Stacking Ensemble Learning for Diagnosing Acute Myocardial Infarction in Mice Using Plasma Raman Spectroscopy and Mass Spectrometry.

Abstract

Cardiac troponin is the reference biomarker for acute myocardial infarction (AMI) but has limited utility in the hyperacute phase. Raman spectroscopy enables rapid, label-free molecular analysis, whereas mass spectrometry offers greater molecular specificity at higher cost. Plasma from 57 coronary-ligated or Sham-operated mice underwent Raman spectroscopy and mass spectrometry. A stacking ensemble model was developed to identify AMI using Raman spectroscopy, mass spectrometry, or fused data. Under nested cross-validation, the Raman spectroscopy model yielded an area under the receiver operating characteristic curve (AUC) of 0.807 (95% CI [0.679, 0.910]). Mass spectrometry and fused models achieved 1.000 (95% CI [1.000, 1.000]), and fusion showed no detectable improvement over mass spectrometry. Cardiac troponin I demonstrated 0.714 (95% CI [0.567, 0.842]), significantly lower than mass spectrometry and fused models (Holm-adjusted p < 0.001) and numerically lower than the Raman spectroscopy model (Holm-adjusted p = 0.357). These findings require confirmation in larger human cohorts.