Field pattern determination method suitable for magnetic resonance scanning
By combining non-machine learning and machine learning methods to determine the field map in magnetic resonance scan, the problems of low accuracy of field maps and inverse phenomena are solved, and the accuracy of fat quantification is improved.
Patent Information
- Application Number
- CN202310592042.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-24
- Publication Date
- 2024-11-26
AI Technical Summary
During the existing fat quantification process in magnetic resonance scans, the accuracy of the field map is not high, resulting in frequent occurrence of inverse phenomena, affecting the diagnostic results.
A field graph determination method combining non-machine learning methods and machine learning methods is adopted to determine the first field graph and the second field graph corresponding to the echo image group, and then calculate the target field graph to reduce the inverse phenomenon.
It improves the accuracy of the field map, reduces the occurrence of inverse phenomena, and enhances the reliability of fat quantitative results.
Smart Images

Figure CN119024240A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of medical technology, and in particular to a method for determining a field map suitable for magnetic resonance scanning. Background Art
[0002] Fat quantification is a major technique used in magnetic resonance imaging (MRI), and the results of fat quantification are an important diagnostic basis in the detection of heart and lung diseases.
[0003] At present, in the process of obtaining fat quantification, the echo image of the part of interest is usually subjected to region growing processing to determine the field map corresponding to the echo image, and then fat quantification is performed based on the field map corresponding to the echo image. In other words, the field map corresponding to the echo image will affect the final fat quantification result.
[0004] However, the accuracy of the field map currently determined by methods such as global iterative filtering or region growing is not high, which leads to the inversion phenomenon of the determined field map. Summary of the invention
[0005] Based on this, it is necessary to provide a method for determining a field map suitable for magnetic resonance scanning that can improve the accuracy of the field map and reduce the reflection phenomenon in order to solve the above technical problems. The method includes:
[0006] Determine an echo image group corresponding to the part of interest; the echo image group includes a plurality of echo image frames;
[0007] Determine a first field image corresponding to the echo image group according to the echo image group and a non-machine learning method;
[0008] Determine a second field image corresponding to the echo image group according to the echo image group and the machine learning method;
[0009] A target field image is determined according to the first field image and the second field image.
[0010] In one embodiment, the machine learning method includes a supervised network; the method further includes:
[0011] Acquire initial training data; the initial training data includes a plurality of echo image groups corresponding to parts of interest;
[0012] Determine a target data set according to the initial training data; the target data set includes a first data set and a second data set; the first data set includes a field map determined based on the non-machine learning method and the initial training data, and the second data set includes a field map determined based on an unsupervised network and the initial training data;
[0013] The initial supervised network is trained according to the initial training data and the target data set to obtain the supervised network.
[0014] In one embodiment, determining a target field image according to the first field image and the second field image comprises:
[0015] Determine a difference value between each first field image value and a corresponding second field image value; the first field image value is a field image value of a first pixel point in the first field image, and the second field image value is a field image value of a second pixel point in the second field image corresponding to the first pixel point;
[0016] The target field image is determined according to the difference between each of the first field image values and the corresponding second field image value.
[0017] In one embodiment, determining the target field image according to the difference between each first field image value and the corresponding second field image value includes:
[0018] The field image value of the first target pixel point is used as the field image value of the pixel point corresponding to the first target pixel point in the target field image; the first target pixel point is the first pixel point whose difference value is less than or equal to the preset difference threshold;
[0019] The field image value of the second target pixel point is used as the field image value of the pixel point corresponding to the second target pixel point in the target field image; the second target pixel point is the first pixel point whose difference value is not less than the preset difference threshold.
[0020] In one embodiment, the non-machine learning method includes a seed point growth algorithm and a water-fat exchange region extraction algorithm; and determining the first field image corresponding to the echo image group according to the echo image group and the non-machine learning method includes:
[0021] Determine a third field image corresponding to the echo image group according to the seed point growth algorithm and the echo image group;
[0022] Determine a fourth field image corresponding to the echo image group according to the water-fat exchange region extraction algorithm and the echo image group;
[0023] The first field picture is determined according to the third field picture and the fourth field picture.
[0024] In one embodiment, determining the third field image corresponding to the echo image group according to the seed point growth algorithm and the echo image group includes:
[0025] Down-sampling each frame of the echo image in the echo image group to obtain an echo image group at each preset scale;
[0026] Determine a first initial field image corresponding to each echo image group at the preset scale;
[0027] Determine a first seed point corresponding to each first initial field image;
[0028] The third field image is determined according to the first seed points corresponding to the first initial field images.
[0029] In one embodiment, determining the third field image according to the first seed points corresponding to the first initial field images includes:
[0030] Performing region growing processing on the first seed points corresponding to the first initial field images to obtain second initial field images at the preset scales corresponding to the first seed points, and performing upsampling processing on the second initial field images to obtain third initial field images at the target scale;
[0031] Determine the second seed points corresponding to each of the third initial field images, and perform region growing processing on the second seed points corresponding to each of the third initial field images to obtain fourth initial field images corresponding to each of the second seed points;
[0032] fusing the fourth initial field images to obtain a fifth initial field image;
[0033] The third field picture is determined according to the fifth initial field picture.
[0034] In one embodiment, determining the fourth field image corresponding to the echo image group according to the water-fat exchange region extraction algorithm and the echo image group includes:
[0035] Processing the echo image group to obtain a sixth initial field image;
[0036] Determine the target water-lipid exchange region in the sixth initial field map;
[0037] The fourth field map is obtained according to the target water-lipid exchange area.
[0038] In one of the embodiments, the region growing process includes a three-dimensional region growing process;
[0039] The three-dimensional region growing process includes the following steps:
[0040] Performing two-dimensional region growing processing on the target region to obtain an intermediate field image of the echo image group corresponding to the target region; the intermediate field image includes each field layer corresponding to the echo image group, and the target region includes the first seed point, the second seed point or the target water-fat exchange region;
[0041] Determine the target field layer of the intermediate field map; the target field layer is the field layer with the largest number of seed points among the field layers;
[0042] The field map value of the target field layer is mapped to other field layers of the intermediate field layer to obtain the field map after the three-dimensional region growth processing; the other field layers are the field layers in each of the field layers except the target field layer.
[0043] In one embodiment, the method further comprises:
[0044] In the case where a second field image corresponding to the echo image group is determined based on the echo image group and the machine learning method, if a fifth field image that modifies the second field image is received, the machine learning method is updated based on the fifth field image and the echo image group.
[0045] The above-mentioned field map determination method applicable to magnetic resonance scanning first determines the echo image group corresponding to the part of interest, and then determines the first field map corresponding to the echo image group according to the echo image group and the non-machine learning method, and determines the second field map corresponding to the echo image group according to the echo image group and the machine learning method, thereby determining the target field map according to the first field map and the second field map. Among them, the echo image group includes multiple echo image frames. Since the first field map and the second field map are both determined based on multiple echo image frames, the accuracy of the first field map and the second field map is improved. Furthermore, since the first field map is determined by the non-machine learning method and the second field map is determined by the machine learning method, the field map determination method applicable to magnetic resonance scanning provided in this embodiment combines the non-machine learning method and the machine learning method, improves the accuracy of the target field map determined according to the first field map and the second field map, thereby reducing the reverse phenomenon of the target field map. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 This is an application environment diagram of the field map determination method applicable to magnetic resonance scanning in an embodiment of the present application;
[0047] Figure 2 It is a schematic flow chart of a method for determining a field map applicable to magnetic resonance scanning in an embodiment of the present application;
[0048] Figure 3 A schematic diagram of a process for obtaining a machine learning method in an embodiment of the present application;
[0049] Figure 4 A schematic diagram of a process for determining a target field map in an embodiment of the present application;
[0050] Figure 5 This is a schematic diagram of another process for determining a target field map in an embodiment of the present application;
[0051] Figure 6 A schematic diagram of a process for determining a first field image in an embodiment of the present application;
[0052] Figure 7 This is a schematic diagram of a process for determining a third field image in an embodiment of the present application;
[0053] Figure 8This is a schematic diagram of another process of determining a third field image in an embodiment of the present application;
[0054] Fig. 9 This is a schematic diagram of a process for determining a third field image in an embodiment of the present application;
[0055] Fig.10 This is a schematic diagram of a process for determining a fourth field image in an embodiment of the present application;
[0056] Fig.11 This is a schematic diagram of a process for determining a fourth field image in an embodiment of the present application;
[0057] Fig.12 A schematic diagram of a process of three-dimensional region growing processing in an embodiment of the present application;
[0058] Fig.13 This is a schematic diagram of a process for determining an echo image group in an embodiment of the present application;
[0059] Fig.14 This is a schematic diagram of the process of the fat quantification method in the embodiment of the present application;
[0060] Fig.15 A schematic diagram of a use process in an embodiment of the present application;
[0061] Fig.16 A schematic diagram of another use process in an embodiment of the present application;
[0062] Fig.17 It is a structural block diagram of a field map determination device suitable for magnetic resonance scanning in an embodiment of the present application;
[0063] Fig.18 Schematic diagram of the fat quantification device in the embodiment of the present application. DETAILED DESCRIPTION
[0064] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0065] Since the frequencies of water and fat in the subject under the action of the transmitting field and / or receiving field are different, this characteristic can be used to determine the corresponding field map according to the echo image of the subject. Each pixel in the echo image corresponds to a pixel in the field map, and each pixel in the field map corresponds to a field map value, and each field map value is used to indicate the frequency difference between water and fat in the echo image corresponding to the pixel. In other words, the field map can be further calculated to solve the water map and fat map.
