Blood flow measurement waveform analysis method and device for bridge blood vessel

Abstract

The invention discloses a blood flow measurement waveform analysis method and device of a bridge blood vessel, and relates to the technical field of machine learning. Comprising the steps of collecting instant blood flow measurement waveforms of bridge blood vessels; dense sampling is conducted on the instant blood flow measurement waveform, and a plurality of time sequence sampling points are obtained; re-sampling the plurality of time sequence characteristics so as to align time sequence sampling points with different sampling intervals to obtain full time sequence information of the bridge blood vessel; constructing a data set based on the full time sequence information; constructing a random forest model, and training the random forest model by using the data set to obtain a bridge vessel patency analysis model; and analyzing the immediate patency of the bridge blood vessel and the postoperative middle and long-term patency by using the bridge blood vessel patency analysis model.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and particularly relates to a method and device for analyzing blood flow measurement waveforms of bypass vessels. Background Art

[0002] Coronary artery bypass grafting (CABG) is a common surgical method for treating coronary heart disease. Doctors intercept a blood vessel from other parts of the patient's own body as a bypass vessel and connect it to both ends of the coronary artery stenosis site, so that blood flow passes through the bypass vessel and perfuses to the distal end of the coronary artery to achieve the purpose of treating myocardial ischemia. According to the different bypass vessel materials selected, it can be divided into internal mammary artery bypass vessel (LIMA) and saphenous vein bypass vessel (SVG). An important problem in bypass surgery is the patency of the bypass vessel, that is, the bypass vessel may become occluded some time after the operation, which will threaten the patient's life and health again. The patency of the bypass vessel can be divided into immediate patency and medium- and long-term patency after the operation. The immediate patency problem refers to the patency problem that occurs within a very short time after the operation (within one week after the operation), mostly caused by the surgical operation method; the medium- and long-term patency refers to the patency problem that occurs more than one year after the operation, and the patient's own physical state and hemodynamic environment account for a large reason. The accurate prediction of these two types of patency is an urgent problem to be solved in clinical practice.

[0003] Currently, for the method of analyzing the patency of bypass vessels, most of them select one or several clinical features to construct a data set, and then use this data set to train a regression model or a network model to achieve the analysis of the patency of bypass vessels, as shown in patents CN114972242B and CN109637657B.

[0004] However, the bypass vessel data is diverse, including but not limited to morphological data, hemodynamic data, etc. These different types of data have differences in structure and information content, which brings challenges to data integration and analysis: and the distribution of bypass vessel patency characteristics in the original data is not synchronous or consistent, which makes it difficult for the model to accurately extract these key information during the training process, increases the difficulty of model training, and reduces the accuracy and reliability of the analysis results output by the model. Summary of the Invention

[0005] Based on this, in order to solve the technical problems in the prior art, the present invention provides a method and device for analyzing blood flow measurement waveforms of bypass vessels.

[0006] The present invention provides a method for analyzing blood flow measurement waveforms of bypass vessels, including:

[0007] Collecting the immediate blood flow measurement waveform data of a plurality of bypass vessels and the corresponding bypass vessel occlusion labels;

[0008] Dense sampling is performed on the instant blood flow measurement waveform data of each bypass vessel according to different sampling intervals to extract the blood flow change information of each bypass vessel within a short time window, and a number of time-series sampling points corresponding to each bypass vessel are obtained; resampling is performed on the number of time-series sampling points corresponding to each bypass vessel to align the instant blood flow measurement waveform data with different sampling intervals on the time scale, and the full time-series information of each bypass vessel is obtained;

[0009] A data set is constructed based on the full time-series information of a number of bypass vessels and the corresponding bypass vessel occlusion labels; a machine learning model is constructed, and the data set is used to train the machine learning model to obtain a bypass vessel blood flow measurement waveform analysis model; the bypass vessel blood flow measurement waveform analysis model is used to analyze the blood flow measurement waveform of the bypass vessel.

[0010] Further, the dense sampling is to perform dense sampling on the waveform signal within a 5s time range in the instant blood flow measurement waveform data.

[0011] Further, the resampling is to align the time-series sampling points of the waveform signal within a 5s time range to 562 resampling points, and the data of the 562 resampling points is used as the input of the machine learning model.

