Nuclear magnetic multi-time sequence image map segmentation method and device and electronic equipment

Abstract

The invention discloses an atlas segmentation method, and particularly relates to a nuclear magnetic multi-time sequence image atlas segmentation method and device and electronic equipment, and the method comprises the steps: carrying out the data splitting of nuclear magnetic multi-time sequence image data, and obtaining first data and second data, the nuclear magnetic multi-time-sequence image data comprises a time sequence dimension, a position dimension, an image height dimension and an image width dimension, the first data comprises the time sequence dimension, the image height dimension and the image width dimension, and the second data comprises the position dimension, the image height dimension and the image width dimension; carrying out cardiac cycle position regression based on the first data to obtain a cardiac cycle position of the image data; performing relative spatial position regression based on the second data to obtain a relative spatial position of the image data; and based on the cardiac cycle position and the relative spatial position of the image data, performing map segmentation on the image data to obtain a map segmentation result. According to the method, a better map segmentation effect can be realized.

Description

Technical Field

[0001] The present invention relates to a spectrum segmentation method, and specifically to a nuclear magnetic resonance multi-time series image spectrum segmentation method, device and electronic equipment. Background Art

[0002] Nuclear Magnetic Resonance Imaging (NMRI), also known as Magnetic Resonance Imaging (MRI), uses the principle of nuclear magnetic resonance (NMR). Based on the different attenuation of the released energy in different structural environments inside the material, the emitted electromagnetic waves are detected by applying an external gradient magnetic field. The position and type of the atomic nuclei that make up the object can be known, and a structural image of the object's interior can be drawn based on this.

[0003] CINE sequence short axis is a common scanning method for cardiac imaging in MRI, usually stored as a DICOM (Digital Imaging and Communications in Medicine) format file. CINE sequence short axis is a multi-time series four-dimensional image, usually containing four dimensions of information, namely time series dimension, position dimension, image height dimension and image width dimension. CINE short axis image format is a combination of time and space dimensions, including dynamic change information within the cardiac cycle.

[0004] When performing atlas segmentation on CINE sequence short-axis images, the image data in the CINE sequence short-axis images (that is, each frame of the time series data) is usually used as input and directly input into the deep learning atlas segmentation model for atlas segmentation, without considering the time series information and scanning information of the data, resulting in poor atlas segmentation effect. Summary of the invention

[0005] The technical problem to be solved by the present invention is that the above-mentioned method commonly used in the prior art does not take into account the timing information and scanning information of the data when performing atlas segmentation, resulting in poor atlas segmentation effect. In order to solve the above-mentioned problem, the present invention provides a method, device and electronic device for nuclear magnetic resonance multi-time series image atlas segmentation.

[0006] The content of the present invention includes:

[0007] In a first aspect, an embodiment of the present invention provides a method for segmenting a multi-time series nuclear magnetic resonance imaging atlas, comprising:

[0008] Splitting the nuclear magnetic resonance multi-time series image data to obtain first data and second data, wherein the nuclear magnetic resonance multi-time series image data includes a time series dimension, a position dimension, an image height dimension, and an image width dimension, the first data includes the time series dimension, the image height dimension, and the image width dimension, and the second data includes the position dimension, the image height dimension, and the image width dimension;

[0009] Perform cardiac cycle position regression based on the first data to obtain the cardiac cycle position of image data, wherein the image data includes the image height dimension and the image width dimension;

[0010] Performing relative spatial position regression based on the second data to obtain the relative spatial position of the image data;

[0011] Based on the cardiac cycle position of the image data and the relative spatial position of the image data, the image data is segmented to obtain a segmentation result.

[0012] Optionally, the NMR multi-time series imaging data is CINE sequence short-axis data, the size of the CINE sequence short-axis data is (P, T, H, W), the size of the first data is (T, H, W), the number of the first data is P, the size of the second data is (P, H, W), and the number of the second data is T, wherein P is used to characterize the size of the position dimension, T is used to characterize the size of the timing dimension, H is used to characterize the size of the image height dimension, and W is used to characterize the size of the image width dimension, and P, T, H, and W are all positive integers.

