Prediction method and device for re-blockage of blood vessel, medium, terminal and program product

Abstract

The embodiment of the invention provides a blood vessel re-occlusion prediction method and device, a medium, a terminal and a program product, and the method comprises the steps: obtaining a sample set, enabling a sample in the sample set to comprise a plurality of risk features, enabling each risk feature to have a plurality of preset sub-feature types, the individual feature data of each risk feature can belong to a single sub-feature type of the risk feature; and inputting the sub-feature type group of at least one part of samples in the sample set and the actual blood vessel re-occlusion value of the input sample into a to-be-trained machine learning model for training to obtain a trained machine learning model, the sub-feature type group is used for representing a combination of sub-feature types of each risk feature of the sample to which the sub-feature type group belongs; and inputting all non-repeated sub-feature type groups into the trained machine learning model to obtain the blood vessel re-occlusion probability of each sub-feature type group. By adopting the technical scheme, the probability of blood vessel re-blockage can be predicted.

Description

Technical Field

[0001] The disclosed embodiments relate to the field of clinical prediction technology, and in particular to a method, device, medium, terminal and program product for predicting vascular reocclusion. Background Art

[0002] Vascular occlusion refers to the situation where blood flow in a blood vessel is blocked or completely stopped, which is mainly caused by thrombosis, inflammation of the blood vessel wall, etc. Stent implantation can quickly restore the patency of blood vessels by inserting stents into blood vessels, so it is widely used. However, studies have found that blood vessels are prone to re-blockage after stent implantation.

[0003] Therefore, how to predict the probability of vascular re-blockage has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the embodiments of the present disclosure provide a method, device, medium, terminal and program product for predicting vascular reocclusion, which can predict the probability of vascular reocclusion.

[0005] In order to solve the above technical problems, the present disclosure provides a method for predicting blood vessel reocclusion, the prediction method comprising:

[0006] Acquire a sample set, wherein the samples in the sample set include multiple risk features, each risk feature has multiple preset sub-feature types, wherein individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature;

[0007] Using the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample, inputting them into the machine learning model to be trained for training, so as to obtain a trained machine learning model, wherein the sub-feature type group is used to represent the combination of sub-feature types of each risk feature of the sample;

[0008] All non-repetitive sub-feature type groups are input into the trained machine learning model to obtain the probability of vascular re-occlusion for each sub-feature type group.

[0009] Optionally, a sample set is obtained, where samples in the sample set contain multiple risk features, including:

[0010] Acquire a sample set, wherein the samples in the sample set include multiple initial risk features;

[0011] The multiple initial risk characteristics of each sample are preprocessed to obtain the multiple risk characteristics of the sample.

[0012] Optionally, preprocessing is performed on each sample using a variety of initial risk features, including:

[0013] The initial individual characteristic data of each initial risk feature in the sample set is interpolated so that each initial risk feature in the sample set has corresponding initial individual characteristic data.

[0014] Optionally, preprocessing is performed on each sample using a variety of initial risk features, including:

[0015] The multiple initial risk characteristics of each sample are transformed and preprocessed to obtain the multiple risk characteristics of the sample.

[0016] Optionally, individual characteristic data of some risk characteristics are continuous; preprocessing multiple initial risk characteristics of each sample also includes:

[0017] The risk characteristics with continuous individual characteristic data are truncated to obtain the individual characteristic data truncation value;

[0018] The individual characteristic data cutoff value is used to divide the individual characteristic data of the risk characteristic so that the individual characteristic data of the risk characteristic can be attributed to a single sub-characteristic type of the risk characteristic.

[0019] Optionally, preprocessing of the multiple initial risk features of each sample may also include:

[0020] A single screening pretreatment method was used to screen and pretreat multiple risk characteristics of each sample;

[0021] Based on the screening results of each sample, various risk characteristics of the sample are obtained.

[0022] Optionally, preprocessing of the multiple initial risk features of each sample may also include:

[0023] Multiple screening and pretreatment methods were used to screen and preprocess multiple risk characteristics of each sample;

[0024] Based on the screening results of various screening pretreatment methods for each sample, various risk characteristics of the sample are determined.

[0025] Optionally, the multiple risk characteristics of the sample are obtained by taking the union of the screening results of various screening preprocessing methods for each sample;

[0026] Alternatively, the multiple risk characteristics of the sample are obtained by taking the intersection of the screening results of various pre-processing methods for each sample.

[0027] Optionally, the samples in the sample set include: unobstructed samples and complex-blocked samples;

[0028] Preprocessing of various initial risk features for each sample also includes:

[0029] The number of the unobstructed samples and the number of the complex-blocked samples are balanced so that the ratio of the number of complex-blocked samples to the number of unobstructed samples after the balancing process is in the range of [0.9, 1].

[0030] Optionally, the number of risk features of the sample is N, and each risk feature has M sub-feature types, and the number of all non-repeating sub-feature type groups is M to the power of N;

[0031] Wherein, N is a positive integer, and N>1; M is a positive integer, and M>1.

[0032] Optionally, the risk feature is selected from: age; cause of lesion; lesion length; preoperative diameter stenosis.

[0033] Optionally, the sample set includes a first part of samples and a second part of samples, and the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample are input into the machine learning model to be trained for training, including:

[0034] Using the sub-feature type group of the first part of samples and the actual value of the vascular re-occlusion of the input sample, input a plurality of preset initial machine learning models for pre-training, and obtain a pre-trained machine learning model corresponding to each initial machine learning model;

[0035] Using the sub-feature type group of the second part of samples, inputting each pre-trained machine learning model for testing, so as to select a target machine learning model from each pre-trained machine learning model;

[0036] The sub-feature type groups of all samples in the sample set and the actual values ​​of vascular re-occlusion of the input samples are input into the target machine learning model for retraining to obtain the trained machine learning model.

[0037] Optionally, the prediction method further includes:

[0038] Creating a re-blockage prediction table, wherein the re-blockage prediction table has a first area and a second area, wherein the first area corresponds to a first prediction result, the first prediction result is used to indicate that the blood vessel will be re-blocked, and the second area corresponds to a second prediction result, which is used to indicate that the blood vessel will remain unobstructed;

[0039] If the blood vessel re-blockage probability of the sub-feature type group is greater than or equal to the re-blockage probability threshold, the sub-feature type group belongs to the first area;

[0040] If the blood vessel re-blockage probability of the sub-feature type group is less than the re-blockage probability threshold, the sub-feature type group belongs to the second area.

