Medical imaging system, method for identifying body position of detection object, and storage medium

Abstract

An embodiment of the present invention provides a method for identifying the body position of a detection object in medical imaging, a medical imaging system, and a computer-readable storage medium. The method comprises: receiving an image group via a trained deep learning network, the image group comprising a plurality of pre-scanned images in a plurality of orientations obtained by pre-scanning the detection object; and outputting the body position information of the detection object via the deep learning network.

Description

Technical Field

[0001] The present invention relates to medical imaging technology, and more particularly to a method for identifying the body position of a detection object in medical imaging and a computer-readable storage medium. Background Art

[0002] In medical imaging technologies such as magnetic resonance imaging (MRI) and computed tomography (CT) imaging, it is usually necessary to set the patient's body position information when performing the scan so as to determine the anatomical section direction of the image (e.g., coronal, sagittal, and transverse). Such body position information includes, for example, the orientation of the head or feet (e.g., whether the head enters the scanning area first or the feet enters the scanning area first when the patient is moved to the scanning area) and the patient's lying position (e.g., supine, prone, left side, and right side). However, the operator performing the imaging scan (e.g., a doctor or technician) hopes to minimize the tediousness and time consumption caused by manual settings, and even more hopes to minimize more serious problems caused by incorrect information input, such as images obtained with erroneous information cannot be used for diagnosis and need to be manually modified again or even require the patient to be scanned again. Summary of the invention

[0003] An embodiment of the present invention provides a method for identifying the body position of a detection object in medical imaging, comprising: receiving an image group via a trained deep learning network, the image group comprising a plurality of pre-scan images in a plurality of orientations obtained by performing a pre-scan on the detection object; and outputting the body position information of the detection object via the deep learning network.

[0004] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a computer, the computer executes the above method.

[0005] An embodiment of the present invention also provides a medical imaging system, comprising a controller unit, wherein the controller unit is used to control a trained deep learning network to receive an image group and receive body position information of a detection object output by the deep learning network, wherein the image group includes multiple pre-scan images in multiple orientations obtained by performing a pre-scan on the detection object via the medical imaging system.

[0006] Other features and aspects will become apparent from the following detailed description, drawings, and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The present invention may be better understood by describing exemplary embodiments of the present invention in conjunction with the accompanying drawings, in which:

[0008] Figure 1is a flow chart of a method for identifying the body position of a detection object in medical imaging according to an embodiment of the present invention;

[0009] Figure 2 An example architecture diagram of a VGG convolutional neural network is shown;

[0010] Figure 3 A schematic diagram showing the use of a separate deep learning network to output body position information;

[0011] Figure 4 A schematic diagram showing outputting body position information using a cascaded first learning network and a second learning network;

[0012] Figure 5 A block diagram showing an example of a magnetic resonance imaging scanning system. DETAILED DESCRIPTION

[0013] The specific embodiments of the present invention will be described below. It should be noted that in the specific description of these embodiments, in order to provide a concise description, it is impossible for this specification to provide a detailed description of all the features of the actual embodiments. It should be understood that in the actual implementation of any embodiment, just as in the process of any engineering project or design project, in order to achieve the specific goals of the developer and to meet system-related or business-related restrictions, various specific decisions are often made, and this will also change from one embodiment to another. In addition, it can also be understood that although the efforts made in this development process may be complex and lengthy, for ordinary technicians in the field related to the content disclosed by the present invention, some changes such as design, manufacturing or production based on the technical content disclosed in this disclosure are just conventional technical means, and should not be understood as insufficient content of this disclosure.

[0014] Unless otherwise defined, the technical or scientific terms used in the claims and the specification shall have the usual meaning understood by persons with ordinary skills in the technical field to which the invention belongs. The words "first", "second" and similar words used in the patent application specification and the claims of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "One" or "one" and other similar words do not indicate a quantitative limitation, but indicate the existence of at least one. "Include" or "comprises" and other similar words mean that the elements or objects appearing before "include" or "comprises" include the elements or objects listed after "include" or "comprises" and their equivalent elements, and do not exclude other elements or objects. "Connected" or "connected" and other similar words are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.

[0015] Some embodiments of the present invention can automatically identify the body position information of the imaged detection object based on deep learning technology, so that the identified body position information can be further used to determine the orientation information of the scanned image during the imaging process. Therefore, during the imaging process of the detection object, the operator does not need to manually set the body position information, which reduces the operating pressure and avoids the wrong body position information being input due to mistakes. These embodiments can be used in magnetic resonance imaging technology. For example, the operation interface of the magnetic resonance imaging system no longer sets a body position selection / input window, but uses the image obtained when performing a pre-scan on the detection object to automatically identify the body position information. These embodiments can also be applied in a similar and reasonable manner to other imaging technologies such as CT, PET, SPECT, etc.

