Densely Connected Convolutional Networks and Signal Quality Analysis to Detect Atrial Fibrillation Using Short Single-Lead ECG Recordings

摘要

The development of new technology such as wearables that record high-qualitysingle channel ECG, provides an opportunity for ECG screening in a largerpopulation, especially for atrial fibrillation screening. The main goal of thisstudy is to develop an automatic classification algorithm for normal sinusrhythm (NSR), atrial fibrillation (AF), other rhythms (O), and noise from asingle channel short ECG segment (9-60 seconds). For this purpose, signalquality index (SQI) along with dense convolutional neural networks was used.Two convolutional neural network (CNN) models (main model that accepts 15seconds ECG and secondary model that processes 9 seconds shorter ECG) weretrained using the training data set. If the recording is determined to be oflow quality by SQI, it is immediately classified as noisy. Otherwise, it istransformed to a time-frequency representation and classified with the CNN asNSR, AF, O, or noise. At the final step, a feature-based post-processingalgorithm classifies the rhythm as either NSR or O in case the CNN model'sdiscrimination between the two is indeterminate. The best result achieved atthe official phase of the PhysioNet/CinC challenge on the blind test set was0.80 (F1 for NSR, AF, and O were 0.90, 0.80, and 0.70, respectively).