Abstract
Diabetic retinopathy (DR) is a chronic disease characterized by progressive vision loss, with its progression posing an increasing risk of impairment among individuals with diabetes. Predicting DR progression over extended periods is essential for guiding patient management and therapeutic strategies. We propose a matrix-based multi-task learning (MTL) framework that redefines progression prediction through three contributions. First, we present an MTL framework that simultaneously predicts disease progression across multiple time points. Second, instead of predicting isolated future states, we predict a matrix of state transition probabilities between successive time points, offering a structured and richer representation of disease evolution. Third, we introduce a progression consistency loss to align predicted transitions with empirical progression dynamics observed in training data. We evaluate the proposed method for predicting DR across one to five years on a large-scale longitudinal dataset. Our results highlight the potential of the matrix-based approach, particularly for long-term DR outcome prediction and early intervention planning.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/OMIA_044.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=zPrUDpAqjY
BibTex
@InProceedings{ZegRac_AMultitask_MICCAISAT2026,
author = { Zeghlache, Rachid AND Conze, Pierre-Henri AND Mutsvangwa, Tinashe Ernest AND Massin, Pascale AND Cochener, Béatrice AND Brahim, Ikram AND Quellec, Gwenolé},
title = { { A Multitask Learning Framework for Predicting Diabetic Retinopathy Progression via State Transition Matrices } },
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}
}