CN-122025150-A - Intelligent evaluation method, device and storage medium before vascular access catheterization

CN122025150ACN 122025150 ACN122025150 ACN 122025150ACN-122025150-A

Abstract

The invention belongs to the technical field of vascular access catheterization evaluation, and discloses an intelligent evaluation method, device and storage medium before vascular access catheterization, wherein the method comprises the steps of firstly acquiring medical data of three characteristics of a patient treatment scheme, vascular conditions and host risk, and constructing a comprehensive feature vector through numerical treatment and standardized treatment; the method comprises the steps of inputting the initial adaptive probability vector into a pre-trained multi-classification gradient lifting decision tree model, carrying out probability calibration, correcting the calibrated probability based on a clinical rule set containing hard constraint rules and soft penalty rules, generating a final recommended probability vector, and finally outputting a recommended list and key feature contribution information according to probability sequencing. The invention realizes quantitative and accurate assessment of the vascular access, avoids selection deviation caused by experience dependence, reduces complication risk, improves clinical decision efficiency and safety, has transparent and interpretable decision process, and meets the requirements of clinical practical application.

Inventors

WU YI
YING LI
ZHOU PING

Assignees

浙江省肿瘤医院

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. An intelligent evaluation method before vascular access catheterization is characterized by comprising the following steps: Acquiring medical data of a patient to be evaluated, wherein the medical data at least comprises a treatment scheme feature set, a blood vessel condition feature set and a host risk feature set; Inputting the comprehensive feature vector into a pre-trained multi-classification gradient lifting decision tree model to obtain initial adaptation probability corresponding to each candidate vascular access type, and forming an initial adaptation probability vector; carrying out probability calibration processing on the initial adaptive probability vector to obtain a calibrated path adaptive probability vector; performing constraint correction on the calibrated path adaptation probability vector based on a preset clinical rule set to obtain a final path recommendation probability vector, wherein the clinical rule set comprises a hard constraint rule and a soft penalty rule; And generating and outputting a blood vessel access recommendation result based on the final access recommendation probability vector.
2. The method for intelligent evaluation before vascular access catheterization according to claim 1, wherein, The treatment scheme feature set comprises a medicine irritation grade, a drug administration mode and an expected treatment period length, the blood vessel condition feature set comprises a peripheral blood vessel diameter, a visual score and past puncture times, and the host risk feature set comprises a blood coagulation related index and a nutritional state index.
3. The method for intelligent evaluation before vascular access catheterization according to claim 2, wherein: the multi-classification gradient lifting decision tree model adopts LightGBM, XGBoost or CatBoost frames, an objective function is a softmax multi-classification function, the output dimension is consistent with the number of candidate vascular access categories, and the candidate vascular access categories at least comprise a central venous catheter, a midline catheter and a central venous catheter which are placed through peripheral veins.
4. The method for intelligent evaluation before vascular access catheterization according to claim 1, wherein, The probabilistic calibration process is an equidistant regression calibration, a logistic regression calibration, or a temperature scaling calibration.
5. The method for intelligent evaluation before vascular access catheterization according to claim 1, wherein, The constraint correction comprises the steps of setting the adaptation probability of the corresponding candidate vascular access type to zero or setting the adaptation probability to be the minimum value when the characteristics of the patient to be evaluated meet the triggering condition of the hard constraint rule, and carrying out weight reduction processing on the adaptation probability of the corresponding candidate vascular access type based on the punishment intensity coefficient and the risk punishment item when the characteristics of the patient to be evaluated meet the triggering condition of the soft punishment rule.
6. The method for intelligent evaluation before vascular access catheterization as defined in claim 5, wherein, The weight reduction processing is realized by multiplying the adaptive probability of each candidate vascular passageway type in the calibrated passageway adaptive probability vector by an exponential decay factor based on the punishment intensity coefficient and the risk punishment item of the candidate vascular passageway type respectively, and normalizing the product result to obtain the corresponding adaptive probability in the final passageway recommendation probability vector.
7. The method for intelligent evaluation before vascular access catheterization according to claim 1, wherein, The blood vessel access recommendation result comprises a recommendation list which is ordered according to the final access recommendation probability and key feature contribution information which is associated with the recommendation result, wherein the key feature contribution information is calculated by a SHAP value or feature gain contribution method and is used for indicating the contribution degree and the contribution direction of input features to the recommendation probability.
8. A vascular access pre-deployment intelligent assessment system, comprising: The system comprises a data acquisition and processing module, a data processing module and a data processing module, wherein the data acquisition and processing module is used for acquiring medical data of a patient to be evaluated, and the medical data at least comprises a treatment scheme feature set, a blood vessel condition feature set and a host risk feature set; The initial probability prediction module is used for inputting the comprehensive feature vector into a pre-trained multi-classification gradient lifting decision tree model to obtain initial adaptation probability corresponding to each candidate vascular access type, and an initial adaptation probability vector is formed; the probability calibration module is used for carrying out probability calibration processing on the initial adaptive probability vector to obtain a calibrated path adaptive probability vector; the rule correction module is used for carrying out constraint correction on the calibrated path adaptation probability vector based on a preset clinical rule set to obtain a final path recommendation probability vector; And the result output module is used for generating and outputting a blood vessel access recommendation result based on the final access recommendation probability vector.
9. An electronic device, the electronic device comprising: and a memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.
10. A computer readable storage medium having stored thereon computer program instructions executable by a processor to implement the method of any of claims 1-7.

