Bo Zhou , Xiaoying Li , Shitong Cheng , Zhiwei Zhou , Hui Kang
{"title":"Patient-based pre-classified real-time quality control with neural network (PCRTQC-NN)","authors":"Bo Zhou , Xiaoying Li , Shitong Cheng , Zhiwei Zhou , Hui Kang","doi":"10.1016/j.plabm.2025.e00506","DOIUrl":null,"url":null,"abstract":"<div><h3>Objectives</h3><div>Patient-based real-time quality control (PBRTQC) is essential for clinical laboratory management but struggles with detecting small systematic errors. This study presents the patient-based pre-classified real-time quality control with neural network (PCRTQC-NN) model, utilizing neural networks to improve error detection by extracting analytical features from testing instruments.</div></div><div><h3>Methods</h3><div>Using PCRTQC's clustering analysis, we pre-classified and processed Na, CHOL, ALT, and CR data from 611,031 patients. A neural network autoencoder, trained using TensorFlow with mean squared error (MSE) as the loss function, extracted the testing instrument's analytical features under error-free conditions. Systematic errors were identified by comparing reconstruction residuals between test and reconstructed data. The average number of patient samples until error detection (ANPed) evaluated the model performance.</div></div><div><h3>Results</h3><div>The PCRTQC-NN's error detection surpasses traditional algorithms Compared to PCRTQC, it reduced the ANPed for ALT by 37 % (constant error, CE) and 22 % (proportional error, PE) at 1 total error allowable (TEa), with comparable results for other analytes. For 0.5 TEa errors, the ANPed for CHOL decreased by 23 % (CE) and 22 % (PE), for ALT by 14 % (CE) and 6 % (PE), and for CR by 4 % (CE) and 9 % (PE), enhancing error detection capabilities for analytes with high inter-individual variability and sensitivity to smaller errors.</div></div><div><h3>Conclusions</h3><div>PCRTQC-NN significantly enhances systematic error detection compared to PCRTQC, leveraging autoencoders to extract analytical features as discrete signals, thus improving SNR for high-variability analytes. It promises improved laboratory efficiency and inter-laboratory standardization via robust feature models. Future multi-center studies will validate broad applicability across diverse settings.</div></div>","PeriodicalId":20421,"journal":{"name":"Practical Laboratory Medicine","volume":"47 ","pages":"Article e00506"},"PeriodicalIF":1.3000,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Practical Laboratory Medicine","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2352551725000599","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"MEDICAL LABORATORY TECHNOLOGY","Score":null,"Total":0}
引用次数: 0
Abstract
Objectives
Patient-based real-time quality control (PBRTQC) is essential for clinical laboratory management but struggles with detecting small systematic errors. This study presents the patient-based pre-classified real-time quality control with neural network (PCRTQC-NN) model, utilizing neural networks to improve error detection by extracting analytical features from testing instruments.
Methods
Using PCRTQC's clustering analysis, we pre-classified and processed Na, CHOL, ALT, and CR data from 611,031 patients. A neural network autoencoder, trained using TensorFlow with mean squared error (MSE) as the loss function, extracted the testing instrument's analytical features under error-free conditions. Systematic errors were identified by comparing reconstruction residuals between test and reconstructed data. The average number of patient samples until error detection (ANPed) evaluated the model performance.
Results
The PCRTQC-NN's error detection surpasses traditional algorithms Compared to PCRTQC, it reduced the ANPed for ALT by 37 % (constant error, CE) and 22 % (proportional error, PE) at 1 total error allowable (TEa), with comparable results for other analytes. For 0.5 TEa errors, the ANPed for CHOL decreased by 23 % (CE) and 22 % (PE), for ALT by 14 % (CE) and 6 % (PE), and for CR by 4 % (CE) and 9 % (PE), enhancing error detection capabilities for analytes with high inter-individual variability and sensitivity to smaller errors.
Conclusions
PCRTQC-NN significantly enhances systematic error detection compared to PCRTQC, leveraging autoencoders to extract analytical features as discrete signals, thus improving SNR for high-variability analytes. It promises improved laboratory efficiency and inter-laboratory standardization via robust feature models. Future multi-center studies will validate broad applicability across diverse settings.
期刊介绍:
Practical Laboratory Medicine is a high-quality, peer-reviewed, international open-access journal publishing original research, new methods and critical evaluations, case reports and short papers in the fields of clinical chemistry and laboratory medicine. The objective of the journal is to provide practical information of immediate relevance to workers in clinical laboratories. The primary scope of the journal covers clinical chemistry, hematology, molecular biology and genetics relevant to laboratory medicine, microbiology, immunology, therapeutic drug monitoring and toxicology, laboratory management and informatics. We welcome papers which describe critical evaluations of biomarkers and their role in the diagnosis and treatment of clinically significant disease, validation of commercial and in-house IVD methods, method comparisons, interference reports, the development of new reagents and reference materials, reference range studies and regulatory compliance reports. Manuscripts describing the development of new methods applicable to laboratory medicine (including point-of-care testing) are particularly encouraged, even if preliminary or small scale.