ORIGINAL RESEARCH article
Front. Artif. Intell.
Sec. Machine Learning and Artificial Intelligence
This article is part of the Research TopicAdvanced Machine Learning Techniques for Single or Multi-Modal Information ProcessingView all 4 articles
Enhancing Accessibility: A Multi-Level Platform for Visual Question Answering in Diabetic Retinopathy for Individuals with Disabilities
Provisionally accepted
- 1Department of Computer Science, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia
- 2King Salman Center for Disability Research, Riyadh, Saudi Arabia
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Individuals with visual disabilities possess impairments that affect their ability to perceive visual information, ranging from partial to complete vision loss. Visual disabilities affect about 2.2 billion people globally. In this paper, we introduce a new multi-level Visual Questioning Answering (VQA) framework for visually disabled people that leverages the strengths of various VQA models of the multi-level components to enhance system performance. The model relies on a bi-level architecture that employs two distinct layers. In the first level, the model classifies the question type. This classification guides the visual question to the appropriate component model in the second level. This bi-level architecture incorporates a switch function that enables the system to select the optimal VQA model for each specific question, hence enhancing overall accuracy. The experimental findings indicate that the multi-level VQA technique is significantly effective. The bi-level VQA model enhances the overall accuracy over the state-of-the-art from 87.41% to 88.41%. This finding suggests the use of multiple levels with different models can boost the VQA systems' performance. This research presents a promising direction for developing advanced, multi-level VQA systems. Future work may explore optimizing and experimenting with various model levels to enhance performance further.
Keywords: Disability-Aware VQA, ELECTRA, Med-VQA, Medical Visual Question Answering, Multi-level VQA, questionanswering, Swin, vision-language models
Received: 12 Jun 2025; Accepted: 07 Oct 2025.
Copyright: © 2025 Al-Ahmadi, Alhadhrami and Alotaibi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Saad Al-Ahmadi, salahmadi@ksu.edu.sa
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.