[0041] Optionally, the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample are input into the machine learning model to be trained for training, including:

[0042] Using the sub-feature type group of at least a part of the samples in the balanced sample set and the actual value of the vascular re-occlusion of the input sample, inputting them into the machine learning model to be trained for training;

[0043] The prediction method further comprises:

[0044] Using the complex blockage prediction table to predict the samples in the sample set before balancing processing, and outputting the corresponding prediction results;

[0045] According to the actual value of blood vessel re-occlusion of the sample before the balancing process and the predicted result, an accuracy model is obtained, and the accuracy model is used to evaluate the accuracy of the re-occlusion prediction table.

[0046] Optionally, the accuracy model has a first parameter, a second parameter, a third parameter, a fourth parameter and a fifth parameter;

[0047] The first parameter is the ratio of the number of samples whose actual values ​​of vascular reocclusion are consistent with the predicted results to the total number of samples in the sample set;

[0048] The second parameter is the ratio of the number of samples where both the actual value of vascular re-blocking and the predicted result are that the vessel is unobstructed to the number of samples where the actual value of vascular re-blocking is that the vessel is unobstructed;

[0049] The third parameter is the ratio of the number of samples where both the actual value of vascular re-blocking and the predicted result are vascular re-blocking to the number of samples where the actual value of vascular re-blocking is vascular re-blocking;

[0050] The fourth parameter is the ratio of the number of samples whose actual values ​​and predicted results are both vascular re-occlusion and the number of samples whose predicted results are vascular re-occlusion;

[0051] The fifth parameter is the ratio of the number of samples whose actual values ​​and predicted results are both unobstructed blood vessels to the number of samples whose predicted results are re-obstructed blood vessels.

[0052] Optionally, the prediction method further includes:

[0053] Performing correlation analysis on the trained machine learning model to determine a correlation model, wherein the correlation model includes one or more of the following:

[0054] The correlation between a single sub-feature type and the probability of vascular reocclusion;

[0055] The correlation between all sub-feature types and the probability of vascular reocclusion

[0056] The correlation between each type and the probability of vascular reocclusion.

[0057] Optionally, the blood vessel comprises a deep vein.

[0058] Optionally, the sample is a sample taken within a preset time after the stent implantation surgery.

[0059] The present disclosure provides a device for predicting blood vessel reocclusion, comprising:

[0060] A sample set acquisition module, which acquires a sample set, wherein the samples in the sample set include multiple risk features, each risk feature has multiple preset sub-feature types, wherein individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature;

[0061] A training module, using a sub-feature type group of at least a part of the samples in the sample set and an actual value of vascular re-occlusion of the input sample, and inputting them into a machine learning model to be trained for training, so as to obtain a trained machine learning model, wherein the sub-feature type group is used to represent a combination of sub-feature types of various risk features of the samples;

[0062] The re-blockage probability determination module inputs all non-repetitive sub-feature type groups into the trained machine learning model to obtain the vascular re-blockage probability of each sub-feature type group.

[0063] An embodiment of the present disclosure provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the prediction method described in any of the above embodiments are executed.

[0064] An embodiment of the present disclosure provides a terminal, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that the processor executes the steps of the prediction method described in any of the above embodiments when running the computer program.

[0065] An embodiment of the present disclosure provides a computer program product, including a computer program, which executes the steps of the prediction method described in any of the above embodiments when the computer program is run.

[0066] Compared with the prior art, the technical solution of the embodiment of the present disclosure has the following advantages:

[0067] In the technical solution of the disclosed embodiment, by obtaining the probability of vascular re-blockage for each sub-feature type group, the probability of vascular re-blockage in patients corresponding to each sub-feature type group can be predicted in advance, so that follow-up treatment suggestions can be made to the patients in advance, thereby reducing the chance of vascular re-blockage.

[0068] Furthermore, the sub-feature type group to which the patient belongs is first determined, and then the region in which the sub-feature type group is located in the re-blockage prediction table is determined. If the sub-feature type group is in the first region, a first prediction result is obtained, predicting that the patient's blood vessel will be re-blocked within a preset time. If the sub-feature type group is in the second region, a second prediction result is obtained, predicting that the patient's blood vessel will remain unobstructed within a preset time. Each sub-feature type group is assigned to the first region and the second region of the re-blockage prediction table so that the probability of re-blockage of the blood vessel of each sub-feature type group corresponds to the first prediction result and the second prediction result, which is conducive to simplifying the doctor's decision-making process, improving execution efficiency, and facilitating patient understanding and communication between doctors and patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the drawings required for use in the embodiments of this specification or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 A schematic diagram showing a flow chart of a method for predicting blood vessel reocclusion in an embodiment of the present disclosure is shown;

[0071] Figure 2 A schematic diagram showing the correlation between a risk feature output by a Boruta method and vascular reocclusion in an embodiment of the present disclosure is shown;

[0072] Figure 3 A schematic diagram of a complex blockage prediction table in an embodiment of the present disclosure is shown;

[0073] Figure 4 A schematic diagram of an accuracy model in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0074] Vascular obstruction refers to the situation where the blood flow in a blood vessel is blocked or completely stopped. During treatment, stents are placed in the blood vessels to quickly restore the patency of the blood vessels, so they are widely used. For example, for iliac vein obstruction, an iliac vein stent can be placed in the iliac vein through iliac vein stent placement (IVSP), thereby quickly restoring the patency of the iliac vein.

[0075] However, studies have found that blood vessels are prone to re-blockage after stent implantation, which can increase the risk of subsequent treatment for patients.

