A method, apparatus, storage medium, and electronic device for identifying motion artifacts

Abstract

Embodiments of the present invention disclose a method, apparatus, storage medium, and electronic device for identifying motion artifacts. The method includes obtaining a coronary artery image to be identified, performing window data screenshot on the coronary artery image along the direction of the coronary artery centerline based on a preset window size and step length to obtain a plurality of window data; identifying motion artifact description parameters of each window data, and determining a motion artifact identification result in the coronary artery image based on the motion artifact description parameters of each window data. The technical solution of this embodiment replaces the situation of overall identification of the entire coronary artery image, has strong pertinence for identifying motion artifacts of window data, less interfering data, and is convenient for improving the identification accuracy of motion artifacts. At the same time, the preset window size and step length are used to balance the computational complexity and signal accuracy in the process of identifying motion artifacts, which is convenient for improving the identification accuracy of motion artifacts.

Description

Technical Field

[0001] Embodiments of the present invention relate to image processing technology, and in particular, to a method, apparatus, storage medium, and electronic device for identifying motion artifacts. Background Art

[0002] Motion artifacts are caused by the autonomous or non-autonomous movement of the target object (such as the scanned human body or animal body) during the acquisition of medical images. Among them, non-autonomous movement is the physiological-related movement of the target object, such as cardiac movement and respiratory movement, etc., which is uncontrollable or incompletely controllable during the medical image acquisition process. Active movement is the conscious active movement of the target object during the medical image acquisition process. Generally, motion artifacts are mainly caused by non-autonomous movement.

[0003] In the current methods for identifying motion artifacts, there is a problem of poor positioning accuracy. Summary of the Invention

[0004] The present invention provides a method, apparatus, storage medium, and electronic device for identifying motion artifacts to solve the problem of improving the identification accuracy of motion artifacts.

[0005] According to one aspect of the present invention, a method for identifying motion artifacts is provided, including:

[0006] Obtain a coronary artery image to be identified, and perform window data screenshot on the coronary artery image along the direction of the coronary artery center line based on a preset window size and step length to obtain a plurality of window data;

[0007] Identify the motion artifact description parameters of each window data, and determine the motion artifact identification result in the coronary artery image based on the motion artifact description parameters of each window data.

[0008] According to another aspect of the present invention, an apparatus for identifying motion artifacts is provided, including:

[0009] A window data interception module, configured to obtain a coronary artery image to be identified, and perform window data screenshot on the coronary artery image along the direction of the coronary artery center line based on a preset window size and step length to obtain a plurality of window data;

[0010] A motion artifact identification module, configured to identify the motion artifact description parameters of each window data, and determine the motion artifact identification result in the coronary artery image based on the motion artifact description parameters of each window data.

[0011] According to another aspect of the present invention, an electronic device is provided, and the electronic device includes:

[0012] At least one processor; and

[0013] A memory communicatively connected to the at least one processor; wherein,

[0014] The memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the at least one processor is enabled to execute the method for identifying motion artifacts according to any embodiment of the present invention.

[0015] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for causing a processor to implement the method for identifying motion artifacts according to any embodiment of the present invention when executed.

[0016] The technical solution provided in this embodiment intercepts window data of the coronary artery image by pre-determining the window size and step length for intercepting data of the coronary artery image, and obtains the identification result of motion artifacts in the coronary artery image by identifying the motion artifacts of the window data, replacing the situation of overall identification of the entire coronary artery image. The identification of motion artifacts of the window data has strong pertinence and less interference data, which is convenient for improving the identification accuracy of motion artifacts. At the same time, the pre-set window size and step length are used to balance the computational complexity and signal accuracy in the process of motion artifact identification, which is convenient for improving the identification accuracy of motion artifacts.

[0017] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a flowchart of the method for identifying motion artifacts provided in the embodiment of the present invention;

[0020] Figure 2 It is a schematic structural diagram of an apparatus for identifying motion artifacts provided in the embodiment of the present invention;

[0021] Figure 3 It is a schematic structural diagram of an electronic device in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] Figure 1 It is a flowchart of the method for identifying motion artifacts provided by an embodiment of the present invention. This embodiment is applicable to the situation of identifying motion artifacts in coronary artery images. This method can be executed by the motion artifact identification device provided by the embodiment of the present invention. The motion artifact identification device can be implemented by software and / or hardware, and the motion artifact identification device can be configured on an electronic computing device. Specifically, it includes the following steps:

[0025] S110. Obtain the coronary artery image to be identified, and perform window data screenshot on the coronary artery image along the direction of the coronary artery center line based on a preset window size and step length, so as to obtain a plurality of window data.

[0026] S120. Identify the motion artifact description parameters of each window data, and determine the motion artifact identification result in the coronary artery image based on the motion artifact description parameters of each window data.

[0027] In this embodiment, the image to be processed is a coronary artery image, that is, an image including the coronary artery part. The acquisition method and image type of the coronary artery image are not limited herein. Exemplarily, the coronary artery image can be, but is not limited to, a CT (Computed Tomography) image, an MRI (Magnetic Resonance Imaging) image, etc. By identifying motion artifacts in the coronary artery image, the positions of the motion artifacts in the coronary artery image are located, facilitating the removal of artifacts from the coronary artery image.

[0028] After obtaining the coronary artery image, preprocessing of the coronary artery image can be performed. The preprocessing includes, but is not limited to, denoising processing, image enhancement processing, etc., and is not limited thereto. By preprocessing, the interference data in the image is reduced, and the image clarity is improved to enhance the recognition accuracy of motion artifacts.

[0029] Obtain a preset window size and step size. The window size is used to define the size of the window for intercepting data in the coronary artery image, and the step size is used to define the unit length of window sliding. By controlling the sliding of the window in the coronary artery image based on the window size and step size, the window data corresponding to each sliding of the window is intercepted for motion artifact recognition. Each window data is part of the data in the coronary artery image. By processing each window data separately, the motion artifact recognition result of the coronary artery image is obtained, replacing the method of overall recognition of the coronary artery image, reducing the amount of data to be processed, and at the same time improving the pertinence and accuracy of motion artifact recognition.

[0030] In this embodiment, based on the preset window size and step size, the window is controlled to slide along the direction of the coronary artery centerline in the coronary artery image, and window data screenshots of the coronary artery image are taken during the sliding of the window to obtain a plurality of window data. Among them, the coronary artery centerline in the coronary artery image can be pre-marked, and the determination method of the coronary artery centerline in the coronary artery image is not limited. Exemplarily, the coronary artery can be identified in the coronary artery image, and the identified coronary artery is subjected to erosion processing to obtain the coronary artery centerline; Exemplarily, the coronary artery image can also be input into a pre-trained coronary artery centerline detection model to obtain the detection result output by the coronary artery centerline detection model, and the detection result can be a coronary artery image including the coronary artery centerline mark.

[0031] For multiple intercepted window data, the motion artifact description parameters of the window data are respectively identified, where the motion artifact description parameters are the response values of the feature information in the window data to the motion artifacts. The motion artifact description parameters corresponding to different types of feature information may be different. Based on the motion artifact description parameters of each window data, the motion artifact recognition result in the coronary artery image is determined. Exemplarily, when the motion artifact description parameter is a specific value, it is determined that there is a motion artifact in the window data, or when there is data fluctuation in the motion artifact description parameter, it is determined that there is a motion artifact in the window data with data fluctuation, that is, the window corresponding to the window data with a motion artifact can be determined as the motion artifact window. Based on the window with a motion artifact, the position information of the motion artifact in the coronary artery image is determined, so as to realize the recognition and positioning of the motion artifact in the coronary artery image.

[0032] In this embodiment, the motion artifact recognition result of the coronary artery image includes the position information of the motion artifact in the coronary artery image. The position information of the motion artifact in the coronary artery image is determined according to the motion artifact window. For example, the position information of each motion artifact window can be determined as the position information of the motion artifact, or the central position of the motion artifact window can be determined as the position information of the motion artifact. It can also be that, based on each motion artifact window, the motion artifact region in the coronary artery image is determined, and the central position information of the motion artifact region is determined as the position information of the motion artifact. Among them, the motion artifact region can be a region formed by continuous motion artifact windows.

