Predicting laboratory solution kit accuracy using artificial intelligence: a data-driven approach

Abstract This study developed and validated an artificial intelligence (AI)-based computational tool for standardizing glucose and urea measurements across different clinical laboratory systems (Biolabo and BioScien) using Biomagreb as the reference standard. Through empirical analysis of parallel-t...

Full description

Saved in:
Bibliographic Details
Main Authors: Abdulwahab AL-Deib, Eshraq Alsherif, Husameddin Abuzgaia, Amel Rabti, Ruwaida Rtemi, Salem Elfard, Sheren Njaim
Format: Article
Language:English
Published: SpringerOpen 2025-08-01
Series:Bulletin of the National Research Centre
Subjects:
Online Access:https://doi.org/10.1186/s42269-025-01351-1
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract This study developed and validated an artificial intelligence (AI)-based computational tool for standardizing glucose and urea measurements across different clinical laboratory systems (Biolabo and BioScien) using Biomagreb as the reference standard. Through empirical analysis of parallel-tested blood samples, four linear regression models were established, demonstrating excellent predictive accuracy (R2 > 0.99, mean absolute error < 1.2%). The glucose conversion models yielded precise transformations (Biolabo: Y = 1.01X−0.03357; BioScien: Y = 1.141X−13.71), while urea models showed robust consistency (Biolabo: Y = 0.9312X + 1.549; BioScien: Y = 0.7073X + 5.252). Validation confirmed near-perfect agreement with manual calculations, with minor discrepancies (≤ 0.5%) attributable only to rounding effects. The AI-driven software implementation enhanced clinical utility by automating conversions, displaying normal ranges, and eliminating human calculation errors. Despite high accuracy, limitations include unit restriction (mg/dL only), single-value processing, and lack of batch export functionality. The study highlights AI’s transformative role in laboratory standardization, particularly through machine learning (ML) techniques like linear regression and random forests. Challenges like data quality, understanding the model, and following regulations were tackled using explainable AI (XAI) and strict validation methods. Future improvements should expand analyte coverage, incorporate mmol/L conversion, and enable batch processing. These findings support AI’s potential to enhance diagnostic reliability and inter-laboratory comparability in clinical settings.
ISSN:2522-8307