應用於睡眠階段判讀之低功耗處理器設計

對睡眠障礙患者而言，Polysomnography(PSG)是目前最好的方式來檢查患者的睡眠階段，並藉此了解患者的睡眠品質(Sleep Qality)，診斷各式睡眠疾病，例如阻塞型睡眠呼吸中止症(Obstructive Sleep Apnea)。然而判別睡眠階段需檢測至少八種不同的生理特徵，包含: 腦電訊號、眼動訊號和肌電訊號，而且人工判讀亦相當花費時間以及高成本，因此許多研究嘗試減少測量的參數以及設計適合自動判讀的演算法。雖然目前已經有很多研究提出了不同的自動判讀睡眠階段方法運用少量的參數做預測，但提出之演算法準確率參差不齊，並且無法兼具普遍性，通常只適用於小樣本內的人群特性。最大的缺點是無法兼顧每一種睡眠階層之準確性，導致使用自動判讀演算法無法做出一個具有說服力的判斷。基於以上理由目前尚無一可以在醫學上被各界認定有效且被廣泛使用之標準自動判讀演算法。而未來，基於醫療勞動人口的缺乏，以及現代人飽受睡眠困擾的現象比以往更加地頻繁，開發出一個兼具完備性以及普遍性的自動睡眠判讀演算法已是勢在必行，並且為了因應睡眠為普遍長時間之連續現象，此演算法還必須兼具低能量消耗的特性，藉以補上醫療人力不足之空窗。本論文提出一個可以使用少量生理訊號進行判讀之演算法，不僅可以擺脫傳統PSG 檢測動輒16 道訊號之困擾，更將實現硬體之成本降至最低。同時本演算法結果兼具低功耗以及高準確度之特性，並且接受NSSR[18]公開資料庫之2,000,000個睡眠片段資料測試，相信比以往提出之演算法更加具備普遍性。一個具備低功耗、高準確度以及普遍性之演算法，搭配對於受試者較為友善之檢測環境，相信能使睡眠醫療更普及，同時也可能是睡眠醫療離院監測的第一步。

關鍵字

睡眠判讀；低功耗處理器

並列摘要

For patients with sleep disorder, Polysomnography (PSG) is the best method to analyze their sleep stage and understand sleep quality. By taking a wholenight examination, the sleep laboratory can diagnose several kinds of sleep diseases such as obstructive sleep apnea. However, PSG requires at least eight different physiological signals, including EEG C3, C4, EOG and EMG, to analyze the sleep stage. Besides, manual interpretation is also costly and time-consuming. Therefore, many researches try to reduce the number of channels required to classify sleep stage automatically. Although there are several works announced that are based on different methods, the results do not reach a level that could be recognized as a general method to distinguish the sleep stage. They may be lack of university, and are only valid to a very small group of peaple. The main drawback is that they may not be able to be equally effective to all the sleep stages and therefore less convicing. As a result,there is not a general automatic sleep stage algorithm serving for medical field. In the near furture, developing a convicing and general automatic sleep stage algorithm is imperative since the shortage of medical personel and the phenomenon that people suffering from sleep disorder are more common than before. In order to monitor sleep stages for duration, the algorithm has to consume low power at the same time. This thesis proposes a algorithm based on few bio-signals, it not only gets rid of the conventional 16 channels of bio-signals that PSG needs but also drops the cost of hardware to the lowest point. The algorithm consumes less power while realizing high accuracy at the same time. We verify the algorithm by nearly 2,000,000 data released from open NSRR database and believe the algorithm we developed is much robuster and more general. A low-power, highly accurate algorithm with a friendly interface to the patients will make sleep treatment more widespread and become the very first step of treatment outside the hospital.

並列關鍵字

Sleep Recognition ； Low-Power Processor

參考文獻

[15] Introduction to the empirical mode decomposition method. https://www.mql5.com/en/articles/439.

[13] Discrete wavelet transform. https://en.wikipedia.org/wiki/Discrete_wavelet_transform.

[29] R. B. Berry, R. Brooks, and C. E. Gamaldo. The AASM Manual for the Scoring of Sleep and Associated Events.

[32] J. W. Cooley and J. W. Tukey. An algorithm for the machine calculation of complex Fourier series. Mathematics of Computation, 19:297–301, 1965.

[34] L. Davies and U. Gather. The identification of multiple outliers. Journal of the American Statistical Association, 88(423):782–792, 1993.

國際替代計量

應用於睡眠階段判讀之低功耗處理器設計

全文下載

主題瀏覽