Reliable Mislabel Detection for Video Capsule Endoscopy Data

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Health & Wellbeing — Medical Devices & Health Technology, Health & Medical Research · Depth: Expert, extended

Summary

A new framework has been introduced for reliable mislabel detection in medical imaging datasets, specifically validated for Video Capsule Endoscopy (VCE) data. This framework addresses the challenge of noisy annotations in medical datasets, which often require specialized physician expertise and suffer from ambiguous class boundaries. The pipeline was tested on the Kvasir-Capsule and Galar datasets, the two largest publicly available VCE datasets. On Kvasir-Capsule, it successfully detected 456 out of 471 (1% noise) and 2262 out of 2360 (5% noise) injected mislabels. When applied to the Galar dataset, the framework filtered 167,709 samples (4.8%) and corrected 31,650 samples (0.9%). Training a neural network on this cleaned Galar dataset boosted anomaly detection performance, achieving an accuracy of 93.83% and an F1-score of 71.58%, significantly surpassing previous baselines of 37.01% and 54.38%. Clinical validation by three experienced gastroenterologists confirmed 78% of 100 suspected mislabels were indeed incorrect.

Key takeaway

For Machine Learning Engineers developing anomaly detection models for Video Capsule Endoscopy, you should prioritize implementing a robust mislabel detection and cleaning pipeline. This approach significantly improves model accuracy and F1-scores compared to training on uncleaned data. By identifying and correcting noisy labels, you can achieve more reliable diagnostic support and enhance overall classification performance for critical medical applications.

Key insights

Cleaning medical imaging datasets of mislabels significantly improves deep neural network anomaly detection performance.

Principles

• Label noise severely impacts deep learning model generalization.
• Medical image annotation is prone to errors and subjectivity.
• Dataset cleaning enhances model performance over robust training.

Method

The pipeline trains a CNN, then a Gaussian Mixture Model (GMM) to determine noise probability. It corrects top k^c labels with highest noise reduction and filters top k^f labels with highest noise probability.

In practice

• Use GMMs to identify noisy samples in medical datasets.
• Prioritize cleaning VCE datasets before model training.
• Validate detected mislabels with clinical expert review.

Topics

• Video Capsule Endoscopy
• Mislabel Detection
• Gaussian Mixture Models
• Medical Imaging
• Anomaly Detection
• Dataset Cleaning

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• RARE disease detection from Capsule Endoscopic Videos based on Vision Transformers — Vision Transformers can be fine-tuned for multi-label classification of gastrointestinal diseases from capsule endoscopic videos.
• Computational Imaging Priors for Wireless Capsule Endoscopy: Monte Carlo-Guided Hemoglobin Mapping for Rare-Anomaly Detection — A Monte Carlo-inspired prior improves wireless capsule endoscopy classification by better distinguishing hemoglobin from confounding factors.
• Medical AI gets 66% worse when you use automated labels for training, and the benchmark hides it! [R][P] — A recent study on fairness in medical segmentation for breast cancer tumors revealed that AI models perform significantly worse for younger patients, a bias attributed to the qualitative nature of the…
• Divide-then-Diagnose: Weaving Clinician-Inspired Contexts for Ultra-Long Capsule Endoscopy Videos — Diagnosis-driven video summarization extracts key evidence from ultra-long capsule endoscopy videos for accurate clinical diagnosis.
• Automated Detection of Dosing Errors in Clinical Trial Narratives: A Multi-Modal Feature Engineering Approach with LightGBM — Combining sparse lexical features with dense semantic embeddings improves clinical dosing error detection.
• Benchmarking Machine Learning Architectures for Antimicrobial Stewardship in Pediatric ICUs — Machine learning performance for pediatric antimicrobial stewardship is driven more by target design and data characteristics than model complexity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.