Related Experiment Video
Updated: May 5, 2026
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
13.4K
A Reproducible COCO-Polygon Quality-Control Pipeline Improves Segmentation Stability in Endoscopic Airway Imaging
Medine Atmaca1, Ilkay Sibel Kervancı2, Necati Olgun1
1Department of Mathematics, Graduate School of Natural and Applied Sciences, Gaziantep University, 27410 Gaziantep, Türkiye.
Diagnostics (Basel, Switzerland)
|May 4, 2026
Summary
A new quality control (QC) workflow for endoscopic airway imaging datasets significantly improves deep learning segmentation accuracy. This systematic annotation refinement enhances reproducibility and spatial stability in airway analysis.
Area of Science:
- Medical Imaging
- Deep Learning
- Computational Anatomy
Background:
- Endoscopic airway imaging is crucial for geometry-based airway assessment in airway management.
- Current COCO-format datasets suffer from fragmented regions and inconsistent polygon annotations, hindering deep learning model reproducibility and spatial stability.
- A systematic annotation quality-control (QC) workflow is proposed to enhance dataset integrity prior to model training.
Purpose of the Study:
- To develop and evaluate a systematic annotation quality-control (QC) workflow for endoscopic airway imaging datasets.
- To improve the integrity, reproducibility, and spatial stability of airway segmentation using deep learning.
- To assess the impact of QC-refined annotations on the performance of a YOLOv8-seg model.
Main Methods:
- Analysis of the phantom subset of the Upper Airway Anatomical Landmark (UAAL) dataset (4526 polygon instances).
- Implementation of a QC pipeline involving polygon structure validation, mask generation at native resolution, and removal of small instances.
- Training a YOLOv8-seg model with both original and QC-refined annotations, followed by performance evaluation using mAP and Dice Similarity Coefficient (DSC).
Main Results:
- Annotation QC improved boundary consistency and reduced mask fragmentation.
- Training with QC-refined annotations led to increased mAP@0.5:0.95 (0.602 to 0.628) and mean DSC (0.823 to 0.830, p < 0.05).
- Pilot evaluation on the UAAL clinical subset showed significant performance improvements in Box mAP@0.5 (0.635 to 0.706) and Mask mAP@0.5 (0.631 to 0.704).
Conclusions:
- Annotation-level QC enhances deep learning segmentation robustness without altering network architecture.
- The proposed QC workflow improves result reproducibility in endoscopic image segmentation.
- This method supports more stable geometry-based airway analysis in deep learning applications.
