Intelligent PICC catheter and positioning system for tumor chemotherapy

Abstract

This invention discloses an intelligent PICC catheter and positioning system for tumor chemotherapy, belonging to the field of biomedical engineering technology. This invention addresses the problem of inaccurate puncture due to the inherent subjectivity in puncture point selection in existing technologies. By assessing the importance of vascular imaging features and blood flow parameters, it can identify the features most influential on predicting puncture results, thereby reducing model complexity and improving model efficiency and generalization ability. Furthermore, by weighting the input data according to feature importance, the random forest model focuses more on important features, thus improving the accuracy and reliability of model predictions. Through the collaborative work of multiple modules, the positioning system is more accurate and efficient in recommending the optimal puncture point, effectively improving the puncture success rate, reducing the risk of complications, providing strong decision support for medical staff, and enhancing the patient's treatment experience and safety.

Description

Technical Field

[0001] This invention relates to the field of biomedical engineering technology, specifically to a smart PICC catheter and positioning system for tumor chemotherapy. Background Technology

[0002] In chemotherapy for cancer, peripherally inserted central venous catheters (PICCs) are a common infusion method. PICCs provide patients with long-term, stable venous access, reducing the pain of repeated punctures and lowering the risk of infection. However, traditional PICC placement techniques have some limitations, mainly in the following aspects:

[0003] Traditional methods rely mainly on the experience and manual operation of medical staff to select puncture sites, lacking precise imaging and blood flow parameter support, which leads to a certain degree of subjectivity in the selection of puncture sites;

[0004] Inaccurate selection of the puncture site may affect the success rate of the puncture, especially for patients with poor vascular conditions. In addition, complications such as hematoma and nerve damage may occur during the puncture process, increasing the patient's pain and treatment risks.

[0005] Therefore, to address the shortcomings of existing solutions, we propose a smart PICC catheter and positioning system for tumor chemotherapy. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent PICC catheter and positioning system for tumor chemotherapy. By assessing the importance of vascular imaging features and blood flow parameters, it can screen out the features most influential on predicting puncture results, thereby reducing model complexity and improving model efficiency and generalization ability. Furthermore, by weighting the input data according to feature importance, the random forest model focuses more on important features, thus improving the accuracy and reliability of model predictions. Through the collaborative work of multiple modules, the positioning system is more accurate and efficient in recommending the optimal puncture point, effectively improving the puncture success rate, reducing the risk of complications, providing strong decision support for medical staff, and enhancing the patient's treatment experience and safety, thus solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A smart PICC catheter positioning system for tumor chemotherapy, implemented via a smart PICC catheter for tumor chemotherapy, is characterized in that the smart PICC catheter positioning system for tumor chemotherapy includes:

[0009] The feature extraction module is configured to collect historical data and extract vascular image features and blood flow parameter features from the historical data. The vascular image features include: extracting the shape, inner diameter and depth of the blood vessels, extracting the centerline and contour of the blood vessels, and calculating the length and curvature of the blood vessels; the blood flow parameter features include: extracting the average, maximum and minimum values ​​of blood flow velocity, and calculating the rate of change and stability index of blood flow velocity.

[0010] The intelligent analysis module is configured to train and test historical data and its vascular imaging features and blood flow parameter features using a random forest model, and output the coordinates of the best puncture point and the reasons for the recommendation. The reasons for the recommendation include the vascular diameter, blood flow velocity and vascular depth as key factors.

[0011] The importance assessment module is configured to, after the random forest model has been trained, output the average impurity reduction value of each feature through the random forest model, normalize the average impurity reduction value of each feature, and obtain the importance result of each feature.

[0012] The data weighting module is configured to perform a weighted average of the input data for each feature based on its importance, resulting in weighted input data. The weights corresponding to each feature are also compensated for in the following ways:

[0013] Retrieve the average reduction in impurity for each feature; obtain the impurity reduction ratio for each feature based on the average reduction in impurity for each feature.