[0012] Further, the machine learning model is a random forest model.

[0013] Further, the training of the machine learning model using the data set specifically includes:

[0014] Initializing the random forest model parameters: setting the splitting criterion to information gain, the maximum number of selected features to the square root of the number of all input features, the minimum number of samples in a leaf to 1, the minimum number of samples for splitting to 2, the random seed number to 1, and the maximum number of leaf nodes and the maximum depth to 300 and 100 respectively;

[0015] Using grid search and k-fold cross-validation to optimize the parameters of the random forest model to obtain the optimal parameter combination.

[0016] The present invention provides a bypass vessel blood flow measurement waveform analysis device, including:

[0017] A data acquisition module for acquiring the instant blood flow measurement waveform data of a number of bypass vessels and the corresponding bypass vessel occlusion labels;

[0018] A feature extraction module, which is used to densely sample the instant blood flow measurement waveform data of each bypass vessel according to different sampling intervals, so as to extract the blood flow change information of each bypass vessel within a short time window, and obtain several corresponding time-series sampling points of each bypass vessel; resample the several corresponding time-series sampling points of each bypass vessel, so as to align the instant blood flow measurement waveform data with different sampling intervals on the time scale, and obtain the full time-series information of each bypass vessel;

[0019] A data analysis module, which is used to construct a data set based on the full time-series information of several bypass vessels and the corresponding bypass vessel occlusion labels; construct a machine learning model, use the data set to train the machine learning model, and obtain a bypass vessel blood flow measurement waveform analysis model; use the bypass vessel blood flow measurement waveform analysis model to analyze the blood flow measurement waveform of the bypass vessel.

[0020] The above at least one technical solution adopted by the present invention can achieve the following beneficial effects:

[0021] In the bypass vessel blood flow measurement waveform analysis method provided by the present invention, the instant blood flow measurement waveform data with the most information is used as the training data. By densely sampling the waveform data, more time points can be obtained, so as to capture more change characteristics existing within a short time window; the purpose of resampling is to align the time series data, so as to make the waveform data under different sampling intervals synchronous, and avoid feature distortion or inaccuracy caused by time inconsistency; the data after dense sampling and resampling has high consistency and time synchronization, which has a significant promoting effect on the training of the subsequent machine learning model and the accuracy of the inference result, so as to obtain a more accurate analysis result. Description of the Drawings

[0022] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute an improper limitation of the present invention. In the drawings:

[0023] Figure 1 is a schematic diagram of the measurement waveform provided by the present invention; Figure 1 in (a) is a schematic diagram of the main unit of TTFM; Figure 1 in (b) is a schematic diagram of the process of measuring the blood flow waveform of the graft using a TTFM probe; Figure 1 in (c) is the measured TTFM waveform diagram;

[0024] Figure 2 is a schematic diagram of the flow chart of a bypass vessel blood flow measurement waveform analysis method provided by the present invention;

[0025] Figure 3A schematic diagram of full time-series information collection based on TTFM waveform provided by the present invention, Figure 3 (a) is the TTFM waveform; Figure 3 (b) is based on Figure 1 (a) The full time series information collected. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in combination with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Currently, a method called point-in-time flow measurement (TTFM) is commonly used in clinical practice to evaluate the blood flow status of the graft vessels after bypass surgery, and is also often used to predict the patency of the graft vessels. The TTFM method is to clamp an ultrasound probe on the graft vessel after coronary artery bypass surgery and before closing the chest cavity to measure the flow waveform of the graft vessel, which is the TTFM waveform. From the TTFM waveform, clinical characteristics such as average flow, pulsatility index (PI), diastolic flow fraction (DF), etc. can be read. These characteristics are currently frequently referenced by doctors for evaluating the flow status of graft vessels. TTFM machine, measurement method, measurement waveform such as Figure 1 As shown. Most doctors directly use the clinical characteristics of the TTFM waveform to make a self-judgment, and there is currently no unified judgment standard. Commonly used clinical characteristics include average flow, PI (calculation method: (maximum flow-minimum flow) / average flow), DF (calculation method: diastolic flow / total flow), and negative flow ratio (negative flow / total flow). It is generally believed that if the average flow is >15ml / min, PI<5, DF>50%, and there is no negative flow, the blood flow state of the graft vessel is considered to be good, otherwise it is considered that there is a problem with the blood flow state of the graft vessel. Some scholars have predicted the immediate patency and medium- and long-term patency of the graft vessel based on the above TTFM waveform clinical characteristics, but the methods used are either to simply find a threshold (cut-off value) for binary classification, or to combine several features to make a discriminant equation, and the prediction effect is not very good.

