Algorithm Research On Detection And Classification Of AECG Beats Based On Mathematical Morphology And SOM Network

Cardiovascular and cerebrovascular diseases have become the first killer for human health. How to effectively prevent, timely diagnose and treat of cardiovascular diseases has become an urgent need to solve the problem. The ECG changes occurred of 24 hours ware recorded Dynamic electrocardiogram (AECG). Earlier or potential ECG information of the cardiovascular and cerebrovascular diseases was captured and evidences for clinical diagnosis and treatment were presented by AECG. Therefore, it has been the focus of attention for many years at home and abroad that how to automatically, rapidly and accurately analyze of the AECG.The important clinical significance and current situation of technologies of AECG automatically analysis was described in the dissertation. In order to improve the anti-interference ability and the classification rate of AECG beats, algorithm on detection and classification of AECG beats based on mathematical morphology and SOM network was proposed, to achieve to accurately detect beats in the noise conditions and rapidly classify beats.Main research contents include: 1. Deviation extraction of peak algorithm for QRS complex based on mathematical morphology was proposed, to achieve to accurately detect beats in the noise conditions2. The fast heart beat classification algorithm based on SOM network was proposed. The morphological characteristics of QRS waves were as the input vectors. On the based of rapid classification, cohesion clustering method was used to merge output over-refinement of clustering of SOM network to achieve the self-learning classification with unsupervised learning.3. Combined with the research of our team, the algorithm had good results on the detection and classification of AECG and was tested by the clinical data.The algorithms were tested by the MIT-BIH database, and had good results. The sensitivity and specificity of QRS waves by PDE algorithm were 99.85%and 99.84%. The 48 pieces of data was classified by SOM network and cohesion clustering method, and just cost 5032ms to finish the classification. Compared with other algorithms, this algorithm had certain advantages.