[0014] Retrieve the original weights corresponding to each feature; wherein, the proportion of the original weights corresponding to each feature is consistent with the importance result corresponding to each feature; based on the original weights of each feature corresponding to each importance result, obtain the standard deviation of the weight values ​​of the original weights corresponding to each feature;

[0015] The standard deviation of the original weight value corresponding to each feature is compared with a preset standard deviation threshold; when the standard deviation of the original weight value corresponding to the feature exceeds the preset standard deviation threshold, the original weight corresponding to the feature is compensated and adjusted, and the original weight is replaced with the compensated and adjusted weight value; the compensation and adjustment of the original weight corresponding to the feature includes:

[0016] Retrieve the medical knowledge graph corresponding to tumor chemotherapy; retrieve the clinical relevance coefficient between feature i and feature j from the medical knowledge graph, wherein the value range of the clinical relevance coefficient is: ;

[0017] The standard deviation of the weight value corresponding to each feature is compared with a preset standard deviation threshold to obtain the standard deviation ratio parameter. ;in, This represents the standard deviation ratio parameter corresponding to the i-th feature; This represents the preset standard deviation threshold; This represents the standard deviation of the weight values ​​corresponding to the i-th feature;

[0018] The original weights of the clinical relevance coefficient pairs between features i and j are adjusted by using the standard deviation ratio parameter corresponding to each feature.

[0019] A smart PICC catheter for tumor chemotherapy includes: a catheter body and an external component, characterized in that one end of the catheter body is connected to the external component to form the overall structure of the smart PICC catheter, a micro magnetic head and a high-frequency linear probe are superimposed and embedded on the tip of the catheter body, and a micro electrode is built into the catheter body.

[0020] The miniature magnetic head is made of high-performance magnetic material and is configured to generate a magnetic field signal;

[0021] The microelectrode is connected to the tip of the catheter body and is configured to collect electrical signals generated when the tip of the catheter body contacts the heart chamber in real time.

[0022] The high-frequency linear probe is configured to acquire real-time images of blood vessel cross-sections and longitudinal sections, and automatically identify blood vessel depth, inner diameter, and blood flow velocity using Doppler mode. Based on the identification results, it predicts and recommends the optimal puncture point.

[0023] Furthermore, the micro magnetic head is connected to the signal transmission interface at the end of the catheter via internal wires, and the magnetic field signal is transmitted to the external sensor. The external sensor consists of multiple highly sensitive magnetic field sensors distributed at specific locations on the patient's body surface. It is configured to sense the magnetic field signal generated by the micro magnetic head in real time, convert the magnetic field signal into an electrical signal, and transmit the electrical signal to the host for processing and analysis wirelessly.

[0024] The microelectrode is connected to an external electrocardiogram (ECG) monitoring device via internal wires. The ECG monitoring device consists of an ECG amplifier, a filter, and a signal processor. It is configured to process the electrical signal generated when the tip of the catheter body contacts the heart chamber in real time. After amplification, filtering, and processing, the signal is converted into an ECG signal and transmitted wirelessly to the host for analysis and judgment.

[0025] Furthermore, the system also includes: a data annotation module and a feature selection module;

[0026] The data annotation module is configured to annotate puncture results and complications in historical data. Puncture result annotation includes: marking the puncture result as successful or unsuccessful; for successful punctures, recording the coordinates of the puncture point; for unsuccessful punctures, recording the reason for the failure; complication annotation includes: recording the occurrence of complications.

[0027] The feature selection module is configured to select highly important features from vascular imaging features and blood flow parameter features based on their importance, and use these features as input features for the random forest model. The labeled puncture results and puncture point coordinates are selected as target variables for the random forest model.

[0028] Furthermore, the system also includes a visual interface and a data processing module:

[0029] The visual interface is configured to display vascular images and recommended puncture points in real time on the monitoring screen, providing visual guidance for the puncture path of the PICC catheter.

[0030] The data processing module is configured to delete duplicate records in historical data, fill in missing blood flow parameters or imaging data, process abnormal data using the Z-score statistical method, and divide the processed historical data into training and testing sets.

[0031] Furthermore, the intelligent analysis module includes:

[0032] The model training module is configured to set the parameters of the random forest model, train the random forest model using training set data, evaluate the performance of the random forest model using test set data, calculate the accuracy, recall, and F-score of the random forest model on the test set, analyze the confusion matrix of the random forest model, and evaluate the predictive ability of the random forest model for different data.