[0033] Optionally, determining the motion artifact position based on the motion artifact window includes: determining continuous motion artifact windows according to the adjacent relationship of each window; for any group of continuous motion artifact windows, based on the sum of the starting position of the window and half of the window depth in at least one continuous motion artifact window, the corresponding motion artifact position is determined. Among them, the number of continuous motion artifact windows is greater than or equal to 1. Exemplarily, n window data are intercepted from the coronary artery image, and the motion artifact description parameters of each window data are 01110010…. The window with the motion artifact description parameter being a specific value (for example, 1) is determined as the motion artifact window, that is, windows 2, 3, 4, and 7 are motion artifact windows. Further, the first group of continuous motion artifact windows includes windows 2, 3, and 4, and the second group of continuous motion artifact windows includes window 7. The first motion artifact position is determined based on the sum of the starting position of the first group of continuous motion artifact windows and half of the window depth, and the second motion artifact position is determined based on the sum of the starting position of the second group of continuous motion artifact windows and half of the window depth.

[0034] It should be noted that the coronary artery image is a two-dimensional image or a three-dimensional image. Correspondingly, the coordinate information of each data point in the window data is two-dimensional coordinates or three-dimensional coordinates. Correspondingly, based on the sum of the starting position and half of the window depth in at least one consecutive motion artifact window, the corresponding motion artifact position can be determined, which can be the sum of the initial coordinate information and half of the window depth in at least one consecutive motion artifact window to obtain the motion artifact position corresponding to the at least one consecutive motion artifact window.

[0035] Optionally, the motion artifact description parameters include one or more of a motion artifact classification prediction value, a signal-to-noise ratio response value, a matrix rank response value, and a frequency domain information response value. Correspondingly, determination rules for the above-mentioned motion artifact description parameters are preset. Based on the motion artifact description parameters to be identified, the corresponding determination rules for the motion artifact description parameters are called to process each window data respectively to obtain the motion artifact description parameters corresponding to each window data. By identifying the above at least one motion artifact description parameter, the motion artifact recognition result in the coronary artery image is determined.

[0036] In some embodiments, the motion artifact description parameter includes a motion artifact classification prediction value. Correspondingly, identifying the motion artifact description parameters of each window data includes: for any window, inputting the window data of the window into a pre-trained neural network model to obtain the motion artifact classification prediction value output by the neural network model. In this embodiment, the window data is processed by the pre-trained neural network model to obtain the motion artifact classification prediction value. Optionally, the motion artifact classification prediction value can be a data identifier corresponding to the prediction type, such as 0 or 1, etc. A motion artifact classification prediction value of 1 can indicate that the classification type of the window data is the motion artifact type, and a motion artifact classification prediction value of 0 can indicate that the classification type of the window data is the non-motion artifact type. Optionally, the motion artifact classification prediction value can be the probability value that there is a motion artifact in the window data.

[0037] The pre-trained neural network model has the function of classifying window data based on motion artifacts. The network structure of the neural network model is not limited in this embodiment. Exemplarily, the network structure of the motion artifact recognition model can be a convolutional neural network model, a recurrent neural network, or a Transformer model, etc. Exemplarily, the convolutional neural network module includes but is not limited to AlexNet, VGG, inception, ResNet, DenseNet, etc., and the recurrent neural network model includes but is not limited to LSTM (Long Short Term, long short-term memory network), GRU (gated recurrent unit) model, etc.

[0038] The neural network model can be obtained through iterative training based on window data and the motion artifact recognition labels corresponding to each window of data. In some embodiments, it can be obtained by training the initial models of at least one network structure to obtain neural network models with different network structures, and screening the final neural network model from multiple neural network models.

[0039] Taking the motion artifact classification prediction value as the data identifier corresponding to the prediction type as an example, based on the motion artifact description parameters of each window of data, determining the motion artifact recognition result in the coronary artery image includes: determining the window with the motion artifact classification prediction value being the first specific value as the motion artifact window, and determining the motion artifact position based on the motion artifact window. Exemplarily, the first specific value can be the motion artifact classification prediction value corresponding to the motion artifact type. For example, the first specific value is 1.

[0040] Taking the motion artifact classification prediction value as the probability value of the existence of motion artifacts in the window data as an example, based on the motion artifact description parameters of each window of data, determining the motion artifact recognition result in the coronary artery image includes: determining the window with the motion artifact classification prediction value greater than the preset threshold as the motion artifact window, and determining the motion artifact position based on the motion artifact window. The preset threshold can be determined in advance. Exemplarily, the preset threshold can be 85% or 95%, etc., and there is no limitation thereto.

[0041] Using the motion artifact classification prediction value as the response value of the window data to the motion artifact to determine the motion artifact window in the coronary artery image, and based on the interception order of each motion artifact window, determining adjacent motion artifact windows as consecutive motion artifact windows. When the adjacent windows of the motion artifact window are all non-motion artifact windows, this motion artifact window serves as a group of consecutive motion artifact windows, that is, each group of consecutive motion artifact windows includes at least one motion artifact window. Determining the corresponding motion artifact position for each consecutive motion artifact window.

[0042] In some embodiments, the motion artifact description parameter includes the signal-to-noise ratio response value; correspondingly, recognizing the motion artifact description parameters of each window of data includes: for any window, determining the signal-to-noise ratio data based on the window data of the window; comparing the signal-to-noise ratio data with the signal-to-noise ratio threshold, and determining the signal-to-noise ratio response value of the window data according to the comparison result.

[0043] Taking the coronary artery image as a CT image as an example, the window data intercepted from the coronary artery image can be the CT data corresponding to the window. The signal-to-noise ratio data of the window data is used to reflect the ratio of the signal to the noise in the window data. Among them, the motion artifacts in the window data are the main factors of the noise. By statistically analyzing the signal-to-noise ratio data in the window data, the signal-to-noise ratio response value of the window data is determined.

[0044] Optionally, the method for determining the signal-to-noise ratio data includes: calculating the standard deviation and the mean for any window data respectively, and determining the signal-to-noise ratio data based on the ratio of the standard deviation to the mean of the window data. Optionally, the method for determining the signal-to-noise ratio data includes: inputting the window data into a pre-trained signal-to-noise ratio relationship model to obtain the signal-to-noise ratio data corresponding to the window data. The signal-to-noise ratio relationship model can be pre-trained to reflect the mapping relationship between the window data and the signal-to-noise ratio information. Since the window data intercepted with different window sizes and step lengths are different, correspondingly, the signal-to-noise ratio relationship model can be based on the mapping relationship between the signal-to-noise ratio data and multiple parameters such as the window size, the step length, and the window data. For any window data, the inputs of the signal-to-noise ratio relationship model include the window data and the corresponding window size and step length. By setting the signal-to-noise ratio relationship model, the signal-to-noise ratio data corresponding to the window data can be quickly determined, simplifying the calculation process of the signal-to-noise ratio data.

[0045] A signal-to-noise ratio threshold is preset, and the signal-to-noise ratio data is compared with the signal-to-noise ratio threshold. When the signal-to-noise ratio data is greater than or equal to the signal-to-noise ratio threshold, the signal-to-noise ratio response value of the window data is determined to be the first response value. When the signal-to-noise ratio data is less than the signal-to-noise ratio threshold, the signal-to-noise ratio response value of the window data is determined to be the second response value. The first response value and the second response value can be numerical identifiers such as 1 and 0. In other embodiments, the signal-to-noise ratio response value can also be other forms of identifiers, which are not limited herein.

