Ventricular assist device and control unit therefor, medical device
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CORE MEDICAL TECHNOLOGY (HK) LTD
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-21
Smart Images
Figure CN122424488A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical device technology, and in particular to a ventricular assist device and its control unit, and a medical device. Background Technology
[0002] Short-term ventricular assist devices (VAMs) are an effective treatment for heart failure and related complications. They provide hemodynamic support to patients in a short period, promoting rapid recovery of cardiac and other vital tissue and organ function. The VAM itself is crucial; its relative position within the heart during operation determines whether the interventional VAM can assist the patient in achieving pumping function.
[0003] Currently, the location of ventricular assist devices can only be quickly determined in catheterization labs or operating rooms equipped with imaging equipment such as digital subtraction angiography (DSA) and ultrasound. However, this reliance on visualization equipment to determine the location poses a risk of radiation exposure for patients and medical staff. Furthermore, the equipment is expensive and the operation is complex, making it difficult to apply in emergency or resource-constrained scenarios where visualization equipment is unavailable. Summary of the Invention
[0004] This application provides a ventricular assist device, its control unit, and a medical device that can correctly identify the location of the ventricular assist device in the patient and whether it is abnormal.
[0005] In a first aspect, embodiments of this application provide a control unit for a ventricular assist device, the control unit being configured to perform the following steps: Multiple target parameter curves are obtained, each target parameter curve being a curve of one operating parameter of the ventricular assist device during the first cycle; n features are extracted from the multiple target parameter curves to obtain a target feature vector. The target feature vector includes m features, which are the features after dimensionality reduction of the n features. Both n and m are integers greater than 1, and n is less than m. The target feature vector is input into the target location prediction model to obtain the predicted location classification.
[0006] Secondly, this application provides a ventricular assist device, the ventricular assist device comprising: impeller; A motor that drives the impeller to rotate; A control unit connected to the motor is configured to perform the following steps: Multiple target parameter curves are obtained, each target parameter curve being a curve of one operating parameter of the ventricular assist device during the first cycle; n features are extracted from the plurality of target parameter curves to obtain a target feature vector. The target feature vector includes m features, which are the features after dimensionality reduction of the n features. Both n and m are integers greater than 1, and n is less than m. The target feature vector is input into the target location prediction model to obtain the predicted location classification.
[0007] Thirdly, embodiments of this application provide a medical device, the medical device including a processor, a memory, a communication interface, and one or more programs, the one or more programs being stored in the memory and configured to be executed by the processor, the programs including instructions for performing some or all of the steps described in the method described in the first aspect above.
[0008] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the method described in the first aspect above.
[0009] The technical solution provided in this application involves a control unit acquiring multiple target parameter curves, each curve representing an operating parameter of the ventricular assist device (VAD) during the first cycle. N features are extracted from these multiple target parameter curves to obtain a target feature vector, which includes m features resulting from dimensionality reduction of the n features. The target feature matrix is then input into a target position prediction model to obtain a predicted position classification. This application directly inputs the real-time operating parameters of the VAD into the target position prediction model, thereby outputting a predicted position of the VAD. This allows for direct estimation of the VAD's position within the patient's body based on its real-time operating parameters without relying on external equipment, reducing equipment costs and operational complexity. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a schematic diagram of the result of a ventricular assist device provided in an embodiment of this application; Figure 2 This is a schematic diagram of a ventricular assist device located inside the heart, as provided in an embodiment of this application; Figure 3This is a schematic diagram of a control unit performing position prediction according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a medical device provided in an embodiment of this application. Detailed Implementation
[0012] To help those skilled in the art better understand the technical solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the description of the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0013] The terms "first," "target," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, software, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but also includes steps or units not listed, or other steps or units inherent to such processes, methods, products, or apparatus.
[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0015] It should be noted that in this application, the terms "proximal" or "proximal" refer to the end or side closer to the surgeon; and "distal" or "distal" refer to the end or side farther from the surgeon.
[0016] Please see Figures 1-2 , Figure 1 This is a schematic diagram of the structure of a ventricular assist device 100 provided in an embodiment of this application. Figure 2This is a schematic diagram illustrating a ventricular assist device 100 located in the normal position of a patient's heart 120, as provided in an embodiment of this application. The ventricular assist device 100 can operate within the patient's heart, partially within the heart, outside the heart, partially outside the heart, partially outside the vascular system, or in any other suitable location within the vascular system. The ventricular assist device 100 can be percutaneously inserted into the aorta 124 via the femoral artery 122 and passes through the aorta 124 into the left ventricle 128. For example, the ventricular assist device 100 can be percutaneously inserted into the aorta 124 via the axillary artery 123 and passes through the aorta 124 into the left ventricle. In other embodiments, the ventricular assist device 100 can also be directly inserted into the aorta 124 and passes through the aorta 124 into the left ventricle 128. During operation, the ventricular assist device 100 pumps blood from the left ventricle 128 into the aorta 124.
[0017] The ventricular assist device 100 includes a cannula 10. The cannula 10 has a proximal end and a distal end, the distal end of the cannula 10 having a fluid inlet 101 and the proximal end of the cannula 10 having a fluid outlet 102, through which blood flows in from the fluid inlet 101 and out from the fluid outlet 102 via the cannula 10.
[0018] The ventricular assist device 100 includes an impeller (not shown). The impeller is located at least partially at the proximal end of the cannula 10, such as at the fluid inlet 101 of the cannula 10, such that when the ventricular assist device 100 is in operation, it drives the impeller to rotate to pump blood from the left ventricle 128 to the aorta 124.
[0019] The ventricular assist device 100 may include a motor (not shown in the figure), which may be located inside or outside the ventricular assist device 100. This embodiment of the application illustrates the example where the motor is located inside the ventricular assist device 100. For instance, the motor is housed in a motor housing 201, with the distal end of the motor housing 201 connected to the proximal end of the cannula 10. The motor drives the drive shaft to rotate, thereby rotating the impeller and realizing the pumping function of the ventricular assist device 100.