[0013] Optionally, performing cardiac cycle position regression based on the first data to obtain the cardiac cycle position of the image data includes:

[0014] The P / 2th first data is input into a pre-trained first deep learning regression model to perform cardiac cycle position regression to obtain a first output vector. The first output vector is used to characterize the cardiac cycle position of the image data corresponding to each moment in the first data. The first output vector is a T-dimensional vector.

[0015] Optionally, performing relative spatial position regression based on the second data to obtain the relative spatial position of the image data includes:

[0016] Input T second data into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector, where the second output vector is used to characterize the relative spatial position of the image data corresponding to each position in the second data, and the second output vector is a P-dimensional vector.

[0017] Optionally, the step of inputting the T second data into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector includes:

[0018] Input T second data into the pre-trained second deep learning regression model to perform relative spatial position regression, and obtain T third output vectors, where the third output vector is a P-dimensional vector;

[0019] An average value of the T third output vectors is determined as the second output vector.

[0020] Optionally, performing atlas segmentation on the image data based on the cardiac cycle position of the image data and the relative spatial position of the image data to obtain the atlas segmentation result includes:

[0021] Inputting the first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and relative spatial position of the target image data into a pre-trained deep learning segmentation model for atlas segmentation to obtain an atlas segmentation result of the target image data;

[0022] Among them, the target image data is the image data corresponding to (p, t), the first image data is the image data corresponding to (p-1, t), the second image data is the image data corresponding to (p, t-1), the third image data is the image data corresponding to (p, t+1), and the fourth image data is the image data corresponding to (p+1, t), p is any position in the position dimension, and t is any moment in the timing dimension.

[0023] Optionally, the step of inputting the first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and relative spatial position of the target image data into a pre-trained deep learning segmentation model for atlas segmentation to obtain the atlas segmentation result of the target image data comprises:

[0024] Inputting the first image data, the second image data, the target image data, the third image data and the fourth image data into the encoding part of the deep learning segmentation model for downsampling processing to obtain a downsampled feature map;

[0025] The down-sampled feature map is multiplied by the cardiac cycle position and the relative spatial position of the target image data respectively, and then spliced ​​to obtain a target feature map;

[0026] The target feature map is input into the decoding part of the deep learning segmentation model for upsampling processing to obtain the atlas segmentation result of the target image data.

[0027] In a second aspect, an embodiment of the present invention further provides a nuclear magnetic resonance multi-time series image atlas segmentation device, comprising:

[0028] A data splitting module, used for splitting the nuclear magnetic resonance multi-time series image data to obtain first data and second data, wherein the nuclear magnetic resonance multi-time series image data includes a time series dimension, a position dimension, an image height dimension and an image width dimension, the first data includes the time series dimension, the image height dimension and the image width dimension, and the second data includes the position dimension, the image height dimension and the image width dimension;

[0029] A first processing module, configured to perform cardiac cycle position regression based on the first data to obtain the cardiac cycle position of image data, wherein the image data includes the image height dimension and the image width dimension;

[0030] A second processing module, used for performing relative spatial position regression based on the second data to obtain the relative spatial position of the image data;

[0031] The atlas segmentation module is used to perform atlas segmentation on the image data based on the cardiac cycle position of the image data and the relative spatial position of the image data to obtain the atlas segmentation result.

[0032] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor; the processor is used to read the program in the memory to implement the steps in the method for segmenting nuclear magnetic resonance multi-time series images as described in the first aspect.

[0033] In a fourth aspect, an embodiment of the present invention provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method for segmenting nuclear magnetic resonance multi-time series images as described in the first aspect.

