Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70028
Title: | 利用1維卷積神經網及切割法可即時適應病患的心電圖分類 Real-Time Patient-Specific ECG classification by 1-D Convolutional Neural Networks and Delineation |
Authors: | Che-Wei Lee 李哲瑋 |
Advisor: | 管傑雄(Chieh-Hsiung Kuan) |
Keyword: | 1-D卷積神經網路,適應病患心電圖分類,ECG切割技術,即時心電圖運算, Convolutional neural networks (CNNs),Patient specific ECG classification,ECG delineation,Real-time heart monitoring, |
Publication Year : | 2018 |
Degree: | 碩士 |
Abstract: | 目標:此篇碩士論文提出一個快速、準確,並可以利用很小的訓練資料量,針對不同病患做參數調整的心電圖分類系統。
方法:本研究架出一種可以適應病患的1-D卷積神經網路,只要神經網路訓練完成,輸入的信號不需做任何前處理,便可單獨使用此模型準確達到心電圖分類最大的兩個流程:特徵提取和分類原則。1-D卷積神經網路能用一個共同資料庫及病患本身,針對每一個病患的心電圖進行訓練,經過訓練的神經網路不僅能大幅提高分類準確率,還能省去手刻特徵的過程。此模型不需大量資料即可訓練完成,藉此設計出架構很小的模型,便可應用於心電圖即時診斷的穿戴式裝置。 結果:此模型的結果利用MIT-BIH心律不整資料庫進行驗證分析。結果比絕大部分最先進的方式好或是比其他利用神經網路的方式計算量少很多。 結論:當一個神經網路訓練完成,使用者即可以運用其處理很長的心電圖訊號,不需要再重新訓練;此方式不僅運算量低,且準確率高,此外,更能針對不同病患之心電圖差異進行調節,應用在不同的資料庫上。 Goal: This thesis presents a fast and accurate patient specific electrocardiogram (ECG) classification and monitoring system with relatively small training data size. Methods: An adaptive implementation of 1-D convolutional neural networks (CNNs) is inherently used to fuse the two major blocks of ECG classification without signal de-noising: feature extraction and classification. Therefore, for each patient, and individual and simple CNN will be trained by using a small common set and a patient specific data, this patient specific feature extraction method can further improve the classification performance. Since this also negates the necessity to extract any hand-crafted manual features, once a dedicated CNN is trained for a specific patient, it can solely be used to classify long ECG data streams in a fast and accurate manner. Using this method the structure of the CNN can be small, allowing this solution to be implemented for real-time ECG monitoring and early alert systems on wearable devices. Results: The results are evaluated over the MIT-BIH arrhythmia data benchmark database, the proposed solution achieves a superior classification performance than most of the state-of-the-art methods, or have much lower computation complexity than other NN solutions for the detection of ventricular ectopic beats and supraventricular ectopic beats. Conclusion: Once a dedicated CNN is trained, it can be used to classify long data streams for an individual patient with extremely low computation complexity with high accuracy. In addition, due to its simple and parameter invariant nature, the proposed system is extremely generic, thus applicable to any ECG dataset. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70028 |
DOI: | 10.6342/NTU201800357 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 電子工程學研究所 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-107-1.pdf Restricted Access | 1.78 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.