[0020] The ventricular assist device 100 includes a catheter 30, the distal end of which is connected to the proximal end of a motor housing 201, through which a drive cable extends. As an example, the catheter 30 may accommodate an electrical connection between the ventricular assist device 100 and an external controller. As an example, the ventricular assist device 100 also includes a distal component 110, such as a pigtail cannula, extending distally away from the distal end of the cannula 10.
[0021] The ventricular assist device 100 also includes a contrast ring (not shown) for detecting the position of the ventricular assist device 100 within the patient's body. This contrast ring is positioned between a fluid inlet 101 and a fluid outlet 102. For example, the contrast ring is positioned midway between the fluid inlet 101 and the fluid outlet 102; alternatively, when the ventricular assist device 100 is in the desired position, the contrast ring is near the aortic valve, such as being within the aortic valve or at a distance of 1 cm from it.
[0022] When the ventricular assist device 100 is used on a patient's left heart, the ventricular assist device 100 can be considered to be in the desired position when it is positioned such that the cannula 10 extends across the patient's aortic valve 126, the distal end of the cannula 10 is located in the patient's left ventricle 128, the proximal end of the cannula 10 is located in the patient's aorta 124, and the distal component 110 is just abutting against or at a predetermined distance from the inner wall of the patient's left ventricle 128.
[0023] It should be noted that the ventricular assist device 100 in this application is not limited to use in the left ventricle, but can also be used in the right ventricle. When the ventricular assist device 100 is used in the right ventricle, the distal end of the cannula 10 has a fluid outlet 101 and the proximal end of the cannula 10 has a fluid inlet 102. Blood flows in from the fluid inlet 102 and flows out from the fluid outlet 101 via the cannula 10. The ventricular assist device 100 is considered to be in the desired position, i.e., the ventricular assist device 100 is in normal position, when the cannula 10 extends across the pulmonary valve 125, the distal end of the cannula 10 is located in the patient's right ventricle 127, the proximal end of the cannula 10 is located in the patient's pulmonary artery 121, and the distal component 110 is just abutting the inner wall of the patient's right ventricle 127. The following description uses the ventricular assist device 100 used in the left ventricle as an example.
[0024] The ventricular assist device 100 also includes one or more pressure sensors disposed on the outer surface of the cannula 10 and positioned corresponding to the fluid outlet 102 and / or fluid inlet 101, for detecting pressure at the fluid outlet 102 and / or the fluid inlet 101 of the ventricular assist device 100. When the ventricular assist device 100 is correctly positioned across the aorta 124, the top (outer surface) of the pressure sensor at the fluid outlet 102 is exposed to the aorta 124, thereby allowing measurement of aortic pressure; the top (outer surface) of the pressure sensor at the fluid inlet 101 is exposed to the left ventricle 128, thereby allowing measurement of left ventricular pressure.
[0025] The ventricular assist device 100 also includes a control unit, which can be used to perform any of the embodiments, aspects, and methods of this application. The control unit may be located inside or outside the ventricular assist device 100. The control unit is used to detect relevant parameters of the ventricular assist device 100 and the patient, and to control the operation of the ventricular assist device 100. For example, the control unit supplies current to a motor through one or more wires and detects the current through a current detection circuit (such as a phase current detection circuit); controls the rotational speed of the ventricular assist device 100 according to received instructions; receives feedback signals from pressure sensors to identify the position of the ventricular assist device 100, and so on.
[0026] The ventricular assist device 100 is implanted into the patient's heart via a guidewire and positioned as desired, with the proximal end of the cannula 10 located in the patient's aorta 124 and the distal end in the patient's left ventricle 128. The position of the ventricular assist device 100 requires imaging equipment such as DSA and ultrasound for detection. However, this method of determining position using visualization instruments poses a risk of radiation exposure for both patients and healthcare workers. Furthermore, the equipment is expensive and the operation is complex, making it unsuitable for emergency situations or resource-constrained scenarios where visualization equipment is unavailable. Post-operatively, patients require the ventricular assist device 100 for support for a period. During this time, the position of the ventricular assist device 100 within the patient cannot be detected in a timely and real-time manner, and patient movement may cause displacement of the ventricular assist device 100, potentially endangering the patient's life.
[0027] Based on this, this application uses the operating parameters of the ventricular assist device 100 and combines them with a clustering method to classify the different positions of the ventricular assist device 100 in the patient's body. The model is trained using the classified training data to obtain a target position prediction model. Thus, by directly inputting the real-time operating parameters of the ventricular assist device 100 into the target position prediction model, the current position prediction of the ventricular assist device 100 can be output. Therefore, the position of the ventricular assist device 100 in the patient's body can be estimated directly based on the real-time operating parameters of the ventricular assist device 100 without relying on external equipment, and the current position can be determined as abnormal.
[0028] Based on the above description, this application will now be described from the perspective of method examples.
[0029] Please see Figure 3 , Figure 3 This is a flowchart illustrating a control unit performing position prediction, provided as an embodiment of this application, applied to the aforementioned ventricular assist device 100. For example... Figure 3 As shown, the method includes the following steps.
[0030] S310. Obtain multiple target parameter curves, each target parameter curve being a curve of one operating parameter of the ventricular assist device during the first cycle.
[0031] After the ventricular assist device 100 is implanted in the target user's body and activated, the control unit can acquire its operating parameters in real time, input the operating parameters into the machine learning model, and predict the current displacement of the ventricular assist device 100 in the target user's body and whether its position is abnormal.
[0032] The operating parameters of the ventricular assist device 100 include current, rotational speed, pump flow rate, left ventricular pressure or aortic pressure, and the pressure difference between the pressure at the fluid inlet 101 and the pressure at the fluid outlet 102. The control unit can acquire multiple operating parameters of the ventricular assist device in real time, and generate corresponding parameter curves for each operating parameter in the first cycle to obtain multiple target parameter curves. These multiple target parameter curves can reflect the current operating status of the ventricular assist device 100, and thus determine the position of the ventricular assist device 100 in the target user's body and whether its position is abnormal based on its operating status.
[0033] S320. Extract n features from the plurality of target parameter curves to obtain a target feature vector. The target feature vector includes m features, which are the features after dimensionality reduction of the n features. Both n and m are integers greater than 1, and n is less than m.
