Design and analysis of miniaturized microstrip Yagi - Uda Antenna

Abstract

This report proposes a method for automatic heartbeat classification into four classes recommended by AAMI EC57:1998 standard, i.e., normal beat(N), ventricular ectopic beat (VEB), supraventricular ectopic beat (SVEB) or fusion of normal and VEB beat. The classification is divided into two steps, the first one is the determination of normality/abnormality of the heartbeat, then identify the type of abnormality if it exists. Data was obtained from two databases: the 44 non-pacemaker recordings of MIT-BIH arrythmia database, and the St Petersburg INCART database which consists of 75 recordings. Two data distributions were used during the implementation, the first is splitting MIT-BIH records into two sets with 22 records for each (train and test sets) then the training set was merged with INCART records without affecting the test set, this distribution is indicated by DSB1. The second distribution is merging the two databases into one set that was split into training set (70%), validation set (10%) and test set (20%), this distribution is indicated by DSB2. A 1D Convolutional Neural Network (1D CNN) and a Convolutional Auto-encoder Neural Network (CANN) were used to implement the proposed method, since the classification is divided into two sub-classifiers, we need two models to implement this technique, the models are indicated by ModelA for abnormality detection and ModelB for abnormality identification. The auto- encoder showed a robust performance for data reconstruction and features extraction. the resulted accuracies vary from 90.00% up to 98.23% by experimenting each architecture for both models (ModelA & ModelB) on both data distributions (DSB1 & DSB2).