CN-122000036-A - Nonlinear algorithm system for accurate chronic disease management of chronic kidney disease

CN122000036ACN 122000036 ACN122000036 ACN 122000036ACN-122000036-A

Abstract

The invention relates to the technical field of medical artificial intelligence and chronic kidney disease management intersection, in particular to a nonlinear algorithm system for accurate chronic kidney disease management. Through the deep coupling of the algorithm and combining two independent quantification formulas, the transformation of CKD chronic disease management from standardized follow-up visit to accurate and dynamic algorithm driving is realized, the core pain points such as large stage evaluation deviation, progress prejudging hysteresis, poor suitability of an intervention scheme, homogenization of a follow-up visit strategy and the like in the traditional CKD management are solved, full-period decision support is provided for a three-hospital nephron physician, an individual chronic disease management path is customized for a CKD patient, and the method is suitable for CKD 1-5 stage full-period patients, and can be widely applied to the CKD special management scenes of three-hospital nephron chronic disease diagnosis and treatment centers, chronic disease management outpatient service of a nephron special department and high-end health management institutions.

Inventors

CHEN MENG
LU YANFANG
GU YUE

Assignees

河南省人民医院

Dates

Publication Date: 20260508
Application Date: 20260325

Claims (8)

1. The nonlinear algorithm system for chronic kidney disease accurate chronic disease management is characterized in that a three-layer architecture of a data layer, an algorithm layer and an application layer is adopted to form a full-flow closed loop of stage evaluation, progress prejudgment, intervention optimization and follow-up management, and the nonlinear algorithm system specifically comprises the following steps: The data layer is used for collecting and preprocessing multi-mode data, wherein the multi-mode data comprises clinical examination data, kidney image data, pathology data, gene data, patient baseline data, intervention execution data and follow-up data, and a standardized data matrix is generated after normalization, noise filtering, missing value filling and structuring treatment; The algorithm layer comprises a multi-mode data fusion kidney disease early screening algorithm, a kidney disease progress nonlinear prejudgment algorithm, a CKD individualized intervention scheme self-adaptive optimization algorithm and a CKD dynamic risk layering follow-up optimization algorithm, and is embedded with a kidney disease early abnormality degree quantization formula and a CKD disease progress risk quantization formula, wherein the four algorithms are deeply coupled, so that accurate stage assessment of the CKD, quantitative prejudgment of the progress risk, generation of the individualized intervention scheme and dynamic follow-up optimization are sequentially realized, and the two special quantization formulas provide quantization support for algorithm core decisions; The application layer comprises a clinical decision terminal, a patient management platform and a data visualization module, and is used for outputting an algorithm result, receiving man-machine interaction feedback, realizing data real-time monitoring and efficiency evaluation, and providing full-period decision support for clinical diagnosis and patient management.
2. The system of claim 1, wherein the multi-modal data fusion kidney disease early screening algorithm adopts a four-layer modeling logic of data layering analysis, feature fusion modeling, abnormality degree quantification and screening result calibration, wherein the data layering analysis extracts core features for three types of data of clinical examination, images and genes respectively, the feature fusion modeling adopts an attention mechanism fusion model to realize cross-modal feature deep coupling, the abnormality degree quantification calculates an abnormality degree coefficient gamma based on a kidney disease early abnormality degree quantification formula, and the result calibration adopts a double-check mechanism of internal cross-check and external clinical review to output a three-level screening result of normal-suspected-abnormal and CKD accurate stage.
3. The system of claim 2, wherein the early stage kidney disease abnormality quantification formula expression is γ = ω - (C ·Wc + I ·Wi + G ·Wg ) ++ (1- Ω) R.A, where ω is the modal weight coefficient (value range [0.6,0.75 ]), C Normalized values for clinical test characteristics Wc For clinical examination of feature weight vectors (satisfy Σwc =1),I Wi for normalizing image features Is the image characteristic weight vector (satisfying Σwi =1),G For the normalization of gene characteristics, wg Is a gene characteristic weight vector (satisfying Σwg =1), R is a cross-modal feature association coefficient (value range [0.8,1.2 ]), a is a crowd adaptation coefficient (value range [0.9,1.1 ]), and the formula realizes early abnormal signal quantification through a three-order logic of 'core feature leading+association correction assisting+crowd adaptation calibration'.
4. The system of claim 1, wherein the kidney disease progression nonlinear prognosis algorithm uses an improved long-short term memory network (LSTM) in combination with a dual-attention mechanism to construct a nonlinear association model, integrates four major progression factors including a core pathological factor, a clinical dynamic factor, an intervention influence factor and a baseline risk factor based on the output result of the MDS-ES algorithm, and uses a multi-task learning framework to synchronously realize progression risk probability prognosis, progression time node prognosis and key driving factor recognition, wherein the prognosis period covers 3-12 months in the future.
5. The system of claim 4, wherein the CKD risk of progression quantification formula is expressed as P = α·(F ·W )·β + (1-α)·(I·T ) Wherein P is For the future t months of disease progression risk probability (value range [0,1 ]), alpha is the weight coefficient of the core factor (value range [0.65,0.8 ]), F Normalized value for core progress factor, W For the core progress factor weight vector (satisfy Σw Beta is a multi-factor nonlinear association intensity coefficient (value range [0.9,1.3 ]), I is an intervention regulation coefficient (value range [0.6,1.1 ]), T The formula is used for adapting different stages of CKD, different intervention effects and multiple prejudgement period requirements for time attenuation factors (a value range [0.8,1.2 ]).
6. The system of claim 1, wherein the CKD personalized intervention scheme adaptive optimization algorithm employs a closed loop logic of individual feature resolution-intervention scheme generation-response feedback modeling-dynamic optimization iteration, individual feature resolution extracts four types of core features of pathological adaptation, physiological tolerance, intervention response and lifestyle adaptation, scheme generation is based on a CKD diagnosis and treat guideline framework, condition control, safety and execution difficulty are balanced through a multi-objective optimization function, adaptation scores are generated by a fuzzy comprehensive evaluation method in response to feedback, and dynamic optimization is performed by a continuous iteration scheme of a feedback-analysis-adjustment-verification mechanism, so that the adaptation scores are more than or equal to 80 minutes.
7. The system of claim 1, wherein the CKD dynamic risk stratification follow-up optimization algorithm is based on three algorithm output results, integrates prognosis feedback factors and patient management capacity factors to construct five main types of core risk factors, adopts a fuzzy clustering algorithm to divide the five main types of core risk factors into three risk levels, wherein weights are distributed into a progression risk probability ratio of 40%, a staging result ratio of 25%, an intervention response score ratio of 20%, a prognosis feedback factor ratio of 10% and a management capacity factor ratio of 5%, and dynamically generates a differentiated follow-up scheme based on the risk levels, including a follow-up period (high risk 1-2 months, stroke risk 3-4 months, low risk 6-12 months), follow-up content and follow-up modes.
8. The system of claim 1, wherein the data layer supports real-time interaction of data with hospital information systems, laboratory information systems, image archiving and communication systems, gene detection platforms, and patient home monitoring terminals, and the preprocessing flow comprises Z-score normalization, min-max normalization, outlier rejection of isolated forest algorithm, missing value filling of KNN algorithm, and non-text data structured extraction of natural language processing technology.