[0034] After acquiring multiple target parameter curves of the ventricular assist device 100, multiple features are extracted from each target parameter curve. These multiple features may include the maximum value, minimum value, mean, kurtosis, skewness, autocorrelation, variance, etc. of the target parameter curve. The multiple features of the multiple target parameter curves are used to generate a feature vector, which is a 1×n matrix.
[0035] To analyze the current position of the ventricular assist device 100 within the target user's body from n features extracted from multiple target parameter curves, the control unit can perform dimensionality reduction on the extracted n features. This retains the core patterns related to the positional state of the ventricular assist device 100 from multiple features, enabling intuitive feature analysis and improving the model's prediction accuracy and speed. For example, the control unit can use the UMAP algorithm to perform dimensionality reduction on the extracted n features, simplifying high-dimensional feature data into two-dimensional low-dimensional data, facilitating subsequent model analysis and prediction. Here, n can be greater than 10, and m can be greater than 1 and less than or equal to 3, such as n taking the value 130 and m taking the value 2.
[0036] Furthermore, to ensure the accuracy of operating parameters, the control unit can preprocess the features, such as filtering invalid features. For example, features with too small a variance are deleted, as a small variance indicates that the target parameter curve does not change in the first period, thus failing to provide location judgment information. Highly correlated features are also deleted (one of two highly correlated features is removed), as these features are redundant and will interfere with subsequent analysis results. Data standardization involves standardizing the retained valid features, adjusting the values of all features to the same scale to prevent single features from obscuring location correlation patterns due to differences in numerical scale.
[0037] S330. Input the target feature vector into the target location prediction model to obtain the predicted location classification.
[0038] After extracting features from multiple target parameter curves of the ventricular assist device 100, the feature vector composed of these features is input into a pre-trained target location prediction model, which outputs the predicted location classification of the ventricular assist device 100. Based on this predicted location classification, it can be determined whether the current location of the ventricular assist device 100 within the target user's body is abnormal. This target location prediction model can be a random forest model.
[0039] The ventricular assist device 100 within the patient's body can be categorized in several ways, such as normal, slightly deep, slightly shallow; normal, slightly deep, too deep, slightly shallow, too shallow; normal, slightly deep, too deep, slightly shallow; etc. "Normal" indicates that the distance between the contrast ring on the ventricular assist device 100 and the aortic valve is appropriate, such as [-1cm, 1cm], [0cm, 1cm]. "Slightly deep" indicates that the distance between the contrast ring on the ventricular assist device 100 and the aortic valve is large, but the fluid outlet 102 of the ventricular assist device 100 is still within the aorta, such as [-2cm, -1cm] or [-3cm, -2cm]. "Too deep" indicates that the distance between the contrast ring on the ventricular assist device 100 and the aortic valve is large, and the fluid outlet 102 of the ventricular assist device 100 is located directly within the aortic valve or the left ventricle, such as [-3cm, -2cm]. Slightly shallow indicates that the distance between the contrast ring on the ventricular assist device 100 and the aortic valve is close, but the fluid inlet 101 of the ventricular assist device 100 is still in the left ventricle, such as [1cm, 2cm] or [2cm, 3cm]. Too shallow indicates that the distance between the contrast ring on the ventricular assist device 100 and the aortic valve is close, and the fluid inlet 101 of the ventricular assist device 100 is located directly in the aortic valve or in the aorta, such as [2cm, 3cm].
[0040] Furthermore, when the target location prediction model outputs the current location of the ventricular assist device 100 within the target user's body, the control unit can classify abnormalities and issue alarms based on the predicted location.
[0041] For example, when the current position of the ventricular assist device 100 in the patient's body is classified as normal, slightly deep, or slightly shallow, if the target position prediction model outputs "normal," the current position of the ventricular assist device 100 can be determined to be qualified, and no alarm is required. If the target position prediction model outputs "slightly deep," the current position of the ventricular assist device 100 can be determined to be abnormal, and an alarm can be triggered in time. The control unit can issue an alarm and display a prompt such as "Position slightly deep, please move the blood pump outward by 1cm-2cm" on the controller or remote terminal (monitor or monitoring platform) connected to the ventricular assist device 100. If the target position prediction model outputs "slightly shallow," the current position of the ventricular assist device 100 can be determined to be abnormal, and an alarm can be triggered in time. The control unit can issue an alarm and display a prompt such as "Position slightly shallow, please move the blood pump inward by 1cm-2cm" on the controller or remote terminal connected to the ventricular assist device 100.
[0042] When the current position of the ventricular assist device 100 within the patient's body is classified as normal, slightly deep, too deep, slightly shallow, or too shallow, if the target position prediction model outputs "normal," the current position of the ventricular assist device 100 is considered acceptable, and no alarm is required. If the target position prediction model outputs "slightly deep," the current position of the ventricular assist device 100 is considered abnormal, but the ventricular assist device 100 can still continue to operate. The control unit can issue an alarm and display a message such as "Slightly deep position, which will affect the operation of the blood pump. Please move the blood pump outward by 1cm-2cm" on the controller or remote terminal connected to the ventricular assist device 100. If the target position prediction model outputs "too deep," the current position of the ventricular assist device 100 is considered abnormal. If the position of device 100 is abnormal and the ventricular assist device 100 is unable to pump blood, the control unit immediately issues an alarm and displays a message such as "Position too deep, blood pump cannot operate, immediately move the blood pump outward by 2cm-3cm" on the controller or remote terminal connected to the ventricular assist device 100. If the target position prediction model outputs "slightly shallow," it can be determined that the current position of the ventricular assist device 100 is abnormal, but the ventricular assist device 100 can still continue to operate. An alarm can be issued in time, and the control unit can issue an alarm and display a message such as "Position slightly shallow, which will affect the operation of the blood pump, please move the blood pump inward by 1cm-2cm" on the controller or remote terminal connected to the ventricular assist device 100. If the target position prediction model outputs "too shallow," it can be determined that the current position of the ventricular assist device 100 is abnormal and the ventricular assist device 100 is unable to pump blood. The control unit immediately issues an alarm and displays a message such as "Position too shallow, blood pump cannot operate, immediately move the blood pump inward by 2cm-3cm" on the controller or remote terminal connected to the ventricular assist device 100.
