A method for interactive representation of medical time-series data medical features

Abstract

The invention relates to a medical feature interactive representation method of medical time series data, belonging to the technical field of artificial intelligence data processing. It includes the following steps: S1 obtains the original medical feature vector x; S2 pairs the i-th medical feature value x in x i Generate the embedding vector e through the embedding mechanism i ; S3 interacts with embedding vectors of different medical features r i，j ; S4 assigns attention weights α to different interactions with respect to their features i，j ; S5 multiplies and adds the i-th medical feature and all other medical features to obtain its interaction representation c i ;S6 will e i with c i Perform linear compression after splicing to obtain a comprehensive representation of the i-th medical feature f i ; S7 splices the comprehensive representation of all features to obtain the comprehensive representation of x. The present invention can deal with each medical feature separately so as to retain the individuality of the feature, and at the same time allow it to be more closely integrated with the interaction; therefore, the method of the present invention can effectively describe different abnormal patterns, so as to better distinguish patients with different health status.

Description

technical field [0001] The invention relates to a data processing method, in particular to a medical feature interactive representation method of medical time series data, and belongs to the technical field of artificial intelligence data processing. Background technique [0002] Healthcare analytics aims to improve patient management by analyzing various healthcare data through a data-driven approach to aid healthcare decisions and provide personalized treatment recommendations. With the progress of health informatization and the development of big data, international scientific researchers have paid more and more attention to and participated in the preservation and mining of medical big data. Electronic health records (EHR) are used for One of the important data sources for healthcare analytics. Electronic health records contain various forms of information, such as demographic information (such as age, gender, height, admission and discharge time, death, etc.), dynamic ...

AI Technical Summary

Problems solved by technology

This rough division cannot establish a close relationship between the original features and their corresponding feature interactions;

[0006] 2. These models do not consider the time series properties of EHR data, and only using two separate scalars as the final representation of medical data at each moment is not enough to capture the complex relationship between medical features at a specific moment;

But for the medical features of numerical features (such as vital signs, laboratory results, etc.), the linear embedded expression has inevitable limitations: first, the embedding vector has a strong relationship with the proportion of the feature, so it is necessary for all Standardization of medical characteristics

In addition, features with opposite values ​​will be embedded into opposite vectors (i.e., vectors with the same size but opposite directions), which have small differences in expression and are relatively weak in the ability to capture abnormal patterns

Images