Method and device for improving rotation robustness of aortic valve CTA image classifier
Through adversarial training technology and axial rotation attack algorithm, an adversarial sample set is generated and adversarial training is solved, and the aortic valve CTA image classifier's performance degradation under rotation transformation is significantly improved, which is suitable for existing systems.
Patent Information
- Application Number
- CN202411783760.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-06
- Publication Date
- 2025-03-28
AI Technical Summary
The performance of existing aortic valve CTA image classifiers has decreased under rotational transformation, resulting in reduced accuracy and robustness, and the existing methods are highly complex and not highly applicable.
By designing an axial rotation attack algorithm, the training sample set is axially rotated, and the adversarial sample set is generated, and it is added to the training sample set to form a rotation pool, and the aortic valve CTA image classifier is trained against each other to improve its rotation robustness.
It significantly improves the rotational robustness and accuracy of the aortic valve CTA image classifier, without the need for geometric feature extraction or complex modification of the classifier architecture, is highly applicable and has important clinical application value.
Smart Images
Figure CN119723181A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent medical technology, and in particular to a method, device and computer-readable storage medium for improving the rotational robustness of an aortic valve CTA image classifier. Background Art
[0002] Aortic valve CT angiography (CTA) images are an important diagnostic tool in cardiac surgery, used to display the morphology and function of the aortic valve. Through CTA images, doctors can intuitively observe the anatomical structure of the aortic valve, the movement of the valve, and the direction of related blood vessels, which is of great significance for diagnosing aortic valve disease, assessing the severity of the disease, and formulating surgical plans.
[0003] However, in practical applications, CTA image data is often affected by geometric transformations such as rotation and translation, which mainly come from changes in patient position, angle adjustment of scanning equipment, and motion artifacts during image acquisition. These geometric transformations may cause changes in the direction and position of the image, thereby affecting the performance of the image classifier and reducing its accuracy and robustness.
[0004] At present, in response to this problem, researchers have proposed some methods to improve the robustness of image classifiers. For example, the geometric descriptor extraction method enhances the robustness of the classifier by extracting the geometric features of the image, such as shape and edges. Although these methods have improved the performance of the classifier to a certain extent, they require complex feature extraction and processing, which increases the complexity of the system. In addition, the architecture modification method adjusts and optimizes the existing classifier network architecture to adapt to the geometric transformation of the image. These methods usually require a lot of computing resources and time, and are difficult to promote in practical applications. Summary of the invention
[0005] The embodiments of the present application provide a method and device for improving the rotation robustness of an aortic valve CTA image classifier, thereby solving the problem that existing methods require complex feature extraction and processing, or require complex modifications to the classifier architecture, resulting in high system complexity and poor applicability.
[0006] In a first aspect, an embodiment of the present application provides a method for improving the rotation robustness of an aortic valve CTA image classifier, comprising the following steps:
[0007] Acquire aortic valve CTA image data, and pre-process the acquired data to form a training sample set;
[0008] Extracting aortic valve features of the training sample set;
[0009] According to the extracted aortic valve features, an axial rotation attack algorithm is designed; the training sample set is subjected to axial rotation transformation by the axial rotation attack algorithm to generate an adversarial sample set with different rotation angles;
[0010] Adding the adversarial sample set to the training sample set to form a rotation pool;
[0011] The aortic valve CTA image classifier is adversarially trained through the rotation pool to improve the rotation robustness of the aortic valve CTA image classifier.
[0012] Preferably, the aortic valve characteristics include the morphology of the aortic valve, the position of the valve and the direction of the blood vessels.
[0013] Preferably, a pre-trained convolutional neural network is used to extract high-level semantic features of the training sample set to achieve aortic valve feature extraction.
[0014] Preferably, the axial rotation attack algorithm comprises the following steps:
[0015] (A) Determination of the rotation center
[0016] In the aortic valve CTA image, the rotation center point is the center of the blood vessel or the center of the aortic valve according to the aortic valve characteristics;
[0017] (B) Determination of rotation angle
[0018] The rotation angle is determined according to the aortic valve characteristics of the aortic valve CTA image and the rotation robustness required by the aortic valve CTA image classifier;
[0019] (C) Rotation Matrix Determination
[0020] The rotation matrix is used to transform the coordinates of each pixel in the image. For a three-dimensional aortic valve CTA image, the rotation matrix R is as follows:
[0021]
[0022] Where θ is the rotation angle;
[0023] (D) Coordinate transformation
[0024] For each pixel in the aortic valve CTA image, the rotation matrix R is applied to perform coordinate transformation and calculate the new coordinate after rotation.