[0076] The disclosed embodiment provides a method for predicting vascular reocclusion, comprising the following steps: obtaining a sample set, wherein the samples in the sample set contain multiple risk features, each risk feature has multiple preset sub-feature types, wherein the individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature; using the sub-feature type group of at least a part of the samples in the sample set and the actual value of vascular reocclusion of the input sample, inputting into the machine learning model to be trained for training, so as to obtain the trained machine learning model, wherein the sub-feature type group is used to represent the combination of sub-feature types of each risk feature of the sample; inputting all non-repeating sub-feature type groups into the trained machine learning model, so as to obtain the probability of vascular reocclusion of each sub-feature type group. By adopting the above prediction method, by obtaining the probability of vascular reocclusion of each sub-feature type group, the probability of vascular reocclusion of the patient corresponding to each sub-feature type group can be predicted in advance, so that subsequent treatment suggestions can be made to the patient in advance, thereby reducing the probability of vascular reocclusion.

[0077] In order to make the above-mentioned objects, features and beneficial effects of the present disclosure more obvious and easy to understand, the specific embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0078] Combined with reference Figure 1 ,in, Figure 1 1 is a flow chart of a method for predicting blood vessel reocclusion in an embodiment of the present disclosure. The method for predicting blood vessel reocclusion may include steps S1 to S3:

[0079] Step S1: obtaining a sample set, wherein the samples in the sample set include multiple risk features, each risk feature has multiple preset sub-feature types, wherein individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature;

[0080] Step S2: using the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample, inputting them into the machine learning model to be trained for training, so as to obtain a trained machine learning model, wherein the sub-feature type group is used to represent the combination of sub-feature types of each risk feature of the sample;

[0081] Step S3: inputting all non-repetitive sub-feature type groups into the trained machine learning model to obtain the probability of blood vessel re-occlusion for each sub-feature type group.

[0082] This application is mainly used for a prediction scenario of vascular reocclusion based on a machine learning model. The following is an explanation of some terms in this application:

[0083] 1) Risk characteristics are individual characteristics associated with vascular reocclusion, and the individual characteristics may include, for example, human characteristics, and the human characteristics may include age, lesion etiology, lesion length, diameter stenosis, etc.

[0084] 2) The individual characteristic data is the value corresponding to the risk characteristic of the sample. For example, when the risk characteristic is age, the individual characteristic data of age is the age value. For example, when the risk characteristic is the cause of the lesion, the individual characteristic data of the cause of the lesion is thrombosis (PT) or non-thrombosis (NT).

[0085] 3) Sub-feature type is a secondary classification under risk feature, and individual feature data of risk feature can be attributed to a single sub-feature type of the risk feature.

[0086] 4) A sub-feature type group is a collection of sub-feature types corresponding to each risk feature. Different sub-feature type groups may have the same number of sub-feature types or different numbers of sub-feature types.

[0087] In some embodiments, different sub-feature types within a sub-feature type group may belong to different risk features.

[0088] 5) The actual value of blood vessel reocclusion is the result of whether the patient has undergone stent implantation surgery at a preset time in the past. The statistics include the results of whether the blood vessel is reoccluded or not, including the two values ​​of blood vessel reocclusion and blood vessel remaining unobstructed;

[0089] 6) The probability of vascular reocclusion is the probability that the patient may experience vascular reocclusion within a preset time in the future after stent implantation surgery.

[0090] In some embodiments, the blood vessel may be a blood vessel that can undergo a stent placement procedure, for example, the blood vessel may be a deep vein such as the iliac vein, femoral vein or jugular vein.

[0091] It should be noted that, in the following description, the iliac vein is used as an example for predicting the probability of blood vessel re-occlusion, but this does not constitute a limitation on the types of blood vessels that can be predicted by the prediction method.

[0092] In some embodiments, each sample in the sample set has undergone a stent implantation surgery before a first preset time. Accordingly, the prediction method is used to predict the probability of a patient experiencing vascular re-blockage within a second preset time after the stent implantation surgery, and the first preset time is the same as the second preset time.

[0093] For example, the samples in the sample set had undergone stent implantation surgery 12 months ago. Accordingly, the prediction method is used to predict the probability of vascular re-occlusion in patients within 12 months after the stent implantation surgery.

[0094] In some embodiments, the samples in the sample set may include unobstructed samples and re-blocked samples.

[0095] Among them, the unobstructed samples had undergone stent implantation surgery before the preset time, and the blood vessels remained unobstructed at the time of statistics; the re-blocked samples had undergone stent implantation surgery before the preset time, and the blood vessels had been re-blocked at the time of statistics.

[0096] In some embodiments, obtaining a sample set may include: obtaining a sample set, where samples in the sample set contain multiple initial risk features.

[0097] The sample set includes multiple samples, and each sample has a consistent number and type of initial risk features.

[0098] The initial risk characteristics have a certain correlation with vascular reocclusion, such as a positive correlation or a negative correlation.

[0099] The initial risk characteristics are human characteristics obtained by direct detection, such as age (Age), lesion cause (Lesion etiology), lesion length (Lesion length), diameter stenosis (Diameter stenosis), etc., as well as stent characteristics obtained directly, such as stent length (Stent length), etc.

[0100] The initial risk feature has corresponding initial individual characteristic data, for example, when the initial risk feature is age, the initial individual characteristic data of age is an age value. For example, when the initial risk feature is the cause of the lesion, the initial individual characteristic data of the cause of the lesion is thrombosis (PT) or non-thrombosis (NT).

[0101] In a specific implementation, the multiple initial risk features of each sample in the sample set may be preprocessed so that each sample in the sample set contains multiple risk features. The preprocessing may include at least one of interpolation processing, conversion processing, truncation processing, screening processing, balancing processing, and the like.

[0102] In some embodiments, the initial individual characteristic data of each initial risk feature in the sample set may be interpolated so that each initial risk feature in the sample set has corresponding initial individual characteristic data.

[0103] Specifically, the individual characteristic data of the initial risk characteristics in the sample set can be obtained through archives, follow-up visits and re-examinations. Some of the initial risk characteristics in the sample set have the problem of missing individual characteristic data. The missing individual characteristic data of some initial risk characteristics are supplemented by interpolation processing, which can reduce the probability of missing initial individual characteristic data of each initial risk characteristic in the sample set, thereby improving the accuracy of subsequent training.

[0104] In some embodiments, the interpolation process may be implemented by a multiple interpolation method.

[0105] Specifically, the multiple imputation method is a commonly used and reliable statistical method, which can be implemented through the 'mice' package of R language.

