Electrocardiogram diagnosis method and device based on artificial rule enhanced neural network

Abstract

The invention discloses an electrocardiogram diagnosis method based on an artificial rule enhanced neural network. The electrocardiogram diagnosis method comprises the steps that an ECG signal training sample is obtained; constructing a training model comprising a deep learning module and a rule reasoning module, inputting the ECG signal training sample into the deep learning module to obtain a first anomaly prediction probability vector, and inputting the ECG signal training sample into the rule reasoning module to obtain a second anomaly prediction probability vector for performing probability prediction on the electrocardiogram diagnosis tag, fusing to obtain a final abnormal prediction probability vector for carrying out probability prediction on the electrocardio diagnosis tag; an ECG signal training sample is input into the training model, and parameters of the training model are optimized through a total loss function to obtain a multi-lead electrocardiogram signal diagnosis model; during application, the ECG signal is input into the multi-lead electrocardiogram signal diagnosis model to obtain the prediction probability of the electrocardiogram diagnosis tag of the ECG signal. The method can be used for accurately and quickly diagnosing the electrocardiosignals.

Description

technical field [0001] The invention belongs to the field of medical data processing, and in particular relates to an electrocardiogram diagnosis method and device based on an artificial rule-enhanced neural network. Background technique [0002] According to the data provided by the China Tele-ECG Diagnostics Group, there are about 250 million ECG examinations and 35 million dynamic ECG examinations in my country every year, but only about 36,000 people are really proficient and engaged in ECG examinations. The supply and demand are seriously unbalanced. Medical institutions, especially lower-level hospitals and community ECG diagnostic staff cannot meet the needs, and many people have limited ability to read ECG, resulting in non-standard ECG measurement, untimely and inaccurate diagnosis. [0003] At present, there are several prominent problems in ECG labeling, including: 1. Heavy workload: conventional standard ECG examinations involve 12 or 18 leads, and each lead can ...

AI Technical Summary

Problems solved by technology

These methods are often based on traditional signal recognition methods, which cannot effectively identify ECG signals.

The specific recognition difficulties include: 1. Difficult recognition of heart beat segmentation points: Heart beats of patients with serious diseases do not have clear characteristics of P, Q, R, S, and T points, which will make heart beats difficult to be recognized; 2. Waveform superposition is complicated : Patients with heart disease often have more than one pathological manifestation, so there is the possibility of superposition of various variant signals; 3. Difficulty in joint diagnosis of multiple heartbeats: some arrhythmia diseases need to be diagnosed by combining multiple heartbeats, and the model needs to be accurate. While having sufficient recognition ability, it also needs to have a large receptive field to complete the joint diagnosis of multi-cardiac beats;

[0005] Based on the problem of weak positioning and difficult diagnosis of ECG labeling algorithms in the market, it is urgent to design a system that can accurately and quickly diagnose ECG signals, which will help colleges and universities in the field of medical telecommunications to diagnose, reduce the pressure on doctors, and provide great convenience. Meet the needs of patients for ECG diagnosis

Images