[0033] Furthermore, the system also includes: a practical application module and a user feedback module;

[0034] The practical application module is configured to apply the trained random forest model to actual chemotherapy. By acquiring vascular images and blood flow parameters in real time, the random forest model outputs the optimal puncture point coordinates and the reasons for the recommendation, and displays the recommended puncture point and reasons on a visual interface, providing visual guidance for PICC catheter positioning.

[0035] The user feedback module is configured to adjust the parameters of the random forest model and optimize its performance based on feedback from medical staff and real-time data.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] In this invention, the importance of vascular imaging features and blood flow parameter features is assessed to identify the features most influential on predicting puncture results, thereby reducing model complexity and improving model efficiency and generalization ability. Furthermore, the input data is weighted and averaged according to feature importance, making the random forest model focus more on important features, thus improving the accuracy and reliability of model predictions. Through the collaborative work of multiple modules, the positioning system is more accurate and efficient in recommending the optimal puncture point, effectively improving the puncture success rate, reducing the risk of complications, providing strong decision support for medical staff, and enhancing the patient's treatment experience and safety. Attached Figure Description

[0038] Figure 1 This is a structural diagram of the intelligent PICC catheter for tumor chemotherapy of the present invention;

[0039] Figure 2 This is a magnified view of the intelligent PICC catheter for tumor chemotherapy of the present invention.

[0040] Figure 3 This is a flowchart of the intelligent PICC catheter positioning process for tumor chemotherapy according to the present invention.

[0041] In the figure: 1. Catheter body; 2. External components; 3. Miniature magnetic head; 4. High-frequency linear probe; 5. Miniature electrode. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] To address the shortcomings of existing techniques, which rely heavily on the experience and manual operation of medical personnel to select puncture sites, lacking precise imaging and blood flow parameter support, and thus introducing subjectivity into puncture site selection; the inaccuracy of puncture site selection may affect the success rate of puncture, especially in patients with poor vascular conditions; furthermore, complications such as hematoma and nerve damage may occur during the puncture process, increasing patient suffering and treatment risks, please refer to [link to relevant documentation]. Figures 1-3 This embodiment provides the following technical solution:

[0044] The intelligent PICC catheter for tumor chemotherapy includes: a catheter body 1 and an external component 2. One end of the catheter body 1 is connected to the external component 2 to form the overall structure of the intelligent PICC catheter. A micro magnetic head 3 and a high-frequency linear probe 4 are superimposed and embedded on the tip of the catheter body 1. The catheter body 1 also has a built-in microelectrode 5. The micro magnetic head 3 is made of high-performance magnetic material and is configured to generate a stable magnetic field signal. Its size is only at the millimeter level, which does not affect the overall flexibility and permeability of the catheter. The micro magnetic head 3 is connected to the signal transmission interface at the end of the catheter through internal wires and transmits the magnetic field signal to an external sensor. The external sensor consists of multiple high-sensitivity magnetic field sensors distributed at specific locations on the patient's body surface, such as the chest and shoulder, forming a three-dimensional magnetic field sensing network. It is configured to sense the magnetic field signal generated by the micro magnetic head 3 in real time, convert the magnetic field signal into an electrical signal, and transmit the electrical signal to the host for processing and analysis wirelessly. The microelectrode 5 is connected to the tip of the catheter body 1 and is configured to acquire the electrical signal generated when the tip of the catheter body 1 contacts the heart chamber in real time. The microelectrode 5 is connected to an external electrocardiogram (ECG) monitoring device via internal wires. The ECG monitoring device consists of an ECG amplifier, a filter, and a signal processor, and is configured to process the electrical signal generated when the tip of the catheter body 1 contacts the heart chamber in real time. After amplification, filtering, and processing, a clear ECG signal is generated and transmitted wirelessly to the host for analysis and judgment. The high-frequency linear probe 4 is configured to acquire cross-sectional and longitudinal images of blood vessels in real time and automatically identify the vessel depth, inner diameter, and blood flow velocity using Doppler mode. Based on a random forest model, the identification results are predicted to recommend the optimal puncture point; for example, selecting a vein with an inner diameter ≥3mm and a blood flow velocity of 20-40cm / s.

