Artificial intelligence-based venous catheter matching method and system

Abstract

The present invention relates to the field of medical assistance technology and discloses an artificial intelligence-based venous catheter matching method and system. The method comprises the following steps: obtaining water content data of a patient's upper and lower limbs in real time through a human body composition analysis device, and simultaneously collecting blood draw thrombosis index data of the patient; performing data cleaning on the collected water content data and thrombosis index data to remove outliers and noise, and obtaining data to be analyzed after normalization; constructing an SVCS risk prediction model, inputting the patient's data to be analyzed, and outputting an SVCS risk probability value through the SVCS risk prediction model; and generating a catheter placement site allocation recommendation through a catheter matching model based on the SVCS risk probability value and the patient's clinical characteristics. The method combines the water content data of the patient's upper and lower limbs and the thrombosis index of the patient to perform a comprehensive and accurate assessment of the patient before catheterization, and judges in advance whether the patient is suitable for upper body catheterization, thereby improving the safety and effectiveness of catheterization.

Description

Technical Field

[0001] The present invention relates to the field of medical assistance technology, and in particular to an artificial intelligence-based venous catheter matching method and system. Background Art

[0002] In clinical medicine, intravenous catheterization (such as CVC tube) is a commonly used treatment method used to provide patients with long-term venous access. Currently, according to industry experience, patients are routinely assigned upper body catheterization. However, during subsequent examinations, such as CT examinations, patients are not immediately examined upon admission. Patients may be found to have superior vena cava syndrome (SVCS). In this case, the upper body catheterization is no longer usable and needs to be replaced with a lower body catheterization. This not only causes the patient's upper body to be intubated, increasing their pain and infection risk, but also wastes medical resources. Therefore, there is an urgent need for a method that can accurately assess the patient's condition before catheterization and rationally allocate the catheterization site. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design an artificial intelligence-based venous catheter matching method and system.

[0004] A first aspect of the present invention provides an artificial intelligence-based venous catheter matching method, the method comprising the following steps:

[0005] The body composition analysis device is used to obtain the water content data of the patient's upper and lower limbs in real time, and the patient's blood test thrombosis index data is collected at the same time;

[0006] The collected water content data and thrombosis index data were cleaned to remove outliers and noise, and the data to be analyzed were obtained after normalization;

[0007] Construct an SVCS risk prediction model, input the patient's data to be analyzed, and output the SVCS risk probability value through the SVCS risk prediction model;

[0008] Based on the SVCS risk probability value and the patient's clinical characteristics, a catheter placement matching model was used to generate a catheter placement recommendation.

[0009] Optionally, in a first implementation of the first aspect of the present invention, the data to be analyzed is obtained by performing data cleaning on the collected water density data and thrombosis index data to remove outliers and noise and normalizing the data, including:

[0010] The Isolation Forest algorithm was used to detect outliers in the water density data and thrombosis index data. Multiple random trees were constructed and the path length of each data sample in the tree was calculated to evaluate the degree of anomaly.

[0011] Based on the preset anomaly score threshold, abnormal data points that deviate from the normal range are screened out. For the detected abnormal data points, a generative adversarial network is used to repair them.

[0012] The minimum-maximum normalization method was used to normalize the water density data and thrombosis index data to obtain the data to be analyzed.

[0013] Optionally, in a second implementation of the first aspect of the present invention, the SVCS risk prediction model is constructed using an MC-GNN neural network, and the MC-GNN neural network includes two GCN layers and one fully connected layer;

[0014] The input in channel 1 is the node feature matrix and adjacency matrix of the topological graph, and the spatial correlation features between indicators are extracted through the first layer of GCN;

[0015] The input in channel 2 is the node feature matrix and time series adjacency matrix of the time series graph, and the time series features of the indicators are extracted through the first layer of GCN;

[0016] The output features of the two channels are concatenated for feature fusion, and finally the SVCS risk probability value is output through the fully connected layer.

[0017] Optionally, in a third implementation of the first aspect of the present invention, the topological graph nodes include upper limb water ratio, lower limb water ratio and thrombosis index, and the edges between the nodes represent the correlation between the indicators, wherein the correlation is determined by calculating the Pearson correlation coefficient, and undirected edges are established between indicators whose absolute value of the correlation coefficient is greater than 0.3, and the weight is the correlation coefficient value.

