Abstract
Visual impairment categorization is an important functional assessment task for disability certification and social welfare support. Despite of conventional subjective assessments, visual electrophysiology has recently attracted increasing attention for this task as it provides an objective and effective measure of visual function by recording stimulus-evoked neural responses. However, electrophysiological recordings remain susceptible to various artifacts and signal quality issues, which can compromise the reliability of signal-based categorization. To address these challenges, we propose a multi-modal structure-function interaction framework that integrates fundus imaging and visual electrophysiology for visual impairment categorization. The framework introduces an Existence-Constrained Consistency Gate to suppress features derived from unavailable channels and emphasize signal tokens consistently supported by local waveform morphology and global temporal patterns. It further employs Top-K sparse cross-attention to link electrophysiological function with fundus-derived structural evidence, allowing structural tokens to selectively retrieve the most relevant functional responses. We also collect a multi-modal dataset from 425 subjects, including paired bilateral RGB-IR fundus images and 5-channel ERG/VEP recordings, with 143 severe, 112 moderate, and 170 normal cases. Experimental results show that the proposed framework outperforms single- and multi-modal baselines, demonstrating the effectiveness of structure-function interaction for this task.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/OMIA_008.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=zuBOjSryOC
BibTex
@InProceedings{LiuJin_StructureFunction_MICCAISAT2026,
author = { Liu, Jing AND Yao, Chenglin AND Han, Zaidao AND Zhao, Zhongwei AND Higashita, Risa AND Liu, Jiang},
title = { { Structure-Function Linking with Gated Sparse Cross-Attention for Visual Impairment Categorization Using Fundus Imaging and Visual Electrophysiology } },
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}
}