[0045] In practice, firstly, check whether the main body 1 of the intelligent PICC catheter and its external components 2 are intact and ensure that all connections are secure; secondly, conduct a preliminary assessment of the patient, including vascular conditions and skin condition at the puncture site, to ensure that the patient is suitable for PICC placement; thirdly, place the high-frequency linear probe 4 at the patient's puncture site, such as the elbow crease or upper arm, ensuring the probe position is stable; turn on the ultrasound equipment to acquire cross-sectional and longitudinal images of the blood vessels in real time; and automatically identify the depth, inner diameter, and blood flow velocity of the blood vessels through Doppler mode, and select those meeting the criteria based on the identification results. For blood vessels, such as veins with an inner diameter ≥3mm and a blood flow velocity of 20-40cm / s, the random forest model predicts and recommends the coordinates of the optimal puncture point based on vascular characteristics, such as inner diameter, blood flow velocity, and vascular depth. The recommended puncture point is displayed on the visualization interface, and a visual guide for the puncture path is provided. Based on the puncture point recommended by the system, the puncture point is marked on the patient's skin using a marker pen. The location is disinfected and local anesthetized. The puncture needle is used to puncture at the marked puncture point. After confirming successful puncture, the tip of the intelligent PICC catheter body 1 is inserted into the puncture needle and slowly advanced into the catheter body 1 along the puncture needle to ensure that the catheter body 1 smoothly enters the blood vessel.

[0046] A smart PICC catheter positioning system for tumor chemotherapy includes:

[0047] The feature extraction module is configured to collect historical data and extract vascular image features and blood flow parameter features from the historical data. Vascular image features include: extracting the shape, inner diameter, and depth of the blood vessels; extracting the centerline and contour of the blood vessels using an edge detection algorithm; and calculating the length and curvature of the blood vessels using a Gaussian curvature calculation method; directly measuring the inner diameter of the blood vessels from cross-sectional images; and extracting the depth information of the blood vessels from longitudinal section images using image processing algorithms. Blood flow parameter features include: extracting the average, maximum, and minimum values ​​of blood flow velocity over a period of time, and calculating the rate of change and stability indices of blood flow velocity, including: calculating the rate of change of blood flow velocity using a sliding window method and calculating the time derivative using a difference method; using the standard deviation to evaluate the stability of blood flow velocity; and analyzing the frequency characteristics of blood flow velocity using Fourier transform.

[0048] The intelligent analysis module is configured to train and test historical data and its vascular imaging features and blood flow parameters using a random forest model, and output the coordinates of the optimal puncture point and the reasons for the recommendation. The reasons for the recommendation include vascular diameter, blood flow velocity, and vascular depth as key factors. The intelligent analysis module includes:

[0049] The model training module is configured to set parameters for the random forest model, such as the number of decision trees, maximum depth, and minimum sample split. It trains the random forest model using training set data and evaluates its performance using test set data, ensuring the model's generalization ability and stability. It calculates the random forest model's accuracy, recall, and F1 score on the test set, analyzes the confusion matrix, and assesses the model's predictive ability across different datasets. For example: Vessel diameter: Vessels with an inner diameter ≥3mm are easier to puncture and have more stable blood flow. Blood flow velocity: Vessels with blood flow velocities between 20-40cm / s have better drug infusion effects and reduce the risk of backflow. Vessel depth: Vessels at appropriate depths are easier to control during puncture, reducing the risk of complications.

[0050] The importance assessment module is configured to, after the random forest model has been trained, calculate the reduction in impurity for each feature at the split node in each decision tree, average the reduction in impurity across all decision trees to obtain the average reduction in impurity for each feature, and normalize the average reduction in impurity for each feature to obtain the importance result for each feature. For example, assuming the importance of feature A is 0.5, feature B is 0.3, and feature C is 0.2, then when inputting data, the weight of feature A is 0.5, feature B is 0.3, and feature C is 0.2.

