A method, system, device, and storage medium for predicting peripheral venous iodinated contrast extravasation for contrast-enhanced CT examinations

Abstract

The application discloses a method, system, device and storage medium for predicting peripheral vein iodine contrast agent extravasation of enhanced CT examination, and relates to the technical field of medical data processing. The method comprises the following steps: acquiring environmental temperature difference characteristic data, waiting time characteristic data, vein vessel characteristic data, puncture consumable characteristic data and contrast agent injection parameter data of a to-be-examined object; attenuating and reducing the weight of the environmental temperature difference characteristic data based on the waiting time characteristic data to obtain dynamic temperature difference calibration characteristic data; extracting coupling risk characteristic data by crossing the vein vessel and the puncture consumable characteristic data; generating a target characteristic vector by fusion and calculating a prediction probability value through a pre-trained extravasation risk prediction model; determining a dynamic early warning threshold value according to the contrast agent injection parameter data; comparing the prediction probability value with the dynamic early warning threshold value to generate early warning prompt information and outputting the early warning prompt information. The application avoids the false negative or false positive caused by the fixed threshold under complex working conditions, and improves the accuracy of the contrast agent extravasation prediction.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a method, system, device, and storage medium for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examinations. Background Technology

[0002] With the rapid development of medical imaging technology, enhanced CT scans have become a core tool for clinical disease diagnosis and staging. During the examination, high-pressure, rapid injection of iodine contrast agent is essential for obtaining high-quality images, but it can easily lead to complications such as extravasation of the contrast agent into peripheral veins. Accurately predicting and warning of the risk of contrast agent extravasation is crucial for ensuring patient safety and improving the quality of medical care.

[0003] Currently, methods for predicting iodine contrast agent extravasation mainly rely on static analysis of patients' basic physiological data and routine puncture data. By calculating the predicted probability of extravasation and comparing it with a preset fixed risk threshold, early warning information is generated to prevent the risk of extravasation during the examination.

[0004] However, relying solely on fixed risk thresholds for comparison and judgment makes it difficult to accurately match complex and ever-changing actual injection conditions (such as different high-pressure and high-flow-rate states), resulting in the generated warning information failing to truly reflect dynamic risks, thereby reducing the accuracy of contrast agent extravasation prediction. Summary of the Invention

[0005] To address the aforementioned shortcomings, a method, system, device, and storage medium for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examinations are provided.

[0006] In a first aspect, the present invention provides a method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination, the method comprising: Acquire environmental temperature difference characteristics, waiting time characteristics, venous blood vessel characteristics, puncture consumable characteristics, and contrast agent injection parameter data of the subject to be examined; Based on the waiting time feature data, the environmental temperature difference feature data is attenuated and weighted to obtain dynamic temperature difference calibration feature data. The attenuation and weighting process is used to characterize the logical relationship that the stress effect of environmental temperature difference on blood vessels weakens with the increase of waiting time. Feature cross-processing is performed on venous vessel feature data and puncture consumable feature data to extract coupling risk feature data characterizing the hydrodynamic resistance of puncture consumables in venous vessels; The target feature vector is generated by splicing and fusing other basic feature data that did not participate in feature cross-processing from dynamic temperature difference calibration feature data, coupling risk feature data, venous blood vessel feature data and puncture consumable feature data. The target feature vector is input into a pre-trained extravasation risk prediction model, which outputs the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination. Based on the contrast agent injection parameter data, a dynamic warning threshold matching the current injection status is determined. The dynamic warning threshold decreases as the injection flow rate in the contrast agent injection parameter data increases. The predicted probability value is compared with the dynamic early warning threshold. Based on the comparison result, a corresponding early warning message for the risk of extravasation is generated and output to the interactive terminal.

[0007] By employing the above technical solution, environmental temperature difference characteristic data, waiting time characteristic data, venous vessel characteristic data, puncture consumable characteristic data, and contrast agent injection parameter data of the subject to be examined are obtained. Based on the waiting time characteristic data, the environmental temperature difference characteristic data is attenuated and weighted to obtain dynamic temperature difference calibration characteristic data, thus restoring the physical law of the weakening effect of temperature difference on vascular stress over time. Feature cross-processing is performed on the venous vessel characteristic data and puncture consumable characteristic data to extract coupling risk characteristic data, quantifying the hydrodynamic resistance generated by the puncture consumables within the blood vessel. The dynamic temperature difference calibration characteristic data, coupling risk characteristic data, and venous vessel characteristic data and puncture consumable characteristic data that did not participate in feature cross-processing are then analyzed. Other basic feature data are spliced ​​and fused to generate a target feature vector. This vector is then used in conjunction with preset feature weight parameters to perform linear algebra calculations to obtain an initial predicted value. A preset nonlinear activation function is then used to numerically map and transform the initial predicted value to obtain a predicted probability value. This transforms multidimensional data into a continuous probability value, objectively reflecting the tendency for extravasation. A dynamic warning threshold that decreases with increasing flow rate is determined based on contrast agent injection parameter data, ensuring the judgment criteria align with actual injection conditions. The predicted probability value is compared with the dynamic warning threshold, and corresponding extravasation risk warning information is generated and output based on the comparison results. This avoids missed or false alarms caused by fixed thresholds under complex operating conditions, improving the accuracy of contrast agent extravasation prediction.

