Machine Learning Approach to Enhance the Performance of MNP-Labeled Lateral Flow Immunoassay

Abstract

Abstract The use of magnetic nanoparticle (MNP)-labeled immunochromatography test strips (ICTSs) is very important for point-of-care testing (POCT). However, common diagnostic methods cannot accurately analyze the weak magnetic signal from ICTSs, limiting the applications of POCT. In this study, an ultrasensitive multiplex biosensor was designed to overcome the limitations of capturing and normalization of the weak magnetic signal from MNPs on ICTSs. A machine learning model for sandwich assays was constructed and used to classify weakly positive and negative samples, which significantly enhanced the specificity and sensitivity. The potential clinical application was evaluated by detecting 50 human chorionic gonadotropin (HCG) samples and 59 myocardial infarction serum samples. The quantitative range for HCG was 1–1000 mIU mL−1 and the ideal detection limit was 0.014 mIU mL−1, which was well below the clinical threshold. Quantitative detection results of multiplex cardiac markers showed good linear correlations with standard values. The proposed multiplex assay can be readily adapted for identifying other biomolecules and also be used in other applications such as environmental monitoring, food analysis, and national security.

Keywords

• Point-of-care testing
• Immunochromatography test strips
• Magnetic nanoparticles
• Machine learning
• Support vector machine