[0066] Currently, in the process of obtaining fat quantification, the echo image of the part of interest is usually subjected to global iterative filtering or region growing and other methods to determine the field map corresponding to the echo image, and then fat quantification is performed based on the field map corresponding to the echo image.
[0067] However, the region growing process may cause the miscalculated field map to be transmitted in space, resulting in a large area of inversion phenomenon, that is, the part that should be water is calculated as fat, and the part that should be fat is calculated as water. Therefore, the accuracy of the field map currently determined is not high, which makes the inversion phenomenon easy to occur. Based on this, it is necessary to provide a field map determination method suitable for magnetic resonance scanning. The following will introduce the field map determination method suitable for magnetic resonance scanning.
[0068] Figure 1 FIG. 1 is an application environment diagram of a method for determining a field map for magnetic resonance scanning in an embodiment of the present application. In an embodiment of the present application, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 1 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store relevant data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a field map determination method suitable for magnetic resonance scanning is implemented.
[0069] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0070] This embodiment uses the method applied to a server as an example for illustration. It is understandable that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers, laptops, smart phones, and tablet computers. The server can be implemented as an independent server or a server cluster consisting of multiple servers.
[0071] Figure 2 FIG. 1 is a flow chart of a method for determining a field map suitable for magnetic resonance scanning in an embodiment of the present application. The method can be applied to Figure 1 In the computer device shown, in one embodiment, as Figure 2 As shown, the following steps are included:
[0072] S201, determining an echo image group corresponding to a site of interest; the echo image group includes a plurality of echo image frames.
[0073] In this embodiment, the computer device first determines the echo image group corresponding to the part of interest, wherein the part of interest may be the abdomen, legs, elbows, etc. of the subject, or the whole body area of the subject.
[0074] The echo image group includes a plurality of echo image frames. Optionally, the echo image group is a plurality of echo image frames acquired by the subject in the same scanning device in the same scanning time. The scanning device includes but is not limited to an MRI (Magnetic Resonance Imaging) device.
[0075] For example, the computer device can obtain echo image frame 1 at time 1, echo image frame 2 at time 2, and echo image frame 6 at time 6 of the MRI device in a relaxation process, and use echo image frame 1 to echo image frame 6 as an echo image group. The following example will be taken as an example in which the echo image group includes 6 echo image frames, and it can be understood that the echo image frames can also be set to other numbers.
[0076] Further optionally, the computer device can directly receive the echo image group sent by the scanning device; the computer device can also receive each scanning image sent by the scanning device and then process each scanning image to obtain the echo image group; the computer device can also store the echo image group corresponding to the part of interest in advance and directly call it when it is necessary to determine the target field map.
[0077] It can be understood that in order to control variables, in some embodiments, the resolution, size and number of slices of the plurality of echo image frames are generally kept the same.
[0078] S202: Determine a first field image corresponding to the echo image group according to the echo image group and a non-machine learning method.
[0079] In this embodiment, after the computer device determines the echo image group corresponding to the part of interest, it determines the first field image corresponding to the echo image group based on the echo image group and a non-machine learning method.
[0080] Among them, the non-machine learning method may include but is not limited to at least one of a nonlinear least squares method, a seed point growth algorithm, and a water-lipid exchange area extraction algorithm.
[0081] Multiple echo image frames generally correspond to a first field image. Exemplarily, the computer device can use the nonlinear least squares method to solve the 6 echo image frames in the echo image group to obtain a first field image. Assuming that the size of the 6 echo image frames includes 300 pixels, that is, each of the 6 echo image frames includes pixel 1 to pixel 300, then the first field image also includes pixel 1 to pixel 300, and each pixel has a corresponding field image value. Among them, pixel 1 refers to the pixel in the first row and first column, pixel 2 refers to the pixel in the first row and second column, and so on.
[0082] S203: Determine a second field image corresponding to the echo image group according to the echo image group and a machine learning method.
[0083] In this embodiment, the computer device also determines the second field image corresponding to the echo image group according to the echo image group and the machine learning method.
[0084] The machine learning method may be, but is not limited to, based on at least one of a supervised network and an unsupervised network.
[0085] Taking the example of a machine learning method including a supervised network, the supervised network may be a network trained based on a plurality of echo image group samples and corresponding field image labels. Exemplarily, the computer device may obtain the echo image group samples corresponding to the part of interest, and determine the field image labels corresponding to the echo image group samples, and then input the echo image group samples corresponding to the part of interest and the corresponding labels into the initial supervised network, and when the error between the field image predicted by the initial supervised network and the field image labels corresponding to the echo image group samples is less than a preset error, the training is stopped to obtain the above-mentioned supervised network.
[0086] A method for a computer device to determine a field image label corresponding to an echo image group sample is: after the computer device determines the initial field image label corresponding to the echo image group sample using a preset algorithm, the initial field image label is manually corrected by a computer device or other electronic device to determine the field image label corresponding to the echo image group sample. For example, echo image group sample 1 includes 6 echo image frame samples, and the computer device determines an initial field image label 1 corresponding to the echo image group sample 1 using a preset algorithm, and then the doctor adjusts the water-fat separation part on the initial field image label 1 to obtain the field image label 1, so that the computer device determines the field image label 1 corresponding to the echo image group sample 1.
[0087] Furthermore, the computer device can be built with the supervised network obtained after the above training. After the echo image group is determined, the echo image group is input into the supervised network, and the second field image corresponding to the echo image group can be determined. Similar to the principle of S201, multiple echo image frames generally correspond to one second field image. Similarly, if the six echo image frames all include pixel points 1 to pixel points 300, the second field image also includes pixel points 1 to pixel points 300, and each pixel point has a corresponding field image value.
[0088] It can be understood that the above example is to execute S202 first and then execute S203. S202 and S203 can also be executed at the same time, or restrictively execute S203 and then execute S202.
[0089] S204: Determine a target field image according to the first field image and the second field image.
[0090] In this embodiment, after S201 to S203, the computer device can determine the target field image according to the first field image and the second field image, wherein the field image value in the first field image and the field image value in the second field image are complex numbers.
[0091] Optionally, the computer device may average the field image values at corresponding positions in the first field image and the second field image to obtain the target field image. For example, assuming that both the first field image and the second field image include pixel points 1 to 300, the field image value corresponding to pixel point 1 in the first field image is A, and the field image value corresponding to pixel point 1 in the second field image is B, then the computer device will calculate the average value C of A and B as the field image value corresponding to pixel point 1 in the target field image, and so on. The computer device determines the field image values of 300 pixels in the target field image, that is, determines the target field image.
[0092] In some embodiments, the computer device may also calculate a weighted average of the field image values at corresponding positions in the first field image and the second field image, or the computer device may only consider the field image values of the preset areas in the first field image and the second field image to determine the target field image. Of course, the computer device may also determine the target field image using the first field image and the second field image in other ways, which is not limited in this embodiment.