[0025] Furthermore, the rotation angle range is set to [-30°, 30°] with a step size of 5°.
[0026] Furthermore, the axial rotation attack algorithm further comprises the following steps:
[0027] (E) Image interpolation
[0028] Since the new coordinates after rotation may not be integers, an interpolation method is used to determine the pixel value at the new coordinates;
[0029] Furthermore, the axial rotation attack algorithm further comprises the following steps:
[0030] (F) Image resizing
[0031] Rotation may cause the image's bounds to change; the image is resized to fit the new bounds.
[0032] Preferably, when the aortic valve CTA image classifier is subjected to adversarial training through the rotation pool, an optimization algorithm Adam based on a first-order gradient is adopted to accelerate the convergence speed of the adversarial training.
[0033] In a second aspect, an embodiment of the present application provides a device for improving the rotation robustness of an aortic valve CTA image classifier, comprising:
[0034] A data acquisition and preprocessing module is used to acquire aortic valve CTA image data and preprocess the acquired data to form a training sample set;
[0035] A feature extraction module, used to extract aortic valve features of the training sample set;
[0036] An adversarial sample generation module is used to design an axial rotation attack algorithm based on the extracted aortic valve features; and to perform an axial rotation transformation on the training sample set through the axial rotation attack algorithm to generate an adversarial sample set with different rotation angles;
[0037] A rotation pool construction module, used for adding the adversarial sample set to the training sample set to form a rotation pool;
[0038] The adversarial training module is used to perform adversarial training on the aortic valve CTA image classifier through the rotation pool to improve the rotation robustness of the aortic valve CTA image classifier.
[0039] In a third aspect, an embodiment of the present application provides a device for improving the rotational robustness of an aortic valve CTA image classifier, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for improving the rotational robustness of an aortic valve CTA image classifier when executing the program.
[0040] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the above-mentioned method for improving the rotation robustness of the aortic valve CTA image classifier are implemented.
[0041] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:
[0042] 1. The present invention utilizes adversarial training technology, designs an axial rotation attack algorithm and a rotation pool, generates diversified training data, and significantly improves the rotation robustness and accuracy of the aortic valve CTA image classifier.
[0043] 2. Compared with traditional methods, the present invention does not require geometric feature extraction or complex modifications to the classifier architecture. It can be directly applied to the existing aortic valve CTA image classifier system, has good versatility and applicability, and has important clinical application value and promotion potential.
[0044] 3. The method of the present invention is easy to implement, simple and efficient, and does not require geometric feature extraction or complex modifications to the classifier architecture, which greatly reduces the complexity of system development and also reduces the system deployment cost. This provides a simple and effective solution for the clinical application of aortic valve CTA image classification in cardiac surgery.
[0045] 4. The present invention targets the specific technical field of aortic valve CT angiography images, and makes full use of the unique features of aortic valve CT angiography images, such as the morphology of the aortic valve, the position of the valve and the direction of the blood vessels, to design an axial rotation attack algorithm, thereby greatly improving the accuracy of the classifier on this type of medical imaging data. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0047] Figure 1 This is a flow chart of the method for improving the rotation robustness of the aortic valve CTA image classifier provided in Example 1 of the present application;
[0048] Figure 2 This is a flow chart of the axial rotation attack algorithm provided in Example 1 of the present application;
[0049] Figure 3 This is a structural block diagram of a device for improving the rotational robustness of an aortic valve CTA image classifier provided in Example 2 of the present application;
[0050] Figure 4 This is a structural block diagram of the device for improving the rotational robustness of the aortic valve CTA image classifier provided in Example 3 of the present application.
[0051] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0052] The embodiments of the present application provide a method and device for improving the rotation robustness of an aortic valve CTA image classifier, thereby solving the problem that existing methods require complex feature extraction and processing, or require complex modifications to the classifier architecture, resulting in high system complexity and poor applicability.
[0053] In order to better understand the above technical solution, exemplary embodiments will be described in detail here, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are only examples of devices consistent with some aspects of the present application as detailed in the attached claims.