Description

Intelligent evaluation method, device and storage medium before vascular access catheterization Technical Field The invention relates to the technical field of vascular access tube placement evaluation, in particular to an intelligent evaluation method, device and storage medium before vascular access tube placement. Background Vascular access is an indispensable key link in clinical treatment, and especially for patients needing long-term administration and infusion of stimulatory drugs, reasonable selection of access types directly affects treatment effect and patient safety. In the current clinical practice, the vascular access selection mainly depends on the clinical experience of medical staff, and the lack of uniform and quantitative evaluation standards leads to significant differences in judgment of different operators, so that the objectivity and scientificity of the selection are difficult to ensure. The traditional blood vessel access assessment method mostly adopts a static questionnaire or a single factor judgment mode, and cannot comprehensively integrate the core requirements of the treatment scheme such as drug irritation, administration mode, treatment period and the like, the peripheral blood vessel diameter of a patient, the visual degree, the past blood vessel conditions such as feudal provincial and the like, and meanwhile, the influence of host risk factors such as blood coagulation function, nutrition state and the like on the access safety is ignored. The evaluation mode of the prior art on one side is easy to cause improper access selection, which not only can obviously increase the occurrence risk of complications such as thrombus, infection, leakage and the like, but also can cause the problems of early tube drawing, repeated puncture and the like of the access, thereby not only increasing the pain and the economic burden of patients, but also increasing the workload of clinical care. In addition, the existing evaluation system can only judge whether a single channel is applicable or not, quantitative comparison results of multiple candidate channels cannot be provided, and part of AI evaluation models have the problem of opaque black boxes in the decision process, so that the credibility and controllability requirements of clinical decisions are difficult to meet, and the wide landing of the AI evaluation models in clinical scenes is limited. Disclosure of Invention In view of the above, the present invention provides a method, an apparatus and a storage medium for intelligent evaluation before vascular access catheterization, so as to at least partially solve the above technical problems. The invention provides an intelligent evaluation method before vascular access catheterization, which comprises the following steps: Acquiring medical data of a patient to be evaluated, wherein the medical data at least comprises a treatment scheme feature set, a blood vessel condition feature set and a host risk feature set; Inputting the comprehensive feature vector into a pre-trained multi-classification gradient lifting decision tree model to obtain initial adaptation probability corresponding to each candidate vascular access type, and forming an initial adaptation probability vector; carrying out probability calibration processing on the initial adaptive probability vector to obtain a calibrated path adaptive probability vector; performing constraint correction on the calibrated path adaptation probability vector based on a preset clinical rule set to obtain a final path recommendation probability vector, wherein the clinical rule set comprises a hard constraint rule and a soft penalty rule; And generating and outputting a blood vessel access recommendation result based on the final access recommendation probability vector. In one possible embodiment, the treatment regimen feature set comprises a drug irritation level, a mode of administration, and an expected treatment cycle length, the vascular condition feature set comprises a peripheral vascular diameter, a visual score, and a past number of punctures, and the host risk feature set comprises a coagulation-related indicator, a nutritional status indicator. In one possible embodiment, the multi-classification gradient-lifted decision tree model employs a LightGBM, XGBoost or CatBoost framework, the objective function is a softmax multi-classification function, and the output dimension is consistent with the number of candidate vascular access categories, and the candidate vascular access categories at least include a central venous catheter, a midline catheter, and a central venous catheter placed through a peripheral vein. In one possible embodiment, the probabilistic calibration process is equidistant regression calibration, logistic regression calibration, or temperature scaling calibration. In one possible embodiment, the constraint correction comprises the steps of setting the adaptation probability of the corresponding candidate v