Automated Progressive Label Refinement for Nasogastric Tube Segmentation in Chest X-Rays

IEEE Access 2026 Open access

Read full text
Authors: Kanghee Lee, Doohee Lee, Inseo Park, Kyoung Min Moong, Gwiseong Moon, Dae-Hyun Park, Han-Gil Jang, Ji Young Hong, Jeongwon Heo, Hongseok Ko, Woo Jin Kim, Hyun-Soo Choi
+1 more Yoon Kim
DOI: 10.1109/access.2026.3717048
Cited by: 0

Abstract

To address the "annotation bottleneck" from prevalent under-annotation in non-expert labels, this study introduces a fully automated Progressive Label Refinement strategy. Our key innovation, the Prediction-to-Annotation Ratio (PAR), quantifies spatial discrepancies to automatically guide the correction of omitted segments and improve nasogastric (NG) tube segmentation accuracy. On multi-center internal and external MIMIC-CXR datasets, our method substantially improved segmentation accuracy, measured by the Dice Similarity Coefficient (DSC). Intermediate refinement thresholds (

Fields of Study & Subjects

Radiomics and Machine Learning in Medical Imaging Lung Cancer Diagnosis and Treatment COVID-19 diagnosis using AI

Publication details

Journal / source
IEEE Access
Volume
14
Pages
118591–118605
Publication date
1 January 2026
Language
EN