[0016] As discussed herein, deep learning techniques (also known as deep machine learning, hierarchical learning, or deep structured learning, etc.) can use deep learning networks (e.g., artificial neural networks) to process input data to identify information of interest. Deep learning networks can be completed using one or more processing layers (e.g., input layer, normalization layer, convolution layer, pooling layer, output layer), etc., and different deep learning network models can have different numbers and functions of processing layers), where the configuration and number of layers allow the deep learning network to handle complex information extraction and modeling tasks. Specific parameters of the network (also known as "weights" or "biases") are usually estimated through a so-called learning process (or training process). The learned or trained parameters usually result in (or output) a network corresponding to different levels of layers, so that different aspects of the initial data or the output of the previous layer can usually represent a hierarchy or cascade of layers. In the process of image processing or reconstruction, this can be characterized as different layers relative to different levels of features in the data. Therefore, processing can be performed in layers, that is, earlier or higher-level layers can correspond to extracting "simple" features from the input data, followed by layers that combine these simple features into features that exhibit higher complexity. In practice, each layer (or more specifically, each "neuron" in each layer) may employ one or more linear and / or nonlinear transformations (so-called activation functions) to process the input data into an output data representation. The number of "neurons" may be constant between layers, or may vary from layer to layer.

[0017] As discussed herein, as part of the initial training of a deep learning process for solving a particular problem, a training data set includes known input values ​​(e.g., a medical image with RGB depth information or a pixel matrix that transforms a medical image into coordinates) and the desired (target) output value of the final output of the deep learning process (e.g., the position information of the detected object in the image or the image category information including the position information). In this way, the deep learning algorithm can process the training data set (in a supervised or guided manner or in an unsupervised or unguided manner) until the mathematical relationship between the known input and the desired output is identified and / or the mathematical relationship between the input and output of each layer is identified and characterized. The learning process typically utilizes (part of) the input data and creates a network output for the input data, then compares the created network output with the desired output of the data set, and then uses the difference between the created and desired outputs to iteratively update the parameters (weights and / or bias) of the network. Typically, the stochastic gradient descent (SGD) method can be used to update the parameters of the network, but those skilled in the art should understand that other methods known in the art can also be used to update the network parameters. Similarly, a separate validation data set can be used to validate the trained learning network, where both known inputs and expected outputs are known, and the network output can be obtained by providing the known inputs to the trained learning network, which is then compared with the (known) expected output to validate previous training and / or prevent over-training.

[0018] Figure 1 FIG. 1 is a flow chart of a method 100 for identifying the body position of a subject in medical imaging according to some embodiments of the present invention. Figure 1 As shown, the method 100 includes steps S102 and S104.

[0019] In step S102, an image group is received via a trained deep learning network, where the image group includes a plurality of pre-scan images at a plurality of positions obtained by performing a pre-scan on a detection object.

[0020] Taking magnetic resonance imaging as an example, usually, a pre-scan (also called a positioning scan or a basic scan) performed on the detection object can reconstruct a series of images. This series of images, for example, includes 15 (or other numbers) of pre-scan images, and each pre-scan image has identifiable orientation information (for example, based on pre-set body position information of the detection object). Those skilled in the art should understand that the orientation information is used to indicate the anatomical plane of the human body on which the image is located, which may include the three basic orientations of coronal, sagittal and transverse.

[0021] The multiple orientations in step S102 include coronal, sagittal and transverse orientations, that is, the image group received by the deep learning network includes images in three orientations, namely, coronal pre-scan images, sagittal pre-scan images and transverse pre-scan images. For example, the above-mentioned 15 pre-scan images may include 5 coronal images, 5 sagittal images and 5 transverse images.

[0022] The trained deep learning network has the ability to identify the body position of the detection object according to the set of pre-scan images. In step S104, the body position information of the detection object is output via the deep learning network.

[0023] Step S104 may specifically include: determining one of a plurality of predetermined posture types as the posture information of the detected object. For example, the deep learning network may process the received plurality of pre-scanned images to determine which of the posture types has the highest probability of the detected object's posture belonging to.

[0024] The deep learning network can be completed through the preparation of training data, the selection and construction of the network model, and the training, testing and optimization of the network.