[0106] In some embodiments, a plurality of initial risk features of each sample may be pre-processed by transformation to obtain a plurality of risk features of the sample.

[0107] Specifically, among the multiple risk characteristics of the sample, some risk characteristics are initial risk characteristics, and the individual characteristic data of these risk characteristics are the corresponding initial individual characteristic data. For example, when the risk characteristic is age, the individual characteristic data of age is consistent with the initial individual characteristic data, both of which are age values; the remaining risk characteristics are converted from the initial risk characteristics according to relevant formulas, and the individual characteristic data of these risk characteristics are converted from the corresponding initial individual characteristic data according to relevant formulas. For example, when the risk characteristic is body mass index (Body Mass Index, BMI), the BMI index is the ratio of weight to height squared, the individual characteristic data of the BMI index is the value of the BMI index, the individual characteristic data of weight is the weight value, the individual characteristic data of height is the height value, and the value of the BMI index is the ratio of weight value to height squared.

[0108] Compared with direct use, the risk characteristics formed after the transformation of some initial risk characteristics can more comprehensively reflect the correlation with vascular re-occlusion.

[0109] In some embodiments, after the conversion preprocessing, each sample has a consistent number and type of risk features.

[0110] In some embodiments, the number of risk features of each sample after conversion preprocessing is consistent with, greater than, or less than the number of initial risk features before conversion preprocessing.

[0111] In some embodiments, the risk feature has a plurality of preset sub-type features, and the individual feature data of the risk feature can be attributed to a single sub-feature type of the risk feature.

[0112] Specifically, the subtype feature of the risk feature is used to form a subtype feature group of the sample to which it belongs together with the subtype features of other risk features of the sample to which it belongs.

[0113] In some embodiments, at least part of the risk features are categorical variables, and individual characteristic data of the categorical variables are selected from a limited number of values.

[0114] For example, the risk feature may be the cause of the lesion, and the value of the cause of the lesion may be non-thrombotic (NT) and thrombotic (PT). The lesion length has a preset first sub-feature type and a second sub-feature type. When the value of the cause of the lesion is non-thrombotic (NT), it belongs to the first sub-feature type, and when the value of the cause of the lesion is thrombotic (PT), it belongs to the second sub-feature type.

[0115] For example, the risk feature may be diameter stenosis, and the value of diameter stenosis may be occlusion and stenosis. Diameter stenosis has a preset first sub-feature type and a second sub-feature type. When the value of diameter stenosis is occlusion, it belongs to the first sub-feature type, and when the value of diameter stenosis is stenosis, it belongs to the second sub-feature type.

[0116] In some embodiments, at least part of the risk feature is a continuous variable, and the individual feature data of the risk feature can be selected within a continuous interval. For example, the risk feature can be age, and the value of age can be selected within the interval of 1 to 100 years old. For example, the risk feature can be lesion length, and the value of lesion length can be selected within the interval of 1 millimeter (mm) to 500 mm.

[0117] In some embodiments, the risk characteristic is determined to be a continuous variable, and the risk characteristic is truncated to obtain an individual characteristic data truncation value, wherein the individual characteristic data truncation value is used to divide the individual characteristic data of the corresponding risk characteristic so that the individual characteristic data of the risk characteristic can be attributed to a single sub-characteristic type of the risk characteristic.

[0118] For example, the risk feature is lesion length, and the lesion length has a preset first sub-feature type and a second sub-feature type. The individual feature data cutoff value corresponding to the lesion length is 110 millimeters (mm). When the lesion length value is less than 110 mm, the lesion length value belongs to the first sub-feature type of the lesion length. When the lesion length value is greater than or equal to 110 mm, the lesion length value belongs to the second sub-feature type of the lesion length.

[0119] For example, the risk feature is age, and age has a preset first sub-feature type and a second sub-feature type. The individual feature data cutoff value corresponding to age is 55 years old. When the age value is less than 55 years old, the age value belongs to the first sub-feature type of age. When the age value is greater than or equal to 55 years old, the age value belongs to the second sub-feature type of age.

[0120] By setting the individual feature data cutoff value, the individual feature data of the risk feature is attributed to a single sub-feature type of the risk feature, so that the risk feature can be converted from a continuous variable to a categorical variable. Compared with a continuous variable, converting the risk feature into a categorical variable is beneficial to reducing the amount of subsequent calculations and improving the training efficiency of the machine learning model.

[0121] In some embodiments, the R language 'cutoff' package can be used to obtain the individual characteristic data cutoff value of the risk feature. Among them, the R language 'cutoff' package is a tool for determining the best split point (cutoff value) in the data (in the embodiment of the present disclosure, the individual characteristic data cutoff value is the best split point for the individual characteristic data), especially in classification, regression analysis and bioinformatics research. These split points are usually used to convert continuous variables into categorical variables for statistical modeling or data analysis.

[0122] In some embodiments, the multiple risk features of each sample may be screened and preprocessed to obtain the multiple risk features of the sample. Through screening and preprocessing, the number of risk features used to train the machine learning model can be reduced, thereby reducing the amount of later calculations and improving the training efficiency of the machine learning model.

[0123] In some embodiments, after screening pre-processing, the number of risk features possessed by each sample is consistent, and the types of risk features possessed by each sample are consistent.

[0124] In some embodiments, the screening pretreatment method can be selected from: Boruta method.

[0125] Specifically, the correlation between risk features and vascular reocclusion can be statistically analyzed according to the Boruta method, which is a feature selection algorithm that is mainly used to screen out the risk features that are most relevant to the target variable from a high-dimensional data set. This method is based on random forests and evaluates the importance of risk features by comparing the importance of actual risk features with randomly generated "shadow features".

[0126] In a specific example, referring to Figure 2 , Figure 2A schematic diagram showing the correlation between the risk characteristics output by the Boruta method and the reocclusion of blood vessels in an embodiment of the present disclosure is shown. Figure 2 The four most relevant risk features were screened out ( Figure 2 The four risk features are shown in green in the figure. The order of the four risk features from high to low is: cause of lesion, preoperative diameter stenosis, age, and lesion length.

[0127] In some embodiments, the screening preprocessing method can be selected from: Cox regression.