[0008] In a second aspect, the present invention provides a system for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examinations, the system comprising: The data acquisition module is used to acquire environmental temperature difference characteristics, waiting time characteristics, venous blood vessel characteristics, puncture consumable characteristics, and contrast agent injection parameter data of the subject to be examined. The temperature difference calibration module is used to perform attenuation and weight reduction processing on the environmental temperature difference feature data based on the waiting time feature data to obtain dynamic temperature difference calibration feature data. The attenuation and weight reduction processing is used to characterize the logical relationship that the stress effect of environmental temperature difference on blood vessels weakens as the waiting time increases. The feature cross-processing module is used to perform feature cross-processing on venous vessel feature data and puncture consumable feature data to extract coupling risk feature data that characterizes the hydrodynamic resistance of puncture consumables in venous vessels. The feature fusion module is used to splice and fuse other basic feature data that did not participate in feature cross-processing from dynamic temperature difference calibration feature data, coupling risk feature data, venous blood vessel feature data, and puncture consumables feature data to generate a target feature vector. The probability prediction module is used to input the target feature vector into the pre-trained extravasation risk prediction model and output the predicted probability value of the subject's peripheral venous iodine contrast agent extravasation. The threshold determination module is used to determine a dynamic warning threshold that matches the current injection status based on the contrast agent injection parameter data. The dynamic warning threshold decreases as the injection flow rate in the contrast agent injection parameter data increases. The early warning control module is used to compare the predicted probability value with the dynamic early warning threshold, generate corresponding extravasation risk early warning information based on the comparison result, and output it to the interactive terminal.

[0009] Thirdly, the present invention provides an electronic device, including a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the above-described method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination when executing the computer program.

[0010] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when read and run by a processor, implements the above-described method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination. Attached Figure Description

[0011] Figure 1 A flowchart illustrating a method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination, provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the logic processing for generating target feature vectors and predicting extravasation risk, provided for an embodiment of the present invention. Figure 3 This is a schematic diagram of the architecture of a system for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination, provided by an embodiment of the present invention. Detailed Implementation

[0012] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0013] Example 1 like Figure 1 As shown, a method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination includes: Acquire environmental temperature difference characteristics, waiting time characteristics, venous blood vessel characteristics, puncture consumable characteristics, and contrast agent injection parameter data of the subject to be examined; Based on the waiting time feature data, the environmental temperature difference feature data is attenuated and weighted to obtain dynamic temperature difference calibration feature data. The attenuation and weighting process is used to characterize the logical relationship that the stress effect of environmental temperature difference on blood vessels weakens with the increase of waiting time. Feature cross-processing is performed on venous vessel feature data and puncture consumable feature data to extract coupling risk feature data characterizing the hydrodynamic resistance of puncture consumables in venous vessels; The target feature vector is generated by splicing and fusing other basic feature data that did not participate in feature cross-processing from dynamic temperature difference calibration feature data, coupling risk feature data, venous blood vessel feature data and puncture consumable feature data. The target feature vector is input into a pre-trained extravasation risk prediction model, which outputs the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination. Based on the contrast agent injection parameter data, a dynamic warning threshold matching the current injection status is determined. The dynamic warning threshold decreases as the injection flow rate in the contrast agent injection parameter data increases. The predicted probability value is compared with the dynamic early warning threshold. Based on the comparison result, a corresponding early warning message for the risk of extravasation is generated and output to the interactive terminal.

[0014] Specifically, the electronic device first acquires data on the environmental temperature difference characteristics, waiting time characteristics, venous vessel characteristics, puncture consumable characteristics, and contrast agent injection parameters of the subject to be examined. This data can be automatically retrieved from the hospital information system, radiology information system, electronic medical record system, and IoT temperature and humidity sensors deployed in the examination room via communication interfaces. Among these, the environmental temperature difference characteristics are used to characterize the temperature changes in the environment surrounding the subject. Different temperature difference calculation methods are adopted according to different application scenarios and data acquisition conditions: In the conventional initial screening scenario where real-time temperature sensing data is lacking, the environmental temperature difference characteristic data is the diurnal temperature difference on the day of examination; in the accurate prediction scenario where real-time environmental monitoring is available and dynamic attenuation calibration is required, the environmental temperature difference characteristic data is the change in environmental temperature experienced by the patient before and after entering the examination room (i.e., the difference between the examination room temperature and the outdoor ambient temperature). For example, when the outdoor temperature is 5℃ and the examination room temperature is 22℃, the environmental temperature difference characteristic data is recorded as 17℃; the waiting time characteristic data is the waiting time from the patient entering the examination room to the start of the puncture operation, in minutes; the venous vessel characteristic data includes the diameter of the vein at the puncture site, the vascular elasticity score, and the vascular wall fragility assessment value; the puncture consumable characteristic data includes the needle gauge (e.g., 20G, 22G) and catheter material type of the indwelling needle; the contrast agent injection parameter data includes the injection flow rate (in mL / s), the total injection volume, and the upper limit of the injection pressure.

[0015] In the specific implementation, in order to facilitate the subsequent model calculation, the core parameters collected above are standardized and quantified and assigned values. The specific assignment rules are as follows: (1) Environmental temperature difference characteristic data (X1): According to the preset conditions of the current examination scenario, the day-night temperature difference value of the examination day is taken in the conventional initial screening scenario, and the difference between the examination room temperature and the outdoor ambient temperature is taken in the accurate prediction scenario, which is an integer; (2) Indwelling needle type in puncture consumable characteristic data (X2): Straight indwelling needle is assigned a value of 1, and Y-type indwelling needle is assigned a value of 2; (3) Indwelling needle insertion depth in puncture consumable characteristic data (X3): Complete insertion is assigned a value of 0, and incomplete insertion is assigned a value of 1; (4) Target vein puncture history within 24 hours in the vein vascular characteristic data (X4): No puncture history is assigned a value of 0, and puncture history (including blood collection, infusion, etc.) is assigned a value of 1; (5) Vein quality grading in the vein vascular characteristic data (X5): Good vein quality is assigned a value of 0, average quality is assigned a value of 1, poor quality is assigned a value of 2, and very poor quality is assigned a value of 3; (6) Indwelling needle model in the puncture consumables characteristic data (X6): 18G indwelling needle is assigned a value of 3, 20G indwelling needle is assigned a value of 2, and 22G indwelling needle is assigned a value of 1.

