Automatic Transformer-Based Grading of Multiple Retinal Inflammatory Signs on Fluorescein Angiography

Abstract

Background: Grading fluorescein angiography (FA) in the context of uveitis is complex, often leading to the oversight of retinal inflammation in clinical studies. This study aims to develop an automated method for grading retinal inflammation.

Methods: Patients from Jules-Gonin Eye Hospital with active or resolved uveitis who underwent FA between 2018 and 2021 were included. FAs were acquired using a standardized protocol, anonymized, and annotated following the Angiography Scoring for Uveitis Working Group criteria, focusing on four inflammatory signs of the posterior pole. Intergrader agreement was assessed by four independent graders. Four deep learning transformer models were developed, and their performance was evaluated using the Ordinal Classification Index (1-OCI), accuracy, F1 scores, and Kappa scores. Grad-CAM analysis was employed to visualize model predictions.

Findings: A total of 543 patients (1042 eyes, 40987 images) were included in the study. The models closely matched expert graders in detecting vascular leakage (F1-score=0.87, 1-OCI=0.89), capillary leakage (F1-score=0.86, 1-OCI=0.89), macular edema (F1-score=0.82, 1-OCI=0.86), and optic disc hyperfluorescence (F1-score=0.72, 1-OCI=0.85). Saliency analysis confirmed that the models focused on relevant retinal structures. The mean intergrader agreement across all inflammatory signs was F1-score=0.78 and 1-OCI=0.83.

Interpretation: We developed a vision transformer-based model for the automatic grading of retinal inflammation in uveitis, utilizing the largest dataset of FAs in uveitis to date. This approach provides significant clinical benefits for the objective evaluation of uveitis and paves the way for future advancements, including the identification of novel biomarkers through the integration of clinical data and other imaging modalities.

Funding: This work was supported by the Swiss National Science Foundation (2018DRI13 to Thomas J. Wolfensberger), which provided access to the SOIN infrastructure. The Claire and Selma Kattenburg Foundation (Kattenburg Prize 2021 to Florence Hoogewoud and Mattia Tomasoni) and the Schmieder-Bohrisch Foundation (Schmieder-Bohrisch Prize 2023 to Mattia Tomasoni) who financially supported the development and analysis of the AI model. Additionally, the AURIS Foundation, the W. & E. Grand dʼHauteville Foundation, the Ingvar Kamprad Fund, the Fleurette Wagemakers Foundation, the Kononchuk Family grant, the Blatter Family grant, and the Rhumatismes-Enfants-Suisse Foundation supported the acquisition of clinical data through the JIR-cohort.

Declaration of Interest: I declare, on behalf of all co-authors, that we have no competing interests.Funding agencies did not influence any phase of this study, including conceptualisation, data curation, formal analysis, investigation, methodology, supervision, validation, or visualizations documented.

Ethical Approval: The study was conducted in accordance with the principles of the Declaration of Helsinki and received approval from the Institutional Review Board (CER-VD n° 2018–02161) for retrospective analysis up to 2018. For patients included after this period, informed consent was obtained, allowing the use of their data for research purposes (CER-VD PB 2016−00868).

Keywords: fluorescein angiography, retinal inflammation, vasculitis, macular edema, papillitis, vascular leakage, Deep Learning, optic disc hyperfluorescence, capillaropathy, disease grading, Uveitis, transformers, ordinal classification index, inter-grader agreement