[0128] Specifically, the correlation between risk characteristics and vascular reocclusion can be statistically analyzed according to Cox regression. Cox regression (Cox Proportional Hazards Regression), also known as the proportional hazard model, is a commonly used survival analysis method used to study the impact of variables (reocclusion) that affect time-to-event data.

[0129] In a specific example, the analysis results are shown in Table 1. The P value under the multi-feature analysis obtained by Cox regression is used to characterize the correlation between the risk feature and vascular re-occlusion. The smaller the P value, the greater the correlation between the risk feature and vascular re-occlusion. The larger the P value, the smaller the correlation between the risk feature and vascular re-occlusion. Among them, age, lesion cause and lesion length with a P value less than 0.05 under the multi-feature analysis are screened as the risk features with the highest correlation.

[0130] Table 1

[0131]

[0132] It should be noted that according to the expression habits of medical statistics, if the P value is greater than 0.05, it will be displayed as blank.

[0133] In some embodiments, a single screening preprocessing method can be used to screen and preprocess multiple risk features of each sample; based on the screening results of each sample, multiple risk features of the sample are obtained. This helps to reduce the amount of calculation and improve the screening efficiency.

[0134] In some embodiments, multiple screening preprocessing methods can be used to screen and preprocess multiple risk characteristics of each sample respectively, and then based on the screening results of various screening preprocessing methods of each sample, the multiple risk characteristics of the sample are determined.

[0135] For example, the multiple risk characteristics of the sample can be obtained by taking the union of the screening results of various screening preprocessing methods for each sample. Determining the multiple risk characteristics of the sample by taking the union of multiple screening results obtained by multiple screening preprocessing methods can reduce the defects of a single screening preprocessing method and reduce the probability of screening out risk characteristics with strong actual correlation.

[0136] For example, the multiple risk characteristics of the sample can be obtained by taking the intersection of the screening results of various pre-processing methods for each sample. Using the intersection of multiple screening results obtained by multiple screening pre-processing methods to determine the multiple risk characteristics of the sample can reduce the defects of a single screening pre-processing method and reduce the probability of screening out risk characteristics with weak actual correlation.

[0137] It should be noted that according to statistical analysis, within a preset time, for example, within 1 year, the ratio of patients whose blood vessels become re-blocked to those whose blood vessels remain patent is less than 0.1.

[0138] In some embodiments, the number of the unobstructed samples and the number of the complex-blocked samples may be balanced so that the ratio of the number of complex-blocked samples to the number of unobstructed samples after the balancing process is in the range of [0.9, 1].

[0139] Through balancing processing, the difference in the number of re-blocked samples and the patent samples can be reduced, and the ratio of the number of re-blocked samples to the number of patent samples can be increased. Compared with the probability of vascular re-blockage in each sub-feature type group output by the machine learning model trained with the samples in the sample set before balancing processing, the probability of vascular re-blockage in each sub-feature type group output by the machine learning model trained with the samples in the sample set after balancing processing is higher.

[0140] In some embodiments, the balancing processing step may include: obtaining a balancing multiple, wherein the balancing multiple is the integer part of the ratio of the number of unobstructed samples to the number of complex-blocked samples in the sample set before the balancing processing; based on the complex-blocked samples in the sample set before the balancing processing, simulating and generating new complex-blocked samples, wherein the number of complex-blocked samples after the balancing processing is the product of the balancing multiple and the number of complex-blocked samples in the sample set before the balancing processing.

[0141] For example, there are 148 unobstructed samples and 14 complex-blocked samples in the sample set before balancing treatment, and the balancing multiple is 10. According to the data characteristics of the 14 complex-blocked samples, 126 samples are simulated and generated using the SMOTE method. In the sample set after balancing treatment, the number of unobstructed samples is 148, and the number of complex-blocked samples is 140.

[0142] Among them, SMOTE (Synthetic Minority Oversampling Technique) is a sampling method used to process unbalanced data sets. It increases the number of minority class samples by generating synthetic samples, which can improve the performance of machine learning models in classification tasks, especially the prediction ability of minority classes.

[0143] In some embodiments, the number of risk features of the sample is N, and each risk feature has M sub-feature types, and the number of all non-repetitive sub-feature type groups is M to the power of N, where M is a positive integer and M>1; N is a positive integer and N>1.

[0144] For example, the sample has four risk characteristics A, B, C, and D. Risk characteristic A has a first sub-feature type A1 and a second sub-feature type A2. Risk characteristic B has a first sub-feature type B1 and a second sub-feature type B2. Risk characteristic C has a first sub-feature type C1 and a second sub-feature type C2. Risk characteristic D has a first sub-feature type D1 and a second sub-feature type D2. Then the number of all non-repeating sub-feature type groups is 2 to the fourth power. The 14 sub-feature type groups are [A1+B1+C1+D1], [A1+B1+C1+D2], [A1+B1+C2+D1], [A1+B1+C2+D2], [A1+B2+C1+D1], [A1+B2+C1+D2], [A1+B2+C2+D1], [A1+B2+C2+D2], [A2+B1+C1+D1], [A2+B1+C1+D2], [A2+B1+C2+D1], [A2+B1+C2+D2], [A2+B2+C1+D1], [A2+B2+C1+D2], [A2+B2+C2+D1], [A2+B2+C2+D2].

[0145] In some embodiments, the machine learning model to be trained can be selected from: LR (Logistic Regression) model, RF (Random Forest) model, SVM (Support Vertor Machine) model, KNN (K-Nearest Neighbors) model and RT (Regression Tree) model and other models applied to medical binary classification problems.

[0146] In some embodiments, the samples in the sample set are divided into a first portion of samples and a second portion of samples.

[0147] The sub-feature type group of the first part of the samples and the actual value of the vascular re-blockage of the input sample are used to input multiple preset initial machine learning models for pre-training, and the pre-trained machine learning models corresponding to each initial machine learning model are obtained; the sub-feature type group of the second part of the samples are used to input each pre-trained machine learning model for testing, so as to select a target machine learning model from each pre-trained machine learning model. The target machine learning model is used to continue the subsequent operation.