[0016] After data collection, the environmental temperature difference feature data is attenuated and weighted based on the waiting time feature data to obtain dynamic temperature difference calibration feature data. The attenuation and weighting process adopts an exponential decay form, which can be expressed as: Dynamic temperature difference calibration feature data = Environmental temperature difference feature data × e^(−λ × Waiting time feature data), where λ is the attenuation coefficient, which can be preset based on clinical experience, for example, 0.05. This formula restores the law that the degree of vascular temperature stress gradually recovers as the waiting time increases after the patient enters the examination room. The longer the waiting time, the lower the value of the dynamic temperature difference calibration feature data.

[0017] Based on this, feature cross-processing is performed on venous vessel characteristic data and puncture consumable characteristic data to extract coupled risk characteristic data that characterizes the hydrodynamic resistance of puncture consumables within the venous vessel. Feature cross-processing generates new cross-feature terms by calculating the pairwise product of each component in the venous vessel characteristic data and each component in the puncture consumable characteristic data. For example, multiplying the tube diameter value by the needle gauge value can quantify the magnitude of the fluid resistance generated by a specific specification indwelling needle on the vessel wall after insertion into a vessel of a specific diameter, forming coupled risk characteristic data. Compared with analyzing the characteristics of the vessel or consumables separately, this step can more accurately reflect the comprehensive impact of the actual puncture state on the risk of extravasation.

[0018] Subsequently, the dynamic temperature difference calibration feature data, coupling risk feature data, venous blood vessel feature data, and puncture consumable feature data that did not participate in feature cross-processing are spliced ​​and fused together, and arranged in a fixed dimensional order into a one-dimensional target feature vector to ensure that the input form of subsequent calculations is consistent. The target feature vector is input into a pre-trained extravasation risk prediction model. The model performs the following calculations to output the predicted probability value of extravasation of peripheral vein iodine contrast agent in the subject under examination: First, the target feature vector and the preset feature weight parameters are subjected to linear algebra operations to calculate the initial predicted value representing the extravasation tendency. The specific calculation form is: initial predicted value = W^T × target feature vector + b, where W is the feature weight parameter vector obtained by the extravasation risk prediction model in advance through historical sample data, and b is the bias term. Then, the initial predicted value is numerically mapped and transformed by the nonlinear activation function preset by the extravasation risk prediction model to obtain the predicted probability value of extravasation of peripheral vein iodine contrast agent in the subject under examination. The nonlinear activation function can be the sigmoid function, that is, the predicted probability value = 1 / (1+e^(−initial predicted value)), which maps the initial predicted value to the interval (0,1) to obtain a continuous probability value.

[0019] After obtaining the predicted probability value, a dynamic warning threshold matching the current injection status is determined based on the contrast agent injection parameter data. The dynamic warning threshold decreases as the injection flow rate in the contrast agent injection parameter data increases. For example, when the injection flow rate is 2 mL / s, the dynamic warning threshold can be set to 0.70; when the injection flow rate is 3 mL / s, the dynamic warning threshold decreases to 0.60 accordingly; when the injection flow rate reaches 5 mL / s, the dynamic warning threshold further decreases to 0.50 to match the higher sensitivity to extravasation risk under high flow rate injection conditions. Finally, the predicted probability value is compared with the dynamic warning threshold. If the predicted probability value is greater than or equal to the dynamic warning threshold, a high-risk extravasation warning message is generated and output, prompting the operator to take appropriate protective measures for the patient; if the predicted probability value is less than the dynamic warning threshold, a low-risk extravasation warning instruction is generated and output, allowing the enhanced CT examination to continue according to the normal procedure.

[0020] Example 2 As a preferred embodiment of the present invention, based on Example 1, the environmental temperature difference feature data is attenuated and weighted based on the waiting time feature data to obtain dynamic temperature difference calibration feature data, including: Extract the initial value of the ambient temperature difference from the ambient temperature difference feature data, and the waiting time in the constant temperature zone from the waiting time feature data; Obtain the preset baseline thermal stress half-life parameters, and calculate the dynamic decay coefficient based on the natural exponential decay logic using the waiting time in the isothermal zone and the baseline thermal stress half-life parameters; The initial value of the ambient temperature difference is multiplied by the dynamic attenuation coefficient to obtain dynamic temperature difference calibration feature data that has been smoothed and weighted down over time.

[0021] Specifically, firstly, the initial value of the environmental temperature difference is extracted from the environmental temperature difference characteristic data. The initial value of the environmental temperature difference is the instantaneous temperature drop experienced by the patient when entering the constant temperature examination area from the outside. For example, when the outdoor temperature is 4℃ and the constant temperature of the examination room is 22℃, the initial value of the environmental temperature difference is recorded as 18℃. This value reflects the maximum heat stress intensity that the patient's blood vessels experience the moment they enter the examination area. At the same time, the waiting time in the constant temperature area is extracted from the waiting time characteristic data. The waiting time in the constant temperature area is the time that the patient stays in the constant temperature examination area from the time they enter until the puncture operation begins, in minutes. For example, if the patient waits in the waiting area for 12 minutes before the puncture, the waiting time in the constant temperature area is set to 12.

[0022] After extracting the above two data, the preset baseline heat stress half-life parameter is obtained. This parameter is an objective physiological parameter obtained by statistical regression fitting based on a large number of historical patients' vascular ultrasound observation data and waiting time. It represents the time required for the human blood vessel heat stress intensity to decay to half of the initial value in a constant temperature environment. It can be pre-configured based on prior knowledge of clinical physiology, for example, set to 10 minutes. Based on the natural exponential decay logic, the dynamic decay coefficient is calculated using the waiting time in the constant temperature zone and the baseline heat stress half-life parameter. The calculation formula is: dynamic decay coefficient = e^(−ln(2) / T_half×t), where T_half is the baseline heat stress half-life parameter and t is the waiting time in the constant temperature zone. Taking T_half=10 minutes and t=12 minutes as an example, the dynamic decay coefficient = e^(−ln(2) / 10×12)≈0.44, indicating that after 12 minutes of waiting in the constant temperature zone, the heat stress intensity has decayed to about 44% of the initial value.

