Identifying Air Bubble Errors in Arterial Blood Gas Tests Using Machine Learning Models

DOI

10.2991/978-94-6239-789-7_39How to use a DOI?

Keywords

arterial blood gas; air bubble contamination; pre-analytical error; clinical decision support; machine learning; error-based modeling

Abstract

Among the pre-analytical errors that compromise arterial blood gas (ABG) testing, air bubble contamination stands out for its frequency and its ability to produce systematic, clinically dangerous distortions. Atmospheric air entering a collection syringe raises PO₂, suppresses PCO₂, and shifts pH toward alkalosis — a pattern that can be mistaken for genuine respiratory alkalosis. Addressing this problem, we built and tested a detection framework rooted in gas-diffusion physiology, working with 800 expert-annotated ABG samples. Deviations in pH, PO₂, and PCO₂ from their respective reference values were computed, and a composite rule requiring any two of three abnormal thresholds to be met simultaneously was applied. The rule yielded sensitivity of 87.6% and specificity of 91.2%, separating true contamination from real respiratory disorders with clinically acceptable accuracy. Feeding these error-derived features into SVM, Random Forest, and Logistic Regression classifiers raised accuracy by 5.2–5.6%; alone, they accounted for 34.2% of feature importance in the Random Forest model. Electrolyte and metabolic parameters, as expected from diffusion physics, did not differ between contaminated and clean samples, lending additional confidence to the specificity of the detected signal. Fast enough for point-of-care use and transparent enough for bedside explanation, the method illustrates why encoding known data-quality problems directly into model pipelines pays dividends in clinical AI.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - S. Ramadoss
AU  - A. Kumaravel
PY  - 2026
DA  - 2026/10/08
TI  - Identifying Air Bubble Errors in Arterial Blood Gas Tests Using Machine Learning Models
BT  - Proceedings of theInternational Conference on Innovation and Practices in Artificial Intelligence (ICIPAI 2026)
PB  - Atlantis Press
SP  - 597
EP  - 613
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-789-7_39
DO  - 10.2991/978-94-6239-789-7_39
ID  - Ramadoss2026
ER  -