Search

CN-122000035-A - Intelligent algorithm system for kidney disease full-period accurate diagnosis and treatment and management

CN122000035ACN 122000035 ACN122000035 ACN 122000035ACN-122000035-A

Abstract

The invention relates to the technical field of medical information, in particular to an intelligent algorithm system for kidney disease full-period accurate diagnosis and treatment and management. Through the deep coupling of a core algorithm, the transformation of kidney disease diagnosis and treatment from 'empirical driving' to 'algorithm driving' is realized by combining two independent quantitative formulas, the core pain points of high early diagnosis and treatment rate, delayed progress prejudging, homogenized treatment scheme, inaccurate follow-up visit management and the like in traditional diagnosis and treatment are solved, accurate decision support is provided for a clinician, full-period individualized diagnosis and treatment service is provided for a patient, and the method is suitable for diagnosis and treatment scenes of various kidney diseases such as Chronic Kidney Disease (CKD), acute Kidney Injury (AKI), nephrotic syndrome and the like. The core innovation of the invention is the breakthrough of the algorithm modeling logic and the solving process, a scientific decision system is constructed through a proprietary quantitative formula, the industry blank of the cooperative innovation of the kidney disease full-period algorithm is filled, and the diagnosis and treatment intelligent upgrading is promoted.

Inventors

  • CHEN MENG
  • LU YANFANG
  • GU YUE

Assignees

  • 河南省人民医院

Dates

Publication Date
20260508
Application Date
20260325

Claims (10)