[0034] The beneficial effect of the present invention is that, in an embodiment of the present invention, the nuclear magnetic multi-time series imaging data is firstly split according to the time series dimension and the position dimension to obtain two three-dimensional data, namely the first data and the second data, and then the cardiac cycle position regression is performed based on the first data to obtain the cardiac cycle position of the image data, and the relative spatial position regression is performed based on the second data to obtain the relative spatial position of the image data; finally, based on the cardiac cycle position of the image data and the relative spatial position of the image data, the image data is subjected to atlas segmentation to obtain the atlas segmentation result. Through the above method, the cardiac cycle position of the image data and the relative spatial position of the image data are fully utilized in the process of atlas segmentation, the accuracy of atlas segmentation is improved, and a better atlas segmentation effect is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Attached Figure 1 A flow chart of a method for segmenting a multi-time series nuclear magnetic resonance imaging spectrum provided by an embodiment of the present invention;

[0036] Attached Figure 2 A schematic diagram of the structure of a deep learning segmentation model provided by an embodiment of the present invention;

[0037] Attached Figure 3 A schematic diagram of a device for segmenting a nuclear magnetic resonance multi-time series image atlas provided by an embodiment of the present invention;

[0038] Attached Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In the embodiments of the present application, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In the embodiments of the present application, the term "multiple" refers to two or more, and other quantifiers are similar. The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first" and "second" are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more.

[0040] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0042] The present application provides a method, device and electronic device for segmenting a multi-time series of nuclear magnetic resonance images, aiming to improve the effect of image segmentation. Figure 1 , Figure 1 The present invention provides a flow chart of a method for segmenting a multi-time series of nuclear magnetic resonance images. The method specifically includes the following steps:

[0043] Step 101, splitting the nuclear magnetic resonance multi-time series image data to obtain first data and second data, the nuclear magnetic resonance multi-time series image data includes a time dimension, a position dimension, an image height dimension and an image width dimension, the first data includes the time dimension, the image height dimension and the image width dimension, and the second data includes the position dimension, the image height dimension and the image width dimension.

[0044] Step 102: Perform cardiac cycle position regression based on the first data to obtain the cardiac cycle position of image data, wherein the image data includes the image height dimension and the image width dimension.

[0045] Step 103: Perform relative spatial position regression based on the second data to obtain the relative spatial position of the image data.

[0046] Step 104: performing atlas segmentation on the image data based on the cardiac cycle position of the image data and the relative spatial position of the image data to obtain an atlas segmentation result.

[0047] Optionally, in some embodiments, the nuclear magnetic resonance multi-time series image data is CINE sequence short axis data. CINE sequence short axis data is a multi-time series four-dimensional image. Optionally, in some embodiments, the size of the CINE sequence short axis data is (P, T, H, W), the size of the first data is (T, H, W), the number of the first data is P, the size of the second data is (P, H, W), and the number of the second data is T, wherein P is used to characterize the size of the position dimension, T is used to characterize the size of the timing dimension, H is used to characterize the size of the image height dimension, and W is used to characterize the size of the image width dimension, and P, T, H, and W are all positive integers.

[0048] The short-axis data of the CINE sequence is multiple time-series images collected at multiple locations, usually stored in Dicom format. In specific implementation, the short-axis data of the CINE sequence can be regarded as a (P, T, H, W) multi-dimensional matrix, where P is the number of slices, T is the number of time points, H is the height of the image data, and W is the width of the image data.

[0049] As a specific embodiment, in step 101, the short-axis data of the CINE sequence is split based on the position tag (PositionTag) of the Dicom data to obtain P three-dimensional data, namely the first data (T, H, W). At the same time, since the image acquisition is controlled by the signal, each position is at the same position in the cardiac cycle at every moment, so the short-axis data of the CINE sequence can be split according to the information of the time series dimension, and the short-axis data of the CINE sequence is split into T three-dimensional data, namely the second data (P, H, W).

[0050] Optionally, in some embodiments, step 102 includes:

[0051] The P / 2th first data is input into a pre-trained first deep learning regression model to perform cardiac cycle position regression to obtain a first output vector. The first output vector is used to characterize the cardiac cycle position of the image data corresponding to each moment in the first data. The first output vector is a T-dimensional vector.

[0052] As an optional implementation, the number of first data is P. In this embodiment, only the P / 2th data is input into the first deep learning regression model for one inference to obtain a first output vector.

[0053] Specifically, the P / 2th first data is input into the first deep learning regression model for cardiac cycle position regression. The end-diastole-end-systole-end-diastole is agreed to be a cardiac cycle and defined as 0.0-0.5-1.0. The first output vector obtained is a T-dimensional vector (T values ​​of 0.0-1.0). The tth element in the T-dimensional vector is used to represent the cardiac cycle position of the image data corresponding to time t.