Description

Nonlinear algorithm system for accurate chronic disease management of chronic kidney disease Technical Field The invention relates to the technical field of medical information, in particular to a nonlinear algorithm system for accurate chronic disease management of chronic kidney disease. Background The key difficulties in Chronic Kidney Disease (CKD) chronic disease management are long disease course, dynamic change of disease condition, significant multi-factor cross influence and individual difference, and the algorithm short plate in the prior art has become a core bottleneck for restricting the improvement of management efficiency. Traditional CKD management relies on doctor experience to formulate follow-up and intervention schemes, lacks standardized algorithm modeling support, has weak coupling analysis capability on multidimensional data such as eGFR dynamic change, pathological damage degree, complications regulation and control effect and the like, and is easy to cause stage evaluation deviation and progression risk misjudgment. The existing risk pre-judging model adopts a single-index linear frame, cannot describe complex nonlinear association of pathological injury, therapeutic intervention and disease progress, has remarkable pre-judging hysteresis, and is difficult to avoid acute exacerbation risks in advance. The intervention scheme is limited to guideline standardization recommendation, and is lack of an adaptive optimization algorithm based on individual pathological characteristics and treatment response of patients, so that the homogeneity scheme is poor in suitability. The follow-up management adopts a fixed period mode, a personalized adjustment algorithm based on dynamic risk layering is not adopted, and the data and early diagnosis and treatment and intervention effects are fractured. Meanwhile, the prior art has no exclusive quantification formula for supporting core decision, the adaptability of a general algorithm is poor, a full-flow closed-loop system of 'evaluation-prejudgment-intervention-follow-up' is not formed, a set of intelligent system for accurately managing chronic disease of CKD taking an original algorithm as a core and an exclusive formula as a support is needed, and a short plate of the technology is broken. Disclosure of Invention The invention aims to provide a nonlinear algorithm system for accurate Chronic Kidney Disease (CKD) management, which is characterized by adopting a three-layer architecture of a data layer, an algorithm layer and an application layer, wherein the algorithm layer integrates four primitive core algorithms and two exclusive quantization formulas to form a whole-flow closed loop of 'stage evaluation-progress prejudgment-intervention optimization-follow-up management', and specifically comprises the following steps: The data layer is used for collecting and preprocessing multi-mode data, wherein the multi-mode data comprises clinical examination data, kidney image data, pathology data, gene data, patient baseline data, intervention execution data and follow-up data, and a standardized data matrix is generated after normalization, noise filtering, missing value filling and structuring treatment; The algorithm layer comprises a multi-mode data fusion kidney disease early screening algorithm, a kidney disease progress nonlinear prejudgment algorithm, a CKD individualized intervention scheme self-adaptive optimization algorithm and a CKD dynamic risk layering follow-up optimization algorithm, and is embedded with a kidney disease early abnormality degree quantization formula and a CKD disease progress risk quantization formula, wherein the four algorithms are deeply coupled, so that accurate stage assessment of the CKD, quantitative prejudgment of the progress risk, generation of the individualized intervention scheme and dynamic follow-up optimization are sequentially realized, and the two special quantization formulas provide quantization support for algorithm core decisions; The application layer comprises a clinical decision terminal, a patient management platform and a data visualization module, and is used for outputting an algorithm result, receiving man-machine interaction feedback, realizing data real-time monitoring and efficiency evaluation, and providing full-period decision support for clinical diagnosis and patient management. Furthermore, the multi-mode data fusion kidney disease early screening algorithm adopts a four-layer modeling logic of data layering analysis, feature fusion modeling, abnormality quantification and screening result calibration, wherein the data layering analysis respectively extracts core features aiming at three types of data of clinical examination, images and genes, the feature fusion modeling adopts an attention mechanism fusion model to realize cross-mode feature deep coupling, the abnormality quantification is based on a kidney disease early abnormality quantification formula to calculate an abnormal