Machine Learning-Based Accurate Analysis of cTnI With Alphalisa Immunoassay Enabled Centrifugal Microfluidic System

Cardiac troponin (cTnI) is a biomarker with high sensitivity and specificity for acute myocardial infarction (AMI). Rapid and accurate detection of cTnI can effectively reduce the mortality of AMI. Aiming at the problems of complex operation and low sensitivity of traditional methods used to detect cTnI, an Alphalisa immunoassay enabled centrifugal microfluidic system (AIECMS) is designed to detect cTnI quickly with high sensitivity, and good accuracy is achieved in the linear range of 0.1 ng/mL-50 ng/mL. However, in order to realize the detection of hypersensitive cTnI (the definition standard of weak positive and negative is 0.08 ng/mL), it is necessary to further improve the accuracy of qualitative detection. Since the signal curve of the system for reagents of low concentration range is relatively close, the system can not accurately distinguish weak positive and negative samples, which is easy to cause misjudgment of detection results. In order to solve this problem, this paper proposes to apply machine learning to the signal processing detected by AIECMS for the first time. Firstly, different pre-processing is done according to the characteristics of biological signals; Secondly, different machine learning algorithms are used to train and test the data, and the classification of four clinically significant concentrations (0.02 ng/mL, 0.04 ng/mL, 0.08 ng/mL and 0.1 ng/mL) is realized. Finally, combining the performance of various algorithms, algorithm cost and clinical requirements for the accuracy of low concentration classification, we choose random forest (accuracy 92%) to accurately distinguish the weak positive and negative samples of cTnI.