[0054] Optionally, in some embodiments, step 103 includes:

[0055] Input T second data into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector, where the second output vector is used to characterize the relative spatial position of the image data corresponding to each position in the second data, and the second output vector is a P-dimensional vector.

[0056] Specifically, it is agreed that the uppermost and lowermost ends of the heart are a relative position period and defined as 0.0-1.0, and the second data is input into the second deep learning regression model for relative spatial position regression, and the second output vector obtained is a P-dimensional vector (P values ​​of 0.0-1.0), and the p-th element in the P-dimensional vector is used to represent the relative spatial position of the image data corresponding to position p.

[0057] As an optional implementation, any one of the T second data is input into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector.

[0058] Optionally, as another optional implementation, the step of inputting the T second data into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector includes:

[0059] Input T second data into the pre-trained second deep learning regression model to perform relative spatial position regression, and obtain T third output vectors, where the third output vector is a P-dimensional vector;

[0060] An average value of the T third output vectors is determined as the second output vector.

[0061] The number of second data is T. The T second data are respectively input into the second deep learning regression model for T inferences to obtain T third output vectors. The third output vector is a P-dimensional vector. The average of the T results is taken at each position p. The final P-dimensional result is the second output vector.

[0062] In a specific implementation, the first deep learning regression model is constructed based on ResNet, and / or the second deep learning regression model is constructed based on ResNet. The specific structures of the first deep learning regression model and the second deep learning regression model are not limited here, and the execution order of step 102 and step 103 is not limited here.

[0063] In step 104, a deep learning segmentation model is constructed for performing cardiac atlas segmentation to achieve the segmentation of five regions: left ventricular chamber (endothelium), left ventricular myocardium (epicardium), right ventricular chamber (endothelium), right atrium, and left atrium. As an optional implementation, step 104 includes:

[0064] Inputting the first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and relative spatial position of the target image data into a pre-trained deep learning segmentation model for atlas segmentation to obtain an atlas segmentation result of the target image data;

[0065] Among them, the target image data is the image data corresponding to (p, t), the first image data is the image data corresponding to (p-1, t), the second image data is the image data corresponding to (p, t-1), the third image data is the image data corresponding to (p, t+1), and the fourth image data is the image data corresponding to (p+1, t), p is any position in the position dimension, and t is any moment in the timing dimension.

[0066] Specifically, the image data corresponding to the current position (p, t) at the current moment is segmented, and the data before and after the current time are added, and the data before and after the current moment are added, that is, input = [(p-1, t), (p, t-1), (p, t), (p, t+1), (p+1, t)], forming data of size (5, H, W), and the output is the (H, W, N)-dimensional segmentation result corresponding to the current position (p, t) at the current moment, where N = 5, the 5 regions of the spectrum segmentation, and then reduced to (H, W) by ArgMax, where ArgMax is a maximum probability selection mechanism, that is, the largest value is selected from N values ​​and its subscript is output. For example: [0.1, 0.15, 0.5, 0.25, 0.05], after ArgMax, 2 is output. (The subscript starts from 0, so the subscript corresponding to the position of 0.5 is 2).

[0067] See also Figure 2 Optionally, in some embodiments, the first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and relative spatial position of the target image data are input into a pre-trained deep learning segmentation model for atlas segmentation to obtain the atlas segmentation result of the target image data, including:

[0068] Inputting the first image data, the second image data, the target image data, the third image data and the fourth image data into the encoding part of the deep learning segmentation model for downsampling processing to obtain a downsampled feature map;

[0069] The down-sampled feature map is multiplied by the cardiac cycle position and the relative spatial position of the target image data respectively, and then spliced ​​to obtain a target feature map;

[0070] The target feature map is input into the decoding part of the deep learning segmentation model for upsampling processing to obtain the atlas segmentation result of the target image data.