[0025] The training data may be a medical scan image, such as a magnetic resonance pre-scan image, and specifically, may include, for example, images obtained when a magnetic resonance pre-scan is performed on multiple regions of interest (such as the abdomen, chest, heart, head, etc.) of the human body in multiple known (or determined) body positions. In one embodiment, the above-mentioned deep learning network is obtained by training data using these images as inputs to the network model and the corresponding multiple known body positions as outputs of the network model.

[0026] For example, when performing data training, a plurality of commonly used posture types can be used as the output of the deep learning network, wherein each posture type includes a combination of the orientation of the head or feet of the detected object and one of supine, prone, left side, or right side. For example, these posture types may include, for example, Figure 3There are 8 body positions shown, among which HFS stands for Head First, Supine; HFP stands for Head First, Prone; HFDR stands for Head First, Right Decub; HFDL stands for Head First, Left Decub; FFS stands for Feet First, Supine; FFP stands for Feet First, Prone; FFDR stands for Feet First, Right Decub; FFDL stands for Feet First, Left Decub. "Feet First" means that when the subject is on the examination bed, his feet are close to the scanning area (such as the scanning cavity), and when the examination bed is moved so that the subject on it enters the scanning cavity from the outside, the subject's feet enter the scanning area first, and the same applies to "Head First".

[0027] In some embodiments, the deep learning network is trained based on a VGGNet (Visual Geometry Group Network) convolutional neural network or other well-known models. For example, a VGG-11 network can be selected to implement embodiments of the present invention. Figure 2The schematic diagram of the VGG-11 network architecture is shown, which specifically includes an input layer, an output layer, 5 groups of convolutional layers (a total of 8 convolutional layers) located between the input layer and the output layer, and 3 fully connected layers, where each group of convolutional layers is also connected to a pooling layer. The size of each convolution kernel is 3*3 (pixels), the size of each pooling kernel is 2*2 (pixels), and the number of convolutional kernels in each group of convolutional layers is 64, 128, 256 512, and 512 respectively. During training, a large number of image groups are input into the deep learning network, wherein each image group includes multiple images with the same known body position type (for example, each image group is an image group obtained by a pre-scan or scan), and the size of each image can be, for example, 224*224*3, and the known body position type is correspondingly set as the output of the deep learning network, (or as described below, a large number of images with known image categories are input into the deep learning network, and the known image category is correspondingly used as the output of the deep learning network), and the weights at each pixel in each layer and other required parameters are determined by convolution learning to identify the mathematical relationship between the known input and output and / or to characterize the mathematical relationship between the input and output of each layer, so as to establish a deep learning network that can be used to identify body position (or image category). Among them, the image data output by the upper level is convoluted by each convolution layer to obtain the feature map of the convolution layer. The feature map of the upper convolution layer is subjected to, for example, maximum pooling by the pooling layer to extract the main features. The upper-layer feature map is integrated through the fully connected layer, and the judgment result obtained by, for example, logistic regression processing, i.e., body position information or the following image category, is output through the output layer.

[0028] During the learning process, the network output (position information or image category) corresponding to the known input (pre-scanned image group) can be compared with the actual information (known position information or image category), and the difference is the loss function. The network parameters (weights and / or biases) can be iteratively updated through the loss function to continuously reduce the loss function, thereby training a neural network model with higher accuracy. In some embodiments, there are many functions that can be used as loss functions, including but not limited to mean squared error, cross entropy error, etc.

[0029] Figure 3 A schematic diagram showing the use of a separate deep learning network to output body position information is shown. Figure 3 As shown, in one embodiment, during, for example, a magnetic resonance scan, a series of pre-scan images obtained can be input into the input layer of a deep learning network, and then these images are processed based on parameters such as weights determined during the training process, so that the body position type of the subject currently being scanned is output in the output layer.

[0030] Figure 4 FIG. 4 shows a schematic diagram of outputting body position information using a cascaded first learning network and a second learning network; Figure 4 As shown, the above-mentioned deep learning network may include a first learning network 402 and a second learning network 404. The first learning network 402 may be the above-mentioned VGG-11 architecture or a similar architecture, and the second learning network 404 may be a classifier with a relatively simple architecture.

[0031] The first learning network 402 is used to receive the above-mentioned image group and output the image category of each pre-scan image. The image category includes a combination of at least one of the above-mentioned multiple orientations (coronal, sagittal and transverse) and at least one of the multiple body position types (HFS, HFP, HFDR, HFDL, FFS, FFP, FFDR, FFDL). That is, through the first learning network 402, it can be determined which orientations the pre-scan image may be and the possible body positions of the detection object therein.