[0148] It should be noted that the performance of the machine learning model was evaluated by the sample operating characteristic curve (ROC) and precision-recall curves, and was evaluated using indicators including area under the curve (AUC), accuracy, specificity, sensitivity, positive predictive value (PPV) and negative predictive value (NPV).

[0149] In some embodiments, after determining the target machine learning model, the sub-feature type group of all samples in the sample set and the actual value of the vascular occlusion of the input sample can be used to input the target machine learning model for retraining to obtain the trained machine learning model.

[0150] Specifically, increasing the number of training samples can further improve the reliability of machine learning models.

[0151] It should be noted that in some other embodiments, after determining the target machine learning model, the sub-feature type group of the second part of samples and the actual value of the vascular occlusion of the input sample can also be used to input the target machine learning model for retraining to obtain the trained machine learning model. This can reduce the amount of calculation.

[0152] In some embodiments, the prediction method may further include: creating a complex congestion prediction table.

[0153] Specifically, the re-blockage prediction table has a first area and a second area, wherein the first area corresponds to a first prediction result, the first prediction result is used to characterize that the blood vessel will undergo re-blockage, and the second area corresponds to a second prediction result, which is used to characterize that the blood vessel will remain unobstructed; if the probability of re-blockage of the blood vessel in the sub-feature type group is greater than or equal to a re-blockage probability threshold, the sub-feature type group belongs to the first area; if the probability of re-blockage of the blood vessel in the sub-feature type group is less than the re-blockage probability threshold, the sub-feature type group belongs to the second area.

[0154] In actual use, the sub-feature type group to which the patient belongs is first determined, and then the area in the re-blockage prediction table where the sub-feature type group is located is determined. If the patient is in the first area, a first prediction result is obtained, predicting that the patient's blood vessels will be re-blocked within a preset time. If the patient is in the second area, a second prediction result is obtained, predicting that the patient's blood vessels will remain unobstructed within a preset time. Each sub-feature type group is assigned to the first area and the second area of ​​the re-blockage prediction table so that the probability of re-blockage of the blood vessels of each sub-feature type group corresponds to the first prediction result and the second prediction result, which is conducive to simplifying the doctor's decision-making process, improving execution efficiency, and facilitating patient understanding and communication between doctors and patients.

[0155] In some embodiments, each sub-feature type group in the re-blockage prediction table is arranged in descending order according to the probability of blood vessel re-blockage.

[0156] Specifically, each sub-feature type group belonging to the first area is arranged in descending order according to the probability of blood vessel re-blocking, and each sub-feature type group belonging to the second area is arranged in descending order according to the probability of blood vessel re-blocking.

[0157] In some embodiments, similar items may be merged between some sub-feature type groups within the first region, and similar items may be merged between some sub-feature type groups within the second region.

[0158] In a specific scenario, refer to Figure 3 , Figure 3 A schematic diagram of a complex congestion prediction table in an embodiment of the present disclosure is shown.

[0159] The re-blockage probability threshold is set to 50%, where the area greater than or equal to 50% is the first area, and the area less than 50% is the second area. A1 is the first sub-feature type of the cause of the lesion, PT belongs to A1, A2 is the second sub-feature type of the cause of the lesion, and NT belongs to A2. B1 is the first sub-feature type of diameter stenosis, occlusion belongs to B1, B2 is the second sub-feature type of diameter stenosis, and stenosis belongs to B2. C1 is the first sub-feature type of age, and age values ​​less than 55 years old belong to C1. C2 is the second sub-feature type of age, and age values ​​greater than or equal to 55 years old belong to C2. D1 is the first sub-feature type of lesion length, and lesion length values ​​less than 110 mm belong to D1. D2 is the second sub-feature type of lesion length, and lesion length values ​​greater than or equal to 110 mm belong to D2.

[0160] Among them, [A1+B1] is the merged group of four sub-feature type groups [A1+B1+C1+D1], [A1+B1+C1+D2], [A1+B1+C2+D1] and [A1+B1+C2+D2], [A2+B1+D2] is the merged group of two sub-feature type groups [A2+B1+C1+D2] and [A2+B1+C2+D2], [A1+B2+D2] is the merged group of two sub-feature type groups [A1+B 2+C1+D2] and [A1+B2+C2+D2], [A2+B2+D2] is the merged group of two sub-feature type groups [A2+B2+C1+D2] and [A2+B2+C2+D2], and [A2+D1] is the merged group of four sub-feature type groups [A2+B1+C1+D1], [A2+B1+C2+D1], [A2+B2+C1+D1] and [A2+B2+C2+D1].

[0161] In some embodiments, the prediction method may further include: using the complex blockage prediction table to predict samples in the sample set before balancing processing, and outputting corresponding prediction results; obtaining an accuracy model based on the actual value of vascular complex blockage of the sample before balancing processing and the corresponding prediction results, and the accuracy model is used to evaluate the accuracy of the complex blockage prediction table.

[0162] In a specific scenario, refer to Figure 4 , Figure 4 A schematic diagram of an accuracy model in an embodiment of the present disclosure is shown.

[0163] The accuracy model has: a first unit, which is used to characterize the number of samples in which both the actual value of vascular reocclusion and the predicted result are vascular patency; a second unit, which is used to characterize the number of samples in which both the actual value of vascular reocclusion and the predicted result are vascular reocclusion; a third unit, which is used to characterize the number of samples in which the actual value of vascular reocclusion is vascular reocclusion and the predicted result is vascular patency; a fourth unit, which is used to characterize the number of samples in which the actual value of vascular reocclusion is vascular patency and the predicted result is vascular reocclusion;

[0164] Among them, the accuracy model also has: a first parameter, which is used to characterize the ratio of the number of samples whose actual values ​​and predicted results of vascular re-blockage are consistent, to the total number of samples in the sample set; a second parameter, which is used to characterize the ratio of the number of samples whose actual values ​​and predicted results of vascular re-blockage are both unobstructed, to the number of samples whose actual values ​​of vascular re-blockage are unobstructed; a third parameter, which is used to characterize the ratio of the number of samples whose actual values ​​and predicted results of vascular re-blockage are both vascular re-blockage, to the number of samples whose actual values ​​of vascular re-blockage are vascular re-blockage; a fourth parameter, which is used to characterize the ratio of the number of samples whose actual values ​​and predicted results of vascular re-blockage are both vascular re-blockage, to the number of samples whose predicted results are vascular re-blockage; a fifth parameter, which is used to characterize the ratio of the number of samples whose actual values ​​and predicted results of vascular re-blockage are both unobstructed, to the number of samples whose predicted results are vascular re-blockage.