[0043] In this application, the control unit directly determines the position of the ventricular assist device 100 through its own operating parameters, enabling real-time position determination without relying on external equipment, thus reducing equipment costs and operational complexity. Furthermore, it can output clear position adjustment prompts for abnormal positions, guiding operators to gradually adjust the position of the ventricular assist device 100 until it is in a normal position. This eliminates the need for additional manual judgment, directly assisting surgical procedures and improving operational accuracy and safety.
[0044] In one possible example, regarding the training method of the target location prediction model, the control unit is specifically configured to: acquire t training datasets, each training dataset comprising (k-1)*2 j +1 training data sets, each training data set including multiple target parameter curves when the distance between the contrast ring on the ventricular assist device and the aortic valve is L, where t is a positive integer, j is a natural number, and k is the length of the cannula 10 of the ventricular assist device 100; (k-1)*2 of the t training data sets j +1 training data sets are divided into p target training sets and q target test sets, where p and q are positive integers; the p target training sets are used to train the location prediction model to be trained, thereby obtaining the target location prediction model; the q target test sets are used to verify whether the target location prediction model is qualified; if the target location prediction model is unqualified and falls under the first case, then let t = t + 1, and repeat the above steps until the target location prediction model is qualified; if the target location prediction model is unqualified and falls under the second case, then let j = j + 1, and repeat the above steps until the target location prediction model is qualified.
[0045] In this application, the operating parameters of the ventricular assist device 100 when it is in different positions of the patient are first used to train the position prediction model to be trained, so that the model can fully learn the characteristic rules of the corresponding operating parameters when the ventricular assist device 100 is in different positions, thereby realizing real-time prediction of the position of the ventricular assist device 100.
[0046] In this process, multiple operational parameters of the ventricular assist device 100 (VAD 100) at different distances between the contrast ring on the VAD and the aortic valve are collected through experiments or tests, forming a training dataset with distance and position labels. A time-sliding window is used to segment the operational parameter curves of the VAD 100, dividing the curves at the same position into multiple target parameter curves. This ensures that each training dataset includes multiple training data sets of the VAD 100 within different time windows, and each training data set includes multiple target feature curves within those time windows. The operational parameter curves of the VAD 100 within each time window are used as samples containing statistical regularities to generate one of the training data sets for training the position prediction model. In other words, one training data set corresponds to multiple time windows, ensuring that the sample data effectively reflects the correlation between position and operational parameters.
[0047] To accurately predict the position of the ventricular assist device 100, the collected training data set should cover as many locations of the ventricular assist device 100 as possible within the patient's body. Therefore, k is used as the maximum distance the imaging ring can move. For example, k can be set to 6cm, 6.5cm, 7cm, 7.5cm, etc. In other words, with the distance between the imaging ring and the aortic valve at 0 as coordinate point 0, and the direction of the imaging ring towards the left ventricle as the direction of the abscissa (i.e., the movement of the imaging ring towards the aorta is the positive direction of the abscissa, and the movement of the imaging ring towards the left ventricle is the negative direction of the abscissa), the maximum distance the imaging ring can move within the patient's body is k.
[0048] With a maximum movable distance of k for the contrast ring, curves of multiple operating parameters of the ventricular assist device 100 are collected when the device is located at different positions (distance L between the contrast ring and the aortic valve). This results in a training data set covering the operating data of the ventricular assist device 100 corresponding to the movement range [0, k] of the contrast ring. Where L = +(i-1) / j, or L=-k / 2j+(i-1) / j, where i=1,2,3,…,k. j is the number of iterations required to refine the location acquisition granularity for retraining the target location prediction model. For example, when k=7 and j=1, L=-3, -2, -1, 0, 1, 2, 3, multiple target parameter curves of the ventricular assist device 100 can be acquired as training data at distances of -3cm, -2cm, -1cm, 0cm, 1cm, 2cm, and 3cm between the contrast ring and the aortic valve, respectively. Alternatively, when k=7 and j=1, and L=-3.5, -2.5, -1.5, 0.5, 1.5, 2.5, and 3.5, multiple target parameter curves of the ventricular assist device 100 can be collected as training data when the distance between the contrast ring and the aortic valve is -3.5cm, -2.5cm, -1.5cm, -0.5cm, 0.5cm, 1.5cm, 2.5cm, and 3.5cm, respectively.
[0049] To improve the prediction accuracy of the target location prediction model, this application divides the collected training data into training data and test data proportionally. The training data is used to train the untrained location prediction model to obtain the target location prediction model. After training, the test data is used to verify the performance of the target location prediction model. Only when the target location prediction model passes verification can it be used to determine the actual location of the ventricular assist device 100. If the target location prediction model fails verification, further adjustments are made based on the model's characteristics: either increasing the sample size to retrain a new model, refining the granularity of the ventricular assist device's location acquisition within the patient to retrain a new model, or a combination of both, until the target location prediction model passes verification.
[0050] In this application, the position distribution of the ventricular assist device 100 is classified according to its operating parameters. This not only determines the position of the ventricular assist device 100 in the patient's body, but also determines whether the current position is abnormal. In case of abnormality, a dynamic adjustment alarm prompt can be output according to the position classification to guide the operator to gradually adjust to the normal position, reduce the risk and cost of surgical operation, and improve the operability and safety of interventional surgery.
[0051] To ensure reliable classification of the location of the ventricular assist device 100, this application uses cluster analysis to classify the operating parameters of the ventricular assist device at different locations, so that the location classification results are consistent with the intuitive distance perception of clinical operation and improve the practicality of location judgment.
[0052] Optionally, in the (k-1)*2 of the t training datasets j Regarding the division of +1 training data sets into p target training groups and q target test groups, the control unit is specifically used to: respectively process the (k-1)*2 of the t training datasets.j Feature extraction is performed on +1 training data sets to obtain t*((k-1)*2) j +1) target feature matrices; use clustering algorithm to analyze t*((k-1)*2 j +1) of the target feature matrices are classified to obtain the number of classifications for the t training datasets and the target location classification for each training data group; based on the number of classifications and the target location classification, the (k-1)*2 of the t training datasets are... j +1 training data sets are divided into p target training sets and q target test sets.