[0023] Finally, the initial value of the ambient temperature difference is multiplied by the dynamic attenuation coefficient to obtain the dynamic temperature difference calibration feature data after time-dimension smoothing and weight reduction. Taking the above example, the dynamic temperature difference calibration feature data = 18 × 0.44 ≈ 7.92℃. This value is significantly lower than the initial value of the ambient temperature difference of 18℃, which truly restores the physiological process of the gradual reduction of vascular heat stress over time during the patient's waiting period. This makes the temperature difference feature used for subsequent prediction closer to the vascular state at the actual moment of the puncture operation.

[0024] Example 3 As a preferred embodiment of the present invention, based on Example 1, feature cross-processing is performed on venous vessel characteristic data and puncture consumable characteristic data to extract coupling risk characteristic data characterizing the hydrodynamic resistance of puncture consumables within venous vessels, including: Extract vein quality grading indicators from vein vascular feature data, and indwelling needle type and insertion depth status from puncture consumables feature data; Based on the venous quality grading index and the indwelling needle type index, a basic fluid dynamics resistance matrix is ​​constructed; When the insertion depth is determined to be incomplete and the vein quality grading index is lower than the preset quality standard, the resistance penalty mechanism is triggered. The preset nonlinear mapping table is queried according to the vein quality grading index to obtain the corresponding nonlinear penalty coefficient. The worse the vein quality, the nonlinear penalty coefficient increases nonlinearly. By using a nonlinear penalty coefficient to weight and correct the basic fluid dynamics resistance matrix, coupled risk characteristic data can be extracted.

[0025] Specifically, firstly, vein quality grading indicators are extracted from the vein vascular characteristic data. These indicators are derived based on a comprehensive assessment of the vein diameter, elasticity, and wall fragility at the puncture site, and are represented using a graded coding system. For example, vein quality is divided into grades 1 to 4, from best to worst. Vessels with wide diameters, good elasticity, and low wall fragility correspond to grade 1, while vessels with narrow diameters, poor elasticity, and high wall fragility correspond to grade 4. Simultaneously, indwelling needle model indicators and insertion depth status are extracted from the puncture consumables characteristic data. The indwelling needle model indicator is the needle gauge code, such as 18G, 20G, or 22G. The insertion depth status refers to the position of the indwelling needle tip after the puncture operation, with values ​​indicating either complete insertion or incomplete insertion.

[0026] After extracting the above data, a basic hydrodynamic resistance matrix is ​​constructed based on the vein quality grading index and the indwelling needle model index. The row dimension of the matrix corresponds to each level of the vein quality grading index, and the column dimension corresponds to each model and specification of the indwelling needle model index. The value of each element in the matrix represents the basic hydrodynamic resistance estimate of the puncture state under the combination of the vein quality level and the indwelling needle model. For example, the basic resistance estimate corresponding to the combination of a level 4 vein and a 20G indwelling needle is higher than that of the combination of a level 1 vein and a 22G indwelling needle. This matrix is ​​obtained by performing statistical regression fitting on a large amount of clinical puncture data in advance and is stored in the system in the form of preset parameters.

[0027] After constructing the basic fluid dynamics resistance matrix, the insertion depth status and vein quality grading index are further judged to meet the preset joint penalty conditions: if the insertion depth status is incomplete insertion and the vein quality grading index is lower than the preset quality standard, the resistance penalty mechanism is triggered, generating a nonlinear penalty coefficient greater than 1. The preset quality standard can be configured to level 3, that is, when the vein quality grading index reaches level 3 or 4, it is considered to be below the standard; the nonlinear penalty coefficient is calculated in the form of square penalty, which can be expressed as: nonlinear penalty coefficient = 1 + α × (vein quality grading index value)², where α is the penalty intensity hyperparameter. This hyperparameter and the square penalty form are determined based on the fitting of the pipe wall shear stress data of the computational fluid dynamics simulation model under different pipe diameter and needle gauge combinations, objectively reflecting the nonlinear abrupt change law of fluid resistance. The coefficient can be preset to 0.3. Taking a vein quality grading index of level 4 and an insertion depth of incomplete insertion as an example, the nonlinear penalty coefficient = 1 + 0.3 × 4² = 5.8. This square form ensures that the magnitude of the resistance penalty increases rapidly in a nonlinear manner when the vein quality is poor or the needle tip is not in place, thus more realistically reflecting the sharp increase in actual puncture risk. Conversely, if the combined penalty conditions are not met (i.e., the insertion depth is complete, or the vein quality grading index is not lower than the preset quality standard), the resistance penalty mechanism is not triggered, and the nonlinear penalty coefficient is set to 1, meaning that no additional correction is applied to the basic hydrodynamic resistance matrix.

[0028] Finally, the basic hydrodynamic resistance matrix is ​​weighted and corrected using a nonlinear penalty coefficient. The element values ​​in the matrix corresponding to the current puncture state are multiplied by the nonlinear penalty coefficient to extract coupled risk feature data. This data is output in scalar (single numerical value) form, which comprehensively reflects the actual hydrodynamic resistance level under the combined effect of the puncture consumable model, blood vessel quality and needle tip positioning state, providing a more accurate feature input for subsequent prediction of extravasation risk.

[0029] Example 4 As a preferred embodiment of the present invention, based on Example 1, the target feature vector is input into a pre-trained extravasation risk prediction model, and the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject to be examined is output, including: Obtain the pre-calibrated multidimensional feature weight vector and global intercept constant of the extravasation risk prediction model; The feature values ​​of each dimension in the target feature vector are multiplied one by one with the corresponding weight values ​​in the multidimensional feature weight vector; The summation of the products obtained by multiplying each term yields a comprehensive feature weighting value. The weighted value of the comprehensive features is added to the global intercept constant to obtain the initial predicted value that characterizes the tendency of exudation. Based on the initial prediction value, the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination is determined.