  1. 1. An intelligent algorithm system for kidney disease full-period accurate diagnosis and treatment and management 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 major original core algorithms, namely a multi-mode data fusion kidney disease early screening algorithm, a kidney disease progress nonlinear prejudgment algorithm, an individual treatment scheme self-adaptive optimization algorithm and a prognosis follow-up visit intelligent optimization algorithm, and two original quantification formulas are embedded as core decision basis to realize kidney disease early screening, disease progress prejudgment, individual treatment scheme generation and prognosis follow-up visit management full-flow closed-loop energization; The data layer constructs a multi-mode data acquisition and preprocessing module, integrates clinical examination data, image data, pathology data, gene data, patient baseline data and follow-up data, and generates a standardized data matrix after normalization processing, noise filtering, missing value filling and structural extraction; the algorithm layer realizes the deep coupling of four core algorithms through a data interaction interface, a parameter linkage mechanism and a result feedback link, and the two original quantization formulas respectively provide quantization support for an MDS-ES algorithm and an NPP-PP algorithm for an early abnormal degree quantization formula of kidney disease and a risk quantization formula of kidney disease progression; The application layer comprises a clinical decision terminal, a patient management platform and a data visualization module, and provides targeted functions for doctors, patients and management staff respectively, so that the clinical transformation and the floor application of algorithm results are realized.
  2. 2. The intelligent algorithm system according to claim 1, wherein the multi-modal data fusion kidney disease early stage screening algorithm adopts a four-layer architecture of data layering analysis-feature fusion modeling-abnormality quantification-screening result calibration, and the kidney disease early stage abnormality quantification formula expression is gamma=ω (C) ·Wc + I ·Wi + G ·Wg ) + (1-ω)·R·A; Wherein, gamma is the early abnormality degree coefficient of kidney disease, the value range is [0,1], omega is the modal weight coefficient, the value range is [0.6,0.75], C Normalized value, wc, for the ith clinical test feature Is a clinical test feature weight vector and satisfies Σwc =1;I Wi for the j-th image feature normalization value Is an image characteristic weight vector and meets the requirement of Sigma Wi =1;G Normalized value for the kth gene, wg Is a gene characteristic weight vector and satisfies Σwg 1;R is the cross-modal feature correlation coefficient, the range of values [0.8,1.2], and a is the population adaptation coefficient, the range of values [0.9,1.1].
  3. 3. The intelligent algorithm system according to claim 1, wherein the kidney disease progression nonlinear prognosis algorithm is modeled by a modified long-short term memory network (LSTM) combined with a dual-attention mechanism, and the kidney disease progression risk quantification formula is expressed as P=μ (P ·W + P_c·W_c + P_i·W_i) + (1-μ)·T·(1 - P_b·W_b); Wherein P is the probability of the disease progression of the kidney disease, the value range is [0,1], mu is the weight coefficient of the core factor, the value range is [0.65,0.8], P As the comprehensive value of the core pathological factors, W Is a core pathological factor weight vector and satisfies ΣW P_c is a clinical dynamic factor integrated value, W_c is a clinical dynamic factor weight vector and satisfies ΣW_c=1, P_i is an intervention influence factor integrated value, W_i is an intervention influence factor weight vector and satisfies ΣW_i= 1;T is a time sequence correction coefficient, a value range [0.85,1.15], P_b is a baseline risk factor integrated value, and W_b is a baseline risk factor weight vector and satisfies ΣW_b=1.
  4. 4. The intelligent algorithm system according to claim 1, wherein the personalized treatment scheme adaptive optimization algorithm adopts a four-layer architecture of 'individual feature analysis-scheme generation modeling-curative effect prejudgement and risk assessment-dynamic optimization calibration', and introduces a personalized treatment scheme comprehensive proper formulation quantization formula, wherein the expression is S=α.F+β (E- λ.R) + (1- α - β) D; Wherein S is the comprehensive fit degree score of the treatment scheme, the value range is [0,10], alpha is the individual characteristic fit degree weight coefficient, the value range is [0.35,0.45], F is the individual characteristic fit degree coefficient, beta is the treatment effect risk balance item weight coefficient, the value range is [0.4,0.5], E is the treatment effect pre-judgment score, lambda is the risk weight coefficient, the value range is [1.2,1.5], R is the adverse reaction risk score, and D is the dynamic feedback calibration coefficient.
  5. 5. The intelligent algorithm system according to claim 1, wherein the prognosis follow-up intelligent optimization algorithm adopts a four-layer architecture of follow-up risk stratification-strategy generation optimization-data acquisition analysis-dynamic iteration adjustment, and a follow-up comprehensive risk quantification formula is introduced, wherein the expression is R_total=ω 1 ·R_b + ω 2 ·R_t + ω 3 . R_h+epsilon. K; Wherein R_total is a follow-up comprehensive risk score, the value range is [0,10], omega 1 、ω 2 、ω 3 is a disease condition basic risk, a treatment response risk and a home management risk weight coefficient respectively, omega 1 +ω 2 +ω 3 =1 is satisfied, R_b is a disease condition basic risk score, R_t is a treatment response risk score, R_h is a home management risk score, epsilon is a dynamic correction coefficient, the value range is [0.85,1.15], K is a strategy adaptation calibration coefficient, and the value range is [0.9,1.1].
  6. 6. The intelligent algorithm system according to claim 2, wherein the feature fusion modeling of the MDS-ES algorithm adopts a three-order strategy of local feature enhancement-cross-modal fusion-feature dimension reduction optimization, the local feature enhancement determines each feature importance weight through a hierarchical analysis method, the core feature weight is assigned with 0.6-0.8, the secondary feature weight is assigned with 0.2-0.4, the cross-modal fusion builds an attention mechanism fusion model, the correlation coefficient among different modal features is calculated through a self-attention network, and the feature dimension reduction optimization adopts a method of combining Principal Component Analysis (PCA) with sparse matrix decomposition.
  7. 7. The intelligent algorithm system according to claim 3, wherein the dual-attention mechanism of the NPP-PP algorithm comprises a time sequence attention mechanism and a factor-associated attention layer, the time sequence attention mechanism gives high attention weight to key time nodes of clinical dynamic factors, the factor-associated attention layer calculates association coefficients among different types of factors to generate a multi-factor nonlinear association matrix, and the algorithm synchronously realizes three core tasks of progression risk probability prejudgement, progression time node prejudgement and key driving factor identification by adopting a multi-task learning framework.
  8. 8. The intelligent algorithm system according to claim 4, wherein the scheme generation modeling of the ATO-TS algorithm adopts an improved collaborative filtering algorithm in combination with clinical guideline constraint, a basic scheme library is constructed based on clinical guideline, effective scheme parameters are extracted by matching similar historical cases through the collaborative filtering algorithm, 2-3 sets of candidate schemes are generated by combining individual characteristic adjustment scheme details, the multi-task learning framework synchronously outputs a curative effect pre-judging result and an adverse reaction risk assessment result, a curative effect and risk weight are determined by adopting a hierarchical analysis method, and an optimal scheme is screened.
  9. 9. The intelligent algorithm system according to claim 5, wherein the follow-up risk stratification of the PFO-FU algorithm is based on a three factor system of "basic risk of illness-treatment response risk-home management risk", weighting is given by using a hierarchical analysis method, the basic risk of illness is 0.5, the treatment response risk is 0.3, the home management risk is 0.2, comprehensive follow-up risk grades are calculated and classified into high-risk, medium-risk and low-risk grades, and personalized follow-up strategies are customized based on the grades.
  10. 10. A kidney disease full cycle accurate management method based on the intelligent algorithm system of any one of claims 1-9, comprising the steps of: Step 1, multi-mode data are acquired through a data layer multi-channel, a standardized data matrix is generated through preprocessing, and the standardized data matrix is input into an algorithm layer; Step 2, performing hierarchical analysis, feature fusion, abnormality quantification and result calibration on the data by using an MDS-ES algorithm, and outputting an early screening result; Step 3, the NPP-PP algorithm is based on the screening result, a multi-dimensional progress factor is integrated, and a disease progress risk assessment result is output through nonlinear association modeling and quantitative formula calculation; step 4, the ATO-TS algorithm integrates individual characteristics to generate candidate treatment schemes based on screening and prejudging results, and the optimal schemes are screened through curative effect and risk assessment and pushed to an application layer; and 5, based on treatment scheme and curative effect feedback, the PFO-FU algorithm customizes the follow-up strategy in a layered manner, acquires follow-up data and dynamically iterates and optimizes to form full-period closed-loop management.