[0053] In each category, the classification ratio of the same target location in the target training group to the target test group is 7:3 or 8:2.
[0054] Specifically, for each training data set, features such as maximum, minimum, mean, kurtosis, skewness, autocorrelation, and variance are extracted from each target parameter curve. These features are then used to generate a feature matrix, thus obtaining the feature matrix for that training data set. To retain the core patterns related to the positional state of the ventricular assist device 100 from multiple features, dimensionality reduction is performed on each feature matrix. The UMAP algorithm is used to reduce the dimensionality of the extracted n features, simplifying the high-dimensional feature data into two-dimensional low-dimensional data, resulting in the target feature matrix for each training data set. The k-means clustering algorithm is then used on the dimensionality-reduced two-dimensional target feature matrix for automatic grouping, obtaining the number of categories and the positions of the ventricular assist devices 100 included in each category.
[0055] For example, if the distance between the annular ring and the aortic valve in the collected training data group is -3cm, -2cm, -1cm, 0cm, 1cm, 2cm, and 3cm respectively, when the number of groups obtained by the k-means clustering algorithm is 3, the 3 groups are classified as: too deep, normal, and too shallow; when the number of groups is 5, the 5 groups are classified as: too deep, slightly deep, normal, slightly shallow, and too shallow.
[0056] Furthermore, the control unit determines the category based on the number of classification groups and the average distance between the corresponding radiolucent rings and the aortic valve in each category, or it determines the category based on the number of classification groups and the descending order of the average distance between the corresponding radiolucent rings and the aortic valve in each category. For example, in a classification group of 3, if the average distance between all radiolucent rings and the aortic valve in that category is in the range of [-1, 1], then the category is determined to be normal; if the average distance between all radiolucent rings and the aortic valve in that category is less than -2, then the category is determined to be too deep; if the average distance between all radiolucent rings and the aortic valve in that category is greater than 2, then the category is determined to be too shallow. In a classification group of 5, if the mean distance between all the visible rings and the aortic valve in the classification is within [-1, 1], then the classification is determined as normal; if the mean distance between all the visible rings and the aortic valve in the classification is less than -1 and greater than or equal to -2, then the classification is determined as slightly deep; if the mean distance between all the visible rings and the aortic valve in the classification is less than -2, then the classification is determined as too deep; if the mean distance between all the visible rings and the aortic valve in the classification is greater than 1 and less than or equal to 2, then the classification is determined as slightly shallow; if the mean distance between all the visible rings and the aortic valve in the classification is greater than 2, then the classification is determined as too shallow.
[0057] In this application, to ensure the model can accurately distinguish between different categories of positional states, training data sets whose positions belong to the same category are proportionally divided into a target training group and a target test group. Specifically, (k-1)*2 j +1 training datasets are divided into p target training groups and q target test groups. In each location classification, the ratio of training datasets to target test groups is 7:3 or 8:2, i.e., p:q = 7:3 or 8:2. The target training groups are used to train the location prediction model, and then the target test groups are used to verify the model's performance. If the model passes verification, it can be used for subsequent location predictions. If the model fails verification, either t = t+1 is used to increase the training dataset to increase the sample size, or j = j+1 is used to increase the number of training datasets to refine the location acquisition granularity. The model is then retrained using the same method until it passes verification.
[0058] The target location prediction model is a random forest model. It takes the target feature vectors of p target training groups as input and the location classification as output to train the location prediction model. During the training process, the location prediction model is continuously updated with the actual location classification of the p target training groups, so that the model can fully learn the feature patterns of the running data corresponding to different location classifications and obtain the target location prediction model.
[0059] Optionally, in verifying the qualification of the target location prediction model using the q target test groups, the control unit is specifically configured to: input the q target test groups into the target location prediction model to obtain q predicted location classifications; calculate the average discrimination and confusion matrix between the q predicted target location classifications and the target location groups of the q target test groups respectively; if the average discrimination is greater than or equal to a first threshold and the overall accuracy of the confusion matrix is greater than or equal to a first value, then determine that the target location prediction model is qualified.
[0060] To improve the prediction accuracy of the target location prediction model, after training the model with p target training sets, q target test sets were used to verify its suitability. The performance of the target location prediction model was evaluated by calculating the average discrimination index and confusion matrix, verifying its ability to predict the location of ventricular assist devices. Furthermore, when the performance of the target location prediction model did not meet the requirements, the training dataset was further optimized by increasing the sample size or optimizing the granularity of location acquisition, allowing the model to learn more refined location boundary features, ultimately resulting in a high-precision and highly reliable target location prediction model.
[0061] Specifically, the target feature matrices corresponding to the q target test groups are input into the target location prediction model, which outputs q predicted location classifications. The average discrimination and confusion matrix are calculated based on the output q predicted location classifications and the actual location classifications of the q target test groups. The discrimination index indicates the current location prediction model's ability to distinguish between different location classifications, with a value between 0.5 and 1; the closer the average discrimination index is to 1, the stronger the model's discrimination ability. The confusion matrix visually presents the difference between the model's predictions and the actual results; the diagonal of the confusion matrix represents the number of correctly classified samples, and the off-diagonal represents the number of misclassified samples. Analyzing the average discrimination index and confusion matrix allows for the evaluation of the current target location prediction model's prediction accuracy.
[0062] In this application, the control unit can determine whether the target location prediction model is qualified and the reason for its failure based on the average discrimination and the overall accuracy of the confusion matrix. If the average discrimination is greater than or equal to a first threshold and the overall accuracy of the confusion matrix is greater than or equal to a first value, the target location prediction model is determined to be qualified; otherwise, the target prediction model is determined to be unqualified and needs to be retrained. There are two situations in which the target location prediction model is unqualified: the first situation is that the model fails to learn the relationship between the position of the ventricular assist device 100 and the operating parameters of the ventricular assist device 100; the second situation is that the granularity of the location acquisition in the training dataset is too coarse, resulting in a large deviation in the classification boundary determination.