[0018] Optionally, in a fourth implementation of the first aspect of the present invention, the value of each indicator in the timing diagram at different time points is used as a node, and directed edges are established between the nodes in chronological order, with the weight being the inverse of the time interval.

[0019] Optionally, in a fifth implementation of the first aspect of the present invention, generating a catheter placement site allocation recommendation using a catheter placement matching model based on the SVCS risk probability value and the patient's clinical characteristics includes:

[0020] The patient's SVCS risk probability value and clinical characteristic data are integrated and converted into a state vector, which is then input into the catheter matching model.

[0021] The catheter matching model uses a multi-objective reinforcement learning framework, treating catheter site selection as a decision-making action, with catheter success rate, complication risk, operation time, and patient comfort as core optimization objectives;

[0022] Based on the state vector and preset weights, the comprehensive score of each catheterization site is calculated to obtain the catheterization site allocation recommendation.

[0023] Optionally, in the sixth implementation of the first aspect of the present invention, the clinical characteristic data includes at least age, gender, BMI, disease diagnosis, allergy history, vascular condition level and infusion drug characteristics, and the clinical characteristic data is converted into numerical indicators to obtain a state vector containing 20-dimensional features.

[0024] A second aspect of the present invention provides an artificial intelligence-based venous catheter matching system, the system comprising:

[0025] The data acquisition module is used to obtain the water content data of the patient's upper and lower limbs in real time through the body composition analysis equipment, and simultaneously collect the patient's blood test thrombosis index data;

[0026] The data processing module is used to clean the collected water content data and thrombosis index data, remove outliers and noise, and obtain the data to be analyzed after normalization;

[0027] The risk prediction module is used to build an SVCS risk prediction model, input the patient's data to be analyzed, and output the SVCS risk probability value through the SVCS risk prediction model;

[0028] The catheter matching module is used to generate catheter site allocation recommendations through the catheter matching model based on the SVCS risk probability value and patient clinical characteristics.

[0029] The third aspect of the present invention provides an artificial intelligence-based venous catheter matching device, which includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory to enable the artificial intelligence-based venous catheter matching device to perform each step of the artificial intelligence-based venous catheter matching method as described in any one of the above items.

[0030] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, implement the various steps of the artificial intelligence-based venous catheter matching method as described in any of the above items.

[0031] In the technical solution provided by the present invention, water content data of the patient's upper and lower limbs are obtained in real time through a human body composition analysis device, and blood drawing thrombosis index data of the patient are collected simultaneously; the collected water content data and thrombosis index data are cleaned to remove outliers and noise, and the data to be analyzed is obtained after normalization; an SVCS risk prediction model is constructed, the patient's data to be analyzed is input, and an SVCS risk probability value is output through the SVCS risk prediction model; based on the SVCS risk probability value and the patient's clinical characteristics, a catheterization site allocation recommendation is generated through a catheterization matching model; the present invention combines the patient's upper and lower limb water content data and thrombosis index to conduct a comprehensive and accurate assessment of the patient before catheterization, and judge in advance whether the patient is suitable for upper body catheterization, thereby avoiding the white cannulation phenomenon caused by problems found in subsequent examinations, reducing the patient's pain and the waste of medical resources, and improving the safety and effectiveness of catheterization. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0033] Figure 1 A flowchart of an artificial intelligence-based venous catheter matching method provided in an embodiment of the present invention;

[0034] Figure 2 A schematic diagram of the structure of an artificial intelligence-based venous catheter matching system provided in an embodiment of the present invention;

[0035] Figure 3 A schematic diagram of the structure of an artificial intelligence-based venous catheter matching device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar items and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that shown or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, product, or device.

[0037] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1A flowchart of an artificial intelligence-based venous catheter matching method provided in an embodiment of the present invention includes the following steps:

[0038] Step 101: obtaining water content data of the patient's upper and lower limbs in real time through a body composition analysis device, and simultaneously collecting blood draw data for thrombosis measurement;

[0039] In this embodiment, a bioelectrical impedance body composition analyzer is used to simultaneously measure the bioelectrical impedance values ​​of both upper and lower limbs through 8-point contact electrodes, and the water content of each part is calculated based on the whole body fluid distribution model;

[0040] Thrombosis indicators include D-dimer (D-Dimer), fibrinogen (FIB), prothrombin time (PT), activated partial thromboplastin time (APTT), antithrombin III (AT-III) and platelet count (PLT). Blood was collected from the elbow vein using a vacuum tube with a blood volume of 5 ml.

