Preprocessing and Pattern Recognition for Single-Lead Cardiac Dynamic Model

Objective: Since time-domain/frequency-domain domain data can only provide limited cardiac features, this reduces the recognition accuracy of pathological variation signals. Therefore, this study designed a new feature recognition system to identify patients. Methods: This research first designed a new single-lead cardiac dynamic model, which can be used to distinguish different types of diseases, and then used the new dynamic model to assist wavelet analysis. Subsequently, this research proposed a new discriminant for baseline drift noise, and employed a new dynamic model to assist in eliminating sampling noise and motion artifact noise. This research also designed a new dual-threshold algo-rithm. Afterwards, this research designed a new kurtosis-skewness clipper capable of clipping disease-damaged signals. Since the neural network needs to use samples of equal length, this research designed a new Signal stretch-feature Integrator, which can automatically select the best interpolation method. Finally, this research designed a new automatic machine learning model that can automatically build a neural network and complete the signal identification using only P wave/QRS wave/T wave.Results: Dynamic modeling enables intuitive comparison of signal processing effects. After preprocessing, the baseline drift of the signal was effectively removed and the noise was reduced. After clipping the signal damaged by the disease, the highest accuracy of individual discrimination reached 99.8951 % by using the automatic machine learning model.Conclusion: After being analyzed and processed by the new dynamic model, the recognition accuracy can reach 99.4895% by using P wave / QRS wave / T wave alone.