Summary of Comoto: Unpaired Cross-modal Lesion Distillation Improves Breast Lesion Detection in Tomosynthesis, by Muhammad Alberb et al.
CoMoTo: Unpaired Cross-Modal Lesion Distillation Improves Breast Lesion Detection in Tomosynthesis
by Muhammad Alberb, Marawan Elbatel, Aya Elgebaly, Ricardo Montoya-del-Angel, Xiaomeng Li, Robert Martí
First submitted to arxiv on: 24 Jul 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel framework, CoMoTo, to improve lesion detection in Digital Breast Tomosynthesis (DBT) by leveraging mammography data. The authors develop two novel components: Lesion-specific Knowledge Distillation (LsKD) and Intra-modal Point Alignment (ImPA). LsKD selectively distills lesion features from a mammography teacher model to a DBT student model, disregarding background features. ImPA further enriches LsKD by ensuring the alignment of lesion features within the teacher before distilling knowledge to the student. The proposed framework outperforms traditional pretraining and image-level KD methods, achieving a 7% improvement in Mean Sensitivity under low-data settings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us get better at finding breast tumors using special X-ray pictures called Digital Breast Tomosynthesis (DBT). Right now, DBT is very good at finding tumors but takes longer to look at the pictures. The problem is that we don’t have enough information about how to make DBT better because it’s hard and expensive to get more data. One way to solve this issue could be to use information from a different type of picture called mammography, which is easier to take. This paper shows us how to do just that by creating a new system called CoMoTo. It uses two special techniques: one helps the DBT pictures learn from the mammography pictures and the other makes sure those pictures are looking at the same things. The result is that we can find tumors better using DBT, which could help doctors treat breast cancer more effectively. |
Keywords
» Artificial intelligence » Alignment » Knowledge distillation » Pretraining » Student model » Teacher model
