Analysis and Recognizing of Inspiratory Flow Limitation Signals During Sleep

• Main Authors: Po-Chung Shih, 施博鐘
• Other Authors: Kang-Ping Lin
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/57644286748811021385

碩士 === 中原大學 === 通訊工程碩士學位學程 === 101 === Abstract Obstructive Sleep Apnea (OSA) is an intermittent respiratory arrest disease due to airway obstructed during sleep. Flow limitation is an important symptom of OSA, but the method of detecting flow limitation was not clear now. The signal sources were Polysomnography (PSG) records of clinical patients. 47 different categories of respiratory flow signals identified as RERA respiratory flow were captured. Among those, 8 cases were marked normal, and 39 were marked as flow limitation which can be further categorized into mild and severe obstruction. 3 kinds of methods were used as detection standards: (1) Flattening index (F.I) feature: an accepted standard for the majority of studies. Sample the middle portion of the inspiratory wave to calculate mean deviations, which were divided by its mean to acquire sum of errors. Weighted F.I was F.I after weighted improvements. (2) Value Weight: Weighted data above average of inspiratory waveform. (3)Time Weight: Weighted data from the later part of inspiratory waveform. Four alternative detection methods were proposed in this study: (1) First Order Detection, (2) Second Order Detection, and (3) Third Order Detection. The three methods were calculating the fitting curve of data and obtained the mean absolute error. The forth method was Weighted Curve Detection. It was similar to the Third Order Curve but the curve was weighted at the beginning, the middle and the end to achieve optimal fitting curves. Various levels of noises were added in the amplitude and time of 47 cases of respiratory flow. More than 1400 acquired inspiratory waveforms were analyzed using the proposed 7 detection methods and the performances were compared. 7 methods obtained consistent results in normal cases. However, in the parts that were difficult to recognize, the results from Time Weight, First Order Detection and Second Order Detection were insignificant due to the larger differences with the clinical diagnosis. F.I, Value Weight and Third Order Detection were more consistent with the actual results. Third Order Detection could achieve 99% of accuracy.