GDT-Net:多级特征提取网络,用于精确诊断心房动和心室动
Zhihao Wang1, Mujun Zang2, Tong Liu2
1School of Computer and Artificial Intelligence, Ludong University, Yantai, 264025, Shandong, China.
Computational biology and chemistry
|October 14, 2025
概括
通过对心电图信号进行强有力的分类,GDT-Net增强了用于诊断心房动 (AFIB) 和心室动 (VFIB) 的深度学习,即使是复合噪声. 这种新型的神经网络实现了高精度,改善了心血管疾病的检测.
科学领域:
- 心脏病学 心脏病学
- 生物医学工程 生物医学工程
- 人工智能的人工智能
背景情况:
- 正确诊断心房动 (AFIB) 和心室动 (VFIB) 对心血管健康至关重要.
- 深度学习模型对心电图分析有前途,但与复合噪声作斗争.
- 需要强大的心电图分类来可靠地自动诊断心律失常.
研究的目的:
- 开发一个新的神经网络,GDT-Net,用于在复合噪声下进行可靠的心电图分类.
- 在具有挑战性的条件下提高AFIB和VFIB检测的诊断准确度.
- 提高自动心血管疾病诊断的可靠性.
主要方法:
- 提出GDT-Net,一个神经网络,用于ECG信号分析的G,D和T模块.
- G模块:用于特异性特征提取的分组卷积.
- 模块D:密集连接的架构,用于空间相关性和多诊断.
- T模块:用于时间依赖和远程特征建模的变压器编码器.
主要成果:
- 在MIT-BIH数据集上,GDT-Net获得了高的F1分数:99.46% (内科患者) 和97.20% (内科患者).
- 在复合噪声条件下,大多数子集的F1分数保持在90%以上.
- 在AFIB和VFIB分类表现方面显著改善.
结论:
- GDT-Net为ECG分类提供了强大而有效的解决方案,特别是在复合噪声下.
- 拟议的网络增强了AFIB和VFIB等关键心律失常的自动诊断.
- GDT-Net的表现突显了其在心血管诊断中可靠临床应用的潜力.
相关概念视频
Dysrhythmias III: Characteristics of Dysrhythmias
416
Dysrhythmias, also known as arrhythmias, are irregular heart rhythms that result from abnormal electrical activity in the heart, affecting its ability to circulate blood efficiently. Tachyarrhythmias, a subset of dysrhythmias, are characterized by abnormally fast heart rates exceeding 100 beats per minute. Here are some types of tachyarrhythmias with their distinct ECG features:Sinus Tachycardia:Sinus tachycardia presents a regular heart rhythm with an increased rate of 101-180 beats per...
416
ECG Interpretation of Arrhythmias II: Atrial, Junctional and Ventricular Arrhythmias
465
Arrhythmia is a condition characterized by an irregular heart rhythm, with ECG changes that differ based on its origin and nature. The types of arrhythmias discussed below include atrial, junctional, and ventricular arrhythmias.Atrial ArrhythmiasPremature Atrial Complexes (PACs): PACs are early atrial beats caused by stress, caffeine, alcohol, electrolyte imbalances, hypoxia, hyperthyroidism, or certain medications (e.g., bronchodilators and decongestants). The ECG shows early P waves with an...
465