[0046] Based on the motion artifact description parameters of the respective window data, determining the motion artifact recognition result in the coronary artery image includes: determining the window with the signal-to-noise ratio response value being the second specific value as the motion artifact window, and determining the motion artifact position based on the motion artifact window. The second specific value can be one of the first response value and the second response value. Optionally, the second specific value is the first response value, and the window corresponding to the signal-to-noise ratio data greater than or equal to the signal-to-noise ratio threshold is determined as the motion artifact window.

[0047] Based on the interception order of the respective motion artifact windows, adjacent motion artifact windows are determined as consecutive motion artifact windows, and the sum of the starting position of the window and half of the window depth of each group of consecutive motion artifact windows is determined as the corresponding motion artifact position.

[0048] In some embodiments, the motion artifact description parameter includes a matrix rank response value; correspondingly, identifying the motion artifact description parameters of each window data includes: for any window, determining the matrix rank response value corresponding to the window data. In this embodiment, the window data is converted into a data matrix, that is, each data value in the window data is determined as a factor at the corresponding position in the matrix. The data matrix corresponding to the window data can be decomposed by singular value decomposition to obtain the matrix rank response value corresponding to the window data. Optionally, for any window, perform singular value decomposition on the window data of the window to obtain the decomposition matrix corresponding to each window data; based on the singular value threshold and each singular value in the decomposition matrix, determine the matrix rank response value of the window. Exemplarily, perform singular value decomposition based on the following formula: X t = UΣV T , where X t is the data matrix corresponding to the t-th window data, U is a matrix of a set of output orthogonal singular vectors, V T is a matrix of a set of input orthogonal singular vectors, Σ is a matrix arranged in descending order of the singular value magnitudes of the diagonal elements, that is, the decomposition matrix, and the diagonal elements included in this decomposition matrix Σ are respectively singular values, for example, q1, q2... q N , where q1 is the largest singular value. The matrix rank response value of the window is determined by comparing the preset singular value threshold with each singular value in the decomposition matrix respectively, determining the number of singular values greater than the singular value threshold, and determining this number of singular values greater than the singular value threshold as the matrix rank response value. Exemplarily,

[0049]

[0050] wherein, the middle matrix in the above singular value decomposition is the decomposition matrix, and it can be seen that the singular values in the decomposition matrix include and 1. If the singular value threshold is 2, then the singular values are all less than the singular value threshold (that is, the number is 0), and the matrix rank response value is 0; if the singular value threshold is 1.5, then the singular value is greater than the singular value threshold, and the singular value 1 is less than the singular value threshold, and the matrix rank response value is 1.

[0051] Perform singular value decomposition on each window data to obtain the matrix rank response value corresponding to each window data. If the matrix rank response value of X t is greater than or less than the matrix rank response value of X t-1 , it indicates that the above two window data are different in visual content, and this difference in visual content is caused by the existence of motion artifacts. Correspondingly, when the matrix rank response values corresponding to adjacent windows are different, there are motion artifacts in one or more of the two adjacent window data.

[0052] Optionally, determining the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of each window data includes: according to the adjacent relationship of each window, comparing the matrix rank response values of adjacent windows, determining the window with a fluctuating matrix rank response value as the motion artifact window, and determining the motion artifact position based on the motion artifact window. For any window, compare the matrix rank response value of this window with the matrix rank response value of the previous window and / or the matrix rank response value of the next window respectively. Determine the window with a different matrix rank response value from the previous window and / or the next window as the motion artifact window. In some embodiments, if the matrix rank response value of the current window is different from the matrix rank response values of both the previous window and the next window, determine the current window as the motion artifact window.

[0053] In some embodiments, according to the interception order of each window, compare the matrix rank response values of each window in sequence. For example, it can be based on the matrix rank response value and window order of each window to draw a matrix rank response value curve, and determine the window with a fluctuating matrix rank response value in the matrix rank response value curve as the motion artifact window.

[0054] Based on the determined motion artifact window, determine the recognition result of the motion artifact, that is, determine the position information of the motion artifact. Specifically, based on the interception order of each motion artifact window, determine adjacent motion artifact windows as continuous motion artifact windows, and based on the sum of the starting position and half of the window depth in the continuous motion artifact windows, determine the corresponding motion artifact position.

[0055] In some embodiments, the motion artifact description parameter includes the frequency domain information response value; correspondingly, identifying the motion artifact description parameter of each window data includes: for any window, determining the frequency domain statistical information of the window data corresponding to this window, comparing based on the frequency domain statistical information and the frequency domain threshold, and obtaining the frequency domain information response value corresponding to this window data. According to the characteristics of motion artifacts, including that the low-frequency information in the window data of motion artifacts changes greatly and the high-frequency information changes little.

[0056] In some embodiments, the window data of the window is subjected to frequency-domain conversion to obtain a frequency-domain image, and high-frequency information and low-frequency information are extracted from the frequency-domain image; low-frequency statistical information is determined based on the low-frequency information in the frequency-domain image, high-frequency statistical information is determined based on the high-frequency information in the frequency-domain image, and the low-frequency threshold is compared with the low-frequency statistical information, and the high-frequency information is compared with the high-frequency statistical information. The frequency-domain information response value of the window is determined according to the comparison result of the high-frequency information and the comparison result of the low-frequency information. Among them, through the frequency-domain transformation method, the window data is converted into a frequency-domain image. Exemplarily, the frequency-domain transformation method includes but is not limited to discrete cosine transform, discrete Fourier transform, discrete wavelet transform, etc. The frequency-domain image obtained through frequency-domain conversion and ZIG-ZAG sorting includes high-frequency information and low-frequency information. The high-frequency information and low-frequency information can be extracted by dividing the frequency-domain image into regions. Exemplarily, the upper left corner region of the frequency-domain image is used as the low-frequency region, and the lower right corner region is used as the high-frequency region. Correspondingly, the low-frequency information is extracted based on the low-frequency region to obtain low-frequency statistical information, and the high-frequency information is extracted based on the high-frequency region to obtain high-frequency statistical information. Among them, the division of the low-frequency region and the high-frequency region can be preset. The pixel data of each pixel point in the low-frequency region can be determined as low-frequency information, and the sum value or average value of the low-frequency information of each pixel point can be determined as low-frequency statistical information; the pixel data of each pixel point in the high-frequency region can be determined as high-frequency information, and the sum value or average value of the high-frequency information of each pixel point can be determined as high-frequency statistical information.

[0057] In some embodiments, it may also be to pre-create a frequency-domain relationship model, which is used to reflect the mapping relationship between window data, window size and step size and frequency-domain statistical information. For any window data, the window data and the corresponding window size and step size of the window data are input into the frequency-domain relationship model to obtain the frequency-domain statistical information corresponding to the window data under the window size and step size. In some embodiments, the frequency-domain relationship model may include a high-frequency relationship model and a low-frequency relationship model, and the frequency-domain statistical information includes high-frequency statistical information and low-frequency statistical information. The frequency-domain relationship model can be matched with the current window size and current step size for data interception of the coronary artery image. By calling the frequency-domain relationship model, the frequency-domain statistical information corresponding to the window data can be obtained respectively. By pre-creating the frequency-domain relationship model, the window data can be directly input into the frequency-domain relationship model to quickly obtain the frequency-domain statistical information output by the frequency-domain relationship model, without separately performing frequency-domain image conversion on the window data, and the extraction and processing of high-frequency information and low-frequency information, which simplifies the determination process of the frequency-domain statistical information.

[0058] The frequency-domain information response value includes a low-frequency response value and a high-frequency response value. Comparing the low-frequency statistical information with the low-frequency threshold, when the low-frequency statistical information is greater than or equal to the low-frequency threshold, determining the low-frequency response value as the first low-frequency response value; when the low-frequency statistical information is less than the low-frequency threshold, determining the low-frequency information response value as the second low-frequency response value. Comparing the high-frequency statistical information with the high-frequency threshold, when the high-frequency statistical information is greater than or equal to the high-frequency threshold, determining the high-frequency response value as the first high-frequency response value; when the high-frequency statistical information is less than the high-frequency threshold, determining the high-frequency response value as the second high-frequency response value.

