Hemodialysis risk prediction system based on combination of dynamic and static data and depth auto-encoder

Abstract

The invention discloses a hemodialysis risk prediction system based on combination of dynamic and static data and a depth auto-encoder. The system is based on a depth auto-encoder.Relevance between high-dimension EHR time sequence data is modeled by utilizing blood pressure, weight and other sign data and hemodialysis treatment related data such as dialysis mode and membrane area collected in the hemodialysis treatment process of a patient, and hemodialysis risk of the patient is predicted; a multi-input model is constructed by combining static and dynamic data, the dynamic data is partially introduced into an LSTM auto-encoder network, a time interval between time sequence data is considered, the static data is integrated as additional input, and a multi-layer neural network is introduced; and the model is trained only by using survival samples, so that the model prediction performance is improved and the influence on model training caused by artificial data amplification is reduced on the premise that the number of dead samples is small or dead samples are not amplified.

Description

technical field [0001] The invention belongs to the technical field of medical treatment and machine learning, and in particular relates to a hemodialysis risk prediction system based on the combination of dynamic and static data and a deep self-encoder. Background technique [0002] End-stage renal disease (End-Stage Renal Disease, ESRD), commonly known as uremia, is the end stage of chronic kidney disease, requiring renal replacement therapy for long-term survival, and most patients choose dialysis for lack of suitable kidney sources. L. The early mortality rate of hemodialysis patients is relatively high, and with periodic hemodialysis treatment, the mortality rate is gradually decreasing, but the three-year mortality rate of hemodialysis patients is still as high as 43%. Although the current research has found some high-risk factors related to death, such as age, serum albumin, blood phosphorus, residual renal function, cardiovascular complications, etc., how to evaluat...

AI Technical Summary

Problems solved by technology

[0009] Aiming at the existing problems, the present invention utilizes the time series data in EHRs to build a hemodialysis risk prediction system to solve the problem of insufficient training of the classification model due to data imbalance caused by lack of follow-up

Images