Abstract
Data collection and annotation have been a longstanding bottleneck for computer vision applications, particularly in the medical domain, where data is limited by high annotation costs and low disease prevalence. Computer-aided detection and diagnosis (CADe/x) require substantial data for clinical relevance, yet current approaches rely on manual collection of high-quality images or post hoc keyframe selection from video, both time-consuming and tied to subjective quality assessment. To accelerate CADe development, we propose a quality-controlled keyframe selection algorithm that samples clinically relevant and diverse frames from video, leveraging endoscopic foundation model features and a computer-aided quality system to reject low-quality frames. We evaluate our approach by training downstream CADe models exclusively on the extracted frames for Barrett’s neoplasia detection. Results show an improvement in Area Under the Receiver Operating Characteristic curve of up to 4.7% over uniform sampling, alongside an 80% reduction in data requirements at similar performance. The code is available at github.com/HaikoMid/GastroKey.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DEMI_014.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=4E9iNDF4aq
BibTex
@InProceedings{MidHai_QualityControlled_MICCAISAT2026,
author = { Middeljans, Haiko AND Jaspers, Tim J. M. AND Jong, Martijn R. AND van Eijck van Heslinga, Rixta A. H. AND Slooter, Florance C. AND de Groof, Albert J. AND Bergman, Jacques J. AND de With, Peter H. N. AND van der Sommen, Fons},
title = { { Quality-Controlled Keyframe Sampling using Endoscopic Foundation Models } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17276},
month = {pending},
page = {pending}
}