Structure is Supervision: Multiview Masked Autoencoders for Radiology
MVMAE is a self-supervised framework for radiology that uses multiple X-ray views from a single study to learn view-invariant representations by combining masked image reconstruction with cross-view alignment. MVMAE-V2T extends this by incorporating radiology reports as auxiliary supervision during pretraining, leading to better performance and label efficiency, especially in low-label settings.