[0028] Based on the above research status, most of the current methods for predicting the patency of graft vessels use one or several clinical features and select a threshold to make a judgment on patency. A few combine several clinical features to make a discriminant formula or logistic regression formula for binary classification. The prediction results of these methods are not very ideal.

[0029] In this context, the present invention provides a method for analyzing the blood flow measurement waveform of a bypass vessel. Instead of extracting clinical features from the TTFM waveform, the method uses the full-time series information as features and selects the random forest as the prediction model to construct a bypass vessel patency prediction model. It can realize the patency prediction of different types of bypass vessels (LIMA, SVG) at different postoperative periods (immediate patency, medium- and long-term patency), and can solve the above-mentioned defects.

[0030] Example 1

[0031] Figure 2 The flowchart of the blood flow measurement waveform analysis method of the bypass vessel in this embodiment is shown. Specifically combined with Figure 2 The method is described in detail as follows, including the following steps:

[0032] S1: Collect the immediate blood flow measurement waveform data of a number of bypass vessels and the corresponding bypass vessel occlusion labels;

[0033] Before closing the chest after coronary artery bypass grafting, clamp the bypass vessel with an ultrasonic probe to collect the immediate blood flow measurement waveform of the bypass vessel, as shown in (c) in Figure 1 (a) in Figure 3 shown.

[0034] S2: Densely sample the immediate blood flow measurement waveform data of each bypass vessel according to different sampling intervals to extract the blood flow change information of each bypass vessel within a short time window, and obtain a number of time series sampling points corresponding to each bypass vessel; resample the number of time series sampling points corresponding to each bypass vessel to align the immediate blood flow measurement waveform data with different sampling intervals on the time scale, and obtain the full time series information of each bypass vessel.

[0035] After the TTFM waveform measurement, the full time series information refers to densely sampling the entire 5s waveform signal, as shown in (b) in Figure 3 shown, and using the sampled time series features as the features of the prediction model. Since the sampling intervals of different TTFM waveforms may be different, the number of sampling points within 5s will vary. The resampling technique is a method of reallocating and aligning the sampling points of the waveform. In the present invention, the resampling technique is used to align the time series features of all waveform signals within the 5s time range to 562 sampling points, which are the 562 features input to the prediction model. In this way, instead of extracting clinical features in the traditional sense from the TTFM waveform, using the full time series information can almost not lose the waveform information to achieve a better prediction effect.

[0036] S3: Construct a data set based on the full-time series information of several bypass vessels and the corresponding bypass vessel occlusion labels; construct a machine learning model, use the data set to train the machine learning model, and obtain a bypass vessel blood flow measurement waveform analysis model; use the bypass vessel blood flow measurement waveform analysis model to analyze the blood flow measurement waveform of the bypass vessel.

[0037] S301: Random forest model construction.

[0038] The present invention uses random forest as an artificial intelligence method for predicting the patency of bypass vessels. Random forest is an ensemble learning algorithm that constructs multiple decision trees and takes the mode of their outputs as the final output. Decision trees are based on a tree structure and divide the feature space to generate decision rules and solve classification problems. Different from building a large decision tree using the entire training set, random forest uses different subsets and feature attributes to build multiple small decision trees, and combines the results of multiple decision trees to enhance the model effect.

