorescein angiography (FA) has long been a cornerstone for evaluating retinal vascular leakage in diseases like uveitis, diabetic retinopathy, and macular degeneration, but its interpretation relies on subjective grading that can vary between clinicians. With the emergence of artificial intelligence (AI), there is a push to transform this qualitative assessment into objective, quantifiable metrics. We conducted a comprehensive literature search using PubMed, Embase, and Scopus, combining keywords and MeSH terms related to fluorescein angiography leakage, artificial intelligence, and retinal vascular diseases. Studies were included if they assessed FA leakage using manual, semi-automated, or AI-based methods and were peer-reviewed, published in English, and focused on human subjects. Our review charts the evolution from manual grading to modern machine learning techniques that segment and measure leakage using various angiograms. These AI-based approaches enable standardized, reproducible leakage indices that correlate with disease severity, inform treatment decisions, stratify high-risk patients, and facilitate sensitive monitoring of therapeutic response. We also introduce the concept of "minimal residual disease" in this context. By moving from coarse, subjective estimations to precise digital biomarkers, AI-driven FA leakage quantification promises to improve clinical care and research endpoints in retinal disease.