Reproduce “AME-CAM: Attentive Multiple-Exit CAM for Weakly Supervised Segmentation on MRI Brain Tumor” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yu-Jen Chen, Xinrong Hu, Yiyu Shi, Tsung-Yi Ho
AME-CAM proposes an attentive multiple-exit Class Activation Mapping approach for weakly-supervised brain tumor segmentation using MRI, trained with only image-level labels. The method produces more accurate and comprehensive tumor masks, achieving a Dice score of 0.862 on the BraTS 2021 T2-FLAIR modality, outperforming other weakly-supervised methods and even a supervised baseline.
Key insights
- Leveraging activation maps from multiple network depths, combined with an intelligent attentive aggregation mechanism, effectively addresses the low-resolution and inaccuracy challenges of CAM-based WSSS in medical imaging.
- Unsupervised contrastive learning provides a robust method for guiding the attention network to disentangle foreground and background features, demonstrating its utility for refining weakly-supervised segmentation.
- A comprehensive aggregation of features across a full spectrum of resolutions, rather than relying on single-exit or a subset, is crucial for generating accurate and complete tumor localization in complex medical images.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.