[0093] The field map determination method suitable for magnetic resonance scanning provided in the present embodiment first determines the echo image group corresponding to the part of interest, and then determines the first field map corresponding to the echo image group according to the echo image group and the non-machine learning method, and determines the second field map corresponding to the echo image group according to the echo image group and the machine learning method, so as to determine the target field map according to the first field map and the second field map. Among them, the echo image group includes multiple echo image frames. Since the first field map and the second field map are both determined based on multiple echo image frames, the accuracy of the first field map and the second field map is improved. Furthermore, since the first field map is determined by the non-machine learning method and the second field map is determined by the machine learning method, the field map determination method suitable for magnetic resonance scanning provided in the present embodiment combines the non-machine learning method and the machine learning method, improves the accuracy of the target field map determined according to the first field map and the second field map, and can reduce the reverse phenomenon of the target field map.
[0094] Figure 3 This is a flow chart of a method for obtaining a machine learning method in an embodiment of the present application, referring to Figure 3 This embodiment relates to an optional implementation method of how to obtain a machine learning method. Based on the above embodiment, the machine learning method includes a supervised network; the above field map determination method suitable for magnetic resonance scanning includes the following steps:
[0095] S301, acquiring initial training data; the initial training data includes a plurality of echo image groups corresponding to parts of interest.
[0096] In this embodiment, when a supervised network needs to be trained, the computer device first needs to obtain initial training data. The initial training data includes a plurality of echo image groups corresponding to the parts of interest. Similarly, the parts of interest may include but are not limited to at least one of the head, legs, and elbows of the subject, or may be the whole body area of the subject.
[0097] Optionally, the initial training data may be the echo image groups corresponding to the same part of interest of different subjects, for example, the initial training data includes echo image group 1 of the abdomen of subject 1 and echo image 2 of the abdomen of subject 2, etc. The initial training data may also be the echo image groups corresponding to different subjects at different parts of interest, for example, the initial training data includes echo image group 1 of the abdomen of subject 1, echo image group 2 of the legs of subject 1, echo image 3 of the abdomen of subject 2, echo image 4 of the legs of subject 2, etc.
[0098] S302, determining a target data set based on the initial training data; the target data set includes a first data set and a second data set; the first data set includes a field map determined based on a non-machine learning method and the initial training data, and the second data set includes a field map determined based on an unsupervised network and the initial training data.
[0099] In this embodiment, after the computer device obtains the initial training data, it will determine the target data set according to the initial training data. Determining the target data set is also a process of determining the label corresponding to the initial training data.
[0100] The method of determining the field map using deep learning can be divided into two types of algorithms: supervised and unsupervised. For example, supervised algorithms can use U-net networks for training, and can obtain better training results when the amount of training data is small. However, supervised algorithms have the problem that the gold standard is difficult to obtain, while unsupervised algorithms can be trained without providing a gold standard, but unsupervised algorithms often require a large amount of training data, and the accuracy and reliability of training are not high. Among them, the gold standard is the label corresponding to the training data.
[0101] Therefore, in this embodiment, on the one hand, the computer device determines a first data set based on the non-machine learning method and the initial training data. The first data set includes gold standard data, that is, the computer device uses the field map calculated by the non-machine learning method as the gold standard.
[0102] On the other hand, the computer device determines a second data set based on the unsupervised network and the initial training data, and the second data set does not contain gold standard data. In other words, the computer device not only obtains part of the gold standard through non-machine learning methods, but also obtains part of the field map through the unsupervised network to make up for the inaccuracy of using non-machine learning methods or unsupervised networks alone, so as to enrich the diversity of the gold standard and thus improve the accuracy of the machine learning method.
[0103] The process of the computer device determining the first data set based on the non-machine learning method and the initial training data is similar to the process principle of S202. The non-machine learning method may include but is not limited to at least one of a nonlinear least squares method, a seed point growth algorithm, and a water-lipid exchange area extraction algorithm.
[0104] Exemplarily, assuming that the initial training data includes echo image group 1 of the abdomen of subject 1, echo image group 2 of the legs of subject 1, echo image 3 of the abdomen of subject 2, and echo image 4 of the elbow of subject 2, the computer device can use a seed point growth algorithm to determine field image A corresponding to echo image group 1, field image B corresponding to echo image group 2...field image D corresponding to echo image group 4, and determine the first data set according to field image A, field image B...field image D. Optionally, the computer device can directly use field images A to D as the first data set, and the computer device can also determine the first data set after calibrating and screening field images A to D.
[0105] Before the computer device determines the second data set based on the unsupervised network and the initial training data, it is first necessary to train the unsupervised network. Exemplarily, the computer device inputs the initial training data into the initial unsupervised network, and when the loss function of the initial unsupervised network meets the loss threshold, the computer device stops training to obtain the unsupervised network used in the end.
[0106] Furthermore, the computer device inputs the initial training data into the trained unsupervised network to obtain a field map predicted by the unsupervised network for the initial training data, so as to determine the second data set.
[0107] Continuing with the above example, the computer device can input the echo image group 1 of the abdomen of the subject 1, the echo image group 2 of the legs of the subject 1, the echo image 3 of the abdomen of the subject 2, and the echo image 4 of the elbow of the subject 2 into the unsupervised network, respectively, to obtain the field map A*, field map B*...field map D* output by the unsupervised network respectively. Then, the computer device determines the second data set based on the field maps A*~field map D*. Similarly, the computer device can directly use the field maps A*~field map D* as the second data set, or determine the second data set after calibrating and screening the field maps A*~field map D*.
[0108] Further optionally, when determining the target data set, the computer device may control the ratio of the first data set to the second data set, and the ratio may be set as required. For example, the computer device obtains the second data set based on 80% of the initial training data, and obtains the first data set based on the remaining 20% of the initial training data. In this way, the diversity and accuracy of the target data set are further improved, and the generation efficiency of the target data set is improved.
[0109] S303: Train the initial supervised network according to the initial training data and the target data set to obtain a supervised network.
[0110] In this embodiment, after S301 and S302, the computer device also obtains the label corresponding to the initial training data. Then, the computer device can train the initial supervised network according to the initial training data and the target data set to obtain a supervised network.
[0111] Optionally, the computer device inputs the initial training data and the corresponding target data set into the initial supervised network, for example, inputting the echo image group 1 of the abdomen of the subject 1 and the corresponding field map A*, and so on. When the error between the field map predicted by the initial supervised network and the field map in the target data set is less than a preset error, the training is stopped to obtain the above-mentioned supervised network.
[0112] In this way, after the computer device obtains the supervised network, it can be put into subsequent use to input the echo image group corresponding to the part of interest into the supervised network to obtain the second field image output by the supervised network.
[0113] In this embodiment, when the machine learning method includes a supervised network, initial training data including a plurality of echo image groups corresponding to the parts of interest are obtained, and a target data set is determined based on the initial training data, so as to train the initial supervised network based on the initial training data and the target data set to obtain a supervised network. Since the target data set includes a first data set and a second data set, and the first data set includes a field map determined based on a non-machine learning method and the initial training data, and the second data set includes a field map determined based on an unsupervised network and the initial training data, the unsupervised network can be used to supplement the data set determined by the non-machine learning method, thereby improving the generation speed, accuracy and diversity of the target data set.
[0114] Figure 4 This is a schematic diagram of a process for determining a target field map in an embodiment of the present application, referring to Figure 4 This embodiment relates to an optional implementation method of how to determine the target field image according to the first field image and the second field image. Based on the above embodiment, the above S204, determining the target field image according to the first field image and the second field image, includes the following steps:
[0115] S401, determining the difference between each first field image value and the corresponding second field image value; the first field image value is the field image value of the first pixel point in the first field image, and the second field image value is the field image value of the second pixel point in the second field image corresponding to the first pixel point.
[0116] In this embodiment, the above step S204 is continued. After the computer device determines the first field image and the second field image, the first field image value corresponding to each first pixel point in the first field image and the second field image value corresponding to each second pixel point in the second field image are determined.
[0117] Furthermore, the computer device determines the difference between each first field image value and the corresponding second field image value. That is, the computer device traverses each first pixel point in the first field image, determines the first field image value of each first pixel point, and determines the second field image value of the second pixel point in the second field image corresponding to the first pixel point, thereby determining the difference between the first field image value and the second field image value.
[0118] The second pixel corresponding to the first pixel refers to the pixel in the second field image at the same position as the first pixel in the first field image. For example, the first pixel in the first row and first column in the first field image corresponds to the second pixel in the first row and first column in the second field image.