[0059] Optionally, determining the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of the respective window data includes: comparing the frequency-domain information response values of adjacent windows according to the adjacent relationship of each window, determining the window with fluctuating frequency-domain information response values as the motion artifact window, and determining the motion artifact position based on the motion artifact window. Determining the motion artifact window based on the frequency-domain information response value corresponding to each window, for example, determining the window with a change in the response value as the motion artifact window. Optionally, determining the window with a change in the low-frequency response value and / or the high-frequency response value as the motion artifact window. Determining the motion artifact position based on the motion artifact window in the coronary artery image. In this embodiment, the motion artifact windows are not continuous, that is, only one motion artifact window is included in the continuous motion artifact windows, and the sum of the starting position of the motion artifact window and half of the window depth is determined as the position of the motion artifact.

[0060] Based on the above embodiments, the motion artifact description parameters include one or more of the motion artifact classification prediction value, the signal-to-noise ratio response value, the matrix rank response value, and the frequency domain information response value; correspondingly, determining the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of each window data includes: respectively determining one or more of the motion artifact classification prediction value, the signal-to-noise ratio response value, the matrix rank response value, and the frequency domain information response value based on the window data; respectively determining the motion artifact sub-recognition result based on any one of the determined motion artifact classification prediction value, signal-to-noise ratio response value, matrix rank response value, and frequency domain information response value, and determining the target recognition result of the motion artifact based on multiple motion artifact sub-recognition results. In this embodiment, one or more of the above multiple motion artifact description parameters are respectively determined for each window data, the motion artifact sub-recognition result of the coronary artery image is respectively determined based on each type of motion artifact description parameter, and the target recognition result of the motion artifact is determined based on the motion artifact sub-recognition results respectively determined by each motion artifact description parameter. Exemplarily, it can be obtained by performing a fusion process on multiple motion artifact sub-recognition results, where the fusion process can perform a weighted process on multiple motion artifact sub-recognition results, or perform a merging process and a deduplication process on multiple motion artifact sub-recognition results to obtain the target recognition result. By determining different types of motion artifact description parameters for the coronary artery image, the motion artifact is recognized from the dimension of different feature information, and the target recognition result of the motion artifact is determined based on multiple motion artifact sub-recognition results, thereby improving the recognition accuracy of the motion artifact in the coronary artery image.

[0061] The technical solution provided in this embodiment intercepts the window data of the coronary artery image by pre-determining the window size and step length for data interception of the coronary artery image, and obtains the motion artifact recognition result in the coronary artery image through the recognition of the motion artifact in the window data, replacing the situation of overall recognition of the entire coronary artery image. The recognition of the motion artifact in the window data has strong pertinence and less interference data, which is convenient for improving the recognition accuracy of the motion artifact. At the same time, the pre-set window size and step length are used to balance the computational complexity and signal accuracy in the motion artifact recognition process, which is convenient for improving the recognition accuracy of the motion artifact.

[0062] Based on the above embodiments, the following influencing factors exist in the process of identifying motion artifacts: window size, step size, and motion artifact description parameters. Motion artifacts often occur around the coronary artery. Window data is obtained by intercepting along the tangential direction of the coronary artery centerline. Among them, the window is a three-dimensional window, and the window size reflects the sensitivity of the judgment of motion artifacts. If the window is too large, the cross-section of the lumen cannot be observed completely. If the window is too small, the signal intensity of the window data is insufficient. The step size reflects the accuracy of motion artifact positioning. If the step size is too large, the positioning accuracy error is large. If the step size is too small, the computational complexity is high. Calculating the motion artifact description parameters for each window data affects the recognition accuracy of motion artifacts. In this embodiment, during the process of identifying motion artifacts, each influencing factor is obtained through pre-training and verification. Among them, the influencing factors include window size, step size, and the determination rule for determining the motion artifact description parameters. The determination rule for determining the motion artifact description parameters can be a motion artifact recognition model. Different motion artifact description parameters can correspond to different types and forms of motion artifact recognition models.

[0063] Correspondingly, the above method further includes: pre-determining a motion artifact recognition strategy, where the motion artifact recognition strategy includes window size, step size, and a motion artifact recognition model. It should be noted that the determination methods of the window size, step size, and motion artifact description parameters in the above embodiments can be included in the pre-determined motion artifact recognition strategy. By pre-training and verification to determine the above motion artifact recognition strategy, the influencing factors in the process of identifying motion artifacts are screened, avoiding the negative impact of any influencing factor on the identification of motion artifacts, and ensuring the recognition accuracy of motion artifacts.

[0064] Optionally, pre-determining a motion artifact recognition strategy includes: obtaining motion artifact training samples and multiple groups of image interception parameter combinations including window size and step size; based on any image interception parameter combination, intercepting window data from the motion artifact training samples along the coronary artery centerline in the motion artifact training samples, and training a motion artifact recognition model corresponding to the image interception parameter combination based on the intercepted window data and the motion artifact labels corresponding to each window data; obtaining motion artifact verification samples, verifying the motion artifact recognition models corresponding to each image interception parameter combination based on the motion artifact verification samples, and determining the artifact recognition accuracy rates of each motion artifact recognition model; determining the motion artifact recognition models that meet the artifact recognition accuracy rate screening conditions, and the image interception parameter combinations corresponding to the selected motion artifact recognition models as the motion artifact recognition strategy.

[0065] The motion artifact training samples and motion artifact validation samples can be pre-acquired coronary artery images. Motion artifact sample images are obtained, and each motion artifact sample image is correspondingly set with a motion artifact label. The motion artifact sample images are divided into training samples and validation samples. For example, they are randomly selected from the set of motion artifact sample images based on a preset ratio to obtain a motion artifact training sample set and a motion artifact validation sample set. Among them, there may be partial overlap between the motion artifact training sample set and the motion artifact validation sample set. Multiple motion artifact recognition strategies are trained through the motion artifact training sample set, and the multiple motion artifact recognition strategies obtained through training are optimized and screened through the motion artifact validation sample set to obtain the final motion artifact recognition strategy.

[0066] Multiple combinations of image cropping parameters are preset to respectively determine the motion artifact recognition strategies corresponding to each combination of image cropping parameters. Among them, the window size and / or the step size are different in different combinations of image cropping parameters. For the window size and step size in each combination of image cropping parameters, the window is controlled to slide along the coronary artery centerline direction to crop multiple window data. The motion artifact recognition model is trained through the window data and the motion artifact label corresponding to the window data. In this embodiment, the model type of the motion artifact recognition model is not limited. Exemplarily, the motion artifact recognition model may include, but is not limited to, machine learning models such as neural network models, relationship models obtained by data fitting, etc., as long as it has the function of recognizing motion artifacts in the window data. It should be noted that different types of motion artifact recognition models can be obtained through different training methods.

[0067] For each combination of image cropping parameters, one or more types of motion artifact recognition models are trained to obtain one or more motion artifact recognition models corresponding to each combination of image cropping parameters. The recognition accuracy of each motion artifact recognition model is verified through the motion artifact validation sample set to screen out the optimal motion artifact recognition strategy.

[0068] In some embodiments, based on the pre-set artifact recognition accuracy screening conditions, the motion artifact recognition models are screened, and the screened motion artifact recognition model and the corresponding combination of image cropping parameters are determined as the motion artifact recognition strategy. Specifically, the motion artifact recognition model with the maximum artifact recognition accuracy and the combination of image cropping parameters corresponding to the motion artifact recognition model with the maximum artifact recognition accuracy are determined as the motion artifact recognition strategy. By training different motion artifact recognition models based on different combinations of image cropping parameters and screening the motion artifact recognition model with the maximum artifact recognition accuracy, while ensuring the motion artifact recognition accuracy, the influence of the selection of window size and step size on the motion artifact recognition accuracy is avoided, and at the same time, the difficulty of selecting the window size and step size is reduced.

