Abstract
Estimating the visual field (VF) from a single colour fundus photograph (CFP) is attractive for low-cost glaucoma screening, yet standard pipelines train on one noisy VF per eye and discard longitudinal records. Longitudinal sample expansion (LE) reuses post-baseline CFPs; the harder question is how to supervise them under short, noisy follow-up. We propose anchor-progression denoising (APD): visit 1 remains the raw measured anchor, while follow-up-only progression rates are empirical-Bayes (EB) shrunk to match the GRAPE test protocol. On the published eye-level benchmark with three visits, APD gives a small but seed-consistent gain over LE, supported by no-shrink and full-trajectory EB references. The effect is regime-specific: it appears under short eye-level follow-up, disappears with longer follow-up, and is not retained under subject-level splitting. APD is therefore a supervision-design diagnostic, not a clinically validated performance improvement; inference remains single-image.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/OMIA_015.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=ahjRdy6ivV
BibTex
@InProceedings{GwaSeu_When_MICCAISAT2026,
author = { Gwak, Seunghyun AND Kang, Myungjoo},
title = { { When to Denoise Longitudinal Visual Field Labels: Anchor-Preserving Supervision under Short Follow-up } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17270},
month = {pending},
page = {pending}
}