[0030] Specifically, the process begins by acquiring pre-calibrated multidimensional feature weight vectors and a global intercept constant. These vectors are obtained through the following offline training phase: collecting electronic medical records and injection records of patients undergoing historical enhanced CT examinations as sample data; extracting environmental temperature difference, waiting time, vein characteristics, puncture consumables, and contrast agent injection parameters as feature data, and extracting whether extravasation actually occurred during the examination as the true label; training the model on the sample data using a multivariate logistic regression algorithm, employing backward elimination to screen feature dimensions that significantly influence the prediction results, and finally fitting the multidimensional feature weight vector and global intercept constant. The dimensions of the multidimensional feature weight vector are strictly aligned with the dimensions of the target feature vector, and each weight value in the vector reflects the contribution of the corresponding feature dimension to the risk of extravasation; the global intercept constant is the bias term obtained during model training. Both are stored in the system as preset parameters and are directly accessed during the inference phase.

[0031] like Figure 2As shown, when predicting the risk of extravasation, the system first combines the dynamic temperature difference calibration feature data, coupled risk feature data, and other basic feature data obtained in the previous steps to construct a one-dimensional target feature vector with a unified dimension. After acquiring the parameters, the system performs a product operation on each dimension of the target feature vector with the corresponding weight value in the multidimensional feature weight vector. That is, the first dimension feature value of the target feature vector is multiplied by the first dimension weight value of the multidimensional feature weight vector, the second dimension feature value is multiplied by the second dimension weight value, and so on, until all dimensions have completed a one-to-one product operation. For example, if the target feature vector contains 8 dimensions, the product operation will produce 8 corresponding product results. Each product result independently quantifies the contribution of that dimension feature to the initial predicted value under the current sample.

[0032] After the item-by-item multiplication operation is completed, all product results are summed to obtain a comprehensive feature weighted value. This summation operation integrates the independent contribution components of each dimension into a scalar, comprehensively reflecting the overall extravasation tendency of the target feature vector under the current feature weight parameter configuration. Subsequently, the comprehensive feature weighted value is added to the global intercept constant to obtain the initial predicted value representing the extravasation tendency. In a preferred implementation scenario based on large-sample clinical data validation, the multidimensional feature weight vector and the global intercept constant are determined through multivariate logistic regression (backward elimination method).

[0033] Specifically, in order to ensure that the prediction model can accurately adapt to the complex features after dynamic calibration and cross-coupling, the multidimensional feature weight vector and global intercept constant in this embodiment are obtained by retraining and fitting the model using a large sample of historical clinical data containing dynamic temperature difference calibration feature data and coupled risk feature data through a multi-factor logistic regression algorithm (backward elimination method).

[0034] During inference calculations, the quantified and feature-processed target feature vector is substituted into the linear prediction equation to calculate the initial predicted value (i.e., the logarithmic odds value) characterizing the extravasation tendency. In a basic implementation, the multidimensional feature weight vector and global intercept constant are configured as follows: intercept term is -1.512, weight of environmental temperature difference feature data (X1) is 0.117, weight of indwelling needle type (X2) is 1.224, weight of indwelling needle insertion depth (X3) is -1.813, weight of target venous puncture history within 24 hours (X4) is 2.904, weight of venous quality grading (X5) is 0.944, and weight of indwelling needle model (X6) is -0.536. In a preferred implementation that introduces dynamic calibration and cross-coupling features, the specific linear operation process can be expressed as: initial predicted value = W_temp × X_temp + W_couple × X_couple + W_2 × X2 + W_4 × X4 + ... + b.

[0035] Where: b is the global intercept constant obtained from model training; W_temp, W_couple, W_2, W_4, etc. constitute the multidimensional feature weight vector; X_temp is the dynamic temperature difference calibration feature data calculated in Example 2, X_couple is the coupling risk feature data calculated in Example 3, X2 is the indwelling needle type, and X4 is the target vein puncture history within 24 hours. Through the above item-by-item multiplication and summation operations, the independent contribution components of each dimension are integrated into a scalar. For example, assuming that after the above item-by-item multiplication and summation operations, the comprehensive feature weighted value of a certain subject to be examined is 2.442, combined with the global intercept constant -1.512, the initial predicted value obtained after addition is 2.442 + (-1.512) = 0.93. This initial predicted value, as the final output of the linear operation stage, will be further mapped to an extravasation prediction probability value with probabilistic meaning by a nonlinear activation function in subsequent steps.

[0036] Example 5 As a preferred embodiment of the present invention, based on Example 4, the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination is determined based on the initial predicted value, including: The initial predicted value is negative and then used as the exponent of the natural constant for exponentiation to obtain the exponent-converted value. The denominator data is constructed by adding the preset unit constant and the exponential conversion value, and the reciprocal of the denominator data is calculated to obtain the predicted probability value mapped to the value range of zero to one.

[0037] Specifically, the nonlinear activation function used is the Sigmoid function, whose core function is to monotonically compress the initial predicted value, which takes any value in the real number domain, into the numerical range of 0 to 1, so that the output result has a probabilistic mathematical meaning. The entire mapping and transformation process consists of four sequentially executed numerical operation steps, with the initial predicted value of 0.93 obtained in Example 4 serving as a numerical example for the overall explanation.

[0038] First, the initial predicted value is negated by inverting its sign to obtain a negative predicted value. Taking 0.93 as an example, the negative predicted value is -0.93. The purpose of this negation is to construct an exponent term for the subsequent exponentiation operation, so that the larger the initial predicted value, the smaller the subsequent exponent conversion value, thus obtaining a probability output closer to 1 after the final reciprocal operation, which is consistent with the positive relationship of the infiltration tendency intensity. Then, the natural constant e is used as the base and the negative predicted value is used as the exponent to perform an exponentiation operation to obtain the exponent conversion value, i.e., e^(-0.93), which is approximately 0.3946. The magnitude of the exponent conversion value directly determines the magnitude of the subsequent denominator term, thus affecting the final probability.