[0069] In some embodiments, based on the preset artifact recognition accuracy rate and the consumption parameters of the motion artifact recognition process as screening conditions, a motion artifact recognition model is screened, and the screened motion artifact recognition model and the corresponding image capture parameter combination are obtained. For example, a motion artifact recognition strategy with a motion artifact recognition accuracy meeting the accuracy threshold and the minimum consumption parameter can be determined as the target motion artifact recognition strategy. The consumption parameter can include one or more of the processing duration and the amount of calculation in the motion artifact recognition process. Optionally, the consumption parameter can be characterized by a step size. Among them, the smaller the step size, the higher the computational complexity, and the corresponding consumption parameter is larger. Correspondingly, a motion artifact recognition strategy with a motion artifact recognition accuracy meeting the accuracy threshold and the largest step size can be determined as the target motion artifact recognition strategy. Specifically, one or more motion artifact recognition models with an artifact recognition accuracy rate greater than or equal to the artifact recognition accuracy rate threshold are determined as candidate motion artifact recognition models. Correspondingly, the candidate motion artifact recognition models and their corresponding image capture parameter combinations form candidate motion artifact recognition strategies, and the target motion artifact recognition strategy is determined based on the step size (or consumption parameter) among multiple candidate motion artifact recognition strategies. For example, the candidate motion artifact recognition strategy corresponding to the largest step size is determined as the target motion artifact recognition strategy. By increasing the step size, the number of window data captured in the medical tomography image processing process can be reduced, so as to further reduce the amount of calculation in the motion recognition process. The computational complexity is small, taking into account both the recognition accuracy of motion artifacts and the computational complexity.

[0070] Based on the above embodiments, the motion artifact recognition model can include a relationship model. Among them, the relationship model is used to represent the mapping relationship between the input information and the output information. Optionally, the input information can be window data, and the output information can be motion artifact description parameters. By training the relationship model, the corresponding motion artifact description parameters can be output based on the window data, simplifying the determination process of the motion artifact description parameters.

[0071] Based on the above embodiments, the motion artifact recognition model can include a relationship model and a target threshold. The input information of the relationship model can be window data, and the output information can be the feature information used to represent the window data, such as statistical indicators. The target threshold is used to compare with the output information of the relationship model to determine the motion artifact description parameters of the window data.

[0072] In some embodiments, training the motion artifact recognition model corresponding to the image capture parameter combination based on the captured window data and the motion artifact labels corresponding to each window data includes: determining the statistical metrics corresponding to the window data, and constructing a relationship model between the statistical metrics and the window size, stride, and window data based on the corresponding relationship between the window size, stride, window data, and statistical metrics; determining the training recognition results of each window data based on the comparison result between the statistical metrics and the corresponding thresholds, and determining the target threshold corresponding to the statistical metrics based on the motion artifact labels corresponding to each window data and the training recognition results of each window data, where the relationship model and the target threshold constitute the motion artifact recognition model.

[0073] Among them, the statistical metrics include the signal-to-noise ratio of the window data, the frequency domain statistical information of the window data, and the singular values obtained by singular value decomposition of the window data. Different statistical metrics correspond to different motion artifact description parameters. The statistical metrics and window data of each window data are used to create different relationship models. Correspondingly, this relationship model is used to characterize the mapping relationship between the above statistical metrics and window data. The target threshold is used to compare the statistical metrics output by the relationship model to obtain the motion artifact description parameters, and further determine the motion artifact recognition result. The size of the target threshold also determines the accuracy of the motion artifact recognition result. In this embodiment, the relationship model and the target threshold are trained separately to obtain a motion artifact recognition model including the relationship model and the target threshold.

[0074] Taking the training of the motion artifact recognition model based on the signal-to-noise ratio of the window data as an example, for any set of window sizes and strides for the window data intercepted from the motion artifact training samples, calculate the signal-to-noise ratio data of each window data respectively, where the signal-to-noise ratio data of any window data can be determined based on the ratio of the standard deviation and the mean of the window data. Based on the corresponding relationship between the window data, window size, stride, and signal-to-noise ratio data, construct a relationship model between the signal-to-noise ratio data and the window size, stride, and window data. Here, the form of the relationship model is defined. The input information of this relationship model is the window size, stride, and window data, and the output information is the first statistical metric. Further, the window size and stride in this relationship model can be fixed parameters.

[0075] In some embodiments, the relationship model can be a scatter plot model. Correspondingly, the relationship model between the signal-to-noise ratio data and the window size, stride, and window data can be a signal-to-noise ratio scatter plot model. Exemplarily, based on the corresponding relationship between the window size, stride, window data, and signal-to-noise ratio data, construct a signal-to-noise ratio data scatter plot, and construct a signal-to-noise ratio data scatter plot model based on the signal-to-noise ratio data scatter plot, for example, obtain the mapping relationship between the signal-to-noise ratio data and the window size, stride, and window data by fitting the signal-to-noise ratio data scatter plot.

[0076] Train the signal-to-noise ratio threshold, i.e., the target threshold, based on the trained signal-to-noise ratio scatter plot model. This signal-to-noise ratio threshold is obtained through iterative training using the motion artifact labels of each window's data. Specifically, determine the initial threshold and perform iterative optimization on the initial threshold. In each iteration process, compare the current threshold with the statistical metrics corresponding to each window's data, and determine the training recognition result of the motion artifact training samples based on the comparison result; adjust the current threshold based on the training recognition result of the motion artifact training samples and the motion artifact labels, that is, adjust the current threshold by increasing or decreasing it, and determine the training recognition result and the corresponding motion artifact recognition accuracy rate based on the adjusted threshold again. Until the current threshold meets the recognition accuracy, it is determined as the target threshold. For example, determine the threshold corresponding to the optimal motion artifact recognition accuracy rate as the target threshold, or determine the threshold that meets the recognition accuracy as the target threshold.

[0077] Taking the training of the motion artifact recognition model based on the frequency-domain statistical information of window data as an example, for any set of window data intercepted from the motion artifact training samples with a window size and a step size, calculate the frequency-domain statistical information corresponding to each window's data respectively. The frequency-domain statistical information includes high-frequency statistical information and low-frequency statistical information. Among them, the determination method of the frequency-domain statistical information can include: converting the window data into a frequency-domain image, extracting the high-frequency information and low-frequency information in the frequency-domain image, and determining the low-frequency statistical information of the low-frequency information and the high-frequency statistical information of the high-frequency information. Through the frequency-domain transformation method, convert the window data into a frequency-domain image. Exemplarily, the frequency-domain transformation method includes but is not limited to discrete cosine transform, discrete Fourier transform, discrete wavelet transform, etc. The frequency-domain image obtained through frequency-domain conversion includes high-frequency information and low-frequency information. The high-frequency information and low-frequency information can be extracted by dividing the frequency-domain image into regions. Exemplarily, take the upper left corner region of the frequency-domain image as the low-frequency region and the lower right corner region as the high-frequency region. Correspondingly, extract the low-frequency information based on the low-frequency region to obtain the low-frequency statistical information, and extract the high-frequency information based on the high-frequency region to obtain the high-frequency statistical information. Among them, the division of the low-frequency region and the high-frequency region can be preset. Specifically, the pixel data of each pixel point in the low-frequency region can be determined as the low-frequency information, and the sum value or average value of the low-frequency information of each pixel point can be determined as the low-frequency statistical information; the pixel data of each pixel point in the high-frequency region can be determined as the high-frequency information, and the sum value or average value of the high-frequency information of each pixel point can be determined as the high-frequency statistical information. Based on the corresponding relationship between the window size, step size, and window data and the low-frequency statistical information respectively, construct a low-frequency information scatter plot, and based on the corresponding relationship between the window size, step size, and window data and the high-frequency statistical information respectively, construct a high-frequency information scatter plot. The frequency-domain information scatter plot model includes a high-frequency information scatter plot model and a low-frequency information scatter plot model. Correspondingly, the motion artifact recognition model includes a low-frequency information scatter plot model, a low-frequency threshold, a high-frequency information scatter plot model, and a high-frequency threshold.

