MedFG-VQA uses a memory bank of DCT-based low-frequency features and graph-enhanced cross-attention to perform lightweight medical VQA. The approach includes the SynMed-VQA dataset, containing over 2 million synthetic question-answer pairs across nine imaging modalities.
HOW THIS AFFECTS YOU
●
researcherYou can leverage graph-convolutional aggregation to improve visual-textual alignment in low-resource settings.
●
healthThis offers a more computationally efficient path for deploying medical decision support tools.