Research On Intelligent Analysis And Abnormal Detection Technology Of Electrophysiological Signals

Human physiological signals are the concentrated expression of human life information and a window to peep into life phenomena.In-depth research on the theoretical methods of biomedical signal detection and processing is of great significance for understanding the laws of life movement and exploring new methods for disease prevention and treatment.The two most commonly used signals in medicine are Electrocardiogram(ECG)and electroencephalogram(EEG),which are used as an important basis for diagnosing heart and brain diseases.This paper mainly focuses on two typical diseases of arrhythmia and epilepsy,and studies and constructs a high-performance automated auxiliary detection method for ECG and EEG.Automatic detection and classification of arrhythmias as an effective early warning tool has been applied in many wearable devices in recent years.However,different from traditional application scenarios,wearable ECG devices still have some shortcomings in the process of detecting arrhythmia.The challenge for accurate detection of PVC and SPB arises from a variety of anomalous disturbances.Aiming at this problem,this paper proposes a new lightweight Transformer model.This paper adopts a more lightweight structure,namely LCA(light-Conv Attention,LCA)to replace Transformer's self-attention.While using fewer parameters(23.70% of self-attention),LCA has achieved comparable or even better performance than self-attention.This paper also designs a stronger input embadding structure to enhance the weight of the internal morphological features of the heartbeat.The method in this paper has been tested on the China Physiological Signal Challenge 2020 dataset and achieved good performance indicators.The positive predictive value for category S is 93.86%,the sensitivity is 83.00%,and the positive predictive value for category V is 91.58%.sensitivity is 94.47%.Absence epilepsy is one of the most common types of epilepsy and one of the challenges faced by clinical neurologists.It lacks symptoms that are readily observed in traditional epilepsy,such as spasticity and convulsions,and is highly dependent on the detection of spikes and slow waves in electroencephalographic(EEG)signals.In recent years,graphical representations known as complex networks have been increasingly used to characterize one-dimensional EEG signals.However,existing methods often cannot represent spikes and slow waves effectively,and it is difficult to capture the differences between spikes and slow waves and non-spikes and slow waves,such as subtle differences and different shapes.To address this problem,this paper proposes an efficient weighted horizontal view(WLHVG)to represent spikes and slow waves.On this basis,this paper studies a 2D convolutional neural network with attention mechanism to further learn latent features from graph representations and complete the detection task.Extensive experiments on real absence epilepsy EEG datasets show that the WLHVG-2D-CNN method proposed in this paper can accurately detect slow waves.Experiments on the Bonn dataset further show that the method proposed in this paper can be generalized to traditional seizure detection tasks with good detection accuracy.