Core Concepts
US video-based detection using FocusMAE achieves state-of-the-art accuracy for GBC detection.
Abstract
Recent focus on automated Gallbladder Cancer (GBC) detection from ultrasound videos.
Proposal of video-based GBC detection using FocusMAE to address limitations in image-based methods.
Introduction of a novel design, FocusMAE, for biased masking token selection from high-information regions.
Validation of FocusMAE on US video dataset and public CT-based Covid detection dataset.
Comparison with SOTA methods and human experts in GBC detection.
Ablation study on masking ratio, reconstruction loss, backbone choice, and decoder depth for FocusMAE optimization.
Analysis on candidate region selection using RPN for object localization priors.
Stats
この研究では、GBC検出のための新しいデザインであるFocusMAEを導入しています。
USビデオベースの検出は、高い精度を達成します。
研究では、SOTA方法や人間の専門家と比較しています。