[0054] Embodiment 1
[0055] Figure 1 Flow chart of the method for improving the rotational robustness of the aortic valve CTA image classifier provided in this embodiment. The method for improving the rotational robustness of the aortic valve CTA image classifier comprises the following steps:
[0056] Step S1: acquiring aortic valve CT angiography image (CTA image) data, and preprocessing the acquired data to form a training sample set;
[0057] In this example, we obtained an aortic valve CTA image database from a tertiary hospital, which contains CTA image data of 200 patients. These data cover different degrees of aortic valve diseases, such as aortic valve stenosis, aortic valve regurgitation, etc. All image data have been diagnosed and labeled by professional doctors, providing a reliable basis for subsequent model training and evaluation.
[0058] The preprocessing includes performing denoising, normalization and other processing on the original aortic valve CTA image data to form training samples.
[0059] Step S2: extracting the aortic valve features of the training sample set, specifically including extracting features such as the morphology of the aortic valve, the position of the valve, and the direction of the blood vessels.
[0060] In this embodiment, we use a feature extraction method based on deep learning and use a pre-trained convolutional neural network (such as ResNet-50) to extract high-level semantic features of aortic valve CTA images. These features include the shape, size, position of the aortic valve, and the direction of related blood vessels. These features are extracted and quantified to provide effective input for subsequent adversarial training.
[0061] Step S3: designing an axial rotation attack algorithm according to the extracted aortic valve features; performing an axial rotation transformation on the training sample set by using the axial rotation attack algorithm to generate an adversarial sample set with different rotation angles;
[0062] Combination Figure 2 ,The specific process of the axial rotation attack algorithm is as follows:
[0063] (A) Determination of the rotation center
[0064] The rotation center point is usually a feature point of the image. In the aortic valve CTA image, the rotation center point is the center of the blood vessel or the center of the aortic valve for the aortic valve feature. It is determined according to the angle and core elements of different aortic valve CTA images.
[0065] (B) Determination of rotation angle:
[0066] The rotation angle should be set according to the target of the attack and the required rotation robustness. For the aortic valve feature, in this embodiment, the rotation angle range is set to [-30°, 30°], with a step size of 5°, and a total of 13 adversarial samples with different angles are generated.
[0067] (C) Rotation matrix determination:
[0068] The rotation matrix is used to transform the coordinates of each pixel in the image. For a 3D aortic valve CTA image, the rotation matrix R is as follows:
[0069]
[0070] Here, θ is the rotation angle.
[0071] (D) Coordinate transformation:
[0072] For each pixel in the aortic valve CTA image, the rotation matrix R is applied to perform coordinate transformation and calculate the new coordinate after rotation.
[0073] (E) Image interpolation:
[0074] Since the new coordinates after rotation may not be integers, an interpolation method is needed to determine the pixel value at the new coordinates. Common interpolation methods include nearest neighbor interpolation, bilinear interpolation, or cubic spline interpolation.
[0075] (F) Image resizing:
[0076] Rotation may cause the image's bounds to change, so the image needs to be resized to fit the new bounds.
[0077] In this way, the generated adversarial samples simulate the rotation interference that may occur in practical applications and provide targeted training samples for the classifier.
[0078] Step S4: adding the adversarial sample set to the training sample set to form a rotation pool;
[0079] In addition to using the original aortic valve CTA images for training, we also introduce the adversarial sample set generated by the axial rotation attack algorithm into the training sample set to increase the diversity of training data and improve the robustness of the classifier.
[0080] Step S5: The aortic valve CTA image classifier is adversarially trained through the rotation pool. During the adversarial training process, the aortic valve CTA image classifier will face the challenges of training samples and adversarial samples at the same time, so as to learn more robust feature representation to improve the rotation robustness of the aortic valve CTA image classifier.
[0081] In a preferred embodiment, in order to speed up the convergence of adversarial training, we use the first-order gradient-based optimization algorithm Adam. Compared with the traditional gradient descent method, Adam can find the optimal solution faster.
[0082] In this embodiment, the above-mentioned adversarial training framework, including the axial rotation attack algorithm, the rotation pool and the fast first-order optimization algorithm, is designed using the Python language. Of course, other languages can also be used to design the adversarial training framework according to actual conditions.
[0083] After training, we evaluate the performance of the aortic valve CTA image classifier on two datasets:
[0084] (1) Cleanly aligned dataset: contains aortic valve CTA images without rotation transformation.
