CN-121997172-A - Method and device for realizing blood demand prediction based on AI large model

CN121997172ACN 121997172 ACN121997172 ACN 121997172ACN-121997172-A

Abstract

The application provides a method and a device for realizing blood demand prediction based on an AI large model, which can solve the problems of limited accuracy and poor adaptability caused by the fact that the traditional expert system is solidified and cannot self-evolve, thereby accurately realizing blood demand prediction. The method comprises the steps of obtaining characteristic data related to blood consumption requirements of a patient to be predicted, determining whether the patient to be predicted needs blood transfusion according to the characteristic data, determining predicted blood consumption of the patient based on a trained prediction model if the patient to be predicted needs blood transfusion, and determining that the predicted blood consumption of the patient is zero if the patient to be predicted does not need blood transfusion.

Inventors

CHEN QING
ZHANG XUEBING
WANG SHUYA
GONG JING
YU JINFENG

Assignees

南京鼓楼医院
创观（苏州）生物科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. A method for implementing blood demand prediction based on AI large model, comprising: Acquiring characteristic data related to blood demand of a patient to be predicted; Determining whether the patient to be predicted needs blood transfusion according to the characteristic data; If the patient to be predicted is determined to need blood transfusion, determining the predicted blood volume of the patient based on a trained prediction model; and if the patient to be predicted is determined to not need blood transfusion, determining that the predicted blood volume of the patient is zero.
2. The method of claim 1, wherein the predictive model is a regression model that employs a lightweight gradient hoist algorithm, the method further comprising: Constructing a first data set based on characteristic data of a plurality of patients with historical blood transfusion amounts greater than zero and corresponding actual blood use amounts, wherein the characteristic data comprises one or more of the following information: basic information, blood routine examination information and blood coagulation function examination information, wherein the basic information comprises physiological state parameters and/or operation characteristic parameters of a patient; Based on the first data set, training an initial prediction model by taking the actual blood volume as a prediction target to obtain a trained prediction model.
3. The method according to claim 1 or 2, wherein said determining from said characteristic data whether said patient to be predicted requires transfusion comprises: and based on the trained classification model, predicting according to the characteristic data, and determining whether the patient to be predicted needs blood transfusion or not.
4. The method of claim 3, wherein the classification model employs a lightweight gradient hoist algorithm, the method further comprising: constructing a second data set based on historical multiple patient characteristic data and corresponding bi-classification labels for each patient whether transfusion is required, the characteristic data including but not limited to one or more of the following information: basic information, blood routine examination information and blood coagulation function examination information, wherein the basic information comprises physiological state parameters and/or operation characteristic parameters of a patient; Based on the second data set, training the initial classification model by taking the corresponding classification label as a training target to obtain a trained classification model.
5. The method of claim 4, wherein prior to constructing the second dataset based on the historical plurality of patient characteristic data and the corresponding bi-classification labels for each patient for which transfusion is required, the method further comprises: If a plurality of missing feature data with field content missing conditions exist in the feature data of a plurality of historical patients, filling a part of the feature data in the missing feature data, and keeping missing marks in another part of the feature data in the missing feature data without filling.
6. The method according to claim 1 or 2, characterized in that before the acquisition of the characteristic data related to blood demand of the patient to be predicted, the method comprises: Receiving natural language query information input by a user; Analyzing the natural language query information based on a large language model to obtain the intention of the user; If the intention indicates to acquire the predicted blood volume of the patient to be predicted, determining whether the natural language query information contains the identity of the patient to be predicted or not based on the large language model; And if the natural language query information contains the identity of the patient to be predicted, acquiring the characteristic data of the patient to be predicted, which is related to blood consumption requirements, according to the identity analyzed by the large language model.
7. The method of claim 6, wherein the method further comprises: if the natural language query information does not contain the identity of the patient to be predicted, determining whether the natural language query information contains the characteristic data of the patient to be predicted or not based on the large language model; if the natural language query information does not contain the characteristic data of the patient to be predicted, interacting with the user based on the large language model until the identity or the characteristic data of the patient to be predicted is obtained; If the natural language query information contains the characteristic data of the patient to be predicted, determining whether the characteristic data of the patient to be predicted meets the preset condition for subsequent processing or not based on the large language model; if the characteristic data of the patient to be predicted meets the preset condition, performing subsequent processing according to the characteristic data of the patient to be predicted; otherwise, interacting with the user based on the large language model until feature data meeting the preset conditions are obtained.
8. The method according to claim 6, wherein the obtaining the characteristic data of the patient to be predicted related to blood demand according to the identification analyzed by the large language model includes: based on a model context protocol, acquiring the identity of the patient to be predicted, which is analyzed by the large language model; And acquiring the characteristic data corresponding to the identity of the patient to be predicted by calling a data complement interface.
9. The method of claim 2, wherein prior to constructing the second dataset based on the historical plurality of patient characteristic data and the corresponding bi-classification labels for each patient for which transfusion is required, the method further comprises: carrying out importance assessment on various types of characteristics of the characteristic data by using a saprolitic additive interpretation method to obtain a quantized value of the importance of each type of characteristics; and deleting the data corresponding to the type of the features with the quantized values lower than the preset threshold value in the feature data.
10. The device for realizing blood demand prediction based on the AI large model is characterized by comprising a blood consumption prediction module, wherein the blood consumption prediction module comprises a characteristic data acquisition sub-module, a classification sub-module and a blood consumption prediction sub-module; the characteristic data acquisition sub-module is used for acquiring characteristic data related to blood consumption requirements of a patient to be predicted; the classifying sub-module is used for determining whether the patient to be predicted needs blood transfusion or not according to the characteristic data; The blood consumption prediction submodule is used for determining the predicted blood consumption of the patient based on a trained prediction model if the patient to be predicted needs blood transfusion; the blood volume prediction sub-module is further used for determining that the predicted blood volume of the patient is zero if it is determined that the patient to be predicted does not need blood transfusion.

