[论文解读] Recognising Cardiac Abnormalities in Wearable Device Photoplethysmography (PPG) with Deep Learning
本文提出一种基于LSTM网络的深度学习方法,直接从可穿戴光电容积脉搏波(PPG)信号中检测心脏异常,特别是室性早搏(PVCs),在无需ECG数据的情况下,对ECG确诊的PVCs实现了60%的敏感度和77%的特异度,证明了在真实场景中仅使用PPG进行心脏监测的可行性。
Cardiac abnormalities affecting heart rate and rhythm are commonly observed in both healthy and acutely unwell people. Although many of these are benign, they can sometimes indicate a serious health risk. ECG monitors are typically used to detect these events in electrical heart activity, however they are impractical for continuous long-term use. In contrast, current-generation wearables with optical photoplethysmography (PPG) have gained popularity with their low-cost, lack of wires and tiny size. Many cardiac abnormalities such as ectopic beats and AF can manifest as both obvious and subtle anomalies in a PPG waveform as they disrupt blood flow. We propose an automatic method for recognising these anomalies in PPG signal alone, without the need for ECG. We train an LSTM deep neural network on 400,000 clean PPG samples to learn typical PPG morphology and rhythm, and flag PPG signal diverging from this as cardiac abnormalities. We compare the cardiac abnormalities our approach recognises with the ectopic beats recorded by a bedside ECG monitor for 29 patients over 47.6 hours of gold standard observations. Our proposed cardiac abnormality recognition approach recognises 60%+ of ECG-detected PVCs in PPG signal, with a false positive rate of 23% - demonstrating the compelling power and value of this novel approach. Finally we examine how cardiac abnormalities manifest in PPG signal for in- and out-of-hospital patient populations using a wearable device during standard care.
研究动机与目标
- 开发一种无需依赖ECG的自动化方法,用于从PPG信号中检测心脏异常。
- 评估深度学习模型在对比临床ECG金标准的情况下,从可穿戴PPG数据中识别PVCs的性能。
- 分析心脏异常在不同患者群体的PPG信号中的表现形式,涵盖住院和门诊环境。
- 探索仅使用PPG监测在提升长期心脏风险检测和早期预警系统方面的潜力。
提出的方法
- 使用40万条干净的PPG样本对LSTM深度神经网络进行训练,以学习正常PPG形态和节律。
- 当PPG信号与所学正常模式显著偏离时,模型将其标记为潜在心脏异常。
- 通过29名患者(同时佩戴可穿戴PPG和床边ECG)的对比评估,总监测时长为47.6小时,以PVCs作为金标准。
- 方法包含针对高运动状态PPG信号的滤波处理,以确保信号质量并保持与ECG观察结果的一致性。
- 使用基于集合的混淆矩阵评估性能,将PPG检测到的异常与ECG确诊的PVCs进行对比。
- 额外分析汇总了多个临床人群(HDU、AMU、ED、SURGERY、HOME)中基于PPG的异常频率。
实验结果
研究问题
- RQ1在无ECG参考的情况下,基于正常PPG信号训练的深度学习模型能否检测出如PVCs等心脏异常?
- RQ2在真实世界移动监测中,仅使用PPG的检测性能与ECG金标准相比如何?
- RQ3心脏异常在PPG信号中的频率和模式在不同患者群体中如何变化?
- RQ4信号质量和运动伪影对基于PPG的心脏异常检测准确性有何影响?
主要发现
- 当金标准中每分钟至少存在一个PVC时,所提出的方法在PPG信号中检测到60%的ECG确诊PVCs。
- 假阳性率为23%,表明约23%被PPG标记为异常的信号在ECG中并未包含PVCs。
- 当金标准中每分钟包含两个或以上PVCs时,检测敏感度提升至68%,表明在PVC负荷较高时性能更优。
- 住院患者群体(如HDU和AMU)的PPG异常率较高(分别为5.1%和5.8%),反映出患者病情更重。
- 接受手术治疗的患者(SURGERY)异常率最低(1.54%),可能由于健康状况更好或接受更严密的临床监测。
- 急诊科患者(ED)的异常频率变异较大(均值3.17%,标准差7.9%),表明患者病情严重程度差异显著。
更好的研究,从现在开始
从阅读论文到最终审阅,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。