[0119] The difference value between the first field image value and the second field image value may be the difference between the absolute value of the first field image value and the absolute value of the second field image value, or may be the variance between the first field image value and the second field image value. In some embodiments, since the first field image value and the second field image value are two complex numbers, the difference value may also be determined based on the angle corresponding to the first field image value and the angle corresponding to the second field image value. This embodiment does not limit the specific form of the difference value, as long as it can measure the difference between the first field image value and the second field image value.
[0120] For example, assuming that the first field image includes 300 first pixel points 1 to 300, and the second field image includes 300 second pixel points 1 to 300, the first pixel point 1 corresponds to the second pixel point 1, the first pixel point 2 corresponds to the second pixel point 2, and so on. Then the computer device can calculate 300 difference values, that is, the difference value 1 between the first field image value 1 of the first pixel point 1 and the second field image value 1 of the second pixel point 1, the difference value 2 between the first field image value 2 of the first pixel point 2 and the second field image value 2 of the second pixel point 2...the difference value 300 between the first field image value 300 of the first pixel point 300 and the second field image value 300 of the second pixel point 300.
[0121] S402: Determine a target field picture according to a difference value between each first field picture value and a corresponding second field picture value.
[0122] In this embodiment, further, the computer device may determine the target field image according to the difference value between each first field image value and the corresponding second field image value. For example, the computer device compares the difference value with the preset difference value, and if the difference value is less than the preset difference threshold, the average value of the first field image value and the second field image value is calculated, and if the difference value is not less than the preset difference threshold, the weighted average value of the first field image value and the second field image value is calculated, and so on, so as to determine the entire target field image.
[0123] When determining the target field map according to the first field map and the second field map, the embodiment first determines the difference between each first field map value and the corresponding second field map value, and determines the target field map according to the difference between each first field map value and the corresponding second field map value. Since the first field map value is the field map value of the first pixel point in the first field map, and the second field map value is the field map value of the second pixel point in the second field map corresponding to the first pixel point, the determined target field map takes into account the difference between each field map value in the first field map and the second field map, thereby improving the accuracy of the target field map.
[0124] Figure 5 This is another schematic diagram of a process for determining a target field map in an embodiment of the present application, referring to Figure 5 This embodiment relates to an optional implementation method of how to determine the target field image. Based on the above embodiment, the above S402, according to the difference between each first field image value and the corresponding second field image value, determines the target field image, including the following steps:
[0125] S501, taking the field image value of the first target pixel point as the field image value of the pixel point corresponding to the first target pixel point in the target field image; the first target pixel point is the first pixel point whose difference value is less than or equal to a preset difference threshold value.
[0126] S502, taking the field image value of the second target pixel point as the field image value of the pixel point corresponding to the second target pixel point in the target field image; the second target pixel point is a first pixel point whose difference value is not less than a preset difference threshold.
[0127] In this embodiment, after the computer device determines the difference between each first field image value and the corresponding second field image value, it will compare each difference value with a preset difference threshold. If the difference value is less than or equal to the preset difference threshold, it means that based on the pixel point corresponding to the difference value, the difference between the first field image determined based on the non-machine learning method and the second field image determined based on the machine learning method is not large, and the first field image determined by the non-machine learning method shall prevail; if the difference value is not less than the preset difference threshold, it means that based on the pixel point corresponding to the difference value, the difference between the first field image determined based on the non-machine learning method and the second field image determined based on the machine learning method is large, and the second field image determined by the machine learning method shall prevail. Among them, the preset difference threshold can be set according to demand and is a number greater than or equal to 0.
[0128] That is, the computer device uses the field map value of the first target pixel point as the field map value of the pixel point corresponding to the first target pixel point in the target field map, and uses the field map value of the second target pixel point as the field map value of the pixel point corresponding to the second target pixel point in the target field map. The first target pixel point is the first pixel point whose difference value is less than or equal to the preset difference threshold, and the second target pixel point is the first pixel point whose difference value is not less than the preset difference threshold.
[0129] Continuing with the example of S401 above, the computer device determines difference values 1 to 300. If difference value 1 is less than the preset difference threshold, the field image value of the pixel located in the first row and first column of the target field image, that is, the pixel corresponding to the first target pixel, is set to the first field image value 1. If difference value 2 is not less than the preset difference threshold, the field image value of the pixel located in the first row and second column of the target field image, that is, the pixel corresponding to the second target pixel, is set to the second field image value 2. By analogy, the computer device also determines the field image values corresponding to 300 pixels in the target field image.
[0130] In this embodiment, the field map value of the first target pixel point is used as the field map value of the pixel point corresponding to the first target pixel point in the target field map, and the field map value of the second target pixel point is used as the field map value of the pixel point corresponding to the second target pixel point in the target field map. Since the first target pixel point is the first pixel point whose difference value is less than or equal to the preset difference threshold, and the second target pixel point is the first pixel point whose difference value is not less than the preset difference threshold, this embodiment confirms pixel by pixel, and when the difference between the first field map and the second field map is too large, the second field map shall prevail, and when the difference between the first field map and the second field map is not large, the first field map shall prevail, thereby improving the accuracy of the target field map.
[0131] Figure 6 This is a schematic diagram of a process for determining a first field image in an embodiment of the present application, referring to Figure 6 This embodiment relates to an optional implementation method of how to determine the first field image corresponding to the echo image group. Based on the above embodiment, the non-machine learning method includes a seed point growth algorithm and a water-fat exchange area extraction algorithm; the above S202, according to the echo image group and the non-machine learning method, determines the first field image corresponding to the echo image group, including the following steps:
[0132] S601, determining a third field image corresponding to the echo image group according to a seed point growth algorithm and the echo image group.
[0133] S602: Determine a fourth field image corresponding to the echo image group according to the water-fat exchange region extraction algorithm and the echo image group.
[0134] S603: Determine a first field picture according to the third field picture and the fourth field picture.
[0135] In this embodiment, the computer device first determines the third field image corresponding to the echo image group based on the seed point growth algorithm and the echo image group, and determines the fourth field image corresponding to the echo image group based on the water-fat exchange area extraction algorithm and the echo image group.
[0136] Among them, the seed point growth algorithm refers to a method based on multi-resolution local growth and self-checking field map estimation, which downsamples the echo image group at different scales, selects seed points based on the downsampling results, and performs regional growth processing to calculate the field map. The water-fat exchange region extraction algorithm refers to a method of determining the water quality exchange region corresponding to the echo image group, determining the phase value in the determined water quality exchange region, and thus obtaining the phase solution of the whole image to determine the field map.
[0137] The seed point growth algorithm focuses more on the performance of a single pixel, and the water-fat separation often occurs in local areas. The water-fat exchange area extraction algorithm focuses more on the situation in local areas, and the water-fat separation will appear in larger areas.
[0138] Therefore, in this embodiment, a seed point growth algorithm and a water-fat exchange region extraction algorithm are combined to improve the accuracy of the first field map.
[0139] For example, the echo image group 1 includes 6 echo image frames, and the computer device determines a third field according to the seed point growth algorithm and the echo image group 1. Figure 1 , and according to the water-fat exchange area extraction algorithm and echo image group 1, determine a fourth field Figure 1 It can be understood that the six echo image frames all include pixel points 1 to 300, so the third field Figure 1 and the fourth Figure 1 Pixel 1 to pixel 300 are also included.
[0140] Further, the computer device can determine the first field image according to the third field image and the fourth field image. Similar to the principle of S204, optionally, the computer device can average the field image values at corresponding positions in the third field image and the fourth field image to obtain the first field image, or can calculate a weighted average of the field image values at corresponding positions in the third field image and the fourth field image, or the computer device only considers the field image values of the preset area in the third field image and the fourth field image to determine the first field image.
[0141] In other words, the computer device first uses a multi-resolution downsampling technique based on the seed point method of the multi-resolution initial segmentation image to find seed points at different resolutions, and performs region growth at multiple resolutions to calculate the field map value of each point to determine the third field map. Furthermore, the computer device determines the water area and lipid domain based on the water-lipid exchange area method, and determines the field map value in each area to determine the fourth field map. In this way, the computer device can perform a weighted average of the third field map and the fourth field map to obtain the first field map under the non-machine learning method.
[0142] This embodiment determines the third field image corresponding to the echo image group according to the seed point growth algorithm and the echo image group, and determines the fourth field image corresponding to the echo image group according to the water-fat exchange region extraction algorithm and the echo image group, and then determines the first field image according to the third field image and the fourth field image. Since the seed point growth algorithm and the water-fat exchange region extraction algorithm are combined, the accuracy of the first field image is improved.