[0071] In this embodiment, the deep learning segmentation model is built based on UNet. UNet includes an encoding part and a decoding part, and the features of each stage in the encoding part are directly transferred to the corresponding stage in the decoding part through jump connections. The above data input = [(p-1, t), (p, t-1), (p, t), (p, t+1), (p+1, t)] of size (5, H, W) is input into the encoding part of UNet for downsampling processing to obtain a downsampled feature map. In the last downsampling of UNet, the cardiac cycle position (such as 0.34) and relative spatial position (0.56) corresponding to (p, t) are added to UNet. After multiplying the downsampled feature map output by the last downsampling with the cardiac cycle position and the relative spatial position, the obtained target feature map is input into the decoding part for upsampling processing to obtain the spectrum segmentation result. Among them, the cardiac cycle position and the relative spatial position are both a numerical value, which is multiplied by the downsampled feature map, that is, each element of the downsampled feature map is directly multiplied by the numerical value. Cardiac cycle position p t After multiplying with the downsampled feature map, we get the time feature, denoted as f t , relative spatial position p p The position feature is obtained by multiplying it with the downsampled feature map, denoted as f p , f t and f p Through concatenation splicing together, that is, concatenate(f t ,f p ).

[0072] The relaxation and contraction caused by the beating of the heart will bring about structural changes in the same position of the heart (regular enlargement or reduction), so the images at the previous and next moments play a supervisory and constraining role on the current moment, that is, the overall range of the heart at the current moment and the range of each atlas area of ​​the heart can be constrained by the overall range of the heart at the previous and next moments and the range of each atlas area of ​​the heart. The network can extract more information from the images at the previous and next moments to make the current moment partition more accurate; similarly, the heart structure at the upper and lower positions at the same (current) moment can also provide corresponding constraints, because the heart is a continuous whole, and the structure has relevant coherence, which can also play a supervisory and constraining role. Therefore, through the above method, dual position supervision can be achieved, and effective information in the time series dimension and position dimension can be fully utilized.

[0073] The beneficial effect of the present invention is that, in an embodiment of the present invention, the nuclear magnetic multi-time series imaging data is firstly split according to the time series dimension and the position dimension to obtain two three-dimensional data, namely the first data and the second data, and then the cardiac cycle position regression is performed based on the first data to obtain the cardiac cycle position of the image data, and the relative spatial position regression is performed based on the second data to obtain the relative spatial position of the image data; finally, based on the cardiac cycle position of the image data and the relative spatial position of the image data, the image data is subjected to atlas segmentation to obtain the atlas segmentation result. Through the above method, the cardiac cycle position of the image data and the relative spatial position of the image data are fully utilized in the process of atlas segmentation, the accuracy of atlas segmentation is improved, and a better atlas segmentation effect is obtained.

[0074] like Figure 3 As shown, the embodiment of the present invention further provides a nuclear magnetic resonance multi-time series image atlas segmentation device 300, comprising:

[0075] The data splitting module 301 is used to split the nuclear magnetic resonance multi-time series image data to obtain first data and second data, wherein the nuclear magnetic resonance multi-time series image data includes a time series dimension, a position dimension, an image height dimension and an image width dimension, the first data includes the time series dimension, the image height dimension and the image width dimension, and the second data includes the position dimension, the image height dimension and the image width dimension;

[0076] A first processing module 302, configured to perform cardiac cycle position regression based on the first data to obtain the cardiac cycle position of image data, wherein the image data includes the image height dimension and the image width dimension;

[0077] A second processing module 303, configured to perform relative spatial position regression based on the second data to obtain the relative spatial position of the image data;

[0078] The atlas segmentation module 304 is used to perform atlas segmentation on the image data based on the cardiac cycle position of the image data and the relative spatial position of the image data to obtain atlas segmentation results.

[0079] Optionally, the nuclear magnetic resonance multi-time series imaging data is CINE sequence short-axis data, the size of the CINE sequence short-axis data is (P, T, H, W), the size of the first data is (T, H, W), the number of the first data is P, the size of the second data is (P, H, W), and the number of the second data is T, wherein P is used to characterize the size of the position dimension, T is used to characterize the size of the timing dimension, H is used to characterize the size of the image height dimension, and W is used to characterize the size of the image width dimension, and P, T, H, and W are all positive integers.

