Medical instrument adverse event small sample migration identification method

Abstract

The invention discloses a medical instrument adverse event small sample migration identification method, and belongs to the field of small sample migration identification. The method comprises the following steps: extracting adverse event modal features; the method comprises the following steps of: completing semantic alignment by adopting a cross-modal Transform; migration joint optimization is carried out between the source domain and the target domain through compound loss, wherein the compound loss is composed of comparative learning loss, cross-domain distribution adaptation loss and causal regularization loss; pseudo-correlations are suppressed by causal regularization terms based on intervention-observation distribution differences and average causal effects are calculated using a causal attention module to generate an interpretable causal link. According to the method, the early detection accuracy can be remarkably improved and the model interpretation and privacy compliance can be enhanced when the sample size is small, and the method is suitable for a medical instrument supervision and intelligent safety monitoring system.

Images