[0143] Figure 7 This is a schematic diagram of a process for determining a third field image in an embodiment of the present application, referring to Figure 7 This embodiment relates to an optional implementation method of how to determine the third field image corresponding to the echo image group. Based on the above embodiment, the above S601, according to the seed point growth algorithm and the echo image group, determines the third field image corresponding to the echo image group, including the following steps:
[0144] S701 , performing down-sampling processing on each frame of echo image in the echo image group to obtain echo image groups at each preset scale.
[0145] In this embodiment, the computer device first performs down-sampling processing on each frame of echo image in the echo image group to obtain echo image groups at each preset scale.
[0146] Exemplarily, assuming that echo image group 1 includes echo image frames 1 to 6, and echo image frames 1 to 6 are all 1920px×1080px, then after the computer device downsamples each echo image frame in the echo image group, echo image group 2 and echo image group 3 can be obtained, wherein echo image group 2 includes echo image frames 7 to 12 of 960px×1080px, and echo image group 3 includes echo image frames 13 to 18 of 1920px×540px.
[0147] The above is only an example of downsampling processing. This embodiment does not limit the number and setting of the sizes of the downsampling processing, as long as the echo image groups at various preset scales can be obtained.
[0148] S702: Determine a first initial field image corresponding to the echo image group at each preset scale.
[0149] In this embodiment, the computer device determines the first initial field map corresponding to the echo image group at each preset scale. Optionally, the computer device may solve the echo image group at each preset scale using, but not limited to, a nonlinear least squares method to determine the corresponding first initial field map. Continuing with the example of S701 above, the computing device determines the first initial field map corresponding to the echo image group 1. Figure 1 , the first initial field corresponding to echo image group 2 Figure 2 And the first initial field corresponding to echo image group 3 Figure 3 .
[0150] S703: Determine a first seed point corresponding to each first initial field image.
[0151] In this embodiment, since each pixel point of each first initial field image has a field image value, the computer device can select a first seed point in each first initial field image according to the field image value of each pixel point of each first initial field image to determine the first seed point corresponding to each first initial field image. Optionally, the computer device uses the pixel point whose water component signal and fat component signal are close to or equal to 1:1 in the field image value of each pixel point of each first initial field image as the first seed point. For example, the computer device determines the first initial field image. Figure 1 The corresponding 10 first seed points and the first initial field Figure 2 The corresponding 15 first seed points and the first initial field Figure 3 The corresponding 12 first seed points.
[0152] S704: Determine a third field image according to the first seed points corresponding to the first initial field images.
[0153] In this embodiment, the computer device continues to determine the third field map according to the first seed points corresponding to each first initial field map. Optionally, the computer device may perform region growing processing on the first seed points corresponding to each first initial field map to obtain each second initial field map corresponding to each first initial field map. Further, the computer device performs fusion processing on each first initial field map to obtain the third field map. The region growing processing may be a two-dimensional region growing processing or a three-dimensional region growing processing.
[0154] For example, the computer device responds to the first initial field Figure 1 The corresponding 10 first seed points are processed by region growth to obtain the second initial field Figure 1 , for the first initial field Figure 2 The corresponding 15 first seed points are processed by region growth to obtain the second initial field Figure 2 , and for the first initial field Figure 3 The corresponding 12 first seed points are processed by region growth to obtain the second initial field Figure 3Then, the computer device responds to the first initial field Figure 1 ~First initial field Figure 3 The third field image is obtained by fusion.
[0155] In some embodiments, after obtaining each first initial field image corresponding to each second initial field image, the computer device may further perform a plurality of seed point selection and region growing processes to obtain a third field image.
[0156] In this embodiment, each frame of the echo image in the echo image group is downsampled to obtain the echo image group at each preset scale, and the first initial field image corresponding to the echo image group at each preset scale is determined, and then the first seed point corresponding to each first initial field image is determined, so as to determine the third field image according to the first seed point corresponding to each first initial field image. Since the corresponding first seed point is selected based on the echo image group at each preset scale, the accuracy of the third field image determined based on the first seed point corresponding to each first initial field image is improved.
[0157] Figure 8 This is another schematic diagram of a process for determining the third field image in an embodiment of the present application, referring to Figure 8 This embodiment relates to an optional implementation method of how to determine the third field image. Based on the above embodiment, the above S704, according to the first seed points corresponding to each first initial field image, determines the third field image, including the following steps:
[0158] S801, performing region growing processing on the first seed points corresponding to the first initial field images to obtain second initial field images at preset scales corresponding to the first seed points, and performing upsampling processing on the second initial field images to obtain third initial field images at a target scale.
[0159] Fig. 9 This is a schematic diagram of a process for determining the third field image in an embodiment of the present application. Please refer to Fig. 9 In this embodiment, the computer device processes the echo image group (such as Fig. 9 (A)) is downsampled to obtain four echo image groups at preset scales (such as Fig. 9 (B1) to (B4)), and determine that the echo image groups under the four preset scales correspond to the four first initial field images respectively.
[0160] Then, the computer device determines the first seed points corresponding to the four first initial field images (such as Fig. 9 (C1) to (C4)), and perform region growth processing on the first seed points corresponding to each first initial field map to obtain the second initial field map at each preset scale corresponding to each first seed point (as shown in Fig. 9 (as shown in (D1) to (D4)).
[0161] Furthermore, the computer device performs upsampling processing on each second initial field image to obtain a third initial field image at the target scale (such as Fig. 9 (E1) to (E4)). The target scale is set according to the requirements. In this embodiment, the target scale is taken as an example of the size of each echo image frame in the echo image group.
[0162] S802, determining the second seed points corresponding to the third initial field images, and performing region growing processing on the second seed points corresponding to the third initial field images to obtain fourth initial field images corresponding to the second seed points.
[0163] In this embodiment, the principle of determining the second seed point is the same as the principle of determining the first seed point, which will not be repeated here. Fig. 9 After (E1) to (E4), the computer device will continue to determine the second seed point corresponding to each third initial field image. Fig. 9 (E1) corresponds to 20 second seed points, Fig. 9 (E1) corresponds to 18 second seed points, and so on. It should be noted that the numbers of the first seed points and the second seed points mentioned above are only for example.
[0164] Further, the computer device continues to perform region growing processing on the second seed points corresponding to each third initial field image, for example Fig. 9 The 20 second seed points corresponding to (E1) are processed by region growth to obtain Fig. 9 (F1) as shown in the fourth initial field diagram, and so on, the computer device obtains Fig. 9 The fourth initial field pictures of (F1) to (F4).
[0165] S803: Fusing the fourth initial field images to obtain a fifth initial field image.
[0166] In this embodiment, the computer device fuses the fourth initial field images to obtain the following Fig. 9 (G) shows the fifth initial field diagram. Exemplarily, the computer device may Fig. 9 The field image values of the pixels at the same position in (F1) to (F4) are averaged, and so on. Fig. 9 (F1)~(F4) all pixels to determine Fig. 9 (G).
[0167] S804: Determine a third field picture according to the fifth initial field picture.
[0168] In this embodiment, the computer device can determine the third field image according to the fifth initial field image. Optionally, the computer device can directly use the fifth initial field image as the third field image. The computer device can also continue to determine the third seed point corresponding to the fifth initial field image, and perform region growing processing on the third seed point corresponding to each fifth initial field image to obtain the third field image.
[0169] The third scene Fig. 9 (H) shown.
[0170] In some embodiments, the computer device may further process the third field image to obtain a water map and a fat map corresponding to the third field image, such as Fig. 9 (I) shown.
[0171] In this embodiment, the first seed points corresponding to the first initial field images are subjected to region growing processing to obtain the second initial field images at the preset scales corresponding to the first seed points, and the second initial field images are subjected to upsampling processing to obtain the third initial field images at the target scale, and then the second seed points corresponding to the third initial field images are determined, and the second seed points corresponding to the third initial field images are subjected to region growing processing to obtain the fourth initial field images corresponding to the second seed points, and the fourth initial field images are further fused to obtain the fifth initial field image, so as to determine the third field image according to the fifth initial field image. In this way, after repeatedly determining the seed points and performing the region growing processing, the accuracy of the third field image can be improved.
