Polysomnographic monitoring method and system

Abstract

The application discloses a multi-lead sleep monitoring method and system, which synchronously collects physiological signals, obtains standardized data after adaptive filtering, artifact repair and resampling processing, extracts time-frequency domain features, deep convolution features and nonlinear coupling features respectively, constructs a sleep staging model that can cross crowds by adopting MAML meta-learning combined with an adversarial domain adaptation method, completes probabilistic determination of sleep staging by using Bayesian integration of three heterogeneous models of Transformer, LSTM and graph neural network, corrects key indicators such as AHI based on multi-center data, and generates a grading report, and constructs an individual sleep baseline by a Bayesian dynamic model, and outputs an intervention scheme combined with a knowledge graph. The application can effectively improve the cross-population applicability and the detection accuracy of clinical indicators, and can be used for various clinical sleep diagnosis scenes.