[0085] (2) Rotation dataset: The dataset is generated by randomly rotating the original image (angle range [-45°, 45°]).
[0086] Experiments show that the accuracy of the trained aortic valve CTA image classifier on the clear aligned dataset is 98%, and the accuracy on the rotated dataset is 95%. The accuracy of the traditional system on the rotated dataset is only 70%, proving that the method of this embodiment can greatly improve the rotation robustness of the aortic valve CTA image classifier.
[0087] In summary, the core principle of the method for improving the rotation robustness of the aortic valve CTA image classifier proposed in the present invention is to use adversarial training technology to enhance the robustness of the classifier to geometric transformations (such as rotation). In adversarial training, the image classifier model not only needs to learn to extract effective features from normal input images, but also needs to counter adversarial samples generated by the attack algorithm. These adversarial samples simulate the interference conditions that may be encountered in practical applications by performing a series of geometric transformations (such as rotation) on the original images. During the training process, the image classifier will face the challenges of normal images and adversarial samples at the same time, so as to learn more robust feature representations and improve resistance to geometric transformations. In this way, even when encountering geometric transformations such as rotation in practical applications, the classifier can maintain high accuracy and robustness.
[0088] The method for improving the rotation robustness of the aortic valve CTA image classifier proposed in this embodiment can effectively improve the performance of the classifier in rotation transformation scenarios, and provide more reliable support for the diagnosis and treatment decision-making of cardiac surgery. This method does not require complex geometric feature extraction or architecture modification, and can be directly applied to the existing image classifier system, which has important clinical application value.
[0089] In addition, the present invention can also be applied to other types of medical image classification tasks, such as cardiac CT, MRI, etc., and it is only necessary to appropriately adjust the design of feature extraction and adversarial training framework according to the characteristics of different image data. In short, the method provided by the present invention has wide applicability.
[0090] Embodiment 2
[0091] Based on the same inventive concept as the method for improving the rotational robustness of an aortic valve CTA image classifier in the aforementioned embodiment 1, this embodiment further provides a device for improving the rotational robustness of an aortic valve CTA image classifier, such as Figure 3 As shown, the device comprises:
[0092] The data acquisition and preprocessing module 1 is used to acquire aortic valve CT angiography image (CTA image) data and preprocess the acquired data to form a training sample set;
[0093] Feature extraction module 2, used to extract the aortic valve features of the training sample set, specifically including extracting features such as the morphology of the aortic valve, the position of the valve, and the direction of the blood vessels;
[0094] The adversarial sample generation module 3 is used to design an axial rotation attack algorithm according to the extracted aortic valve features; perform axial rotation transformation on the training sample set through the axial rotation attack algorithm to generate an adversarial sample set with different rotation angles;
[0095] A rotation pool building module 4 is used to add the adversarial sample set to the training sample set to form a rotation pool;
[0096] The adversarial training module 5 is used to perform adversarial training on the aortic valve CTA image classifier through the rotation pool. During the adversarial training process, the aortic valve CTA image classifier will face the challenges of training samples and adversarial samples at the same time, so as to learn more robust feature representation to improve the rotation robustness of the aortic valve CTA image classifier.
[0097] The various specific processes and specific examples of the method for improving the rotational robustness of the aortic valve CTA image classifier in the aforementioned embodiment one are also applicable to the device for improving the rotational robustness of the aortic valve CTA image classifier in the present embodiment. Through the detailed description of embodiment one, those skilled in the art can clearly know the implementation method of the device for improving the rotational robustness of the aortic valve CTA image classifier in the present embodiment, so for the sake of brevity of the specification, it will not be described in detail here.
[0098] Embodiment 3
[0099] Based on the same inventive concept as the method for improving the rotational robustness of an aortic valve CTA image classifier in the aforementioned embodiment 1, this embodiment further provides a device for improving the rotational robustness of an aortic valve CTA image classifier, such as Figure 4 As shown, a computer program is stored thereon, and when the program is executed by a processor, the steps of the method for improving the rotation robustness of the aortic valve CTA image classifier described in Example 1 are implemented.
[0100] Among them, Figure 4 In the embodiment of the present invention, a bus architecture (represented by bus 300) is shown, which may include any number of interconnected buses and bridges, and bus 300 links various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are not further described herein. Bus interface 306 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, namely a transceiver, which provides a unit for communicating with various other devices over a transmission medium.