[0172] Fig.10 This is a schematic diagram of a process for determining a fourth field image in an embodiment of the present application, referring to Fig.10 This embodiment relates to an optional implementation method of how to determine the fourth field image corresponding to the echo image group. Based on the above embodiment, the above S602, according to the water-fat exchange area extraction algorithm and the echo image group, determines the fourth field image corresponding to the echo image group, including the following steps:
[0173] S1001, processing the echo image group to obtain a sixth initial field image.
[0174] Fig.11 This is a schematic diagram of a process for determining the fourth field image in an embodiment of the present application. Please refer to Fig.11 In this embodiment, when the computer device uses the water-fat exchange region extraction algorithm to determine the fourth field image corresponding to the echo image group, the echo image group (such as Fig.11 (A)) is processed to obtain the sixth initial field image (such as Fig.11 (B)).
[0175] Optionally, the sixth initial field image includes a water map and a fat map. The computer device can use the least square method to solve the echo image group to determine the local optimal water map (such as Fig.11 (B) Left) and locally suboptimal lipid maps (e.g. Fig.11 (B) as shown on the right) to determine the sixth initial field map.
[0176] S1002, determining a target water-fat exchange region in a sixth initial field map.
[0177] In this embodiment, since the sixth initial field map is determined, the computer device can also determine the area where water and fat are close to each other as the target water quality exchange area, thereby determining the target water-fat exchange area in the sixth initial field map (such as Fig.11 (C) outlined lines). Optionally, the computer device can Fig.11 (B) The water map on the left is used as a reference. Fig.11 (B) The lipid map on the right takes each pixel whose pixel value change during movement is less than the preset change as the target water quality exchange area.
[0178] S1003, obtaining a fourth field map according to the target water-lipid exchange area.
[0179] In this embodiment, further, the computer device can obtain a fourth field map according to the target water-lipid exchange area.
[0180] Optionally, the computer device may directly perform region growing processing on the target water-lipid exchange region to obtain the fourth field map.
[0181] The computer equipment can also perform sealing treatment on the target water-lipid exchange area to obtain Fig.11 (D) The seventh initial field map shown. Based on the seventh initial field map, the computer device can determine which areas are water and which areas are fat. Further, the computer device performs statistical processing based on the seventh initial field map to determine the phase value, and then performs water and fat area field mapping, thereby determining the eighth initial field map. Further, the computer device performs region growth processing based on the eighth initial field map to determine Fig.11 (E) In some embodiments, the computer device may further process the fourth field image to obtain a water map and a fat map corresponding to the fourth field image, such as Fig.11 As shown in (F) and (G).
[0182] This embodiment processes the echo image group to obtain the sixth initial field image, and determines the target water-fat exchange area in the sixth initial field image, so as to obtain the fourth field image based on the target water-fat exchange area. In this way, the fourth field image corresponding to the echo image group can be determined by using the target water quality exchange area.
[0183] Currently, inter-layer information is usually ignored in two-dimensional region growing processing, so in one embodiment, optionally, the region growing processing includes three-dimensional region growing processing. Fig.12 This is a flow chart of a three-dimensional region growing process in an embodiment of the present application. Based on the above embodiment, the three-dimensional region growing process includes the following steps:
[0184] S1201, perform two-dimensional region growing processing on the target area to obtain an intermediate field map of the echo image group corresponding to the target area; the intermediate field map includes each field layer corresponding to the echo image group, and the target area includes the first seed point, the second seed point or the target water-fat exchange area.
[0185] Assume that each echo image frame of the echo image group has 100 image layers, and take the target area as the first seed point as an example. The echo image frames in the echo image group at each preset scale obtained by the computer device also have 100 image layers, and each determined first initial field map also has 100 field layers. Furthermore, the first seed points corresponding to each first initial field map also have 100 field layers. In this way, the target area is subjected to two-dimensional region growth processing, and the intermediate field map of the echo image group corresponding to the target area also has 100 field layers. That is, the intermediate field map includes each field layer corresponding to the echo image group. The second seed point and the target water-fat exchange area are similar, and will not be repeated here.
[0186] S1202, determining a target field layer of the intermediate field map; the target field layer is the field layer with the largest number of seed points among all the field layers.
[0187] Since the information between the field layers of the intermediate field map is independent after the two-dimensional region growing process, the two-dimensional region growing process often results in large differences between the layers, causing water-fat separation in some field layers. Therefore, in this embodiment, the computer device also determines the target field layer of the intermediate field map.
[0188] Optionally, the computer device determines the number of seed points in each field layer of the intermediate field map, and then uses the field layer with the largest number of seed points in each field layer as the target field layer.
[0189] Continuing with the above example, the intermediate field map includes field layers 1 to 100. The computer device traverses each field layer and counts the number of seed points in each field layer. Assuming that field layer 80 has the largest number of seed points, the computer device uses field layer 80 as the target field layer of the intermediate field map.
[0190] S1203, mapping the field map value of the target field layer to other field layers of the intermediate field layer to obtain the field map after the three-dimensional region growing process; the other field layers are the field layers except the target field layer in each field layer.
[0191] Continuing with the above example of S1202, the computer device maps the field map value of the field layer 80 to other field layers of the intermediate field map, namely, field layers 1 to 79 and field layers 81 to 100, to obtain a field map after three-dimensional region growing processing.
[0192] It should be noted that when the computer device maps the field image value of the target field layer to other field layers of the intermediate field image, the field image value at the corresponding position is mapped. For example, the field image value of the first row and first column pixel point in field layer 80 is mapped to the first row and first column pixel point in field layer 1, and so on.
[0193] In this embodiment, a two-dimensional region growth process is performed on the target area to obtain an intermediate field map of the echo image group corresponding to the target area, and a target field layer of the intermediate field map is determined, and then the field map value of the target field layer is mapped to other field layers of the intermediate field map to obtain a field map after the three-dimensional region growth process. Since the intermediate field map includes each field layer corresponding to the echo image group, and the target area includes the first seed point, the second seed point or the target water-fat exchange area, the three-dimensional region growth provided in this embodiment can be applied to the seed point growth algorithm and the water-fat exchange area extraction algorithm. Furthermore, since the target field layer is the field layer with the largest number of seed points in each field layer, and the other field layers are the field layers other than the target field layer in each field layer, the three-dimensional region growth can make full use of the inter-layer information, reduce the phenomenon that the field layers are relatively different after the current two-dimensional region growth process, and the water-fat separation of some field layers is reversed, so as to improve the accuracy of the first field map.
[0194] In one embodiment, optionally, the above-mentioned field map determination method applicable to magnetic resonance scanning further includes the following steps:
[0195] When a second field image corresponding to the echo image group is determined based on the echo image group and the machine learning method, if a fifth field image obtained by modifying the second field image is received, the machine learning method is updated based on the fifth field image and the echo image group.
[0196] In this embodiment, in the process of the computer device using the machine learning method to determine the second field image corresponding to the echo image group, that is, when the second field image corresponding to the echo image group is determined according to the echo image group and the machine learning method, if the computer device receives the fifth field image after modifying the second field image, for example, the user initiates an update operation on the second field image, the computer device will update the machine learning method according to the fifth field image and the echo image group.
[0197] For example, after the computer device inputs the echo image group into the supervised network to obtain the second field image, if the user initiates an update operation on the second field image, the computer device will retrain the supervised network based on the fifth field image and the echo image group to update the supervised network.
[0198] Since this embodiment can update the machine learning method according to the fifth field image and the echo image group when receiving the fifth field image that modifies the second field image, the computer device can update the machine learning method during use to improve the accuracy of the machine learning method, thereby improving the accuracy of the second field image.
[0199] Fig.13 This is a schematic diagram of a process for determining an echo image group in an embodiment of the present application, referring to Fig.13 This embodiment relates to an optional implementation method of how to determine an echo image group. Based on the above embodiment, the above S201, determining the echo image group corresponding to the part of interest, includes the following steps:
[0200] S1301, obtaining an echo image group of the subject.
[0201] S1302: Determine an echo image group corresponding to at least one part of interest based on a deep learning segmentation network and the echo image group of the subject.
[0202] In this embodiment, the computer device can obtain the echo image group of the subject, and then use the deep learning segmentation network to segment the echo image group of the subject to determine the echo image group corresponding to at least one part of interest. For example, the computer device obtains the echo image group A of the whole body of the subject 1, and inputs the echo image group A of the whole body of the subject 1 into the trained deep learning segmentation network to determine the echo image group 1 of the abdomen of the subject 1, the echo image group 2 of the elbow of the subject 1, and the echo image group 3 of the leg of the subject 1.