[0041] The collected data is synchronized with the electronic medical record in real time through the hospital information system to establish a patient-specific data file. The data storage uses encryption technology to ensure patient privacy and security. The storage server is deployed in the hospital's dedicated data center and has disaster recovery and backup functions to prevent data loss.

[0042] In this embodiment, the CVC tube is a slender catheter that is inserted into the central vein of the human body through skin puncture, such as the internal jugular vein, subclavian vein, femoral vein, etc. It is mainly used for clinical operations such as long-term intravenous infusion, nutritional support, hemodynamic monitoring or chemotherapy drug infusion. Its characteristic is that the tip of the catheter is located in the large blood vessels in the chest cavity, such as the superior vena cava or inferior vena cava, which can avoid the stimulation of high-concentration drugs on the peripheral veins. In the traditional catheterization process, the upper body is often preferred for clinical practice because the veins in the upper body are superficial and easy to operate. However, the route may be changed due to the subsequent discovery of superior vena cava syndrome.

[0043] In this embodiment, superior vena cava syndrome is a clinical syndrome caused by external compression or internal obstruction of the superior vena cava, which leads to obstruction of superior vena cava return and blood congestion in the head, neck, and upper limb veins. Typical manifestations include upper limb edema, jugular vein distension, facial swelling, etc., and the essence is water retention caused by venous return obstruction in the upper body. If this risk is not identified before catheterization and the upper body route is selected, the catheter may irritate or aggravate venous obstruction, leading to catheterization failure or serious complications such as thrombosis and infection. Therefore, early evaluation is needed to avoid upper body catheterization.

[0044] In this embodiment, the body composition analysis device is a medical device that assesses body composition using the principle of bioelectrical impedance. A weak current is applied to the body to measure differences in electrical impedance across tissues. This device is commonly used in clinical nutrition assessment and health management. The device used in this solution is similar to a gym's body measurement device, featuring multi-electrode contact measurement capabilities that can measure water content in both the upper and lower extremities. Its core advantage is its non-invasive, rapid detection capability, providing real-time information on water distribution across various body parts.

[0045] In this embodiment, the water content refers to the percentage of water content in a certain part of the human body to the body weight of that part, reflecting the local body fluid distribution status. Under normal physiological conditions, the water content of the lower limbs is usually higher than that of the upper limbs due to the rich muscle tissue and large vascular capacity of the lower limbs. When the superior vena cava is compressed or blocked, the venous return of the upper limbs is obstructed, and water retention causes the water content of the upper limbs to increase, and even exceed that of the lower limbs, such as the upper limb / lower limb water ratio >1.05.

[0046] In this embodiment, thrombosis indicators are biomarkers used to assess human coagulation function and thrombosis risk through blood tests. This solution mainly includes:

[0047] D-dimer: reflects fibrinolytic activity, and an increase indicates the dissolution process after thrombus formation;

[0048] Fibrinogen (FIB): Participates in the coagulation process, and elevated concentrations increase the risk of thrombosis;

[0049] Prothrombin time (PT) and activated partial thromboplastin time (APTT): evaluate the function of extrinsic and intrinsic coagulation pathways. Prolongation or shortening of PT indicates coagulation abnormality.

[0050] Antithrombin III (AT-III): The main anticoagulant in the body; when it decreases, the risk of thrombosis increases;

[0051] Platelet count (PLT): Thrombocytosis may promote thrombosis. Comprehensive analysis of the dynamic changes in these indicators (e.g., continuous increases or exceeding the normal range) can quantitatively assess the patient's risk of venous thrombosis after catheterization and avoid complications such as catheter blockage or thrombus dislodgement due to coagulation abnormalities.

[0052] Step 102: Clean the collected water content data and thrombosis index data to remove outliers and noise, and obtain the data to be analyzed after normalization;

[0053] In this embodiment, the Isolation Forest algorithm is used to detect outliers in the water content data and thrombosis index data, multiple random trees are constructed, and the path length of each data sample in the tree is calculated to evaluate the degree of abnormality; abnormal data points that deviate from the normal range are screened out based on a preset abnormal score threshold, and the detected abnormal data points are repaired using a generative adversarial network; the minimum-maximum normalization method is used to normalize the water content data and thrombosis index data to obtain the data to be analyzed.

