Multi-center clinical data set adaptation equipment based on knowledge awareness

Abstract

The invention discloses a multi-center clinical data set adaptation device based on knowledge perception. The multi-center clinical data set adaptation device comprises a data input unit, a knowledgegraph embedding unit, an antagonistic learning unit based on knowledge perception, and a clinical result prediction unit. The knowledge graph embedding unit comprises a knowledge graph module and a graph convolutional neural network module; the adversarial learning unit based on knowledge perception comprises a patient characterization module, a knowledge characterization module, a patient characterization module based on knowledge perception and a multi-center discriminator module. The clinical result prediction unit is used for fitting the patient feature representation pi output by the patient representation module based on knowledge perception to obtain a prediction result of a clinical target. According to the multi-center clinical data set adaptation device based on knowledge perception, external clinical expert knowledge is introduced, sharing features and center related features of patients in the multi-center clinical data set can be captured at the same time, and therefore the application performance in different clinical environments is improved.

Description

Technical field[0001]The invention relates to a multi-center clinical data set adaptation device based on knowledge perception.Background technique[0002]Many large clinical data sets, especially data sets collected from different clinical research centers, contain a large number of participants with different geographic locations and center-specific characteristics. Using clinical data collected from multiple centers to prove or disprove a hypothesis in different clinical settings is essential for improving patient treatment and care, improving the quality of healthcare management, and conducting effective clinical research. However, in many cases, the clinical data sets collected by multiple centers differ significantly in different centers because of the differences in the genetics, environment, and ethnic distribution of the patient samples.[0003]Several large-scale studies have shown that machine learning models trained on data collected from one clinical center cannot be reliab...

AI Technical Summary

Problems solved by technology

[0006] Although both instance matching and representation learning have their own unique advantages, they both suffer from the same severe limitations that clinical expert knowledge is not utilized in the learning process and application performance in specific clinical settings is limited.

Images