[0039] The number of trees (n estimators) in the random forest has a greater impact on the model performance. Increasing the number of trees can improve the generalization ability of the model. When a certain number is reached, the model classification performance gradually decreases. The bootstrap resampling technique is used to draw samples from the original data set with replacement, and then construct the training sets of multiple decision trees. The splitting criterion is used to measure the quality of tree node splitting. Common measurement criteria include Gini impurity and information gain. The maximum number of selected features (max features) controls the maximum number of features used when splitting each node, which affects the generalization ability of the model. The maximum number of leaf nodes (max leafnodes) limits the number of leaf nodes. The minimum number of samples in a leaf (min samplesleaf) and the minimum number of samples for splitting (min samples split) limit the splitting of leaf nodes. The maximum depth (max depth) determines the maximum allowed depth of the decision tree. The definitions of these parameters are the stopping conditions for model training and jointly affect the fitting ability and generalization ability of the decision tree model.

[0040] S302: Random forest model training.

[0041] The present invention constructs a classification model using the Scikit-learn library of Python and the Pytorch deep learning framework, and uses the methods of grid search and cross-validation to optimize the model parameters. Grid search is a common method for optimizing model hyperparameters. Based on exhaustive search, by specifying a list of candidate values for the parameters, all possible parameter combinations are evaluated and compared to select the best parameter combination. k-fold cross-validation is a common technique for evaluating model performance. It divides the dataset into a training set and a validation set and repeats this division k times. In each division, the training set is used to train the model, and the validation set is used to evaluate the performance of the model. Finally, the average performance of the model is evaluated based on the accuracy results of all cross-validations. In the present invention, 5-fold cross-validation is used to evaluate the model performance according to the number of data samples. By comparing the accuracy results of cross-validation under all parameter combinations, the parameter combination corresponding to the maximum average accuracy is the optimal parameter.

[0042] In the present invention, for the random forest model, the criterion uses information gain, max features is set to the square root of the number of all input features, min samples leaf is set to 1, min samples split is set to 2, the random seed is set to 1, and max leaf nodes and max depth are set to 300 and 100 respectively.

[0043] S303: Optimization of the random forest model.

[0044] Five metrics, namely precision, recall, F1 score, accuracy, and AUC value, are used to evaluate the predictor obtained by training. Precision represents the proportion of samples that are truly positive among the samples predicted as positive by the model. Recall represents the proportion of positive samples that the model can correctly predict among all true positive samples. The F1 score is a metric that comprehensively considers precision and recall and is used to evaluate the comprehensive performance of the classification model. The F1 score is the harmonic mean of precision and recall, that is, the comprehensive performance of the precision and recall of the model. Accuracy represents the proportion of samples predicted correctly by the model among the total number of samples. The AUC value is the area under the ROC curve. The AUC value is usually used to judge the quality of a classifier, and its value range is between 0 and 1. The larger the AUC value, the better the classification performance of the model.

[0045] The effectiveness of the present invention has been verified through experiments. The TTFM waveforms of 796 arterial grafts (LIMA) and 995 venous grafts (SVG), as well as the coronary CTA results one week after surgery, were collected to verify the prediction effect of the model on immediate patency. The TTFM waveforms of 210 arterial grafts and 268 venous grafts, as well as the coronary CTA results one year after surgery, were collected to verify the prediction effect of the model on medium- and long-term patency. The results are shown in Table 1. As can be seen from the table, the method has very good prediction effects on both arterial grafts and venous grafts, for both immediate patency and medium- and long-term patency.

[0046] Table 1 Model Prediction Results

[0047]

[0048] The above-mentioned method for analyzing the blood flow measurement waveform of bypass vessels realizes the prediction of the immediate patency and medium- and long-term patency of bypass vessels after coronary artery bypass grafting surgery, and can achieve a very high accuracy. It can help doctors revise the surgical strategy in a timely manner and formulate a more scientific and reasonable postoperative review and rehabilitation plan for patients.

[0049] The above is the method for analyzing the blood flow measurement waveform of bypass vessels provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding device for analyzing the blood flow measurement waveform of bypass vessels, including:

[0050] A data acquisition module for acquiring the immediate blood flow measurement waveform data of a plurality of bypass vessels and the corresponding bypass vessel occlusion labels.

[0051] A feature extraction module for densely sampling the immediate blood flow measurement waveform data of each bypass vessel according to different sampling intervals to extract the blood flow change information of each bypass vessel within a short time window, obtaining a plurality of corresponding time series sampling points for each bypass vessel; resampling the plurality of corresponding time series sampling points for each bypass vessel to align the immediate blood flow measurement waveform data with different sampling intervals on the time scale, obtaining the full time series information of each bypass vessel.