[0051] The data weighting module is configured to perform a weighted average of the input data for each feature based on its importance, resulting in weighted input data. For example, if feature A has a weight of 0.5, feature B has a weight of 0.3, and feature C has a weight of 0.2, then the weighted input data = (input data of feature A × 0.5) + (input data of feature B × 0.3) + (input data of feature C × 0.2). Further, assume the following features and their importance evaluation results:

[0052] Inner diameter of blood vessel: importance 0.5; Blood flow velocity: importance 0.3; Depth of blood vessel: importance 0.2.

[0053] Original input data: Inner diameter of blood vessel: 5mm, blood flow velocity: 30cm / s, depth of blood vessel: 10mm;

[0054] Weighted input data: Weighted vessel diameter = 5mm × 0.5 = 2.5; Weighted blood flow velocity = 30cm / s × 0.3 = 9cm / s; Weighted vessel depth = 10mm × 0.2 = 2mm;

[0055] The weighted input data vector is: [2.5, 9, 2].

[0056] On the other hand, the weights corresponding to each feature are also compensated in the following ways:

[0057] Retrieve the average impurity reduction value corresponding to each feature;

[0058] Obtain the impurity reduction ratio for each feature based on the average impurity reduction value for each feature;

[0059] Retrieve the original weights corresponding to each feature; wherein, the original weight percentage of each feature is consistent with the importance result of each feature;

[0060] Based on the original weights of each feature corresponding to each importance result, obtain the standard deviation of the weight values ​​of the original weights for each feature;

[0061] The standard deviation of the original weight value corresponding to each feature is compared with a preset standard deviation threshold;

[0062] When the standard deviation of the original weight value corresponding to the feature exceeds the preset standard deviation threshold, the original weight corresponding to the feature is compensated and adjusted, and the original weight is replaced by the compensated and adjusted weight value.

[0063] The technical effects of the above solution are as follows: By calculating the standard deviation of the original weights, the fluctuation of feature weights during multiple training or iterations is monitored. When the fluctuation exceeds a threshold (e.g., a feature's weight fluctuates due to data noise), compensation adjustments are made to bring the weights back to a reasonable range, avoiding model performance fluctuations caused by weight instability. In the PICC catheter positioning system for tumor chemotherapy, ensuring the stability of key feature weights such as "catheter displacement rate" and "vascular access pressure" makes the model output (e.g., catheter position prediction) more reliable. Stable weight allocation makes the model more resistant to data fluctuations (e.g., patient movement, measurement errors). Even if the input data has small noise or outliers, the model prediction results (e.g., catheter positioning accuracy) can still remain relatively accurate due to stable feature weights, improving the system's applicability in complex clinical environments (patient activity, equipment errors, etc.). The true contribution of features to the model is determined based on the average impurity reduction value and impurity reduction ratio. When the standard deviation of the weights exceeds a threshold, it indicates that the weights deviate from the actual importance of the features. Compensation adjustments allow the weights to re-align with the feature importance (e.g., a feature with high importance but a weight suppressed due to fluctuations; compensation increases this weight), ensuring that key features (such as the impact of catheter tip position deviation on positioning) play their due role in the model and optimizing the model's learning effect on PICC catheter positioning and the correlation of tumor chemotherapy parameters. This avoids a single feature weight excessively dominating the model or the unreasonable suppression of secondary feature weights. Through compensation adjustments, the weight ratio of each feature is matched with its actual importance, allowing the model to integrate multi-feature information (such as combining patient physiological parameters and catheter physical parameters), improving the accuracy of the correlation analysis between PICC catheter positioning and chemotherapy efficacy, and assisting clinical decision-making (such as adjusting catheter position and optimizing chemotherapy regimens). During tumor chemotherapy, the patient's physiological state (such as weight changes and changes in vascular elasticity) and equipment status (such as catheter wear and sensor drift) dynamically change, potentially altering feature importance. This compensation mechanism continuously monitors weight fluctuations and adjusts them promptly, allowing the model to adapt to data changes and ensuring the long-term effective operation of the system (e.g., maintaining reliable PICC catheter positioning accuracy during long-term chemotherapy). Clinical practice demands high precision in PICC catheter positioning and accurate monitoring of chemotherapy parameters. Stable and reasonable feature weights ensure that system outputs (such as catheter position reports and chemotherapy efficacy predictions) are more realistic.