[0101] The processor 302 is responsible for managing the bus 300 and general processing, while the memory 304 may be used to store data used by the processor 302 when performing operations.
[0102] Embodiment 4
[0103] Based on the same inventive concept as the method for improving the rotational robustness of the aortic valve CTA image classifier in the aforementioned embodiment 1, this embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method for improving the rotational robustness of the aortic valve CTA image classifier described in embodiment 1 are implemented.
[0104] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0105] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0106] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0108] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.
[0109] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.
Claims
1. A method for improving the rotation robustness of an aortic valve CTA image classifier, characterized in that: The steps include: Acquire aortic valve CTA image data, and pre-process the acquired data to form a training sample set; Extracting aortic valve features of the training sample set; According to the extracted aortic valve features, an axial rotation attack algorithm is designed; the training sample set is subjected to axial rotation transformation by the axial rotation attack algorithm to generate an adversarial sample set with different rotation angles; Adding the adversarial sample set to the training sample set to form a rotation pool; The aortic valve CTA image classifier is adversarially trained through the rotation pool to improve the rotation robustness of the aortic valve CTA image classifier.
2. The method for improving the rotation robustness of an aortic valve CTA image classifier according to claim 1, characterized in that: The aortic valve characteristics include the morphology of the aortic valve, the position of the valve and the direction of the blood vessels.
3. The method for improving the rotation robustness of an aortic valve CTA image classifier according to claim 1, characterized in that: A pre-trained convolutional neural network is used to extract high-level semantic features of the training sample set to achieve aortic valve feature extraction.
4. The method for improving the rotation robustness of an aortic valve CTA image classifier according to claim 1, characterized in that: The axial rotation attack algorithm comprises the following steps: (A) Determination of the rotation center In the aortic valve CTA image, the rotation center point is the center of the blood vessel or the center of the aortic valve according to the aortic valve characteristics; (B) Determination of rotation angle The rotation angle is determined according to the aortic valve characteristics of the aortic valve CTA image and the rotation robustness required by the aortic valve CTA image classifier; (C) Rotation Matrix Determination The rotation matrix is used to transform the coordinates of each pixel in the image. For a three-dimensional aortic valve CTA image, the rotation matrix R is as follows: Where θ is the rotation angle; (D) Coordinate transformation For each pixel in the aortic valve CTA image, the rotation matrix R is applied to perform coordinate transformation and calculate the new coordinate after rotation.
5. The method for improving the rotation robustness of an aortic valve CTA image classifier according to claim 4, characterized in that: The rotation angle range is set to [-30°, 30°] with a step size of 5°.
6. The method for improving the rotation robustness of an aortic valve CTA image classifier according to claim 4, characterized in that: The axial rotation attack algorithm also includes the following steps: (E) Image interpolation Since the new coordinates after rotation may not be integers, an interpolation method is used to determine the pixel value at the new coordinates; and / or (F) Image resizing Rotation may cause the image's bounds to change; the image is resized to fit the new bounds.
7. The method for improving the rotation robustness of an aortic valve CTA image classifier according to claim 1, characterized in that: When the aortic valve CTA image classifier is subjected to adversarial training through the rotation pool, an optimization algorithm Adam based on a first-order gradient is adopted to accelerate the convergence speed of the adversarial training.
8. A device for improving the rotation robustness of an aortic valve CTA image classifier, characterized in that: include: A data acquisition and preprocessing module is used to acquire aortic valve CTA image data and preprocess the acquired data to form a training sample set; A feature extraction module, used to extract aortic valve features of the training sample set; An adversarial sample generation module is used to design an axial rotation attack algorithm based on the extracted aortic valve features; and to perform an axial rotation transformation on the training sample set through the axial rotation attack algorithm to generate an adversarial sample set with different rotation angles; A rotation pool construction module, used for adding the adversarial sample set to the training sample set to form a rotation pool; The adversarial training module is used to perform adversarial training on the aortic valve CTA image classifier through the rotation pool to improve the rotation robustness of the aortic valve CTA image classifier.
9. A device for improving the rotation robustness of an aortic valve CTA image classifier, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for improving the rotation robustness of an aortic valve CTA image classifier as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for improving the rotation robustness of an aortic valve CTA image classifier as described in any one of claims 1 to 7 are implemented.