[0165] In some embodiments, the prediction method may further include: performing correlation analysis on the trained machine learning model to determine a correlation model, wherein the correlation model includes one or more of the following: the correlation size of a single sub-feature type to the probability of vascular re-occlusion; the correlation size of all sub-feature types to the probability of vascular re-occlusion; the correlation size of each risk feature to the probability of vascular re-occlusion.

[0166] In some embodiments, the SHAP (Shapley Additive Explanations) method can be used to explain the machine learning model. SHAP is an important tool for tree-based model methods. SHAP is used to explain the output of the machine learning model and show the positive or negative impact of each sub-feature type on predicting the loss of vascular patency after stent implantation surgery.

[0167] The disclosed embodiment also provides a prediction device corresponding to the above-mentioned blood vessel occlusion prediction method. The blood vessel reocclusion prediction device may include:

[0168] A sample set acquisition module, used to acquire a sample set, wherein the samples in the sample set include multiple risk features, each risk feature has multiple preset sub-feature types, wherein individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature;

[0169] A training module, used to use the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample, and input them into the machine learning model to be trained for training, so as to obtain a trained machine learning model, wherein the sub-feature type group is used to represent the combination of sub-feature types of each risk feature of the sample;

[0170] The reocclusion probability determination module is used to input all non-repetitive sub-feature type groups into the trained machine learning model to obtain the vascular reocclusion probability of each sub-feature type group.

[0171] The sample set is obtained by the sample set acquisition module, wherein the samples in the sample set include multiple risk features, each risk feature has multiple preset sub-feature types, wherein the individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature; the training module uses the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-blockage of the input sample, and inputs them into the machine learning model to be trained for training to obtain the trained machine learning model, wherein the sub-feature type group is used to represent the combination of the sub-feature types of each risk feature of the sample; the re-blockage probability determination module inputs all non-repetitive sub-feature type groups into the trained machine learning model to obtain the vascular re-blockage probability of each sub-feature type group. By using the above prediction device, by obtaining the vascular re-blockage probability of each sub-feature type group, the probability of vascular re-blockage of the patient corresponding to each sub-feature type group can be predicted in advance, so that subsequent treatment suggestions can be made to the patient in advance, thereby reducing the probability of vascular re-blockage.

[0172] The present disclosure also provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned method for predicting blood vessel re-occlusion are executed.

[0173] Computer-readable storage media may include non-volatile memory or non-transitory memory, and may also include optical disks, mechanical hard disks, solid-state disks, etc.

[0174] Specifically, in the embodiments of the present disclosure, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0175] It should also be understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0176] The embodiment of the present disclosure also provides a terminal, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor executes the steps of the above-mentioned blood vessel reocclusion prediction method when running the computer instructions. The terminal may include but is not limited to terminal devices such as mobile phones, computers, and tablet computers, and may also be a server, a cloud platform, etc.

[0177] The present disclosure also provides a computer program product, including a computer program, wherein the computer program runs the steps of the prediction method described in any of the aforementioned embodiments. The specific steps can be found in the aforementioned embodiments and will not be repeated here.

[0178] It is understandable that the term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article indicates that the associated objects before and after are in an "or" relationship.

[0179] It is understood that the "multiple" mentioned herein refers to two or more than two. The first, second, etc. descriptions in the embodiments of the present application are only used for illustration and distinction of the described objects, and have no order, nor do they represent any special limitation on the number of devices in the embodiments of the present application, and cannot constitute any limitation on the embodiments of the present application.

[0180] It can be understood that the above describes multiple embodiment schemes provided by the embodiments of the present disclosure, and the various optional methods introduced in each embodiment scheme can be combined and cross-referenced with each other without conflict, thereby extending a variety of possible embodiment schemes, which can all be considered as embodiment schemes disclosed and disclosed by the present disclosure.

[0181] Although the embodiments of the present disclosure are disclosed as above, the present disclosure is not limited thereto. Any person skilled in the art may make various changes and modifications without departing from the spirit and scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the scope defined by the claims.

Claims

1. A method for predicting blood vessel reocclusion, characterized in that: include: Acquire a sample set, wherein the samples in the sample set include multiple risk features, each risk feature has multiple preset sub-feature types, wherein individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature; Using the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample, inputting them into the machine learning model to be trained for training, so as to obtain a trained machine learning model, wherein the sub-feature type group is used to represent the combination of sub-feature types of each risk feature of the sample; All non-repetitive sub-feature type groups are input into the trained machine learning model to obtain the probability of vascular re-occlusion for each sub-feature type group.

2. The prediction method according to claim 1, characterized in that: A sample set is obtained, where the samples in the sample set contain multiple risk features, including: Acquire a sample set, wherein the samples in the sample set include multiple initial risk features; The multiple initial risk characteristics of each sample are preprocessed to obtain the multiple risk characteristics of the sample.

3. The prediction method according to claim 2, characterized in that: Each sample is preprocessed with multiple initial risk features, including: The initial individual characteristic data of each initial risk feature in the sample set is interpolated so that each initial risk feature in the sample set has corresponding initial individual characteristic data.

4. The prediction method according to claim 2, characterized in that: Each sample is preprocessed with multiple initial risk features, including: The multiple initial risk characteristics of each sample are transformed and preprocessed to obtain the multiple risk characteristics of the sample.

5. The prediction method according to claim 4, characterized in that: Individual characteristic data for some risk characteristics are continuous; Preprocessing of various initial risk features for each sample also includes: The risk characteristics with continuous individual characteristic data are truncated to obtain the individual characteristic data truncation value; The individual characteristic data cutoff value is used to divide the individual characteristic data of the risk characteristic so that the individual characteristic data of the risk characteristic can be attributed to a single sub-characteristic type of the risk characteristic.

6. The prediction method according to claim 5, characterized in that: Preprocessing of various initial risk features for each sample also includes: A single screening pretreatment method was used to screen and pretreat multiple risk characteristics of each sample; Based on the screening results of each sample, various risk characteristics of the sample are obtained.