[0039] After the exponentiation is completed, the converted exponent value is added to a preset unit constant to obtain the denominator data. The unit constant is fixed at 1, so the denominator data = 1 + 0.3946 = 1.3946. This addition operation ensures that the denominator is always greater than 1, thus ensuring that the result of the subsequent reciprocal operation strictly falls within the open interval of 0 to 1. Finally, the reciprocal of the denominator data is taken to obtain the predicted probability value mapped to the value interval of zero to one.

[0040] Based on the linear calculation results of the aforementioned embodiments, the complete mathematical expression for the probability P of iodine contrast agent extravasation between 0 and 1 obtained by transforming through the Sigmoid activation function is: P=1 / {1+exp[−(initial predicted value)]}.

[0041] Furthermore, to simplify clinical operations, this system can convert the aforementioned prediction model into a visual nomogram and output it to a display terminal. Clinical users only need to find the scores on the corresponding coordinate axes for each patient's parameters (including processed dynamic temperature difference and coupling resistance indicators), add all scores to obtain the total score, and directly obtain the probability of extravasation from the total score axis without complex calculations. A single assessment takes very little time. This continuous probability value will be compared with the dynamic early warning threshold in the subsequent risk grading judgment step to determine whether to trigger the corresponding early warning control command.

[0042] Example 6 As a preferred embodiment of the present invention, based on Example 1, a dynamic early warning threshold matching the current injection state is determined according to the contrast agent injection parameter data, including: Extract the baseline injection flow rate from the contrast agent injection parameter data; The threshold offset is calculated based on the baseline injection flow rate. The higher the baseline injection flow rate, the greater the negative increase in the threshold offset. The preset benchmark medium-risk threshold and benchmark high-risk threshold are added to the threshold offset respectively to calculate the dynamic medium-risk threshold and dynamic high-risk threshold. Based on dynamic medium-risk thresholds and dynamic high-risk thresholds, a dynamic early warning threshold system is constructed that includes low-risk, medium-risk, and high-risk intervals.

[0043] Specifically, the baseline injection flow rate is first extracted from the contrast agent injection parameter data. The baseline injection flow rate is the preset contrast agent injection rate of the infusion pump under the current examination protocol, in milliliters per second. This parameter directly reflects the pressure intensity of the contrast agent entering the venous access and is a key injection condition variable that affects the possibility of extravasation. Its value varies depending on the examination site and the contrast protocol, and the common range is 2.0 to 6.0 mL / s.

[0044] After obtaining the baseline injection flow rate, the threshold offset is calculated based on its value. The calculation rule is: threshold offset = −γ × v, where v is the current baseline injection flow rate and γ is the offset sensitivity coefficient, which can be preset to 0.02. This formula ensures that the absolute value of the threshold offset in the negative direction is larger when the baseline injection flow rate is higher. That is, as the injection flow rate increases, the subsequent dynamic threshold will be systematically shifted downward, so that a lower probability value is used as the risk trigger boundary under high flow rate injection conditions, and the early warning mechanism maintains higher sensitivity to high pressure injection conditions. Taking a baseline injection flow rate of 5 mL / s as an example, the threshold offset = −0.02 × 5 = −0.10.

[0045] After calculating the threshold offset, the preset baseline medium-risk threshold and baseline high-risk threshold are added to the threshold offset to obtain the dynamic medium-risk threshold and dynamic high-risk threshold. The baseline medium-risk threshold and baseline high-risk threshold are obtained by statistical calibration of historical clinical data and can be preset to 0.50 and 0.70 respectively. Taking the above example, the dynamic medium-risk threshold = 0.50 + (−0.10) = 0.40, and the dynamic high-risk threshold = 0.70 + (−0.10) = 0.60. Both dynamic thresholds are shifted down by 0.10 probability units compared to the baseline value, reflecting the reasonable tightening of the risk assessment standard in high-flow-rate injection scenarios.

[0046] Finally, a dynamic early warning threshold consisting of three continuous intervals is constructed based on the dynamic medium-risk threshold and the dynamic high-risk threshold: the interval where the predicted probability value is lower than the dynamic medium-risk threshold is defined as the low-risk interval, i.e., the predicted probability value falls within the range of 0 to 0.40; the interval where the predicted probability value is not lower than the dynamic medium-risk threshold but lower than the dynamic high-risk threshold is defined as the medium-risk interval, i.e., the predicted probability value falls within the range of 0.40 to 0.60; and the interval where the predicted probability value is not lower than the dynamic high-risk threshold is defined as the high-risk interval, i.e., the predicted probability value reaches 0.60 or above. The three intervals together constitute a dynamic early warning threshold covering the entire probability range, which can be directly called upon in the subsequent risk classification output step.

[0047] Example 7 As a preferred embodiment of the present invention, based on Example 1, the predicted probability value is compared with the dynamic early warning threshold, and a corresponding extravasation risk early warning message is generated according to the comparison result and output to the interactive terminal, including: Determine the risk range within which the predicted probability value falls under the dynamic early warning threshold; If the risk level is determined to be in the medium-risk range, a medium-risk deceleration prompt message containing suggested flow rate limit parameters will be generated. If the area is determined to be in a high-risk zone, an extremely high-risk warning message will be generated. The generated medium-risk deceleration prompt or extremely high-risk warning prompt will be output to the display interface of the interactive terminal to prompt medical staff to confirm whether it is necessary to manually adjust the injection parameters or interrupt the operation.