[0054] In this embodiment, for water density data, if a certain collection data of a certain limb is missing, and the patient has valid collection data of the same limb within 1 hour before and after, the cubic spline interpolation method is used to fill the data, and a smooth curve is constructed using multiple valid data points at adjacent time points to improve the accuracy of the filled data; if the limb corresponding to the missing data has no valid data within 1 hour before and after, or the amount of missing data exceeds 20% of the total data volume of the indicator, it is marked as data collection failure, triggering the re-collection process, and the medical staff manually confirms and re-measures. For thrombosis index data, if a certain collection data of a certain indicator is missing, and the patient has valid data of the indicator within 48 hours before and after, the time series-based linear regression interpolation method is used to fill the data using the linear relationship between time and index value; if the missing data is the first collection or there is no adjacent valid data, blood sampling and testing must be repeated to avoid subsequent analysis deviations due to missing data.

[0055] In this embodiment, the Isolation Forest algorithm is used for outlier detection. The sample subset size is set to 256, the number of trees is set to 100, and the anomaly score of each sample is calculated. The anomaly score threshold is determined by cross-validation. In the training set, samples with scores higher than the average score plus 3 times the standard deviation are judged as outliers. For the detected outliers, a generative adversarial network is used to repair them. The generator network structure is: input layer → fully connected layer: the number of neurons is 128, the activation function is ReLU → fully connected layer: the number of neurons is 64, the activation function is ReLU → output layer: the dimension is the number of normal data features, the activation function is Tanh, and the output is mapped to the same range as the original data; the discriminator network structure is: input layer → fully connected layer: the number of neurons is 64, the activation function is LeakyReLU → fully connected layer: the number of neurons is 32, the activation function is LeakyReLU → output layer: the dimension is 1, the activation function is Sigmoid, which determines whether the input data is real data or generated data. During the training process, the Adam optimizer is used, the learning rate is 0.0002, the batch size is 32, and the number of training iterations is 5000 times, so that the generator can generate repaired data with a distribution close to the real data and replace outliers.

[0056] Step 103: construct an SVCS risk prediction model, input the patient's data to be analyzed, and output an SVCS risk probability value through the SVCS risk prediction model;

[0057] In this embodiment, the SVCS risk prediction model is constructed using an MC-GNN neural network, which includes two GCN layers and one fully connected layer. The node feature matrix and adjacency matrix of the topological graph are input in channel 1, and the spatial correlation features between indicators are extracted through the first layer of GCN. The node feature matrix and temporal adjacency matrix of the time series graph are input in channel 2, and the time series features of the indicators are extracted through the first layer of GCN. The output features of the two channels are concatenated for feature fusion, and finally the SVCS risk probability value is output through the fully connected layer.

[0058] In this embodiment, the topological graph nodes include upper limb water ratio, lower limb water ratio and thrombosis index, and the edges between the nodes represent the correlation between the indicators, where the correlation is determined by calculating the Pearson correlation coefficient. Undirected edges are established between indicators with an absolute value of the correlation coefficient greater than 0.3, and the weight is the correlation coefficient value.

[0059] In this embodiment, the values ​​of each indicator at different time points in the time series graph are represented as nodes. Directed edges are established between the nodes in chronological order, with weights equal to the inverse of the time intervals. In constructing the time series graph, the values ​​of each indicator at different time points are abstracted as nodes in the graph structure and connected by directed edges in chronological order, forming a chain structure that reflects the dynamic changes of the indicator. The weight of each edge is set to the inverse of the interval between adjacent time points, for example, a weight of 1 for a one-day interval and 0.5 for a two-day interval. The core logic behind this approach is to emphasize the clinical value of recent data through a weight decay mechanism. Values ​​closer to the current time have a greater impact on assessing immediate risk. For example, a recent sudden increase in a thrombosis indicator is more alarming than a mild abnormality from a week ago. This design enables the model to automatically capture both short-term fluctuations and long-term trends in the indicator. Short-term fluctuations, such as a rapid increase in upper limb water content within 24 hours, and long-term trends, such as three consecutive FIB concentrations exceeding the upper limit of normal, provide in-depth temporal features for SVCS risk prediction and thrombosis risk assessment, avoiding misjudgments caused by relying solely on data from a single time point.