[0203] Among them, the deep learning segmentation network is a network determined based on the echo image group of multiple subjects and the segmentation mask, which can be a convolutional neural network (CNN), a recurrent neural network (RNN), or other deep learning networks.
[0204] Furthermore, the computer device can determine the corresponding target field map based on the echo image groups corresponding to different parts of interest. In this way, this embodiment can achieve fat quantification in various parts with high flexibility.
[0205] Based on the same inventive concept, the present application also provides a fat quantification method. Fig.14 This is a schematic diagram of the process of the fat quantification method in the embodiment of the present application. The method can be applied to Figure 1 In the computer device shown, in one embodiment, as Fig.14 As shown, the following steps are included:
[0206] S1401, using the field map determination method applicable to magnetic resonance scanning, determine a target field map of an echo image group corresponding to a part of interest.
[0207] In this embodiment, the computer device uses the above-mentioned field map determination method applicable to magnetic resonance scanning to determine the target field map of the echo image group corresponding to the part of interest. For example, the target field map corresponding to echo image frames 1 to 6 in echo image group 1 is determined. Figure 1 .
[0208] S1402, determining the fat content of the area of interest according to the target field image corresponding to the echo image group.
[0209] In this embodiment, after determining the target field map of the echo image group corresponding to the part of interest, the computer device can determine the fat content of the part of interest according to the target field map.
[0210] Optionally, the computer device may determine the fat content FF of the area of interest using the following equations (1) and (2).
[0211]
[0212]
[0213] In formula (1), n is a natural number, 1≤n≤N, N is the number of echo image frames in the echo image group, which is an integer greater than 1, for example, 6; α p is the content ratio of the pth fat peak, p is 1-P; P is the number of fat peaks; f F,p is the resonance frequency of the pth fat peak; f B is the target field map; TE n is the echo time of the nth echo signal.
[0214] That is to say, after the target field map is determined by the field map determination method applicable to magnetic resonance scanning, the target field map is substituted into equation (1) to solve the water signal component ρ in the target field map. w and lipid signal component ρ f Furthermore, using the water signal component ρ in the target field image w and lipid signal component ρ f, substituting into formula (2), the fat content FF of the area of interest can be determined.
[0215] This embodiment uses the above-mentioned field map determination method applicable to magnetic resonance scanning to determine the target field map of the echo image group corresponding to the part of interest, and determines the fat content of the part of interest based on the target field map corresponding to the echo image group. Since the accuracy of the target field map is improved by using the above-mentioned field map determination method applicable to magnetic resonance scanning, the accuracy of the fat content of the part of interest is also improved.
[0216] In order to more clearly explain the field map determination method and fat quantification method applicable to magnetic resonance scanning provided by the present application, Fig.15 and Fig.16 Provide explanation. Fig.15 This is a schematic diagram of a use process in an embodiment of the present application. Fig.16 This is a schematic diagram of another usage process in an embodiment of the present application.
[0217] Please combine Fig.15 and Fig.16 The computer device can execute the field map determination method and fat quantification method suitable for magnetic resonance scanning according to the following process.
[0218] First, an echo image group of the subject is obtained, and the echo image group of the subject is input into a deep learning segmentation network to obtain an echo image group corresponding to at least one part of interest.
[0219] Then, the computer device determines the first field image based on the echo image group corresponding to the part of interest by using a non-machine learning method on the one hand, and determines the second field image by using a machine learning method on the other hand.
[0220] Among them, non-machine learning methods include seed point generation algorithm and water-fat exchange area extraction algorithm. In the seed point generation algorithm, the computer device will perform downsampling processing, seed point selection and three-dimensional region growth processing to obtain the third field map. In the water-fat exchange area extraction algorithm, the computer device will determine the target water-fat exchange area and perform water and fat area field diagram to determine the fourth field map. Machine learning methods include supervised networks.
[0221] Furthermore, the computer device determines the above-mentioned first image according to the third field image and the fourth field image.
[0222] Finally, the computer device determines the target field map based on the first field map and the second field map, and determines the fat content of the region of interest based on the target field map. It is understandable that if the computer device obtains echo image groups of multiple parts of interest, what is determined is the target field map and fat content corresponding to each part of interest.
[0223] In addition, the computer device also provides a correction function. After obtaining the second field map, the user can manually repair the area where the water and fat are reversed to obtain the fifth field map. Then, the computer device uses the fifth field map to update the machine learning method online, such as updating the supervised model. In this way, not only fat quantification for different parts is achieved, but also the robustness and accuracy of the field map determination process are improved. Please refer to the introduction of each embodiment for the principles of the above steps, which will not be repeated here.
[0224] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0225] Based on the same inventive concept, the embodiment of the present application also provides a field map determination device suitable for magnetic resonance scanning for implementing the field map determination method suitable for magnetic resonance scanning involved above, and a fat quantification device for implementing the fat quantification method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more field map determination devices and fat quantification devices provided below can refer to the above limitations on the field map determination method suitable for magnetic resonance scanning and the fat quantification method, and will not be repeated here.
[0226] Fig.17 FIG. 1 is a structural block diagram of a field pattern determination device suitable for magnetic resonance scanning in an embodiment of the present application. Fig.17 As shown, in an embodiment of the present application, a field map determination device 1700 suitable for magnetic resonance scanning is provided, comprising: a first determination module 1701, a second determination module 1702, a third determination module 1703 and a fourth determination module 1704, wherein:
[0227] The first determination module 1701 is used to determine an echo image group corresponding to the site of interest; the echo image group includes a plurality of echo image frames.
[0228] The second determination module 1702 is used to determine the first field image corresponding to the echo image group according to the echo image group and a non-machine learning method.
[0229] The third determination module 1703 is used to determine the second field image corresponding to the echo image group according to the echo image group and the machine learning method.
[0230] The fourth determining module 1704 is configured to determine a target field image according to the first field image and the second field image.
[0231] The field map determination device suitable for magnetic resonance scanning provided in the present embodiment first determines the echo image group corresponding to the part of interest, and then determines the first field map corresponding to the echo image group according to the echo image group and the non-machine learning method, and determines the second field map corresponding to the echo image group according to the echo image group and the machine learning method, so as to determine the target field map according to the first field map and the second field map. Among them, the echo image group includes multiple echo image frames. Since the first field map and the second field map are both determined based on multiple echo image frames, the accuracy of the first field map and the second field map is improved. Furthermore, since the first field map is determined by the non-machine learning method and the second field map is determined by the machine learning method, the field map determination device suitable for magnetic resonance scanning provided in the present embodiment combines the non-machine learning method and the machine learning method, improves the accuracy of the target field map determined according to the first field map and the second field map, thereby reducing the reverse phenomenon of the target field map.
[0232] Optionally, the machine learning method includes a supervised network; the field map determination device 1700 suitable for magnetic resonance scanning also includes:
[0233] The first acquisition module is used to acquire initial training data; the initial training data includes a plurality of echo image groups corresponding to the parts of interest.
[0234] The fifth determination module is used to determine the target data set based on the initial training data; the target data set includes a first data set and a second data set; the first data set includes a field map determined based on a non-machine learning method and the initial training data, and the second data set includes a field map determined based on an unsupervised network and the initial training data.
[0235] The training module is used to train the initial supervised network according to the initial training data and the target data set to obtain a supervised network.
[0236] Optionally, the fourth determining module 1704 includes:
[0237] The first determining unit is used to determine the difference value between each first field image value and the corresponding second field image value; the first field image value is the field image value of the first pixel point in the first field image, and the second field image value is the field image value of the second pixel point in the second field image corresponding to the first pixel point.
[0238] The second determining unit is used to determine the target field picture according to the difference value between each first field picture value and the corresponding second field picture value.
[0239] Optionally, the second determining unit includes:
[0240] The first determining subunit is used to use the field image value of the first target pixel point as the field image value of the pixel point corresponding to the first target pixel point in the target field image; the first target pixel point is the first pixel point whose difference value is less than or equal to a preset difference threshold.
[0241] The second determining subunit is used to use the field image value of the second target pixel point as the field image value of the pixel point corresponding to the second target pixel point in the target field image; the second target pixel point is the first pixel point whose difference value is not less than a preset difference threshold.
