B-mode ultrasound-based CAD by learning using privileged information with dual-level missing modality completion

Xiao Wang, Xinping Ren, Ge Jin, Shihui Ying, Jun Wang, Juncheng Li, Jun Shi

Abstract: Highlights•A DLMC-LUPI is proposed to complete image- and feature-level EUS data, enhancing single-modal BUS-based CAD performance.•A VAE-based GAN model uses VAE-generated features as constraints to enhance the synthesis quality of EUS images.•We further propose to apply two PIs based on the dual-level modality completion for transfer.