[0060] In this embodiment, a topological graph is constructed: nodes include upper limb water ratio, lower limb water ratio and thrombosis index, among which upper limb water ratio includes left upper limb water ratio and right upper limb water ratio, lower limb water ratio includes left lower limb water ratio and right lower limb water ratio, and thrombosis index includes D-Dimer, FIB, PT, APTT, AT-Ⅲ and PLT, totaling 8 nodes. The edges between nodes represent the correlation between the indicators, which is determined by calculating the Pearson correlation coefficient. Undirected edges are established between indicators with an absolute value of the correlation coefficient greater than 0.3, and the weight is the correlation coefficient value. For example, if there is a significant correlation between the upper limb water ratio and the D-Dimer in the thrombosis index, and the correlation coefficient is >0.3, an edge connection is established between the corresponding nodes to reflect the mutual influence between the indicators.

[0061] In this example, data from 1,000 patients who underwent intravenous catheterization were collected, including 200 patients clinically diagnosed with superior vena cava syndrome (SVCS) and 800 patients who were undiagnosed. The data were divided into training, validation, and test sets to ensure that the proportion of SVCS patients in each set was close to the true distribution. A cross-entropy loss function was used, the AdamW optimizer was used, the learning rate was initially 0.001, decayed by 0.1 every 50 epochs, the batch size was 32, and the total number of training epochs was 200. During training, model performance was evaluated on the validation set every 10 epochs. If the validation set loss did not decrease for three consecutive times, training was terminated early to prevent overfitting. When the probability value output by the SVCS risk prediction model was ≥0.6, the patient was marked as having the possibility of superior vena cava syndrome. Clinical data validation showed that the sensitivity and specificity of the model at this threshold were both ≥85%. For the risk of thrombosis, the normalized data of various thrombosis indicators and clinical guidelines, such as the Wells scoring standard, were integrated into the model, and the thrombosis indicator data were input into the model to output the thrombosis risk probability value, which was divided into three levels: low risk, medium risk, and high risk, providing a basis for catheter matching selection.

[0062] In this embodiment, cleaned and normalized patient data and their time series data are input into the model. The patient data includes the water content of each part of the upper and lower limbs and the value of the thrombosis index. The time series data includes the historical values ​​of the same index at different time points. This generates two types of analytical data: one is spatial data reflecting the current correlation between the various indicators, such as the mutual influence between the water content of the upper and lower limbs and the thrombosis index; the other is time series data reflecting the dynamic changes of the indicators, such as the fluctuation trend of a certain thrombosis index in multiple recent tests.

[0063] Channel 1: Upper / lower limb water ratios and thrombosis indicators serve as nodes in a graph structure. By calculating correlations between indicators, such as whether elevated thrombosis indicators are often associated with abnormal upper limb water ratios, edges are established between closely related nodes to construct a topological graph reflecting the spatial relationships between indicators. The model uses graph convolution operations to extract the interaction characteristics between different indicators. For example, when the upper limb water ratio is abnormally elevated, which thrombosis indicator changes will increase the risk of SVCS?

[0064] Channel 2: The values ​​of each indicator at different time points are used as nodes and connected in chronological order to form a time series graph. The edge weights between nodes are adjusted according to the time interval. The model uses the time series graph convolution operation to capture the change pattern of the indicator over time. For example, whether a continuous decrease in the water content of the lower limbs indicates worsening venous return obstruction;

[0065] The spatial correlation features and temporal trend features extracted from the two channels are concatenated and fused, and then input into a fully connected layer for comprehensive analysis. Finally, an activation function is used to generate an SVCS risk probability value between 0 and 1, which intuitively reflects the patient's likelihood of developing superior vena cava syndrome. When the probability value is ≥0.6, the system marks the patient at risk of SVCS and comprehensively assesses the risk level of thrombosis in combination with thrombosis indicators. At this time, the model will prioritize lower-body catheterization, avoiding the repeated operations caused by directly selecting upper-body catheterization in traditional methods.

[0066] Step 104: Based on the SVCS risk probability value and the patient's clinical characteristics, a catheter placement site allocation recommendation is generated through a catheter placement matching model.

[0067] In this embodiment, the patient's SVCS risk probability value and clinical characteristic data are integrated and converted into a state vector, which is input into the catheter matching model. The catheter matching model adopts a multi-objective reinforcement learning framework, regards the selection of the catheter site as a decision-making action, and takes the catheter success rate, complication risk, operation time, and patient comfort as the core optimization goals. Based on the state vector and preset weights, the comprehensive score of each catheter site is calculated to obtain a catheter site allocation recommendation.