[0048] Specifically, after obtaining the predicted probability value and the dynamic early warning threshold, the first step is to determine the position of the predicted probability value within the three risk intervals defined by the dynamic early warning threshold. The judgment logic is as follows: if the predicted probability value is lower than the dynamic medium-risk threshold, it is classified into the low-risk interval; if the predicted probability value is not lower than the dynamic medium-risk threshold but lower than the dynamic high-risk threshold, it is classified into the medium-risk interval; and if the predicted probability value is not lower than the dynamic high-risk threshold, it is classified into the high-risk interval. The three judgment conditions are mutually exclusive and completely cover the full probability range from 0 to 1, ensuring that any input predicted probability value can be uniquely classified into one of the intervals. Taking the values ​​in Examples 5 and 6 as examples, the predicted probability value is 0.717, the dynamic medium-risk threshold is 0.40, and the dynamic high-risk threshold is 0.60. Since 0.717 is not lower than 0.60, the current object to be inspected is determined to be in the high-risk interval.

[0049] After determining the risk range, differentiated warning message generation logic is triggered based on the specific range. In a specific clinical application scenario, combined with the adjustment of dynamic warning thresholds, the baseline risk grading standard can be set as: low risk (P<0.3), medium risk (0.3≤P<0.7), and high risk (P≥0.7). If the risk range is determined to be low, no active intervention prompt is triggered; the monitoring interface simply maintains its normal state display, and the enhanced CT examination is performed according to the standard procedure, with routine observation during the injection. If the risk range is determined to be medium, a medium-risk deceleration prompt message containing a suggested flow rate limit parameter is generated. The suggested flow rate limit parameter is calculated by multiplying the current baseline injection flow rate by a preset deceleration ratio coefficient. For example, if the current baseline injection flow rate is 5 mL / s and the deceleration ratio coefficient is set to 0.8, the suggested flow rate limit parameter = 5 × 0.8 = 4 mL / s. This medium-risk deceleration prompt message carries this parameter value to prompt medical staff to manually reduce the actual injection flow rate. The jet rate is adjusted to 4 mL / s, and a designated person is instructed to closely monitor the patient to reduce the risk of extravasation while maintaining the examination. If the patient is determined to be in a high-risk zone, an extremely high-risk warning message is generated. This message is used to trigger a strong visual and audible alarm on the interactive terminal, prompting medical staff to immediately manually stop the injection. The message also includes the current predicted probability value and risk level indicator for manual review by medical staff after receiving the message. The system also outputs intervention prompts: talk to the patient and their family to inform them of the risk of extravasation, assess the risks and benefits, and change the intravenous catheter or use other imaging examination methods as appropriate.

[0050] After generating a medium-risk or extremely high-risk extravasation warning message, the message is output to the injection control terminal via a preset communication interface. Upon receiving the medium-risk extravasation warning message, the injection control terminal parses the suggested flow rate limit parameters and highlights them on the display interface, prompting medical staff to confirm whether to manually adjust the injection flow rate. Upon receiving the extremely high-risk warning message, a confirmation pop-up window for interrupting the injection appears on the display interface, prompting medical staff to confirm whether to immediately and manually interrupt the contrast agent injection. Simultaneously, the risk level and current predicted probability value are displayed on the interactive interface of the injection control terminal to help medical staff quickly grasp the current extravasation risk status of the patient and take subsequent clinical measures.

[0051] Example 8 like Figure 3 As shown in the embodiment of the present invention, a system for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination is provided. The system includes: The data acquisition module is used to acquire environmental temperature difference characteristics, waiting time characteristics, venous blood vessel characteristics, puncture consumable characteristics, and contrast agent injection parameter data of the subject to be examined. The temperature difference calibration module is used to perform attenuation and weight reduction processing on the environmental temperature difference feature data based on the waiting time feature data to obtain dynamic temperature difference calibration feature data. The attenuation and weight reduction processing is used to characterize the logical relationship that the stress effect of environmental temperature difference on blood vessels weakens as the waiting time increases. The feature cross-processing module is used to perform feature cross-processing on venous vessel feature data and puncture consumable feature data to extract coupling risk feature data that characterizes the hydrodynamic resistance of puncture consumables in venous vessels. The feature fusion module is used to splice and fuse other basic feature data that did not participate in feature cross-processing from dynamic temperature difference calibration feature data, coupling risk feature data, venous blood vessel feature data, and puncture consumables feature data to generate a target feature vector. The probability prediction module is used to input the target feature vector into the pre-trained extravasation risk prediction model and output the predicted probability value of the subject's peripheral venous iodine contrast agent extravasation. The threshold determination module is used to determine a dynamic warning threshold that matches the current injection status based on the contrast agent injection parameter data. The dynamic warning threshold decreases as the injection flow rate in the contrast agent injection parameter data increases. The early warning control module is used to compare the predicted probability value with the dynamic early warning threshold, generate corresponding extravasation risk early warning information based on the comparison result, and output it to the interactive terminal.

[0052] Example 9 An electronic device, comprising a memory and a processor; Memory, used to store computer programs; A processor, when executing a computer program, to implement a method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination, as described in any of Examples 1-7.

[0053] Example 10 A computer-readable storage medium storing a computer program that, when read and run by a processor, implements a method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination, as described in any one of Examples 1-7.

[0054] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination, characterized in that, The method includes: Acquire environmental temperature difference characteristics, waiting time characteristics, venous blood vessel characteristics, puncture consumable characteristics, and contrast agent injection parameter data of the subject to be examined; Based on the waiting time feature data, the environmental temperature difference feature data is attenuated and weighted to obtain dynamic temperature difference calibration feature data. The attenuation and weighting process is used to characterize the logical relationship that the stress effect of environmental temperature difference on blood vessels weakens with the increase of waiting time. Feature cross-processing is performed on venous vessel feature data and puncture consumable feature data to extract coupling risk feature data characterizing the hydrodynamic resistance of puncture consumables in venous vessels; The target feature vector is generated by splicing and fusing other basic feature data that did not participate in feature cross-processing from dynamic temperature difference calibration feature data, coupling risk feature data, venous blood vessel feature data and puncture consumable feature data. The target feature vector is input into a pre-trained extravasation risk prediction model, which outputs the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination. Based on the contrast agent injection parameter data, a dynamic warning threshold matching the current injection status is determined. The dynamic warning threshold decreases as the injection flow rate in the contrast agent injection parameter data increases. The predicted probability value is compared with the dynamic early warning threshold. Based on the comparison result, a corresponding early warning message for the risk of extravasation is generated and output to the interactive terminal.