[0063] The first scenario where the target location prediction model is deemed unqualified is when the average discrimination is less than a second threshold and the overall accuracy of the confusion matrix is less than a second value. The second scenario where the target location prediction model is deemed unqualified is when the average discrimination is greater than a third threshold and the overall accuracy of the confusion matrix is greater than the second value but less than the first value.
[0064] The following explanation uses a first threshold of 0.9, a second threshold of 0.7, a third threshold of 0.8, a first value of 0.8, and a second value of 0.6 as examples. However, the first threshold, the second threshold, the third threshold, the first value, and the second value can be set according to the actual application scenario or actual needs. For example, the first threshold can be set to 0.85, the second threshold to 0.75, the third threshold to 0.8, the first value to 0.85, and the second value to 0.7. This application does not limit this.
[0065] For example, in the first scenario, the control unit can increase the sample size, i.e., increase the amount of training dataset. When the average discrimination of the target location prediction model is below 0.7, the overall classification accuracy of the confusion matrix is below 0.6, and the confusion matrix shows that the recall of each class is generally low and there is no obvious bias in sample misclassification, it indicates that the target location prediction model has not converged and has not learned the core correlation between the different locations of the ventricular assist device and its operating parameters. In this case, it is necessary to increase the amount of sample data and supplement the sample size of the operating parameters corresponding to the original locations of the ventricular assist device 100. This ensures that the effective sample size of each location covers the fluctuations in the operating parameters of the ventricular assist device 100 and the differences in different experimental conditions, so that the target location prediction model can learn the stable correlation between the location and the operating parameters. Then, the target location prediction model can be retrained based on the supplemented training dataset.
[0066] For the second scenario, the control unit can increase the number of training data sets to refine the granularity of location acquisition. When the average discrimination of the target location prediction model is greater than 0.8, the overall classification accuracy of the confusion matrix is greater than 0.6 and less than 0.8, and the confusion matrix shows that misclassifications are concentrated in adjacent location samples, while the recall rate of samples at classification boundary locations is significantly low, it indicates that the model has learned the core correlation between location and operating parameters. However, the large deviation in classification boundary judgment is caused by the coarse granularity of location acquisition in the training dataset. In this case, the original interval of location acquisition can be reduced to increase the number of operating parameters corresponding to more locations. For example, the original 1cm location acquisition interval can be adjusted to 0.5cm, thereby increasing the original acquisition positions of -3cm, -2cm, -1cm, 0cm, 1cm, 2cm, 3cm to -3cm, - Locations are collected at 2.5cm, -2cm, -1.5cm, -1cm, -0.5cm, 0cm, 0.5cm, 1cm, 1.5cm, 2cm, 2.5cm, and 3cm to allow the location prediction model to learn more refined location boundary features. Based on the refined dataset, the location prediction model is re-clustered and trained, and the classification boundary determination criteria are optimized, thereby improving the prediction accuracy of the target location prediction model.
[0067] Furthermore, if the target location prediction model simultaneously exhibits both of the above conditions—that is, the average discrimination of the target location prediction model is greater than 0.7, the overall classification accuracy of the confusion matrix is less than 0.7, and misclassification involves both irregular random errors and adjacent location bias—then an optimization strategy of increasing the sample size is first implemented to allow the location prediction model to converge and learn the core association rules. Once the average discrimination of the target location prediction model and the overall classification accuracy of the confusion matrix reach the size of the refined location acquisition granularity mentioned above, a refined location acquisition granularity optimization strategy is then implemented to further improve the location prediction performance of the target location prediction model.
[0068] For example, assuming the initial position acquisition interval of the ventricular assist device 100 is 1cm, the current curve, rotation speed curve, and differential pressure curve of the ventricular assist device 100 are acquired at distances of -3cm, -2cm, -1cm, 0cm, 1cm, 2cm, and 3cm between the contrast ring on the ventricular assist device 100 and the aortic valve, respectively, as training data. That is, the initial training dataset includes 7 training data groups. The current curve, rotation speed curve, and differential pressure curve are segmented using time windows, so that each training data group includes current curves, rotation speed curves, and differential pressure curves under multiple time windows. Then, feature extraction and dimensionality reduction are performed on these 7 training data groups to obtain the target feature matrix for each training data group. The k-means clustering algorithm is used to automatically group the dimensionality-reduced two-dimensional data. If the number of groups is 3, the 7 distance positions are automatically classified into three categories based on the mean distance within the group: "too deep (including -3cm, -2cm), normal (including -1cm, 0cm, 1cm), and too shallow (including 2cm, 3cm)". The training data in these three categories were divided into training and test data in an 8:2 ratio. Specifically, in the "Normal" category, 80% of the data in the -1cm, 0cm, and 1cm positions were assigned to training data, and 20% to test data. The split training data was then used to train the target location prediction model, resulting in the target location prediction model. The split test data was then input into the target location prediction model, outputting the predicted location classification. The average discrimination and confusion matrix between the predicted location classification and the actual predicted location classification of the test data were calculated. If the average discrimination was greater than 0.9 and the overall classification accuracy of the confusion matrix was greater than 0.8, the target location prediction model was considered valid and could be directly used for subsequent location prediction of the ventricular assist device 100. If the average discrimination was less than 0.7 and the overall classification accuracy of the confusion matrix was less than 0.6, the target location prediction model was considered invalid and fell into the first category, meaning the target location prediction model had not converged, and the sample size needed to be increased. Therefore, the training dataset is increased from one to two, with each dataset still including training data for the aforementioned seven locations. The grouping, training, and validation processes are then performed as described above. If the average discrimination is greater than 0.8 and the overall classification accuracy of the confusion matrix is greater than 0.6 and less than 0.8, the current target location prediction model is considered unqualified and falls into the second category, meaning the target location prediction model has low accuracy. Therefore, the location acquisition granularity needs to be optimized, i.e., the number of training data groups in each training dataset needs to be increased.Therefore, the original 1cm position acquisition interval was adjusted to 0.5cm, increasing the original 7 acquisition positions (-3cm, -2cm, -1cm, 0cm, 1cm, 2cm, 3cm) to 13 acquisition positions (-3cm, -2.5cm, -2cm, -1.5cm, -1cm, -0.5cm, 0cm, 0.5cm, 1cm, 1.5cm, 2cm, 2.5cm, 3cm). This results in each training dataset containing 13 training data groups, each still including current curves, rotational speed curves, and differential pressure curves under different time windows. The grouping, training, and validation processes were then repeated as described above until the target position prediction model passed validation.