[0080] Optionally, the first processing module 302 is specifically configured to:

[0081] The P / 2th first data is input into a pre-trained first deep learning regression model to perform cardiac cycle position regression to obtain a first output vector. The first output vector is used to characterize the cardiac cycle position of the image data corresponding to each moment in the first data. The first output vector is a T-dimensional vector.

[0082] Optionally, the second processing module 303 is specifically configured to:

[0083] Input T second data into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector, where the second output vector is used to characterize the relative spatial position of the image data corresponding to each position in the second data, and the second output vector is a P-dimensional vector.

[0084] Optionally, the step of inputting the T second data into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector includes:

[0085] Input T second data into the pre-trained second deep learning regression model to perform relative spatial position regression, and obtain T third output vectors, where the third output vector is a P-dimensional vector;

[0086] An average value of the T third output vectors is determined as the second output vector.

[0087] Optionally, the atlas segmentation module 304 is specifically used for:

[0088] Inputting the first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and relative spatial position of the target image data into a pre-trained deep learning segmentation model for atlas segmentation to obtain an atlas segmentation result of the target image data;

[0089] Among them, the target image data is the image data corresponding to (p, t), the first image data is the image data corresponding to (p-1, t), the second image data is the image data corresponding to (p, t-1), the third image data is the image data corresponding to (p, t+1), and the fourth image data is the image data corresponding to (p+1, t), p is any position in the position dimension, and t is any moment in the timing dimension.

[0090] Optionally, the step of inputting the first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and relative spatial position of the target image data into a pre-trained deep learning segmentation model for atlas segmentation to obtain the atlas segmentation result of the target image data comprises:

[0091] Inputting the first image data, the second image data, the target image data, the third image data and the fourth image data into the encoding part of the deep learning segmentation model for downsampling processing to obtain a downsampled feature map;

[0092] The down-sampled feature map is multiplied by the cardiac cycle position and the relative spatial position of the target image data respectively, and then spliced ​​to obtain a target feature map;

[0093] The target feature map is input into the decoding part of the deep learning segmentation model for upsampling processing to obtain the atlas segmentation result of the target image data.

[0094] The NMR multi-time series image atlas segmentation device 300 provided in the embodiment of the present application can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated in this embodiment.

[0095] It should be noted that the division of units in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0096] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0097] like Figure 4 As shown, an embodiment of the present application provides an electronic device 400, comprising: a memory 402, a processor 401, and a program stored in the memory 402 and executable on the processor 401; the processor 401 is used to read the program in the memory 402 to implement the steps in the above-mentioned nuclear magnetic resonance multi-time series image atlas segmentation method.

[0098] An embodiment of the present application also provides a readable storage medium, on which a program is stored. When the program is executed by a processor, the various processes of the above-mentioned nuclear magnetic resonance multi-time series image atlas segmentation method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here. Among them, the readable storage medium can be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical storage (such as compact disk (CD), digital video disc (DVD), Blu-ray Disc (BD), high-definition versatile disc (HVD), etc.), and semiconductor memory (such as read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read only memory (EEPROM), non-volatile memory (NAND FLASH), solid-state drive (Solid State Disk or Solid State Drive, SSD)), etc.

[0099] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0100] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0101] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for segmenting a nuclear magnetic resonance multi-time series image atlas, characterized in that: include: Splitting the nuclear magnetic resonance multi-time series image data to obtain first data and second data, wherein the nuclear magnetic resonance multi-time series image data includes a time series dimension, a position dimension, an image height dimension, and an image width dimension, the first data includes the time series dimension, the image height dimension, and the image width dimension, and the second data includes the position dimension, the image height dimension, and the image width dimension; Perform cardiac cycle position regression based on the first data to obtain the cardiac cycle position of image data, wherein the image data includes the image height dimension and the image width dimension; Performing relative spatial position regression based on the second data to obtain the relative spatial position of the image data; Based on the cardiac cycle position of the image data and the relative spatial position of the image data, the image data is segmented to obtain a segmentation result.

