Construction method of glomerular filtration rate estimation model

Abstract

The invention discloses a construction method of a glomerular filtration rate estimation model, which comprises the following steps: acquiring patient information corresponding to a digital glomerular filtration rate true value, and randomly splitting the patient information into a training set and a test set; performing centralized processing on the patient information of the training set, taking the true value of the glomerular filtration rate as a dependent variable, and taking the patient information subjected to centralized processing as an independent variable; constructing an Ensemble tree model; adding a penalty term to a target function of the Ensemble tree model to control the complexity of the model; performing step-by-step training on the Ensemble tree model to which the penalty term is added; achieving approximation of an original objective function by using second-order expansion; optimizing and removing constant terms; solving a loss function after any bifurcation according to a greedy algorithm; completing construction of the glomerular filtration rate estimation model; and performing summation and averaging on the prediction output of the Ensemble tree model to obtain the glomerular filtration rate.

Description

technical field [0001] The invention relates to the field of intelligent medical technology, in particular to a method for constructing a glomerular filtration rate estimation model. Background technique [0002] Glomerular filtration rate (GFR) refers to the amount of filtrate produced by the two kidneys per unit time (usually 1 minute), and the normal adult is about 80-120ml / min. The ratio of glomerular filtration rate to renal plasma flow is called filtration fraction. The renal plasma flow per minute is about 660ml, so the filtration fraction is 125 / 660×100%≈19%. This result shows that about 1 / 5 of the plasma flowing through the kidney is filtered into the cyst cavity by the glomerulus to form primary urine. Glomerular filtration rate and filtration fraction are measures of kidney function. In this industry, on the one hand, the American Kidney Foundation has always recommended glomerular filtration rate as the most important index for defining, staging and monitoring...

AI Technical Summary

Problems solved by technology

[0005] First, the measurement process of GFR in clinical patients is complicated and expensive, which is not convenient for clinical research and investigation of a large number of people. On the other hand, some existing GFR estimation tools are regression models built on the assumption of linear models. Unable to effectively capture the non-linear relationship between GFR and other clinical indicators

[0006] Second, the existing CDK-EPI equation is developed based on foreign data and is not very suitable for the Chinese population; in addition, various evaluation models often use different model parameters, but clinically it is often necessary to face certain examinations of patients The lack of index items such as , test, etc. makes some models unusable; not only that, even the CKD-EPI equation, which is the most widely accepted clinically, is difficult to take into account the relationship between variables because it uses the basic structure of a linear statistical model. interaction, and the impact of potential nonlinearity on the accuracy of the model; not only that, the traditional GFR evaluation model is cumbersome to use, and it is necessary to manually bring the patient indicators that meet the requirements into the calculation formula and calculate it. Time-consuming and labor-intensive when conducting clinical research

Images