[0078] Train the low-frequency threshold based on the trained low-frequency information scatter plot model, and train the high-frequency threshold based on the trained high-frequency information scatter plot model. That is, the target threshold includes the high-frequency threshold and the low-frequency threshold. Taking the low-frequency threshold as an example, this low-frequency threshold is obtained through iterative training of the motion artifact labels of each window of data. Specifically, determine the initial threshold, perform iterative optimization on the initial threshold. In each iteration process, determine the low-frequency statistical information based on the trained low-frequency information scatter plot model, compare the current threshold with the low-frequency statistical information corresponding to each window of data, and determine the training recognition result of the motion artifact training sample based on the comparison result; adjust the current threshold based on the training recognition result of the motion artifact training sample and the motion artifact label, that is, perform an increase or decrease adjustment on the current threshold, and determine the training recognition result and the corresponding motion artifact recognition accuracy rate based on the adjusted threshold again. Until the current threshold meets the recognition accuracy, it is determined as the low-frequency threshold. For example, determine the threshold corresponding to the optimal motion artifact recognition accuracy rate as the low-frequency threshold, or determine the threshold that meets the recognition accuracy as the low-frequency threshold. Optionally, the high-frequency threshold can also be determined based on the above threshold optimization method, which will not be elaborated here.

[0079] On the basis of the above embodiments, verify the motion artifact recognition model corresponding to each image cropping parameter combination based on the motion artifact verification sample, and determine the artifact recognition accuracy rate of each motion artifact recognition model, including: based on the image cropping parameter combination corresponding to each motion artifact recognition model, perform window data cropping on the motion artifact verification sample, input the cropped window data into the relationship model in the motion artifact recognition model to obtain the statistical indicators corresponding to each window data; compare the target threshold in the motion artifact recognition model with each of the statistical indicators to determine the motion artifact description parameters corresponding to each window, and determine the motion artifact verification result based on the motion artifact description parameters corresponding to each window; determine the artifact recognition accuracy rate of the motion artifact recognition model based on the motion artifact label and the motion artifact verification result of each motion artifact verification sample.

[0080] Among them, the motion artifact description parameter can be the response value of the above statistical index relative to the target threshold. Exemplarily, when the statistical index is greater than or equal to the target threshold, a first response value is generated; when the statistical index is less than the target threshold, a second response value is generated. Here, the first response value and the second response value are motion artifact description parameters. The motion artifact verification result is determined based on the motion artifact description parameters corresponding to each window, that is, when the motion artifact description parameter is a specific value, or when the motion artifact description parameter fluctuates, the window corresponding to the motion artifact description parameter is determined as the motion artifact window, and the motion artifact verification result is determined based on each motion artifact window. The matching degree between the motion artifact verification result and the motion artifact label is determined to obtain the artifact recognition accuracy of the motion artifact recognition model, and the motion artifact recognition model is screened through the artifact recognition accuracy to obtain the motion artifact recognition strategy.

[0081] Based on the above embodiments, the input information of the motion artifact recognition model can be window data, and the output information can be motion artifact description parameters. Among them, the motion artifact description parameters include motion artifact classification prediction values, signal-to-noise ratio response values, matrix rank response values, and frequency domain information response values. Correspondingly, based on the intercepted window data and the motion artifact labels corresponding to each window data, the motion artifact recognition model corresponding to the image interception parameter combination is trained, including: determining the motion artifact description parameters corresponding to the window data, and based on the corresponding relationship between the window size, step size, window data, and motion artifact description parameters, constructing a motion artifact recognition model of the motion artifact description parameters, the window size, and the step size.

[0082] Taking the matrix rank response value as an example, the data matrix corresponding to each window data is subjected to singular value decomposition to obtain the decomposition matrix corresponding to the window data, and the decomposition matrix includes singular values. Each singular value is compared with the singular value threshold to obtain the number of singular values greater than the singular value threshold, that is, the matrix rank response value. Based on the corresponding relationship between the window size, step size, window data, and matrix rank response value, a scatter plot is drawn, and further a matrix rank scatter plot model is constructed as the relationship model between the matrix rank response value, window size, step size, and window data, that is, the motion artifact recognition model. Exemplarily, it can be obtained by data fitting based on the corresponding relationship between the matrix rank response value, window size, step size, and window data. For other motion artifact description parameters, the corresponding motion artifact recognition models can also be obtained based on the above training process.

[0083] For each motion artifact verification sample in the set of motion artifact verification samples, after intercepting window data based on the window size and stride, the window data, window size, and stride are used as input information and input into the above-mentioned motion artifact recognition model to obtain the motion artifact description parameters output by the motion artifact recognition model. Based on the motion artifact description parameters of each window data, the verification recognition results corresponding to each motion artifact verification sample are determined. The verification recognition results of the motion artifact verification samples are compared with the motion artifact labels to determine the motion artifact recognition accuracy of the motion artifact recognition strategy. Based on the motion artifact recognition accuracy, the target motion artifact recognition strategy is determined.

[0084] In some embodiments, the input information of the motion artifact recognition model can be window data, and the output information can be a motion artifact recognition index or a motion artifact classification probability. Correspondingly, the motion artifact recognition model can be a machine learning model such as a neural network model. In this embodiment, the network structure of the neural network model is not limited. Exemplarily, the network structure of the motion artifact recognition model can be a convolutional neural network model, a recurrent neural network, or a Transformer model, etc.

[0085] For the window data intercepted for each combination of image interception parameters, a neural network model of at least one network structure can be trained to obtain at least one trained motion artifact recognition model. For any network type, neural network models with different network depths can be created. Taking a convolutional neural network as an example, neural network models with different numbers of layers can be created, such as three layers, ten layers, fifteen layers, etc., which are not limited herein. By training neural network models with different network structures and / or different network depths, the corresponding motion artifact recognition models are obtained.

[0086] Training the motion artifact recognition model corresponding to the image interception parameter combination based on the intercepted window data and the motion artifact labels corresponding to each window data includes: constructing an initial neural network model of at least one network type, and iteratively executing the following training process until the training condition is met to obtain the motion artifact recognition models corresponding to each network type: inputting the window data into the neural network model of the current iteration to obtain the motion artifact prediction result output by the neural network model, and determining a loss function based on the motion artifact prediction result and the corresponding motion artifact label to adjust the network parameters in the neural network model of the current iteration.

[0087] During the training process, the network parameters in the neural network model are adjusted iteratively, where the network parameters include but are not limited to weights. When the training process meets the training conditions, it is determined that the neural network model of the current iteration is trained and completed, and it is determined as a motion artifact recognition model. The training conditions include one or more of the following: the number of iterative training times meets the preset training times, the prediction accuracy of the model meets the preset accuracy threshold, and the training process reaches the minimum convergence state.

[0088] In each iterative training process, the neural network model in the current iterative process is used to predict the input window data, and the motion artifact prediction result of the window data is obtained. Based on the motion artifact prediction result and the corresponding motion artifact label, a loss function is determined. The loss function includes but is not limited to the exponential loss function, the cross-entropy loss function, and the hinge loss function, etc., and can be determined according to the training requirements.

[0089] The loss function is input back into the neural network model of the current iteration to adjust the network parameters in the neural network model of the current iteration. Specifically, it can be based on the gradient descent method to realize the adjustment of the network parameters of the neural network model. Iteratively execute the above training process until the trained motion artifact recognition model is obtained. The motion artifact recognition model has the function of recognizing motion artifacts in window data.

[0090] The recognition accuracy of one or more motion artifact recognition models obtained by the above training is verified through motion artifact verification samples. Based on the screening conditions, the final motion artifact recognition model is determined, and the screened motion artifact recognition model and its corresponding image capture parameters are combined to form a motion artifact recognition strategy.

