%0 Journal Article
%T 基于12导联心电图的心肌梗死检测与定位研究
Research on Myocardial Infarction Detection and Localization Based on 12-Lead Electrocardiogram
%A 李嘉琦
%A 陈玉恋
%J Modeling and Simulation
%P 1011-1022
%@ 2324-870X
%D 2025
%I Hans Publishing
%R 10.12677/mos.2025.144350
%X 针对心肌梗死智能诊断中的数据利用不足与特征融合低效问题，本文提出多信息融合网络。该网络的局部特征模块采用时序建模技术捕捉心拍级电生理波动，全局特征模块通过空间注意力机制解析12导联间的解剖关联性，充分挖掘多导联心电信号的诊断价值。引入自适应权重机制动态融合局部与全局特征，有效抑制非特异性噪声并降低冗余维度，增强病理特征的表达区分度。基于共享底层特征与独立任务输出的设计，模型在统一框架下同步实现心梗分类与解剖定位，兼顾共性规律与特异性需求。基于真实临床数据的验证表明，MI-Net在分类与定位任务中表现优异，各性能指标均优于传统方法，其决策逻辑通过特征可视化与心电图诊断标准高度一致，可为临床提供高精度、可解释的智能诊断方案。
To address the challenges of insufficient data utilization and inefficient feature fusion in intelligent myocardial infarction diagnosis, this study proposes a multi-information fusion network. The local feature module employs temporal modeling techniques to capture beat-level electrophysiological fluctuations, while the global feature module leverages spatial attention mechanisms to analyze anatomical correlations across 12-lead ECG signals, thereby fully exploiting the diagnostic value of multi-lead information. An adaptive weighting mechanism dynamically fuses local and global features, effectively suppresses non-specific noise and reduces feature redundancy, enhancing the discriminative representation of pathological characteristics. By integrating shared foundational features with task-specific output layers, the model simultaneously achieves myocardial infarction classification and anatomical localization within a unified framework, balancing shared patterns and clinical task-specific requirements. Validation on real-world clinical datasets demonstrates that MI-Net outperforms conventional methods in both classification and localization tasks. Its decision logic, verified through feature visualization techniques, aligns closely with clinical electrocardiogram diagnostic criteria, providing clinicians with an intelligent diagnostic solution characterized by high precision and interpretability.
%K 急性心肌梗死，
%K 类别不均衡样本，
%K 跨导联特征融合，
%K 临床可解释性，
%K 深度神经网络
Acute Myocardial Infarction
%K Category-Imbalanced Samples
%K Cross-Lead Feature Fusion
%K Clinical Interpretability
%K Deep Neural Networks
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=113226