[0068] In this embodiment, the clinical characteristic data includes at least age, gender, BMI, disease diagnosis, allergy history, vascular condition level and infusion drug characteristics. The clinical characteristic data is converted into numerical indicators to obtain a state vector containing 20-dimensional features.

[0069] In this embodiment, the MORL-PPO architecture uses catheter matching selection as the decision action. The state space includes multi-dimensional data such as SVCS risk probability value, thrombosis risk probability value, and patient clinical characteristics. The action space is the optional catheter matching, including upper limb superficial vein, upper limb deep vein, lower limb vein, central vein, etc., a total of 8 types. The reward function is designed as a multi-objective function, including catheterization success rate, complication rate, catheterization time, and patient comfort. The catheterization success rate has a weight of 0.4, the complication rate has a weight of 0.3, with the goal of reducing complications, the catheterization time has a weight of 0.2, with the goal of shortening the operation time, and the patient comfort has a weight of 0.1, which is evaluated through questionnaire scores or physiological indicators.

[0070] Both the policy network and the value network use multi-layer perceptrons, with a 20-dimensional state vector as the input layer, two hidden layers, 128 and 64 neurons respectively, and ReLU activation function. The output layer of the policy network is an 8-dimensional action probability distribution, and the output layer of the value network is a 1-dimensional value estimate.

[0071] Leveraging historical hospital catheterization data, each record includes patient characteristics, catheter matching, catheterization results, and complication occurrences. This allows for offline training, combined with online interaction, to test new strategies in a simulated environment.

[0072] The training parameters are a batch size of 64, a discount factor of γ = 0.99, an entropy regularization coefficient of α = 0.01, the number of updates K = 10, and a learning rate of 0.0003. Through continuous optimization, the model can generate the optimal cannulation match based on the input patient data. In the output of the model, the action probability distribution with the highest probability is selected as the cannulation match.

[0073] In this embodiment, the batch size controls the amount of data for a single training session, the discount factor balances immediate rewards with long-term benefits, the entropy regularization coefficient enhances the diversity of action selection to avoid overfitting, the number of updates is 10 times for parameter optimization per iteration, and the learning rate adjusts the step size of the parameter update. These parameters work together to optimize the model through a reinforcement learning strategy, so that it can dynamically generate the optimal catheterization plan based on patient data and ultimately output the matching result with the highest probability.

[0074] In this embodiment, a tiered reward mechanism is designed based on core clinical needs, including:

[0075] Risk-averse reward: If the patient's SCS risk is ≥ 0.6 and the model chooses a lower body cannulation site, the reward value is +R1; if the thrombosis risk is high and the model chooses a more hemodynamically stable cannulation site, the reward value is +R2;

[0076] Efficiency Optimization Reward: Directly selecting the correct location to avoid secondary catheterization will earn you a reward of +R3; no abnormal alarm is triggered within 3 days after catheterization will earn you a reward of +R4;

[0077] Patient safety reward: Considering the catheter placement site and the patient's mobility, the reward value is +R5;

[0078] Among them, R1: risk aversion reward (lower body catheterization), the trigger condition is when the patient's SCS risk value is ≥0.6 and the model selects lower body catheterization, the lower body catheterization option is preferred in high-risk patients to reduce the risk of complications;

[0079] R2: Risk-averse reward (thrombosis risk), triggered when the thrombosis risk is high and the model selects a more hemodynamically stable catheterization site, guiding the selection of a safer catheterization site in patients with high thrombosis risk to optimize blood flow stability;

[0080] R3: Efficiency optimization reward (direct selection of the correct site), triggered by direct selection of the correct cannulation site, avoiding secondary cannulation, and encouraging operators to accurately select the appropriate site in one go, reducing repeated operations and improving efficiency;

[0081] R4: Efficiency Optimization Reward (no abnormal alarm), triggered by no abnormal alarm within 3 days after catheter placement, emphasizes the stability of operation quality, reduces postoperative complications or monitors abnormal events;

[0082] R5: Patient Safety Reward (comprehensive consideration of mobility), triggered by comprehensive consideration of the catheterization site and the patient's mobility, selecting the safer option, and encouraging consideration of the patient's physical condition during the procedure to improve comfort and safety;

[0083] R1-R5 guides clinical decision-making through quantitative indicators, providing clear behavioral incentives in the three dimensions of risk control, efficiency improvement, and patient safety.

