Machine Learning-Aided Multifunctional Microfluidic Platform with SERF Atomic Magnetometers for Accurate and Sensitive Cardiovascular Biomarker Diagnosis

Early diagnosis of diseases, which usually relies on biomarker detection, can prevent the mass spread of infectious diseases and dramatically reduce mortality rates in patients with different diseases. However, traditional detection methods are limited by their complexity and long time spent associated with multiple handling and transfer steps. The development of simple and automated diagnostic devices for rapid and sensitive biomarker detection is significant. In this work, a machine learning (ML)-aided multifunctional microfluidic platform is developed to realize the sensitive and accurate detection of four cardiovascular disease (CVD) biomarkers. The mixing, incubation, and enrichment of biomarkers are all realized on the single chip. A miniaturized spin exchange relaxation-free (SERF) atomic magnetometer is used to record two-dimensional (2D) magnetic field maps. Then, the magnetic fields in the detection areas are decoupled through peak location and fitting in two directions. Finally, ML algorithms are used to classify and predict the concentrations of CVD biomarkers. Taking cTnI as an example, the results show that the platform enables an efficient classification accuracy of 98.09% and prediction accuracy of 99.99%.