[0069] Alternatively, assuming the initial position acquisition interval of the ventricular assist device 100 is 1 cm, the current curve, rotational speed curve, and differential pressure curve of the ventricular assist device 100 are acquired at distances of -3.5 cm, -2.5 cm, -1.5 cm, -0.5 cm, 0.5 cm, 1.5 cm, 2.5 cm, and 3.5 cm between the contrast ring on the ventricular assist device 100 and the aortic valve, respectively, as training data. That is, the initial training dataset includes 8 training data sets, and the current curve, rotational speed curve, and differential pressure curve are segmented using time windows, so that each training data set includes current curves, rotational speed curves, and differential pressure curves under multiple time windows. Then, feature extraction and dimensionality reduction are performed on these 8 training data sets to obtain the target feature matrix for each training data set. The k-means clustering algorithm is then used to automatically group the dimensionality-reduced 2D data. If the number of groups is 4, the 8 distance locations are automatically classified into four categories based on the mean distance within each group: "Too Deep (including -3.5cm, -2.5cm), Slightly Deep (including -1.5cm), Normal (including -0.5cm, 0.5cm, 1.5cm), and Too Shallow (including 2.5cm, 3.5cm)". The training data in these four categories is then divided into training and testing data in a 7:3 ratio. For example, in the "Normal" category, 70% of the data in the training data sets for the -0.5cm, 0.5cm, and 1.5cm locations is assigned to training data, and 30% to testing data. The divided training data is then used to train the location prediction model to obtain the target location prediction model. The divided testing data is then input into the target location prediction model to output the location prediction classification. Calculate the average discrimination and confusion matrix between the predicted location classification and the actual predicted location classification of the test data. If the average discrimination is greater than 0.9 and the overall classification accuracy of the confusion matrix is greater than 0.8, the current target location prediction model is considered valid and can be directly used for subsequent location prediction of the ventricular assist device 100. If the average discrimination is less than 0.7 and the overall classification accuracy of the confusion matrix is less than 0.6, the current target location prediction model is considered unqualified and falls into the first category, i.e., the target location prediction model has not converged, and the sample size needs to be increased. Therefore, one training dataset is increased to two training datasets, each still including training data for the above 8 locations, and then the grouping-training-validation method is performed as described above. If the average discrimination is greater than 0.8 and the overall classification accuracy of the confusion matrix is greater than 0.6 and less than 0.8, the current target location prediction model is considered unqualified and falls into the second category, i.e., the target location prediction model has low accuracy, and the location acquisition granularity needs to be optimized, i.e., the number of training data groups in each training dataset needs to be increased.Therefore, the original 1cm position acquisition interval was adjusted to 0.5cm, increasing the original 8 acquisition positions (-3.5cm, -2.5cm, -1.5cm, 0.5cm, -0.5cm, 1.5cm, 2.5cm, 3.5cm) to 15 acquisition positions (-3.5cm, -3cm, -2.5cm, -2cm, -1.5cm, -1cm, -0.5cm, 0cm, 0.5cm, 1cm, 1.5cm, 2cm, 2.5cm, 3cm, 3.5cm). This results in each training dataset containing 15 training data groups, each still including current curves, rotational speed curves, and differential pressure curves under different time windows. The same grouping, training, and validation process was then repeated until the target position prediction model passed validation.
[0070] In this application, the position status is determined by the operating parameters of the ventricular assist device 100 itself and a model, and then a clear adjustment prompt is output when the position is abnormal. This not only allows for application in scenarios without visualization equipment, eliminating radiation exposure risks and reducing equipment costs and operational complexity, but also improves the accuracy and safety of the ventricular assist device 100 implantation operation. Furthermore, pre-classifying the position through cluster analysis before training the model can improve the model's prediction accuracy and ensure the reliability of position determination and alarm prompts.
[0071] As can be seen, this application proposes a control unit for a ventricular assist device (VAD). The control unit acquires multiple target parameter curves, each of which represents an operating parameter of the VAD during the first cycle. It extracts n features from these multiple target parameter curves to obtain a target feature vector, which includes m features obtained by dimensionality reduction of the n features. The target feature matrix is input into a target position prediction model to obtain a predicted position classification. This application directly inputs the real-time operating parameters of the VAD into the target position prediction model to output the current position prediction of the VAD. Therefore, it can directly estimate the position of the VAD within the patient's body based on the real-time operating parameters of the VAD 100 without relying on external equipment, reducing equipment costs and operational complexity.
[0072] The above primarily describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the network device includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0073] For example, an embodiment of this application provides a ventricular assist device, the ventricular assist device comprising: impeller; A motor that drives the impeller to rotate; A control unit connected to the motor is configured to perform the following steps: Multiple target parameter curves are obtained, each target parameter curve being a curve of one operating parameter of the ventricular assist device during the first cycle; n features are extracted from the multiple target parameter curves to obtain a target feature vector. The target feature vector includes m features, which are the features after dimensionality reduction of the n features. Both n and m are integers greater than 1, and n is less than m. The target feature vector is input into the target location prediction model to obtain the predicted location classification.
[0074] For example, an embodiment of this application provides a medical device, which includes a processor, a memory, a communication interface, and one or more programs. The one or more programs are stored in the memory and configured to be executed by the processor. The programs include instructions for performing some or all of the steps performed by the control unit.
[0075] For example, embodiments of this application provide a computer-readable storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps performed by a control unit.
[0076] The control unit of each of the above schemes has the function of implementing the corresponding steps executed by the control unit; the function can be implemented by hardware or by hardware executing corresponding software.
