Ballistocardiogram signal atrial fibrillation computer-aided diagnosis method based on transfer learning

Abstract

The invention provides a ballistocardiogram signal atrial fibrillation computer-aided diagnosis method based on transfer learning, wherein the method relates to the field of computer-aided diagnosis technology. The method comprises the steps of preprocessing an ECG signal and a BCG signal which are synchronously measured and the ECG signal in a current atrial fibrillation disease database, and setting the subframe length of all signal data to 24 seconds; then constructing a convolutional neural network which performs atrial fibrillation aided detection; training the constructed convolutional neural network and performing parameter updating; extracting a transfer learning characteristic vector of the actual measured BCG signal and constructing a random forest classifier; using the BCG signal in a testing set as the input of the convolutional neural network, acquiring a transfer learning model characteristic parameter and inputting into the random forest classifier, and determining whether the BCG signal is an atrial fibrillation disease type through the output result of the random classifier. The method of the invention facilitates extension of the BCG signal to a daily applicationfield and supplies a feasible plan for performing heart-related disease aided diagnosis and prediction through the BCG.

Description

technical field [0001] The invention relates to the technical field of computer-aided diagnosis, in particular to a method for computer-aided diagnosis of atrial fibrillation based on transfer learning. Background technique [0002] With the increasing popularity of wearable devices, the field of non-invasive cardiac function assessment has become a research hotspot today. Conventional cardiac function detection methods, including electrocardiogram (ECG), magnetocardiogram, heart sound, and impedance cardiogram, etc., all need to attach electrodes and other detection equipment to the human body surface, which have certain requirements for the monitoring environment, conditions and operators. requirements, and caused great inconvenience to the daily life of the subjects. In particular, paroxysmal atrial fibrillation has the characteristics of uncertain onset time, sudden onset and unobvious clinical manifestations, so a non-contact real-time monitoring method for cardiac fun...

AI Technical Summary

Problems solved by technology

However, the BCG signal has the characteristics of weak amplitude, susceptibility to interference, and the lack of a relatively complete database. Common deep learning models such as convolutional neural networks (Convolutional Neural Networks, CNN) have certain requirements for the amount of input data, and usually require large batches. Therefore, it is not suitable for the analysis of BCG signals; although traditional machine learning methods are suitable for training and debugging of small batches of data, they usually need to extract the waveform characteristics of the signal, while the BCG signal Waveforms vary greatly with different detection devices and have high time complexity, so machine learning methods are not suitable for daily BCG signal processing

Images