Description

Method and device for realizing blood demand prediction based on AI large model Technical Field The application relates to the technical field of data processing, in particular to a method and a device for realizing blood demand prediction based on an AI large model. Background Blood transfusion is an important treatment means in clinical medical treatment, and accurately predicting the blood transfusion quantity of a patient has important significance for reasonable allocation of blood resources, medical cost reduction and treatment effect improvement. Currently, expert systems or rule-driven systems (which may be collectively referred to as predefined rule-based systems) built based on medical expert knowledge are mainly used for blood demand prediction. Such systems judge and predict the blood demand of a patient by presetting a series of rules and thresholds summarized by clinical experience. For example, when the patient's hemoglobin is below a certain fixed value, a transfusion is determined to be required. However, the predefined rules according to the system are relatively solidified, and cannot be self-learned and optimally adjusted from continuously accumulated clinical data, so that the prediction accuracy is limited by expert knowledge of initial setting, and the system is difficult to adapt to complex and changeable clinical actual conditions. Therefore, a solution that can accurately realize blood demand prediction is needed. Disclosure of Invention The application provides a method and a device for realizing blood demand prediction based on an AI large model, which can solve the problems of limited accuracy and poor adaptability caused by the fact that the traditional expert system is solidified and cannot self-evolve, thereby accurately realizing blood demand prediction. In a first aspect, a method for implementing blood demand prediction based on an AI large model is provided, including: Acquiring characteristic data related to blood demand of a patient to be predicted; determining whether the patient to be predicted needs blood transfusion according to the characteristic data; if the patient to be predicted is determined to need blood transfusion, determining the predicted blood volume of the patient based on the trained prediction model; if it is determined that the patient to be predicted does not need blood transfusion, determining that the predicted blood volume of the patient is zero. In one possible design, the predictive model is a regression model that employs a lightweight gradient hoist algorithm, the method further comprising: Constructing a first data set based on characteristic data of a plurality of patients with historical blood transfusion amounts greater than zero and corresponding actual blood use amounts, wherein the characteristic data comprises one or more of the following information: Basic information, blood routine examination information, blood clotting function examination information, wherein the basic information comprises physiological state parameters and/or surgical characteristic parameters of a patient; Based on the first data set, training an initial prediction model by taking the actual blood volume as a prediction target to obtain a trained prediction model. In one possible design, determining from the characterization data whether the patient to be predicted requires transfusion includes: based on the trained classification model, prediction is carried out according to the characteristic data, and whether the patient to be predicted needs blood transfusion or not is determined. In one possible design, the classification model employs a lightweight gradient hoist algorithm, the method further comprising: Constructing a second data set based on historical multiple patient characteristic data and corresponding bi-classification labels for each patient whether transfusion is required, the characteristic data including, but not limited to, one or more of the following information: Basic information, blood routine examination information, blood clotting function examination information, wherein the basic information comprises physiological state parameters and/or surgical characteristic parameters of a patient; Based on the second data set, training the initial classification model by taking the corresponding classification label as a training target to obtain a trained classification model. In one possible design, prior to acquiring the characteristic data relating to blood demand of the patient to be predicted, the method includes: Receiving natural language query information input by a user; analyzing the natural language query information based on the large language model to obtain the intention of the user; If the intention indicates to obtain the predicted blood volume of the patient to be predicted, determining whether the natural language query information contains the identity of the patient to be predicted or not based on the large language model; i