Li Jiajun, Alavi Rashid, Dai Wangde, Matthews Ray V, Kloner Robert A, Pahlevan Niema M
Department of Aerospace and Mechanical Engineering, University of Southern California, Los Angeles, California, USA.
Department of Medical Engineering, California Institute of Technology, Pasadena, California, USA.
FASEB J. 2025 Sep 15;39(17):e71029. doi: 10.1096/fj.202502111R.
Myocardial infarction (MI) is a leading cause of death worldwide and the most common precursor to heart failure, even after initial treatment. Precise evaluation of myocardial injury is crucial for assessing interventions and improving outcomes. Extensive evidence from both preclinical models and clinical studies demonstrates that the extent and severity of myocardial injury (i.e., myocardial infarct size, ischemic risk zone, and no-reflow area) are critical determinants of long-term outcomes post-MI. This study aims to assess whether carotid pressure waveforms, analyzed using an intrinsic frequency (IF)-machine learning (ML) approach, can accurately quantify myocardial injury sizes: myocardial infarct size, ischemic risk zone, and no-reflow area. Acute MI was induced in N = 88 Sprague-Dawley rats using a standard coronary occlusion/reperfusion model. MI-injury sizes were obtained via histopathology. IF metrics were extracted from carotid pressure waveforms post-MI. ML classifiers were developed using 66 rats and externally tested on 22 additional rats. Our best developed model for infarct size achieved an accuracy of 0.95 (specificity = 0.95, sensitivity = 0.96). For the ischemic risk zone, the best model showed an accuracy of 0.85 (specificity = 0.90, sensitivity = 0.80), and for the no-reflow area, we reached an accuracy of 0.88 (specificity = 0.89, sensitivity = 0.86). To conclude, a hybrid physics-based ML approach applied to carotid pressure waveforms successfully classified MI-injury severity. As carotid pressure waveforms can be measured non-invasively and remotely (e.g., via smartphones), this proof-of-concept preclinical study suggests a translational potential for post-MI management, enabling timely interventions, improved patient monitoring, and mitigating adverse outcomes.
心肌梗死（MI）是全球主要的死亡原因，也是心力衰竭最常见的先兆，即使在初始治疗后也是如此。精确评估心肌损伤对于评估干预措施和改善预后至关重要。来自临床前模型和临床研究的大量证据表明，心肌损伤的程度和严重性（即心肌梗死面积、缺血风险区和无复流区）是心肌梗死后长期预后的关键决定因素。本研究旨在评估使用固有频率（IF）-机器学习（ML）方法分析的颈动脉压力波形是否能够准确量化心肌损伤大小，即心肌梗死面积、缺血风险区和无复流区。使用标准冠状动脉闭塞/再灌注模型在N = 88只Sprague-Dawley大鼠中诱导急性心肌梗死。通过组织病理学获得心肌梗死损伤大小。从心肌梗死后的颈动脉压力波形中提取IF指标。使用66只大鼠开发ML分类器，并在另外22只大鼠上进行外部测试。我们针对梗死面积开发的最佳模型准确率达到0.95（特异性 = 0.95，敏感性 = 0.96）。对于缺血风险区，最佳模型的准确率为0.85（特异性 = 0.90，敏感性 = 0.80），对于无复流区，我们达到了0.88的准确率（特异性 = 0.89，敏感性 = 0.86）。总之，应用于颈动脉压力波形的基于物理的混合ML方法成功地对心肌梗死损伤严重程度进行了分类。由于颈动脉压力波形可以通过非侵入性和远程方式（例如通过智能手机）进行测量，这项概念验证临床前研究表明了其在心肌梗死后管理中的转化潜力，能够实现及时干预、改善患者监测并减轻不良后果。