[0084] In this embodiment, historical cases are converted into state-action-reward triplets through the experience replay mechanism to construct a training data set, and an ε-greedy strategy is adopted to balance exploration and utilization: actions are randomly selected with a probability of ε, and the current optimal action is selected with a probability of 1-ε. Priority experience replay is introduced to give higher weights to cases containing serious complications, accelerating the model's learning of high-risk scenarios. The model ultimately outputs the probability distribution of catheterization sites, such as P(upper body) = 0.35, P(lower body) = 0.65, and combines clinical safety thresholds to generate deterministic recommendations, and simultaneously outputs a decision-making basis report containing the weights of various risk factors; the model automatically synchronizes the hospital's latest catheterization complication data every 24 hours to trigger incremental training.

[0085] In this embodiment, the water content of the upper and lower limbs is judged. Generally, the water content of the upper body of a normal person is lower than that of the lower body. Therefore, if the water content of the upper body is higher than that of the lower body, it means that there is a possibility of superior vena cava syndrome. At this time, the upper body is not allocated, but the lower body is allocated. This can prevent the patient from having an extra catheter placed in vain. The indicators of blood drawing and thrombosis detection are introduced together for comprehensive evaluation. The comprehensive evaluation is made in combination with the patient's whole body data and thrombosis indicators, thereby avoiding unnecessary catheterization operations and improving the accuracy and safety of catheterization.

[0086] See also Figure 2 , a schematic diagram of the structure of an artificial intelligence-based venous catheter matching system provided in an embodiment of the present invention, the system includes:

[0087] The data acquisition module is used to obtain the water content data of the patient's upper and lower limbs in real time through the body composition analysis equipment, and simultaneously collect the patient's blood test thrombosis index data;

[0088] The data processing module is used to clean the collected water content data and thrombosis index data, remove outliers and noise, and obtain the data to be analyzed after normalization;

[0089] The risk prediction module is used to build an SVCS risk prediction model, input the patient's data to be analyzed, and output the SVCS risk probability value through the SVCS risk prediction model;

[0090] The catheter matching module is used to generate catheter site allocation recommendations based on the SVCS risk probability value and patient clinical characteristics through the catheter matching model

[0091] Figure 3 This is a schematic diagram of the structure of an artificial intelligence-based intravenous catheter matching device provided by an embodiment of the present invention. This artificial intelligence-based intravenous catheter matching device 300 may vary significantly depending on its configuration or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors), memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage medium 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instructions for operating on the artificial intelligence-based intravenous catheter matching device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, allowing the artificial intelligence-based intravenous catheter matching device 300 to execute the series of instructions stored in the storage medium 330 to implement the method provided in the above embodiment.

[0092] The artificial intelligence-based intravenous catheter matching device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating devices 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. The arrow at the wired or wireless network interface 350 indicates receiving data through the wireless or wired network interface, and the arrow at the input / output interface 360 ​​indicates reading or writing data through the input / output interface. It will be understood by those skilled in the art that Figure 3 The structure of the artificial intelligence-based intravenous catheter matching device shown does not constitute a limitation on the computer device provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0093] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the various steps of the artificial intelligence-based venous catheter matching method provided in the above embodiments.

[0094] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices, and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0095] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0096] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A venous catheter matching method based on artificial intelligence, characterized in that: The method comprises the following steps: The body composition analysis device is used to obtain the water content data of the patient's upper and lower limbs in real time, and the patient's blood test thrombosis index data is collected at the same time; The collected water content data and thrombosis index data were cleaned to remove outliers and noise, and the data to be analyzed were obtained after normalization; Construct an SVCS risk prediction model, input the patient's data to be analyzed, and output the SVCS risk probability value through the SVCS risk prediction model; Based on the SVCS risk probability value and the patient's clinical characteristics, a catheter placement matching model is used to generate a catheter placement recommendation. The SVCS risk prediction model is constructed using the MC-GNN neural network, which includes two GCN layers and one fully connected layer; The input in channel 1 is the node feature matrix and adjacency matrix of the topological graph, and the spatial correlation features between indicators are extracted through the first layer of GCN; The input in channel 2 is the node feature matrix and time series adjacency matrix of the time series graph, and the time series features of the indicators are extracted through the first layer of GCN; The output features of the two channels are spliced ​​for feature fusion, and finally the SVCS risk probability value is output through the fully connected layer; The topological graph nodes include upper limb water ratio, lower limb water ratio and thrombosis index, and the edges between the nodes represent the correlation between the indexes, wherein the correlation is determined by calculating the Pearson correlation coefficient, and undirected edges are established between the indexes with an absolute value of the correlation coefficient greater than 0.3, and the weight is the correlation coefficient value; The values ​​of each indicator at different time points in the time series graph are used as nodes, and directed edges are established between the nodes in chronological order, with the weight being the inverse of the time interval; The patient's SVCS risk probability value and clinical characteristic data are integrated and converted into a state vector, which is then input into the catheter matching model. The catheter matching model uses a multi-objective reinforcement learning framework, treating catheter site selection as a decision-making action, with catheter success rate, complication risk, operation time, and patient comfort as core optimization objectives; Based on the state vector and preset weights, the comprehensive score of each catheterization site is calculated to obtain the catheterization site allocation recommendation.