[0091] The technical solution provided in this embodiment, by presetting multiple combinations of image capture parameters, different combinations of image capture parameters include different window sizes and step lengths. Based on each combination of image capture parameters, the corresponding motion artifact recognition model is trained respectively. The motion artifact recognition accuracy of each motion artifact recognition model is verified based on the motion artifact verification samples, and the target motion artifact recognition model is screened based on the obtained motion artifact recognition accuracy. The target motion artifact recognition model and the corresponding combination of image capture parameters form a motion artifact recognition strategy for recognizing motion artifacts in the medical tomographic images to be processed. As the window size, step length, and motion artifact recognition model that affect the motion artifact recognition accuracy are obtained through the screening of the motion artifact recognition accuracy as a whole, the above factors all meet the requirements of the motion artifact recognition accuracy, which is beneficial to ensuring the accuracy of recognizing motion artifacts in the medical tomographic images to be processed.

[0092] Based on the above embodiments, the embodiments of the present invention further provide a device for recognizing motion artifacts. See Figure 2 ,Figure 2 It is a schematic structural diagram of an apparatus for identifying motion artifacts provided by an embodiment of the present invention. The apparatus includes:

[0093] A window data intercepting module 210, configured to obtain a coronary artery image to be identified, and perform window data screenshot on the coronary artery image along the direction of the coronary artery center line based on a preset window size and step length, so as to obtain a plurality of window data;

[0094] A description parameter determining module 220, configured to identify motion artifact description parameters of each window data;

[0095] A motion artifact identifying module 230, configured to determine a motion artifact identification result in the coronary artery image based on the motion artifact description parameters of each window data.

[0096] Optionally, the motion artifact description parameters include motion artifact classification prediction values;

[0097] The description parameter determining module 220 is configured to: for any window, input the window data of the window into a pre-trained neural network model, and obtain a motion artifact classification prediction value output by the neural network model.

[0098] Optionally, the motion artifact identifying module 230 is configured to:

[0099] Determine a window with a motion artifact classification prediction value being a first specific value as a motion artifact window, and determine a motion artifact position based on the motion artifact window.

[0100] Optionally, the motion artifact description parameters include signal-to-noise ratio response values;

[0101] The description parameter determining module 220 is configured to: for any window, determine signal-to-noise ratio data based on the window data of the window; compare the signal-to-noise ratio data with a signal-to-noise ratio threshold, and determine a signal-to-noise ratio response value of the window data according to the comparison result.

[0102] Optionally, the motion artifact identifying module 230 is configured to:

[0103] Determine a window with a signal-to-noise ratio response value being a second specific value as a motion artifact window, and determine a motion artifact position based on the motion artifact window.

[0104] Optionally, the motion artifact description parameters include matrix rank response values;

[0105] The description parameter determining module 220 is configured to: for any window, perform singular value decomposition on the window data of the window to obtain a decomposition matrix corresponding to each window data; compare a singular value threshold with each singular value in the decomposition matrix, and determine a matrix rank response value of the window.

[0106] Optionally, the motion artifact recognition module 230 is configured to:

[0107] According to the adjacent relationship between windows, compare the matrix rank response values of adjacent windows, determine the windows with fluctuating matrix rank response values as motion artifact windows, and determine the motion artifact positions based on the motion artifact windows.

[0108] Optionally, the motion artifact description parameters include frequency domain information response values;

[0109] The description parameter determination module 220 is configured to: for any window, perform frequency domain conversion on the window data of the window to obtain a frequency domain image, and extract high-frequency information and low-frequency information from the frequency domain image; determine low-frequency statistical information based on the low-frequency information in the frequency domain image, determine high-frequency statistical information based on the high-frequency information in the frequency domain image, compare the low-frequency threshold with the low-frequency statistical information, and compare the high-frequency information with the high-frequency statistical information, and determine the frequency domain information response value of the window according to the comparison results of the high-frequency information and the comparison results of the low-frequency information.

[0110] Optionally, the motion artifact recognition module 230 is configured to: according to the adjacent relationship between windows, compare the frequency domain information response values of adjacent windows, determine the windows with fluctuating frequency domain information response values as motion artifact windows, and determine the motion artifact positions based on the motion artifact windows.

[0111] Optionally, the motion artifact recognition module 230 is configured to: according to the adjacent relationship between windows, determine continuous motion artifact windows; for any set of continuous motion artifact windows, determine the corresponding motion artifact positions based on the sum of the starting position and half of the window depth in at least one continuous motion artifact window.

[0112] Optionally, the motion artifact description parameters include one or more of motion artifact classification prediction values, signal-to-noise ratio response values, matrix rank response values, and frequency domain information response values;

[0113] The description parameter determination module 220 is configured to: respectively determine one or more of motion artifact classification prediction values, signal-to-noise ratio response values, matrix rank response values, and frequency domain information response values based on the window data.

[0114] The motion artifact recognition module 230 is configured to: respectively determine motion artifact sub-recognition results based on any one of the determined motion artifact classification prediction values, signal-to-noise ratio response values, matrix rank response values, and frequency domain information response values, and determine the target recognition result of the motion artifact based on multiple motion artifact sub-recognition results.

[0115] Optionally, the device further includes:

[0116] A motion artifact recognition strategy determination module is used to pre-determine a motion artifact recognition strategy, where the motion artifact recognition strategy includes a window size, a step size, and a motion artifact recognition model.

[0117] Optionally, the motion artifact recognition strategy determination module includes:

[0118] An information acquisition unit is used to acquire motion artifact training samples and multiple groups of image cropping parameter combinations including window sizes and step sizes;

[0119] A model training unit is used to perform window data cropping on the motion artifact training samples along the coronary artery centerline in the motion artifact training samples based on any one of the image cropping parameter combinations, and train a motion artifact recognition model corresponding to the image cropping parameter combination based on the cropped window data and the motion artifact labels corresponding to each window data;

[0120] An artifact recognition accuracy determination unit is used to acquire motion artifact verification samples, verify the motion artifact recognition models corresponding to each image cropping parameter combination based on the motion artifact verification samples, and determine the artifact recognition accuracies of each motion artifact recognition model;

[0121] A motion artifact recognition strategy determination unit is used to determine the motion artifact recognition models that meet the artifact recognition accuracy screening conditions and the image cropping parameter combinations corresponding to the selected motion artifact recognition models as the motion artifact recognition strategy.

[0122] Optionally, the model training unit is used to:

[0123] Determine the statistical indicators corresponding to the window data, and construct a relationship model between the statistical indicators and the window size, step size, and window data based on the corresponding relationship between the window size, step size, window data, and statistical indicators;

[0124] Based on the comparison result between the statistical indicators and the corresponding thresholds, determine the training recognition results of each window data, and determine the target thresholds corresponding to the statistical indicators based on the motion artifact labels corresponding to each window data and the training recognition results of each window data, where the relationship model and the target thresholds constitute the motion artifact recognition model.

[0125] Optionally, the statistical indicators include the signal-to-noise ratio of the window data, the singular values obtained by singular value decomposition of the window data, and the frequency domain statistical information of the window data.

[0126] Optionally, the artifact recognition accuracy determination unit is used to:

[0127] Based on the image capture parameter combinations corresponding to each motion artifact recognition model, window data is captured from the motion artifact verification samples, and the captured window data is input into the relationship model in the motion artifact recognition model to obtain statistical metrics corresponding to each window data;

[0128] Based on the comparison between the target threshold in the motion artifact recognition model and each of the statistical metrics, motion artifact description parameters corresponding to each window are determined, and a motion artifact verification result is determined based on the motion artifact description parameters corresponding to each window;

[0129] Based on the motion artifact labels of each of the motion artifact verification samples and the motion artifact verification result, the artifact recognition accuracy of the motion artifact recognition model is determined.

[0130] Optionally, the model training unit is configured to:

[0131] Construct an initial neural network model of at least one network type, and iteratively execute the following training process until the training conditions are met to obtain a motion artifact recognition model corresponding to each network type:

[0132] Based on the window data being input into the neural network model of the current iteration, a motion artifact prediction result output by the neural network model is obtained, and a loss function is determined based on the motion artifact prediction result and the corresponding motion artifact label to adjust the network parameters in the neural network model of the current iteration.