2. The method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination according to claim 1, characterized in that, The environmental temperature difference feature data is attenuated and weighted based on waiting time feature data to obtain dynamic temperature difference calibration feature data, including: Extract the initial value of the ambient temperature difference from the ambient temperature difference feature data, and the waiting time in the constant temperature zone from the waiting time feature data; Obtain the preset baseline thermal stress half-life parameters, and calculate the dynamic decay coefficient based on the natural exponential decay logic using the waiting time in the isothermal zone and the baseline thermal stress half-life parameters; The initial value of the ambient temperature difference is multiplied by the dynamic attenuation coefficient to obtain dynamic temperature difference calibration feature data that has been smoothed and weighted down over time.

3. The method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination according to claim 1, characterized in that, The process of performing feature cross-processing on venous vessel characteristic data and puncture consumable characteristic data to extract coupling risk characteristic data characterizing the hydrodynamic resistance of puncture consumables within venous vessels includes: Extract vein quality grading indicators from vein vascular feature data, and indwelling needle type and insertion depth status from puncture consumables feature data; Based on the venous quality grading index and the indwelling needle type index, a basic fluid dynamics resistance matrix is ​​constructed; When the insertion depth is determined to be incomplete and the vein quality grading index is lower than the preset quality standard, the resistance penalty mechanism is triggered to generate a non-linear penalty coefficient. By using a nonlinear penalty coefficient to weight and correct the basic fluid dynamics resistance matrix, coupled risk characteristic data can be extracted.

4. The method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination according to claim 1, characterized in that, The step of inputting the target feature vector into a pre-trained extravasation risk prediction model and outputting a predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination includes: Obtain the pre-calibrated multidimensional feature weight vector and global intercept constant of the extravasation risk prediction model; The feature values ​​of each dimension in the target feature vector are multiplied one by one with the corresponding weight values ​​in the multidimensional feature weight vector; The summation of the products obtained by multiplying each term yields a comprehensive feature weighting value. The weighted value of the comprehensive features is added to the global intercept constant to obtain the initial predicted value that characterizes the tendency of exudation. Based on the initial prediction value, the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination is determined.

5. The method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination according to claim 4, characterized in that, The step of determining the predicted probability value of peripheral venous iodine contrast agent extravasation in the subject under examination based on the initial predicted value includes: The initial predicted value is negative and then used as the exponent of the natural constant for exponentiation to obtain the exponent-converted value. The denominator data is constructed by adding the preset unit constant and the exponential conversion value, and the reciprocal of the denominator data is calculated to obtain the predicted probability value mapped to the value range of zero to one.

6. The method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination according to claim 1, characterized in that, The step of determining a dynamic early warning threshold that matches the current injection status based on contrast agent injection parameter data includes: Extract the baseline injection flow rate from the contrast agent injection parameter data; The threshold offset is calculated based on the baseline injection flow rate. The higher the baseline injection flow rate, the greater the negative increase in the threshold offset. The preset benchmark medium-risk threshold and benchmark high-risk threshold are added to the threshold offset respectively to calculate the dynamic medium-risk threshold and dynamic high-risk threshold. Based on dynamic medium-risk thresholds and dynamic high-risk thresholds, a dynamic early warning threshold system is constructed that includes low-risk, medium-risk, and high-risk intervals.

7. The method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination according to claim 1, characterized in that, The step of comparing the predicted probability value with the dynamic early warning threshold, generating corresponding extravasation risk early warning information based on the comparison result, and outputting it to the interactive terminal includes: Determine the risk range within which the predicted probability value falls under the dynamic early warning threshold; If the risk level is determined to be in the medium-risk range, a medium-risk deceleration prompt message containing suggested flow rate limit parameters will be generated. If the area is determined to be in a high-risk zone, an extremely high-risk warning message will be generated. The generated medium-risk deceleration prompt or extremely high-risk warning prompt will be output to the display interface of the interactive terminal to prompt medical staff to confirm whether it is necessary to manually adjust the injection parameters or interrupt the operation.

8. A system for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination, characterized in that, The system includes: The data acquisition module is used to acquire environmental temperature difference characteristics, waiting time characteristics, venous blood vessel characteristics, puncture consumable characteristics, and contrast agent injection parameter data of the subject to be examined. The temperature difference calibration module is used to perform attenuation and weight reduction processing on the environmental temperature difference feature data based on the waiting time feature data to obtain dynamic temperature difference calibration feature data. The attenuation and weight reduction processing is used to characterize the logical relationship that the stress effect of environmental temperature difference on blood vessels weakens as the waiting time increases. The feature cross-processing module is used to perform feature cross-processing on venous vessel feature data and puncture consumable feature data to extract coupling risk feature data that characterizes the hydrodynamic resistance of puncture consumables in venous vessels. The feature fusion module is used to splice and fuse other basic feature data that did not participate in feature cross-processing from dynamic temperature difference calibration feature data, coupling risk feature data, venous blood vessel feature data, and puncture consumable feature data to generate a target feature vector. The probability prediction module is used to input the target feature vector into the pre-trained extravasation risk prediction model and output the predicted probability value of the subject's peripheral venous iodine contrast agent extravasation. The threshold determination module is used to determine a dynamic warning threshold that matches the current injection status based on the contrast agent injection parameter data. The dynamic warning threshold decreases as the injection flow rate in the contrast agent injection parameter data increases. The early warning control module is used to compare the predicted probability value with the dynamic early warning threshold, generate corresponding extravasation risk early warning information based on the comparison result, and output it to the interactive terminal.

9. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to, when executing the computer program, implement the method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when read and run by a processor, implements the method for predicting extravasation of iodine contrast agent in peripheral veins during enhanced CT examination as described in any one of claims 1-7.

Images