2. The method according to claim 1, characterized in that: The nuclear magnetic resonance multi-time series imaging data is CINE sequence short-axis data, the size of the CINE sequence short-axis data is (P, T, H, W), the size of the first data is (T, H, W), the number of the first data is P, the size of the second data is (P, H, W), and the number of the second data is T, wherein P is used to characterize the size of the position dimension, T is used to characterize the size of the timing dimension, H is used to characterize the size of the image height dimension, and W is used to characterize the size of the image width dimension, and P, T, H, and W are all positive integers.

3. The method according to claim 2, characterized in that: The performing cardiac cycle position regression based on the first data to obtain the cardiac cycle position of the image data includes: The P / 2th first data is input into a pre-trained first deep learning regression model to perform cardiac cycle position regression to obtain a first output vector. The first output vector is used to characterize the cardiac cycle position of the image data corresponding to each moment in the first data. The first output vector is a T-dimensional vector.

4. The method according to claim 2, characterized in that: The performing relative spatial position regression based on the second data to obtain the relative spatial position of the image data includes: Input T second data into a pre-trained second deep learning regression model for relative spatial position regression to obtain a second output vector, where the second output vector is used to characterize the relative spatial position of the image data corresponding to each position in the second data, and the second output vector is a P-dimensional vector.

5. The method according to claim 4, characterized in that: The step of inputting the T second data into a pre-trained second deep learning regression model to perform relative spatial position regression to obtain a second output vector includes: Input T second data into the pre-trained second deep learning regression model to perform relative spatial position regression, and obtain T third output vectors, where the third output vector is a P-dimensional vector; An average value of the T third output vectors is determined as the second output vector.

6. The method according to claim 1, characterized in that: The performing atlas segmentation on the image data based on the cardiac cycle position of the image data and the relative spatial position of the image data to obtain the atlas segmentation result includes: Inputting the first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and relative spatial position of the target image data into a pre-trained deep learning segmentation model for atlas segmentation to obtain an atlas segmentation result of the target image data; Among them, the target image data is the image data corresponding to (p, t), the first image data is the image data corresponding to (p-1, t), the second image data is the image data corresponding to (p, t-1), the third image data is the image data corresponding to (p, t+1), and the fourth image data is the image data corresponding to (p+1, t), p is any position in the position dimension, and t is any moment in the timing dimension.

7. The method according to claim 6, characterized in that: The first image data, the second image data, the target image data, the third image data, the fourth image data, and the cardiac cycle position and the relative spatial position of the target image data are input into a pre-trained deep learning segmentation model for atlas segmentation to obtain the atlas segmentation result of the target image data, including: Inputting the first image data, the second image data, the target image data, the third image data and the fourth image data into the encoding part of the deep learning segmentation model for downsampling processing to obtain a downsampled feature map; The down-sampled feature map is multiplied by the cardiac cycle position and the relative spatial position of the target image data respectively, and then spliced ​​to obtain a target feature map; The target feature map is input into the decoding part of the deep learning segmentation model for upsampling processing to obtain the atlas segmentation result of the target image data.

8. A nuclear magnetic resonance multi-time series image atlas segmentation device, characterized in that it includes: A data splitting module, used for splitting the nuclear magnetic resonance multi-time series image data to obtain first data and second data, wherein the nuclear magnetic resonance multi-time series image data includes a time series dimension, a position dimension, an image height dimension and an image width dimension, the first data includes the time series dimension, the image height dimension and the image width dimension, and the second data includes the position dimension, the image height dimension and the image width dimension; A first processing module, configured to perform cardiac cycle position regression based on the first data to obtain the cardiac cycle position of image data, wherein the image data includes the image height dimension and the image width dimension; A second processing module, used for performing relative spatial position regression based on the second data to obtain the relative spatial position of the image data; The atlas segmentation module is used to perform atlas segmentation on the image data based on the cardiac cycle position of the image data and the relative spatial position of the image data to obtain the atlas segmentation result.

9. An electronic device, comprising: A memory, a processor, and a program stored in the memory and executable on the processor; wherein the processor is used to read the program in the memory to implement the steps in the method for segmenting nuclear magnetic resonance multi-time series images as described in any one of claims 1 to 7.

10. A readable storage medium for storing a program, characterized in that: When the program is executed by a processor, the steps in the method for segmenting nuclear magnetic resonance multi-time series images as described in any one of claims 1 to 7 are implemented.

Images