2. The artificial intelligence-based venous catheter matching method according to claim 1, characterized in that: The collected water density data and thrombosis index data are cleaned to remove outliers and noise, and the data to be analyzed is obtained after normalization, including: The Isolation Forest algorithm was used to detect outliers in the water density data and thrombosis index data. Multiple random trees were constructed and the path length of each data sample in the tree was calculated to evaluate the degree of anomaly. Based on the preset anomaly score threshold, abnormal data points that deviate from the normal range are screened out. For the detected abnormal data points, a generative adversarial network is used to repair them. The minimum-maximum normalization method was used to normalize the water density data and thrombosis index data to obtain the data to be analyzed.

3. The artificial intelligence-based venous catheter matching method according to claim 1, characterized in that: The clinical characteristic data at least include age, gender, BMI, disease diagnosis, allergy history, vascular condition level and infusion drug characteristics. The clinical characteristic data are converted into numerical indicators to obtain a state vector containing 20-dimensional features.

4. An artificial intelligence-based intravenous catheter matching system, characterized in that: The system includes: The data acquisition module is used to obtain the water content data of the patient's upper and lower limbs in real time through the body composition analysis equipment, and simultaneously collect the patient's blood test thrombosis index data; The data processing module is used to clean the collected water content data and thrombosis index data, remove outliers and noise, and obtain the data to be analyzed after normalization; The risk prediction module is used to construct an SVCS risk prediction model, input the patient's data to be analyzed, and output the SVCS risk probability value through the SVCS risk prediction model: the SVCS risk prediction model is constructed using an MC-GNN neural network, which includes two GCN layers and one fully connected layer; the node feature matrix and adjacency matrix of the topological graph are input in channel 1, and the spatial correlation features between indicators are extracted through the first layer of GCN; the node feature matrix and time series adjacency matrix of the time series graph are input in channel 2, and the time series features of the indicators are extracted through the first layer of GCN; the output features of the two channels are spliced ​​for feature fusion, and finally the SVCS risk probability value is output through the fully connected layer; the nodes of the topological graph include upper limb water ratio, lower limb water ratio and thrombosis index, and the edges between the nodes represent the correlation between the indicators, wherein the correlation is determined by calculating the Pearson correlation coefficient, and undirected edges are established between indicators with an absolute value of the correlation coefficient greater than 0.3, and the weight is the correlation coefficient value; the value of each indicator at different time points in the time series graph is used as a node, and directed edges are established between the nodes in chronological order, and the weight is the inverse of the time interval; The catheter matching module is used to generate catheter site allocation recommendations based on the SVCS risk probability value and the patient's clinical characteristics through the catheter matching model: the patient's SVCS risk probability value and clinical characteristic data are integrated and converted into a state vector, which is input into the catheter matching model; the catheter matching model adopts a multi-objective reinforcement learning framework, regards the selection of catheter site as a decision-making action, and takes the catheter success rate, complication risk, operation time, and patient comfort as the core optimization goals; based on the state vector and preset weights, the comprehensive score of each catheter site is calculated to obtain the catheter site allocation recommendation.

5. An artificial intelligence-based intravenous catheter matching device, characterized in that: The artificial intelligence-based venous catheter matching device includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the artificial intelligence-based venous catheter matching device performs each step of the artificial intelligence-based venous catheter matching method as described in any one of claims 1-3.

6. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the artificial intelligence-based venous catheter matching method as described in any one of claims 1 to 3 are implemented.

Images