Self-supervised Polyp Re-identification in Colonoscopy

Intrator, Yotam; Aizenberg, Natalie; Livne, Amir; Rivlin, Ehud; Goldenberg, Roman

doi:10.1007/978-3-031-43904-9_57

Search citation statements

Order By: Relevance

Paper Sections

Select...

Masked Siamese Network (Msn): Compactly Representing Medical...1

Baseline Representations: Simclr Inception Deep Image Prior1

Citation Types

Supporting

Mentioning

Contrasting

Year Published

2023

2024

Publication Types

Select...

Book2

Preprint1

Relationship

Self Cite0

Independent3

Authors

Journals

Cited by 3 publications

(2 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We evaluate MSN embeddings against other types of embeddings such as the SimCLR [32] model described in [33] and pretrained image embeddings described in the following section.…”

Section: Masked Siamese Network (Msn): Compactly Representing Medical...mentioning

confidence: 99%

See 1 more Smart Citation

Predicting Generalization of AI Colonoscopy Models to Unseen Data

Shor,

McNeil,

Intrator

et al. 2024

Preprint

View full text Add to dashboard Cite

Generalizability of AI colonoscopy algorithms is important for wider adoption in clinical practice. However, current techniques for evaluating performance on unseen data require expensive and time-intensive labels. We show that a "Masked Siamese Network" (MSN), trained to predict masked out regions of polyp images without labels, can predict the performance of Computer Aided Detection (CADe) of polyps on colonoscopies, without labels. This holds true on Japanese colonoscopies even when MSN is only trained on Israeli colonoscopies, which differ in scoping hardware, endoscope software, screening guidelines, bowel preparation, patient demographics, and the use of techniques such as narrow-band imaging (NBI) and chromoendoscopy (CE). Since our technique uses neither colonoscopy-specific information nor labels, it has the potential to apply to more medical imaging domains.

show abstract

“…We evaluate MSN embeddings against other types of embeddings such as the SimCLR [32] model described in [33] and pretrained image embeddings described in the following section.…”

Section: Masked Siamese Network (Msn): Compactly Representing Medical...mentioning

confidence: 99%

“…The second trained embedding we investigated is the one described in [33], which uses Contrastive Learning (SimCLR) [32] (Figure 2, right). SimCLR learns representations by mapping images to representations that are "closer" in embedding space if they are "closer" in some meaningful way.…”

Section: Baseline Representations: Simclr Inception Deep Image Priormentioning

confidence: 99%