[0077] In embodiments of this application, the control unit may also be a chip or a chip system, such as a system on chip (SoC).
[0078] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a medical device provided in an embodiment of this application. The medical device includes: one or more processors, one or more memories, one or more communication interfaces, and one or more programs; the one or more programs are stored in the memories and configured to be executed by the one or more processors.
[0079] The above procedure includes instructions for performing the following steps: Multiple target parameter curves are obtained, each target parameter curve being a curve of one operating parameter of the ventricular assist device during the first cycle; n features are extracted from the multiple target parameter curves to obtain a target feature vector. The target feature vector includes m features, which are the features after dimensionality reduction of the n features. Both n and m are integers greater than 1, and n is less than m. The target feature vector is input into the target location prediction model to obtain the predicted location classification.
[0080] All relevant content in each scenario involved in the above embodiments can be referenced from the functional descriptions of the corresponding functional modules, and will not be repeated here.
[0081] It should be understood that the aforementioned memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store information about the device type.
[0082] In the embodiments of this application, the processor of the above-described device may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.
[0083] It should be understood that "at least one" in the embodiments of this application refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.
[0084] Furthermore, unless otherwise stated, the ordinal numbers such as "first" and "target" mentioned in the embodiments of this application are used to distinguish multiple objects and are not used to limit the order, timing, priority, or importance of multiple objects. For example, "first information" and "target information" are only used to distinguish different information, and do not indicate differences in the content, priority, sending order, or importance of these two types of information.
[0085] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly manifested as execution by a hardware processor, or as a combination of hardware and software units within the processor. The software units can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor executes the instructions in the memory, combining them with its hardware to complete the steps of the above method. To avoid repetition, detailed descriptions are omitted here.
[0086] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the methods described in the above method embodiments.
[0087] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. This computer program product can be a software installation package.
[0088] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0089] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0090] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.
[0091] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.
[0092] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0093] If the aforementioned integrated units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or TRP, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0094] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include a flash drive, ROM, RAM, disk, or optical disk, etc.
[0095] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A control unit for a ventricular assist device, characterized in that, The control unit is used to perform the following steps: Multiple target parameter curves are obtained, each target parameter curve being a curve of one operating parameter of the ventricular assist device during the first cycle; n features are extracted from the plurality of target parameter curves to obtain a target feature vector. The target feature vector includes m features, which are the features after dimensionality reduction of the n features. Both n and m are integers greater than 1, and n is less than m. The target feature vector is input into the target location prediction model to obtain the predicted location classification.
2. The control unit according to claim 1, characterized in that, Regarding the training method of the target location prediction model, the control unit is specifically used for: Obtain t training datasets, each training dataset consisting of (k-1)*2 j +1 training data set, each training data set including multiple target parameter curves when the distance between the contrast ring on the ventricular assist device and the aortic valve is L, where t is a positive integer, j is a natural number, and k is the length of the cannula of the ventricular assist device; The (k-1)*2 of the t training datasets j +1 training data groups are divided into p target training groups and q target test groups, where p and q are positive integers; The target location prediction model is obtained by training the p target training groups. The target location prediction model is verified to be qualified using the q target test groups; If the target location prediction model is not qualified and is in the first case, then let t=t+1 and repeat the above steps until the target location prediction model is verified to be qualified. If the target location prediction model is not qualified and falls under the second case, then let j = j + 1 and repeat the above steps until the target location prediction model is verified as qualified.
3. The control unit according to claim 2, characterized in that, In the (k-1)*2 of the t training datasets j Regarding the division of +1 training data groups into p target training groups and q target test groups, the control unit is specifically used for: For the (k-1)*2 of the t training datasets respectively j +1 training data sets are used for feature extraction to obtain t*((k-1)*2) j +1) target feature matrices; Use clustering algorithms to analyze t*((k-1)*2) j +1) classify the target feature matrices to obtain the number of classifications of the t training datasets and the target location classification of each training data group; Based on the number of categories and the target location, classify the (k-1)*2 of the t training datasets. j +1 training data sets are divided into p target training sets and q target test sets.
4. The control unit according to claim 3, characterized in that, The ratio of the same target location classification in the target training group and the target test group is 7:3 or 8:
2.
5. The control unit according to claim 3 or 4, characterized in that, In verifying the suitability of the target location prediction model using the q target test groups, the control unit is specifically configured to: Input the q target test groups into the target location prediction model to obtain q predicted location classifications; Calculate the average discrimination and confusion matrix between the q predicted target location classifications and the target location groups of the q target test groups, respectively; If the average discrimination is greater than or equal to the first threshold and the overall accuracy of the confusion matrix is greater than or equal to the first value, then the target location prediction model is deemed to have passed the verification.
6. The control unit according to claim 5, characterized in that, The target location prediction model is unqualified and falls under the first category: the average discrimination is less than the second threshold, and the overall accuracy of the confusion matrix is less than the second value. And / or, the target location prediction model is unqualified and falls under the second condition: the average discrimination is greater than the third threshold, and the overall accuracy of the confusion matrix is greater than the second value and less than the first value.
7. The control unit according to claim 2, characterized in that, The L= +(i-1) / j, or L=-k / 2j+(i-1) / j, i=1,2,3,…,k.
8. A ventricular assist device, characterized in that, The ventricular assist device includes: impeller; A motor that drives the impeller to rotate; A control unit connected to the motor is configured to perform the following steps: Multiple target parameter curves are obtained, each target parameter curve being a curve of one operating parameter of the ventricular assist device during the first cycle; n features are extracted from the plurality of target parameter curves to obtain a target feature vector. The target feature vector includes m features, which are the features after dimensionality reduction of the n features. Both n and m are integers greater than 1, and n is less than m. The target feature vector is input into the target location prediction model to obtain the predicted location classification.
9. A medical device, characterized in that, The method includes a processor, a memory, and a communication interface, wherein the memory stores one or more programs, and the one or more programs are executed by the processor, the one or more programs including instructions for performing the steps of the method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program for electronic data interchange, wherein the computer program causes a computer to perform the steps of the method as described in any one of claims 1-7.