[0242] Optionally, the non-machine learning method includes a seed point growth algorithm and a water-fat exchange region extraction algorithm; the second determination module 1702 includes:
[0243] The third determining unit is used to determine a third field image corresponding to the echo image group according to a seed point growing algorithm and the echo image group.
[0244] The fourth determining unit is used to determine a fourth field image corresponding to the echo image group according to the water-fat exchange region extraction algorithm and the echo image group.
[0245] The sixth determining unit is configured to determine the first field picture according to the third field picture and the fourth field picture.
[0246] Optionally, the third determining unit includes:
[0247] The down-sampling subunit is used to perform down-sampling processing on each frame of echo image in the echo image group to obtain echo image groups under each preset scale.
[0248] The third determining subunit is used to determine the first initial field image corresponding to the echo image group at each preset scale.
[0249] The fourth determining subunit is used to determine the first seed point corresponding to each first initial field image.
[0250] The fifth determining subunit is configured to determine the third field image according to the first seed points corresponding to the first initial field images.
[0251] Optionally, a fifth determination subunit is used to perform region growing processing on the first seed points corresponding to each first initial field map to obtain the second initial field map at each preset scale corresponding to each first seed point, and upsample each second initial field map to obtain a third initial field map at a target scale; determine the second seed points corresponding to each third initial field map, and perform region growing processing on the second seed points corresponding to each third initial field map to obtain a fourth initial field map corresponding to each second seed point; fuse the fourth initial field maps to obtain a fifth initial field map; and determine the third field map based on the fifth initial field map.
[0252] Optionally, the fourth determining unit includes:
[0253] The processing subunit is used to process the echo image group to obtain a sixth initial field image.
[0254] The sixth determination subunit is used to determine the target water-lipid exchange area in the sixth initial field map.
[0255] The seventh determination subunit is used to obtain the fourth field map according to the target water-lipid exchange area
[0256] Optionally, the region growing process includes three-dimensional region growing process.
[0257] The 3D region growing process includes the following steps:
[0258] Perform two-dimensional region growing processing on the target area to obtain an intermediate field map of the echo image group corresponding to the target area; the intermediate field map includes each field layer corresponding to the echo image group, and the target area includes the first seed point, the second seed point or the target water-fat exchange area.
[0259] Determine the target field layer of the intermediate field map; the target field layer is the field layer with the largest number of seed points among all field layers.
[0260] The field map value of the target field layer is mapped to other field layers of the intermediate field map to obtain the field map after three-dimensional region growing processing; other field layers are field layers other than the target field layer in each field layer.
[0261] Optionally, the field map determination device 1700 suitable for magnetic resonance scanning further includes:
[0262] An updating module is used to update the machine learning method according to the fifth field image and the echo image group when a second field image corresponding to the echo image group is determined according to the echo image group and the machine learning method and a fifth field image that is a modification of the second field image is received.
[0263] Optionally, the first determining module 1701 includes:
[0264] The acquisition unit is used to acquire an echo image group of the object under examination.
[0265] A seventh determination unit is used to determine an echo image group corresponding to at least one part of interest according to the deep learning segmentation network and the echo image group of the subject.
[0266] Fig.18 is a structural block diagram of the fat quantification device in the embodiment of the present application, such as Fig.18 As shown, in an embodiment of the present application, a fat quantification device 1800 is provided, comprising: a first determination module 1801 and a second determination module 1802, wherein:
[0267] The first determination module 1801 is used to determine the target field map of the echo image group corresponding to the part of interest by using the field map determination method applicable to magnetic resonance scanning.
[0268] The second determination module 1802 is used to determine the fat content of the part of interest according to the target field image corresponding to the echo image group.
[0269] Each module in the field map determination device or fat quantification device for magnetic resonance scanning can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.
[0270] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program. In one embodiment, the processor further implements the following steps when executing the computer program:
[0271] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.
[0272] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.
[0273] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.
[0274] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0275] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A method for determining a field map suitable for magnetic resonance scanning, characterized in that: The method comprises: Determine an echo image group corresponding to the part of interest; the echo image group includes a plurality of echo image frames; Determine a first field image corresponding to the echo image group according to the echo image group and a non-machine learning method; Determine a second field image corresponding to the echo image group according to the echo image group and a machine learning method; A target field map is determined according to the first field map and the second field map.
2. The method according to claim 1, characterized in that The machine learning method includes a supervised network; the method also includes: Acquire initial training data; the initial training data includes a plurality of echo image groups corresponding to parts of interest; Determine a target data set according to the initial training data; the target data set includes a first data set and a second data set; the first data set includes a field map determined based on the non-machine learning method and the initial training data, and the second data set includes a field map determined based on an unsupervised network and the initial training data; An initial supervised network is trained according to the initial training data and the target data set to obtain the supervised network.
3. The method according to claim 1 or 2, characterized in that: The determining a target field map according to the first field map and the second field map comprises: Determine a difference between each first field image value and a corresponding second field image value; the first field image value is a field image value of a first pixel point in the first field image, and the second field image value is a field image value of a second pixel point in the second field image corresponding to the first pixel point; The target field picture is determined according to a difference value between each of the first field picture values and the corresponding second field picture value.
4. The method according to claim 3, characterized in that The step of determining the target field image according to the difference between each of the first field image values and the corresponding second field image value comprises: Using the field image value of the first target pixel as the field image value of the pixel corresponding to the first target pixel in the target field image; the first target pixel is the first pixel whose difference value is less than or equal to the preset difference threshold; The field map value of the second target pixel point is used as the field map value of the pixel point corresponding to the second target pixel point in the target field map; the second target pixel point is the first pixel point whose difference value is not less than the preset difference threshold.
5. The method according to claim 1 or 2, characterized in that: The non-machine learning method includes a seed point growth algorithm and a water-fat exchange region extraction algorithm; the first field image corresponding to the echo image group is determined according to the echo image group and the non-machine learning method, including: Determine a third field image corresponding to the echo image group according to the seed point growth algorithm and the echo image group; Determining a fourth field image corresponding to the echo image group according to the water-fat exchange region extraction algorithm and the echo image group; The first field picture is determined according to the third field picture and the fourth field picture.
6. The method according to claim 5, characterized in that The step of determining a third field image corresponding to the echo image group according to the seed point growth algorithm and the echo image group includes: Downsampling each frame of the echo image in the echo image group to obtain an echo image group at each preset scale; Determine a first initial field image corresponding to each echo image group at the preset scale; Determine a first seed point corresponding to each of the first initial field images; The third field image is determined according to the first seed points corresponding to the first initial field images.
7. The method according to claim 6, characterized in that The determining the third field image according to the first seed points corresponding to the first initial field images comprises: Performing region growing processing on the first seed points corresponding to the first initial field images to obtain second initial field images at the preset scales corresponding to the first seed points, and performing upsampling processing on the second initial field images to obtain third initial field images at the target scale; Determine the second seed points corresponding to each of the third initial field images, and perform region growing processing on the second seed points corresponding to each of the third initial field images to obtain fourth initial field images corresponding to each of the second seed points; fusing the fourth initial field images to obtain a fifth initial field image; The third field picture is determined according to the fifth initial field picture.
8. The method according to claim 7, characterized in that Determining a fourth field image corresponding to the echo image group according to the water-fat exchange region extraction algorithm and the echo image group includes: Processing the echo image group to obtain a sixth initial field image; Determining a target water-lipid exchange region in the sixth initial field map; The fourth field map is obtained according to the target water-lipid exchange area.
9. The method according to claim 8, characterized in that The region growing process includes a three-dimensional region growing process; The three-dimensional region growing process comprises the following steps: Performing a two-dimensional region growing process on the target region to obtain an intermediate field map of the echo image group corresponding to the target region; the intermediate field map includes each field layer corresponding to the echo image group, and the target region includes the first seed point, the second seed point or the target water-fat exchange region; Determine a target field layer of the intermediate field map; the target field layer is the field layer with the largest number of seed points among the field layers; The field map value of the target field layer is mapped to other field layers of the intermediate field layer to obtain the field map after the three-dimensional region growing process; the other field layers are the field layers other than the target field layer in the field layers.
10. The method according to claim 1, characterized in that The method further comprises: In the case where the second field image corresponding to the echo image group is determined based on the echo image group and the machine learning method, if a fifth field image that is a modification of the second field image is received, the machine learning method is updated based on the fifth field image and the echo image group.