7. The prediction method according to claim 5, characterized in that: Preprocessing of various initial risk features for each sample also includes: Multiple screening and pretreatment methods were used to screen and preprocess multiple risk characteristics of each sample; Based on the screening results of various screening pretreatment methods for each sample, various risk characteristics of the sample are determined.

8. The prediction method according to claim 7, characterized in that: The various risk characteristics of the samples are obtained by taking the union of the screening results of various screening preprocessing methods for each sample; Alternatively, the multiple risk characteristics of the sample are obtained by taking the intersection of the screening results of various pre-processing methods for each sample.

9. The prediction method according to claim 3, characterized in that: The samples in the sample set include: unobstructed samples and re-blocked samples; Preprocessing of various initial risk features for each sample also includes: The number of the unobstructed samples and the number of the complex-blocked samples are balanced so that the ratio of the number of complex-blocked samples to the number of unobstructed samples after the balancing process is in the range of [0.9, 1].

10. The prediction method according to claim 1, characterized in that: The number of risk features of the sample is N, and each risk feature has M sub-feature types, and the number of all non-repeating sub-feature type groups is M to the power of N; Wherein, N is a positive integer, and N>1; M is a positive integer, and M>1.

11. The prediction method according to claim 1, characterized in that: The risk characteristics are selected from: age; cause of lesion; length of lesion; preoperative diameter stenosis.

12. The prediction method according to claim 1, characterized in that: The sample set includes a first part of samples and a second part of samples, and the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample are input into the machine learning model to be trained for training, including: Using the sub-feature type group of the first part of samples and the actual value of the vascular re-occlusion of the input sample, input a plurality of preset initial machine learning models for pre-training, and obtain a pre-trained machine learning model corresponding to each initial machine learning model; Using the sub-feature type group of the second part of samples, inputting each pre-trained machine learning model for testing, so as to select a target machine learning model from each pre-trained machine learning model; The sub-feature type groups of all samples in the sample set and the actual values ​​of vascular re-occlusion of the input samples are input into the target machine learning model for retraining to obtain the trained machine learning model.

13. The prediction method according to claim 1, characterized in that: Also includes: Creating a re-blockage prediction table, wherein the re-blockage prediction table has a first area and a second area, wherein the first area corresponds to a first prediction result, the first prediction result is used to indicate that the blood vessel will be re-blocked, and the second area corresponds to a second prediction result, which is used to indicate that the blood vessel will remain unobstructed; If the blood vessel re-blockage probability of the sub-feature type group is greater than or equal to the re-blockage probability threshold, the sub-feature type group belongs to the first area; If the blood vessel re-blockage probability of the sub-feature type group is less than the re-blockage probability threshold, the sub-feature type group belongs to the second area.

14. The prediction method according to claim 1, characterized in that: Using the sub-feature type group of at least a part of the samples in the sample set and the actual value of the vascular re-occlusion of the input sample, inputting them into the machine learning model to be trained for training, including: Using the sub-feature type group of at least a part of the samples in the balanced sample set and the actual value of the vascular re-occlusion of the input sample, inputting them into the machine learning model to be trained for training; The method further comprises: Using the complex blockage prediction table to predict the samples in the sample set before balancing processing, and outputting the corresponding prediction results; According to the actual value of blood vessel re-occlusion of the sample before the balancing process and the predicted result, an accuracy model is obtained, and the accuracy model is used to evaluate the accuracy of the re-occlusion prediction table.

15. The prediction method according to claim 14, characterized in that: The accuracy model has a first parameter, a second parameter, a third parameter, a fourth parameter and a fifth parameter; The first parameter is the ratio of the number of samples whose actual values ​​of vascular reocclusion are consistent with the predicted results to the total number of samples in the sample set; The second parameter is the ratio of the number of samples where both the actual value of vascular re-blocking and the predicted result are that the vessel is unobstructed to the number of samples where the actual value of vascular re-blocking is that the vessel is unobstructed; The third parameter is the ratio of the number of samples where both the actual value of vascular re-blocking and the predicted result are vascular re-blocking to the number of samples where the actual value of vascular re-blocking is vascular re-blocking; The fourth parameter is the ratio of the number of samples whose actual values ​​and predicted results are both vascular re-occlusion and the number of samples whose predicted results are vascular re-occlusion; The fifth parameter is the ratio of the number of samples whose actual values ​​and predicted results are both unobstructed blood vessels to the number of samples whose predicted results are re-obstructed blood vessels.

16. The prediction method according to claim 1, characterized in that: Also includes: Performing correlation analysis on at least one evaluation curve output by the trained machine learning model to determine a correlation model, wherein the correlation model includes one or more of the following: The correlation between a single sub-feature type and the probability of vascular reocclusion; The correlation between all sub-feature types and the probability of vascular reocclusion The correlation between each type and the probability of vascular reocclusion.

17. The prediction method according to claim 1, characterized in that: The blood vessels include deep veins.

18. The prediction method according to claim 1, characterized in that: The sample is a sample collected within a preset time after the stent implantation operation.

19. A device for predicting blood vessel reocclusion, characterized in that: include: A sample set acquisition module, which acquires a sample set, wherein the samples in the sample set include multiple risk features, each risk feature has multiple preset sub-feature types, wherein individual feature data of each risk feature can be attributed to a single sub-feature type of the risk feature; A training module, using a sub-feature type group of at least a part of the samples in the sample set and an actual value of vascular re-occlusion of the input sample, and inputting them into a machine learning model to be trained for training, so as to obtain a trained machine learning model, wherein the sub-feature type group is used to represent a combination of sub-feature types of various risk features of the samples; The re-blockage probability determination module inputs all non-repetitive sub-feature type groups into the trained machine learning model to obtain the vascular re-blockage probability of each sub-feature type group.

20. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the prediction method according to any one of claims 1 to 18 are executed.

21. A terminal comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor runs the computer program, the processor performs the steps of the prediction method according to any one of claims 1 to 18.

22. A computer program product comprising a computer program, characterized in that When the computer program is executed, the steps of the prediction method according to any one of claims 1 to 18 are executed.

Images