[0064] Specifically, the original weights corresponding to this feature are compensated and adjusted, including:

[0065] Retrieve the medical knowledge graph corresponding to tumor chemotherapy;

[0066] The clinical relevance coefficient between feature i and feature j is retrieved from the medical knowledge graph, wherein the value range of the clinical relevance coefficient is: ;

[0067] The standard deviation of the weight value corresponding to each feature is compared with a preset standard deviation threshold to obtain the standard deviation ratio parameter. ;in, This represents the standard deviation ratio parameter corresponding to the i-th feature; This represents the preset standard deviation threshold; This represents the standard deviation of the weight values ​​corresponding to the i-th feature;

[0068] The original weights of the clinical relevance coefficients between features i and j are adjusted by using the standard deviation ratio parameter corresponding to each feature.

[0069] The adjusted weight values ​​are obtained using the following formula:

[0070]

[0071] in, This represents the adjusted weight value; represents the weight value before adjustment; n represents the number of features that have a clinical relevance coefficient with the i-th feature; This represents the standard deviation ratio parameter corresponding to the i-th feature; This represents the clinical relevance coefficient between the i-th feature and the j-th feature; and These represent the importance results corresponding to the i-th feature and the j-th feature, respectively.

[0072] The technical effects of the above solution are as follows: It retrieves clinical correlation coefficients between features from the medical knowledge graph (such as the correlation between "catheter displacement rate" and "patient body movement amplitude"), allowing weight adjustments to align with clinical reality. In the PICC catheter positioning system for tumor chemotherapy, it incorporates clinical experts' understanding of feature correlations into weight adjustments, making feature weights more consistent with medical logic and improving the model's accuracy in catheter positioning and chemotherapy parameter prediction. This utilizes clinical correlation coefficients. When feature i and feature j have a strong clinical relevance, the compensation adjustment will take into account the importance result of j. , The weights of collaboratively corrected feature i are determined by calculating the standard deviation ratio parameter. The ratio of the standard deviation of the weight value to the threshold is used to accurately identify abnormal fluctuations in feature weights (such as a feature's weight fluctuating due to data noise). When the fluctuation exceeds the threshold, compensation adjustment is triggered to bring the weights back to a reasonable range, ensuring the stability of model outputs (such as catheter location prediction and chemotherapy effect correlation analysis) and improving the system's robustness in complex clinical environments (patient movement, measurement errors, etc.). Considering multiple features that are clinically relevant to the i-th feature and performing synergistic compensation adjustment can effectively improve the rationality and accuracy of weight settings.

[0073] The data annotation module is configured to annotate puncture results and complications in historical data. Puncture result annotation includes: marking the puncture result as successful or unsuccessful. For successful punctures, the coordinates of the puncture point are recorded; for unsuccessful punctures, the reasons for failure are recorded, such as: difficulty in vascular puncture, hematoma, etc. Complication annotation includes: recording the occurrence of complications, such as: hematoma, nerve damage, etc.

[0074] The feature selection module is configured to select features with higher importance from vascular imaging features and blood flow parameter features based on the importance of the features, and use these features as input features for the random forest model. The labeled puncture results and puncture point coordinates are selected as target variables for the random forest model. For example, if features A and B are of higher importance, while feature C is of lower importance, then only features A and B are selected as input data, and feature C is ignored.

[0075] The visual interface is configured to display vascular images and recommended puncture points in real time on the monitoring screen, providing visual guidance for the puncture path of the PICC catheter and helping medical staff to operate accurately.

[0076] The data processing module is configured to delete duplicate records in historical data, fill in missing blood flow parameters or imaging data, and use the Z-score statistical method to process abnormal data. The processed historical data is then divided into training and testing sets, typically in a ratio of 70% training set and 30% testing set.

[0077] The practical application module is configured to apply the trained random forest model to actual chemotherapy. By acquiring vascular images and blood flow parameters in real time, the random forest model outputs the optimal puncture point coordinates and the reasons for the recommendation, and displays the recommended puncture point and reasons on a visual interface, providing visual guidance for PICC catheter positioning.

