A neonatal nutrient solution intelligent proportioning and health guarantee method and system and medium
By constructing an intelligent system for multimodal data fusion and causal inference, the shortcomings of the neonatal nutritional support system in data fusion, phenotypic cluster-guideline binding, and risk assessment have been addressed. This has improved the accuracy and safety of personalized nutritional support, reduced the incidence of adverse events, and increased the reuse rate and efficiency of formulas.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN PROVINCIAL HOSPITAL FOR WOMEN & CHILDREN
- Filing Date
- 2026-05-15
- Publication Date
- 2026-06-12
AI Technical Summary
Existing neonatal nutritional support systems have shortcomings in multimodal data fusion, phenotypic cluster-guideline binding, risk assessment causality, and clinical closure, resulting in low formula reuse rates, poor guideline adaptability, and insufficient accuracy in risk assessment, making it difficult to meet the personalized nutritional needs of newborns.
We construct an intelligent system that combines a clinical validation model, multimodal dynamic weight fusion, phenotypic cluster-guideline binding, and causal inference closed loop. By using an improved variational neural network, LSTM, and XGBoost model, combined with causal forest and logistic regression, we achieve dynamic weight fusion of multimodal data, dynamic adjustment of phenotypic clusters, and causal risk assessment, forming a closed-loop feedback mechanism.
It significantly improves the accuracy and safety of nutritional support, reduces the incidence of adverse events such as hyperglycemia and abnormal liver function, increases the reusability and preparation efficiency of formulas, and ensures fairness in risk assessment for children of different races and disease subtypes.
Smart Images
Figure CN122201639A_ABST
Abstract
Claims
1. A method for intelligent formulation and health protection of neonatal nutrient solution, characterized in that, Includes the following steps: (1) Obtain multimodal physiological data of newborns, including gestational age, birth weight, real-time disease diagnosis, and metabolic indicators; (2) The improved variational neural network is used to learn the feature representation of the multimodal physiological data. The improved variational neural network introduces a modal attention mechanism to dynamically adjust the weights of different physiological indicators according to the real-time disease diagnosis of the child. (3) Cluster the extracted features to form 15-20 phenotypic clusters, each phenotypic cluster corresponding to a three-dimensional label of gestational age-disease-metabolic state; (4) Construct a dynamic mapping relationship model between each phenotypic cluster and the parameter space of multi-regional nutrition guidelines, wherein the guidelines include Chinese, American and European guidelines, and each cluster is preset with parameter offset rules under different guidelines; (5) Match and output the nutrient solution ratio from the standardized formula library based on the newborn's current phenotypic cluster affiliation and the selected regional guidelines; (6) The formula library is updated monthly based on new clinical data to automate guideline adaptation.
2. The method for intelligent preparation and health protection of neonatal nutrient solution according to claim 1, characterized in that, In step (2), the modal attention mechanism is based on the correlation between the indicator and the current disease diagnosis. Dynamic weight allocation The dynamic weight allocation calculation method is as follows: in, Let i be the weight of the i-th indicator. The basic weight of indicator i, This is for adjusting the coefficient.
3. The method for intelligent formulation and health protection of neonatal nutrient solution according to claim 2, characterized in that, When C i When ≥0.8, ; When 0.5≤C i When <0.8, When C i When <0.5, Among them, the basic weight It is 0.
2.
4. The method for intelligent preparation and health protection of neonatal nutrient solution according to claim 1, characterized in that, The clustering uses the K-means algorithm, and the number of clusters k is determined by the elbow rule, with a value ranging from 15 to 20.
5. The method for intelligent preparation and health protection of neonatal nutrient solution according to claim 1, characterized in that, In step (4), the parameter offset rule is adapted to the guide using the following formula: in, These are the basic nutritional parameters for the phenotypic clusters. This is the preset offset for the guide corresponding to the selected region.
6. The method for intelligent preparation and health protection of neonatal nutrient solution according to claim 1, characterized in that, In step (6), the incremental update includes: identifying newly emerging combinations of gestational age-disease-metabolic state; when the frequency of occurrence exceeds a set threshold, generating new phenotypic clusters and establishing corresponding multi-regional guideline parameter offset rules for them.
7. A smart formula mixing and health protection system for newborn nutrition solutions, characterized in that, include: The data acquisition unit is used to acquire multimodal physiological data of newborns; The AI algorithm processing unit is used to perform modal attention weighting and feature extraction, cluster the extracted features into 15-20 phenotypic clusters, and store and execute parameter offset rules for each phenotypic cluster under the guidelines of China, the United States and Europe. Nutrient solution preparation execution unit, used to output nutrient solution ratio based on the current phenotypic cluster and the selected guideline; The health monitoring feedback unit is used to automatically update the parameter offset rules and standardized formula library stored in the clinical phenotype clustering and standardized formula library module of the AI algorithm processing unit every month.
8. The system according to claim 7, characterized in that, The clinical phenotype clustering and standardized formula library module stores parameter offset rules for each phenotype cluster under Chinese, American, and European guidelines, and allows users to select one or customize a regional parameter offset rule.
9. The system according to claim 7, characterized in that, The health monitoring feedback unit includes a model incremental learning submodule, which is configured to automatically trigger a new cluster generation process and initialize multi-region guide parameter offset rules when the frequency of occurrence of a new phenotype combination exceeds a preset threshold.
10. A computer-readable storage medium, characterized in that, It stores a computer program for implementing the intelligent formulation and health protection system for neonatal nutrition solutions as described in any one of claims 7 to 9.