[0052] A data analysis module for constructing a data set based on the full time series information of a plurality of bypass vessels and the corresponding bypass vessel occlusion labels; constructing a machine learning model, training the machine learning model using the data set, obtaining a bypass vessel blood flow measurement waveform analysis model; and analyzing the blood flow measurement waveform of the bypass vessel using the bypass vessel blood flow measurement waveform analysis model.

[0053] For the specific limitations of the blood flow measurement waveform analysis device of the bypass vessel, reference can be made to the limitations of the blood flow measurement waveform analysis method of the bypass vessel in the above text, which will not be elaborated here. Each module in the above blood flow measurement waveform analysis device of the bypass vessel can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in the form of hardware or independent of it, or stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0054] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to memory, storage, database, or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0055] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present invention.

Claims

1. A method for analyzing blood flow measurement waveforms of a bypass vessel, characterized in that, Including: Collecting the immediate blood flow measurement waveform data of several bypass vessels and the corresponding bypass vessel occlusion labels; Performing dense sampling on the immediate blood flow measurement waveform data of each bypass vessel according to different sampling intervals to extract the blood flow change information of each bypass vessel within a short time window, obtaining several corresponding temporal sampling points for each bypass vessel; Resampling the several corresponding temporal sampling points for each bypass vessel to align the immediate blood flow measurement waveform data with different sampling intervals on the time scale, obtaining the full temporal information of each bypass vessel; Constructing a dataset based on the full temporal information of several bypass vessels and the corresponding bypass vessel occlusion labels; Constructing a machine learning model, training the machine learning model using the dataset to obtain a bypass vessel blood flow measurement waveform analysis model; Analyzing the blood flow measurement waveform of the bypass vessel using the bypass vessel blood flow measurement waveform analysis model.

2. The method for analyzing the blood flow measurement waveform of the bypass vessel according to claim 1, wherein The dense sampling is to perform dense sampling on the waveform signal within a 5s time range in the immediate blood flow measurement waveform data.

3. The method for analyzing the blood flow measurement waveform of the bypass vessel according to claim 2, wherein The resampling is to align the temporal sampling points of the waveform signal within a 5s time range to 562 resampling points, and using the data of the 562 resampling points as the input of the machine learning model.

4. The method for analyzing the blood flow measurement waveform of the bypass vessel according to claim 1, characterized in that, The machine learning model is a random forest model.

5. The method for analyzing the blood flow measurement waveform of the bypass vessel according to claim 4, wherein, The training of the machine learning model using the dataset specifically includes: Initializing the random forest model parameters: setting the splitting criterion to information gain, the maximum number of selected features to the square root of all input feature numbers, the minimum number of samples in a leaf to 1, the minimum number of samples for splitting to 2, the random seed number to 1, and the maximum number of leaf nodes and the maximum depth to 300 and 100 respectively; Using grid search and k-fold cross-validation to optimize the parameters of the random forest model to obtain the optimal parameter combination.

6. A blood flow measurement waveform analysis device for a bypass vessel, characterized in that, Including: A data acquisition module for collecting the immediate blood flow measurement waveform data of several bypass vessels and the corresponding bypass vessel occlusion labels; A feature extraction module for performing dense sampling on the immediate blood flow measurement waveform data of each bypass vessel according to different sampling intervals to extract the blood flow change information of each bypass vessel within a short time window, obtaining several corresponding temporal sampling points for each bypass vessel; Resampling the several corresponding temporal sampling points for each bypass vessel to align the immediate blood flow measurement waveform data with different sampling intervals on the time scale, obtaining the full temporal information of each bypass vessel; A data analysis module for constructing a dataset based on the full temporal information of several bypass vessels and the corresponding bypass vessel occlusion labels; Constructing a machine learning model, training the machine learning model using the dataset to obtain a bypass vessel blood flow measurement waveform analysis model; Analyzing the blood flow measurement waveform of the bypass vessel using the bypass vessel blood flow measurement waveform analysis model.

Images