[0078] The user feedback module is configured to adjust the parameters of the random forest model and optimize its performance based on feedback from medical staff and real-time data.

[0079] The beneficial effects achieved by the above are as follows: By assessing the importance of vascular imaging features and blood flow parameter features, the most influential features on predicting puncture results can be selected, thereby reducing model complexity and improving model efficiency and generalization ability; and by weighting the input data according to feature importance, the random forest model pays more attention to important features, thereby improving the accuracy and reliability of model prediction; through the collaborative work of multiple modules, the positioning system is more accurate and efficient in recommending the best puncture point, thereby effectively improving the puncture success rate, reducing the risk of complications, providing strong decision support for medical staff, and improving the patient's treatment experience and safety.

[0080] Working principle: By embedding a miniature magnetic head 3, a high-frequency linear probe 4, and a microelectrode 5 at the tip of the catheter body 1, combined with external sensors and ECG monitoring equipment, vascular images, blood flow parameters, and ECG signals are acquired in real time. The high-frequency linear probe 4 uses Doppler mode to identify vascular features, weights the input data according to feature importance, and predicts the optimal puncture point based on the input data using a random forest model. Combined with a visual interface that displays vascular images and recommended puncture points in real time, it provides medical staff with intuitive puncture path guidance, thereby achieving accurate recommendation of puncture points and precise positioning of the catheter tip, effectively improving the puncture success rate, reducing the risk of complications, providing strong decision support for medical staff, and improving the patient's treatment experience and safety.

[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart PICC catheter positioning system for tumor chemotherapy, implemented via a smart PICC catheter for tumor chemotherapy, characterized in that, A smart PICC catheter positioning system for tumor chemotherapy includes: The feature extraction module is configured to collect historical data and extract vascular image features and blood flow parameter features from the historical data. The vascular image features include: extracting the shape, inner diameter and depth of the blood vessels, extracting the centerline and contour of the blood vessels, and calculating the length and curvature of the blood vessels; the blood flow parameter features include: extracting the average, maximum and minimum values ​​of blood flow velocity, and calculating the rate of change and stability index of blood flow velocity. The intelligent analysis module is configured to train and test historical data and its vascular imaging features and blood flow parameter features using a random forest model, and output the coordinates of the best puncture point and the reasons for the recommendation. The reasons for the recommendation include the vascular diameter, blood flow velocity and vascular depth as key factors. The importance assessment module is configured to, after the random forest model has been trained, output the average impurity reduction value of each feature through the random forest model, normalize the average impurity reduction value of each feature, and obtain the importance result of each feature. The data weighting module is configured to perform a weighted average of the input data for each feature based on its importance, resulting in weighted input data. The weights corresponding to each feature are also compensated for in the following ways: Retrieve the average reduction in impurity for each feature; obtain the impurity reduction ratio for each feature based on the average reduction in impurity for each feature. Retrieve the original weights corresponding to each feature; wherein, the proportion of the original weights corresponding to each feature is consistent with the importance result corresponding to each feature; based on the original weights of each feature corresponding to each importance result, obtain the standard deviation of the weight values ​​of the original weights corresponding to each feature; The standard deviation of the original weight value corresponding to each feature is compared with a preset standard deviation threshold; when the standard deviation of the original weight value corresponding to the feature exceeds the preset standard deviation threshold, the original weight corresponding to the feature is compensated and adjusted, and the original weight is replaced with the compensated and adjusted weight value; the compensation and adjustment of the original weight corresponding to the feature includes: Retrieve the medical knowledge graph corresponding to tumor chemotherapy; retrieve the clinical relevance coefficient between feature i and feature j from the medical knowledge graph, wherein the value range of the clinical relevance coefficient is: ; The standard deviation of the weight value corresponding to each feature is compared with a preset standard deviation threshold to obtain the standard deviation ratio parameter. ;in, This represents the standard deviation ratio parameter corresponding to the i-th feature; This represents the preset standard deviation threshold; This represents the standard deviation of the weight values ​​corresponding to the i-th feature; The original weights of the clinical relevance coefficient pairs between features i and j are adjusted by using the standard deviation ratio parameter corresponding to each feature. A smart PICC catheter for tumor chemotherapy includes: a catheter body (1) and an external component (2), characterized in that one end of the catheter body (1) is connected to the external component (2) to form the overall structure of the smart PICC catheter, a micro magnetic head (3) and a high-frequency linear probe (4) are superimposed and embedded on the tip of the catheter body (1), and a micro electrode (5) is built into the catheter body (1). The micro magnetic head (3) is made of high-performance magnetic material and is configured to generate a magnetic field signal; The microelectrode (5) is connected to the tip of the catheter body (1) and is configured to collect the electrical signal generated when the tip of the catheter body (1) contacts the heart chamber in real time. The high-frequency linear probe (4) is configured to acquire images of the cross-section and longitudinal section of blood vessels in real time, and automatically identify the depth, inner diameter and blood flow velocity of blood vessels through Doppler mode. Based on the random forest model, the identification results are predicted and the best puncture point is recommended.