[0133] The motion artifact recognition device provided by the embodiments of the present invention can execute the motion artifact recognition method provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the motion artifact recognition method.

[0134] Figure 3 FIG. shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0135] As Figure 3As shown, the electronic device 10 includes at least one processor 11 and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0136] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0137] The processor 11 can be various general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the method for identifying motion artifacts.

[0138] In some embodiments, the method for identifying motion artifacts can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for identifying motion artifacts described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the method for identifying motion artifacts in any other appropriate way (for example, by means of firmware).

[0139] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems-on-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0140] The computer program for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The computer program may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0141] In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0142] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0143] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0144] A computing system can include a client and a server. The client and the server are generally far from each other and usually interact through a communication network. The relationship between the client and the server is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0145] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is made herein.

[0146] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0147] Note that the above are only the preferred embodiments of the present invention and the technical principles applied. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein. Various obvious changes, re-adjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments only. Without departing from the concept of the present invention, more other equivalent embodiments can be included, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for identifying motion artifacts, characterized in that Including: Obtain a coronary artery image to be recognized, and perform window data screenshot on the coronary artery image along the direction of the coronary artery centerline based on a preset window size and step length to obtain a plurality of window data; wherein, the coronary artery image is a three-dimensional image; Identify the motion artifact description parameters of each window data, and determine the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of each window data; Wherein, the motion artifact recognition result includes the motion artifact position, and the determination method of the motion artifact position is: Determine consecutive motion artifact windows according to the adjacent relationship of each window. The motion artifact window is window data with a motion artifact description parameter being a specific value, or window data with data fluctuations in the motion artifact description parameter; For any set of consecutive motion artifact windows, determine the corresponding motion artifact position based on the sum of the starting position and half of the window depth in at least one consecutive motion artifact window.

2. The method according to claim 1, characterized in that, The motion artifact description parameter includes a signal-to-noise ratio response value; The identifying the motion artifact description parameters of each window data includes: For any window, determine the signal-to-noise ratio data based on the window data of the window; Compare the signal-to-noise ratio data with a signal-to-noise ratio threshold, and determine the signal-to-noise ratio response value of the window data according to the comparison result; And, the determining the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of each window data includes: Determine the window with the signal-to-noise ratio response value being a second specific value as a motion artifact window, and determine the motion artifact position based on the motion artifact window.

3. The method according to claim 1, wherein The motion artifact description parameter includes a matrix rank response value; The identifying the motion artifact description parameters of each window data includes: For any window, perform singular value decomposition on the window data of the window to obtain a decomposition matrix corresponding to each window data; Compare the singular value threshold with each singular value in the decomposition matrix to determine the matrix rank response value of the window.

4. The method according to claim 3, wherein The determining the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of each window data includes: According to the adjacent relationship of each window, compare the matrix rank response values of adjacent windows, determine the window with matrix rank response value fluctuations as a motion artifact window, and determine the motion artifact position based on the motion artifact window.

5. The method according to claim 1, wherein The motion artifact description parameter includes a frequency domain information response value; The identifying the motion artifact description parameters of each window data includes: For any window, perform frequency domain conversion on the window data of the window to obtain a frequency domain image, and extract high-frequency information and low-frequency information in the frequency domain image; Determine low-frequency statistical information based on the low-frequency information in the frequency domain image, determine high-frequency statistical information based on the high-frequency information in the frequency domain image, compare the low-frequency threshold with the low-frequency statistical information, and compare the high-frequency information with the high-frequency statistical information. Determine the frequency domain information response value of the window according to the comparison results of the high-frequency information and the low-frequency information.

6. The method according to claim 5, characterized in that, The determining the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of each window data includes: According to the adjacent relationship of each window, compare the frequency-domain information response values of adjacent windows, determine the window with fluctuating frequency-domain information response values as the motion artifact window, and determine the motion artifact position based on the motion artifact window.

7. The method according to claim 1, characterized in that, The motion artifact description parameters include one or more of a motion artifact classification prediction value, a signal-to-noise ratio response value, a matrix rank response value, and a frequency-domain information response value; wherein, the determination method of the matrix rank response value includes: performing singular value decomposition on the data matrix corresponding to the data of each window to obtain a decomposition matrix corresponding to the window data, and the decomposition matrix includes singular values; comparing each of the singular values with a singular value threshold, and determining the number of singular values greater than the singular value threshold as the matrix rank response value; Determining the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of the data of each window includes: Determining one or more of a motion artifact classification prediction value, a signal-to-noise ratio response value, a matrix rank response value, and a frequency-domain information response value based on the window data; Determining a motion artifact sub-recognition result based on any one of the determined motion artifact classification prediction value, signal-to-noise ratio response value, matrix rank response value, and frequency-domain information response value, and determining the target recognition result of the motion artifact based on multiple motion artifact sub-recognition results.

8. The method according to claim 1, wherein The method further includes: Pre-determining a motion artifact recognition strategy, which includes a window size and a step size, and a motion artifact recognition model.

9. The method according to claim 8, wherein Pre-determining the motion artifact recognition strategy includes: Obtaining motion artifact training samples and multiple groups of image cropping parameter combinations including window sizes and step sizes; Based on any one of the image cropping parameter combinations, cropping the window data of the motion artifact training sample along the coronary artery center line in the motion artifact training sample, and training a motion artifact recognition model corresponding to the image cropping parameter combination based on the cropped window data and the motion artifact label corresponding to each window data; Obtaining motion artifact verification samples, verifying the motion artifact recognition models corresponding to each image cropping parameter combination based on the motion artifact verification samples, and determining the artifact recognition accuracy of each motion artifact recognition model; Determining the motion artifact recognition model that meets the artifact recognition accuracy screening condition, and the image cropping parameter combination corresponding to the selected motion artifact recognition model as the motion artifact recognition strategy.

10. The method according to claim 9, wherein Training the motion artifact recognition model corresponding to the image cropping parameter combination based on the cropped window data and the motion artifact label corresponding to each window data includes: Determining the statistical index corresponding to the window data, and constructing a relationship model between the statistical index and the window size, step size, and window data based on the corresponding relationship between the window size, step size, window data, and the statistical index; Determining the training recognition result of each window data based on the comparison result between the statistical index and the corresponding threshold, and determining the target threshold corresponding to the statistical index based on the motion artifact label corresponding to each window data and the training recognition result of each window data, wherein the relationship model and the target threshold constitute the motion artifact recognition model.

11. The method according to claim 10, characterized in that The statistical indicators include the signal-to-noise ratio of the window data, the frequency-domain statistical information of the window data, and the singular values obtained by singular value decomposition of the window data.

12. An apparatus for identifying motion artifacts, characterized in that including: A window data intercepting module, configured to obtain a coronary artery image to be recognized, and perform window data screenshot on the coronary artery image along the direction of the coronary artery center line in the coronary artery image based on a preset window size and step length, so as to obtain a plurality of window data; wherein, the coronary artery image is a three-dimensional image; A motion artifact recognition module, configured to recognize the motion artifact description parameters of each window data, and determine the motion artifact recognition result in the coronary artery image based on the motion artifact description parameters of each window data; Wherein, the motion artifact recognition result includes the motion artifact position, and the determination method of the motion artifact position is: Determine continuous motion artifact windows according to the adjacent relationship of each window, where the motion artifact window is window data with a motion artifact description parameter being a specific value or the motion artifact description parameter having data fluctuations; For any set of continuous motion artifact windows, determine the corresponding motion artifact position based on the sum of the starting position and half of the window depth in at least one continuous motion artifact window.

13. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can execute a method for recognizing a motion artifact according to any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions are used to implement a method for recognizing a motion artifact according to any one of claims 1-11 when executed by a processor.

Images