FocusMAE: Gallbladder Cancer Detection from Ultrasound Videos with Focused Masked Autoencoders

In recent years, automated Gallbladder Cancer (GBC) detection has gained theattention of researchers. Current state-of-the-art (SOTA) methodologies relyingon ultrasound sonography (US) images exhibit limited generalization,emphasizing the need for transformative approaches. We observe that individualUS frames may lack sufficient information to capture disease manifestation.This study advocates for a paradigm shift towards video-based GBC detection,leveraging the inherent advantages of spatiotemporal representations. Employingthe Masked Autoencoder (MAE) for representation learning, we addressshortcomings in conventional image-based methods. We propose a novel designcalled FocusMAE to systematically bias the selection of masking tokens fromhigh-information regions, fostering a more refined representation ofmalignancy. Additionally, we contribute the most extensive US video dataset forGBC detection. We also note that, this is the first study on US video-based GBCdetection. We validate the proposed methods on the curated dataset, and reporta new state-of-the-art (SOTA) accuracy of 96.4against an accuracy of 84and 94.7of the proposed FocusMAE on a public CT-based Covid detection dataset,reporting an improvement in accuracy by 3.3code and pretrained models are available at:https://github.com/sbasu276/FocusMAE.