2. The intelligent PICC catheter for tumor chemotherapy according to claim 1, characterized in that, The micro magnetic head (3) is connected to the signal transmission interface at the end of the catheter via an internal wire, and transmits the magnetic field signal to the external sensor. The external sensor consists of multiple highly sensitive magnetic field sensors distributed at specific locations on the patient's body surface. It is configured to sense the magnetic field signal generated by the micro magnetic head (3) in real time, convert the magnetic field signal into an electrical signal, and transmit the electrical signal to the host for processing and analysis wirelessly. The microelectrode (5) is connected to an external electrocardiogram (ECG) monitoring device via an internal wire. The ECG monitoring device consists of an ECG amplifier, a filter, and a signal processor. It is configured to process the electrical signal generated when the tip of the catheter body (1) contacts the heart chamber in real time. After amplification, filtering, and processing, an ECG signal is generated and transmitted to the host for analysis and judgment via wireless means.

3. The intelligent PICC catheter positioning system for tumor chemotherapy according to claim 1, characterized in that, The system also includes: a data annotation module and a feature selection module; The data annotation module is configured to annotate puncture results and complications in historical data. Puncture result annotation includes: marking the puncture result as successful or unsuccessful; for successful punctures, recording the coordinates of the puncture point; for unsuccessful punctures, recording the reason for the failure; complication annotation includes: recording the occurrence of complications. The feature selection module is configured to select highly important features from vascular imaging features and blood flow parameter features based on their importance, and use these features as input features for the random forest model. The labeled puncture results and puncture point coordinates are selected as target variables for the random forest model.

4. The intelligent PICC catheter positioning system for tumor chemotherapy according to claim 1, characterized in that, The system also includes a visual interface and a data processing module: The visual interface is configured to display vascular images and recommended puncture points in real time on the monitoring screen, providing visual guidance for the puncture path of the PICC catheter. The data processing module is configured to delete duplicate records in historical data, fill in missing blood flow parameters or imaging data, process abnormal data using the Z-score statistical method, and divide the processed historical data into training and testing sets.

5. The intelligent PICC catheter positioning system for tumor chemotherapy according to claim 1, characterized in that, The intelligent analysis module includes: The model training module is configured to set the parameters of the random forest model, train the random forest model using training set data, evaluate the performance of the random forest model using test set data, calculate the accuracy, recall and F1 score of the random forest model on the test set, analyze the confusion matrix of the random forest model, and evaluate the predictive ability of the random forest model on different data.

6. The intelligent PICC catheter positioning system for tumor chemotherapy according to claim 1, characterized in that, The system also includes: a practical application module and a user feedback module; The practical application module is configured to apply the trained random forest model to actual chemotherapy. By acquiring vascular images and blood flow parameters in real time, the random forest model outputs the optimal puncture point coordinates and the reasons for the recommendation, and displays the recommended puncture point and reasons on a visual interface, providing visual guidance for PICC catheter positioning. The user feedback module is configured to adjust the parameters of the random forest model and optimize its performance based on feedback from medical staff and real-time data.

Images