Dipartimento di Elettronica, Informazione e Bioingegneria, Politecnico di Milano, P.zza L. da Vinci 32; 20133 Milano, Italy.
Dipartimento di Elettronica, Informazione e Bioingegneria, Politecnico di Milano, P.zza L. da Vinci 32; 20133 Milano, Italy.
Comput Methods Programs Biomed. 2022 Apr;217:106670. doi: 10.1016/j.cmpb.2022.106670. Epub 2022 Feb 3.
The ongoing pandemic proved fundamental is to assess a subject's respiratory functionality and breathing pattern measurement during quiet breathing is feasible in almost all patients, even those uncooperative. Breathing pattern consists of tidal volume and respiratory rate in an individual assessed by data tracks of lung or chest wall volume over time. State-of-art analysis of these data requires operator-dependent choices such as individuation of local minima in the track, elimination of anomalous breaths and individuation of breath clusters corresponding to different breathing patterns.
A semi-automatic, robust and reproducible procedure was proposed to pre-process and analyse respiratory tracks, based on Functional Data Analysis (FDA) techniques, to identify representative breath curve and the corresponding breathing patterns. This was achieved through three steps: 1) breath separation through precise localization of the minima of the volume trace; 2) functional outlier breaths detection according to time-duration, magnitude and shape; 3) breath clustering to identify different pattern of interest, through K-medoids with Alignment. The method was firstly validated on simulated tracks and then applied to real data in conditions of clinical interest: operational volume change, exercise, mechanical ventilation, paradoxical breathing and age.
The total error in the accuracy of minima detection and in was less than 5%; with the artificial outliers being almost completely removed with an accuracy of 99%. During incremental exercise and independently on the bike resistance level, five clusters were identified (quiet breathing; recovery phase; onset of exercise; maximal and intermediate levels of exercise). During mechanical ventilation, the procedure was able to separate the non-ventilated from the ventilatory-supported breathing and to identify the worsening of paradoxical breathing due to the disease progression and the breathing pattern changes in healthy subjects due to age.
We proposed a robust validated automatic breathing patterns identification algorithm that extracted representative curves that could be implemented in clinical practice for objective comparison of the breathing patterns within and between subjects. In all case studies the identified patterns proved to be coherent with the clinical conditions and the physiopathology of the subjects, therefore enforcing the potential clinical translational value of the method.
持续的大流行证明了评估受试者呼吸功能的重要性，在安静呼吸期间测量呼吸模式在几乎所有患者中都是可行的，即使是那些不合作的患者也是如此。呼吸模式由个体的潮气量和呼吸频率组成，通过肺或胸壁容积随时间变化的数据轨迹进行评估。这些数据的最先进分析需要依赖操作人员的选择，例如在轨迹中个体化局部最小值、消除异常呼吸和个体化对应不同呼吸模式的呼吸簇。
提出了一种半自动、稳健且可重复的预处理和分析呼吸轨迹的程序，该程序基于功能数据分析（FDA）技术，以识别代表性的呼吸曲线和相应的呼吸模式。这是通过三个步骤实现的：1）通过精确定位体积轨迹的最小值来进行呼吸分离；2）根据时间持续时间、幅度和形状检测功能异常呼吸；3）通过 K-medoids 与 Alignment 进行呼吸聚类，以识别不同感兴趣的模式。该方法首先在模拟轨迹上进行验证，然后应用于临床感兴趣的真实数据：操作容积变化、运动、机械通气、矛盾呼吸和年龄。
在最小检测精度和时间上的总误差小于 5%；通过人工异常值的去除，准确率达到 99%。在递增运动期间，并且与自行车阻力水平无关，共识别出五个簇（安静呼吸；恢复阶段；运动开始；最大和中等运动水平）。在机械通气期间，该程序能够将未通气与通气支持的呼吸分开，并识别出由于疾病进展导致的矛盾呼吸恶化以及由于年龄导致的健康受试者呼吸模式的变化。
我们提出了一种稳健的自动呼吸模式识别算法，该算法提取了代表性曲线，可在临床实践中实施，用于在个体内和个体间客观比较呼吸模式。在所有案例研究中，所识别的模式都与临床情况和受试者的病理生理学一致，因此加强了该方法的潜在临床转化价值。