Description

Intelligent algorithm system for kidney disease full-period accurate diagnosis and treatment and management Technical Field The invention belongs to the technical field of medical artificial intelligence and kidney disease diagnosis and treatment intersection, and particularly relates to an intelligent algorithm system covering the whole flow of early screening of kidney disease, prognosis of disease progress, generation of an individual treatment scheme and prognosis follow-up visit management. Background The complexity of nephrotic diagnosis and treatment is derived from the nonlinear association and dynamic change characteristics of multi-modal clinical data, and the algorithm short plate in the prior art has become a core bottleneck restricting the improvement of diagnosis and treatment efficiency, so that a targeted breakthrough is needed. Traditional kidney disease diagnosis and treatment rely on personal clinical experience of doctors, lack of standardized and systematic algorithm modeling support, have weak deep coupling analysis capability on multidimensional data such as serum creatinine, urine protein, pathological sections, gene markers and the like, are difficult to capture hidden abnormal signals of early kidney disease, and cause that the failure rate of diagnosis of the kidney disease in the asymptomatic period is high, and the condition is mostly advanced to middle and late stages in the diagnosis. The existing risk pre-judging model generally adopts a single-index linear analysis framework, focuses only on the static change of core indexes such as eGFR and the like, does not construct multi-factor cooperative pre-judging logic, cannot describe complex nonlinear association between factors such as pathological damage, complications, medication intervention and the like and disease progress, has remarkable pre-judging hysteresis, and is difficult to realize early intervention. Treatment scheme generation is limited to standardized recommendation of clinical guidelines, and lack of an adaptive optimization algorithm based on individual differences of patient gene background, physiological characteristics, treatment preference and the like, a homogenization scheme cannot adapt to a complex clinical scene, and adverse reactions or treatment ineffectiveness are easily caused. The follow-up management adopts a fixed period mode, a dynamic risk layering and frequency optimization algorithm is not adopted, the data is split in the early diagnosis and treatment link, and intervention measures cannot be adjusted in real time based on the change of the illness state. Meanwhile, the prior art has no exclusive quantification formula to support a core decision, the adaptability is poor along with a general medical algorithm framework, a screening-prejudging-treating-follow-up full-flow closed-loop algorithm system is not formed, each link operates independently, data are not communicated, and the overall diagnosis and treatment efficiency is difficult to break through. In summary, a set of full-period intelligent algorithm system taking an original algorithm as a core and a dedicated quantization formula as a support is needed in the industry, so that the short plates in the prior art are broken, and the accurate and intelligent transformation of nephrosis diagnosis and treatment is promoted. Disclosure of Invention The invention aims to provide an intelligent algorithm system for full-period accurate diagnosis and treatment and management of kidney diseases, which adopts a three-layer architecture of a data layer, an algorithm layer and an application layer, wherein the algorithm layer integrates four major original core algorithms, namely a multi-mode data fusion kidney disease early screening algorithm, a kidney disease progress nonlinear prejudgment algorithm, an individual treatment scheme self-adaptive optimization algorithm and a prognosis follow-up visit intelligent optimization algorithm, and two original quantification formulas are embedded as core decision basis to realize kidney disease early screening, illness progress prejudgment, individual treatment scheme generation and prognosis follow-up visit management whole-flow closed-loop energization; The data layer constructs a multi-mode data acquisition and preprocessing module, integrates clinical examination data, image data, pathology data, gene data, patient baseline data and follow-up data, and generates a standardized data matrix after normalization processing, noise filtering, missing value filling and structural extraction; the algorithm layer realizes the deep coupling of four core algorithms through a data interaction interface, a parameter linkage mechanism and a result feedback link, and the two original quantization formulas respectively provide quantization support for an MDS-ES algorithm and an NPP-PP algorithm for an early abnormal degree quantization formula of kidney disease and a risk quantization formula of kidney disease p