[0032] During training, the input layer of the first learning network 402 receives multiple images of size 256*256*3, for example, and sets the known image category corresponding to each image in the output layer. The weight of each pixel in each layer and other required parameters are determined through convolutional learning to establish a deep learning network that can be used to identify image categories.

[0033] Therefore, when performing a magnetic resonance scan, a series of pre-scan images can be input into the input layer of the first deep learning network, and then these images are processed based on parameters such as weights determined during the training process, so that the image category of each pre-scan image is output in the output layer.

[0034] The above image category can be one of several known image categories in the following list:

[0035] Category A: a combination of transverse, HFS, and FFS; it means that the pre-scan image is a transverse image, in which the body position of the detection object is HFS or FFS.

[0036] Category B: A combination of transverse position, HFP, and FFP; it means that the pre-scan image is a transverse position image, in which the body position of the detection object is HFP or FFP.

[0037] Category C: a combination of transverse, HFDR, and FFDL; it means that the pre-scan image is a transverse image, in which the body position of the detection object is HFDR or FFDL.

[0038] Category D: combination of transverse, HFDL, and FFDR; it means that the pre-scan image is a transverse image, in which the body position of the detection object is HFDL or FFDR.

[0039] Category E: combination of sagittal, coronal, HFS, HFP, HFDR, and HFDL; it means that the pre-scan image is a sagittal image or a coronal image, and the body position of the detection object is HFS, HFP, HFDR, or HFDL;

[0040] Category F: combination of sagittal, coronal, FFS, FFP, FFDR, and FFDL; it means that the pre-scan image is a coronal image or a sagittal image, and the body position of the detection object therein is FFS, FFP, FFDR, or FFDL.

[0041] The above-mentioned division method of image categories is conducive to obtaining accurate body position information. Of course, the embodiments of the present invention can also use image categories with other numbers and division methods to train the above-mentioned first learning network and second learning network, so as to obtain more accurate body position information or achieve other beneficial effects.

[0042] During training, the second learning network 404 can receive known image types of image groups generated during multiple pre-scan processes, and set the output layer to perform each pre-scan to detect the body position of the object, and determine the weights at each pixel in each layer and other required parameters through convolutional learning to establish a second learning network that can be used to identify body positions based on image categories.

[0043] Therefore, after the image category of each pre-scan image is output to the second learning network through the first learning network 402, the second learning network 404 can output the body position information of the detected object.

[0044] First, the input image group is analyzed by the first learning network 402 to obtain the image category of each pre-scanned image, and at least the possible orientation and body position type of each pre-scanned image can be obtained according to the image category, and then the image category of each input pre-scanned image is further analyzed according to the second learning network 404 to obtain the correct body position information of the detected object. In this way, the accuracy of body position recognition can be greatly improved, for example, at least the problem of reduced accuracy caused by excessive calculation can be avoided.

[0045] For example, if the first learning network 402 receives 10 pre-scan images and outputs 4 image categories of the first to tenth pre-scan images belonging to category D and 6 belonging to category E, the second learning network 404 can output the accurate body position of the detected object, namely HFDL.

[0046] Further, the second learning network 404 is used to extract a smaller number of image categories from the received image categories and correct them to be consistent with the larger number of image categories from the received image categories. This can be achieved by setting a preprocessing layer in the second learning network 404. By first outputting the image category of each pre-scanned image, the second learning network 404 has the opportunity to correct the wrong image category, so that the accuracy of the body position information finally outputted is higher.

[0047] For example, if the first learning network 402 receives 10 pre-scan images, and outputs 3 image categories of the first to tenth pre-scan images belonging to category D, 6 belonging to category E, and 1 belonging to category C, then the second learning network 404 corrects the above category C to category D or category E through preprocessing, and processes the preprocessed data to output the accurate body position of the detected object, namely HFDL.

[0048] Further, the second learning network 404 determines the orientation in the above-mentioned smaller number of image categories, and corrects it to the larger number of image categories including the orientation. That is, the above-mentioned smaller number of image categories that are modified have the same orientation as the larger number of image categories. For example, since the orientation in category C is the transverse position, it is corrected to category D that also points to the transverse position, rather than category E that points to the sagittal position.

[0049] In the medical imaging scanning process, after the pre-scan is completed, a formal scan is performed on the test object to obtain multiple medical diagnostic images of the test object. Therefore, after the body position information of the test object is identified based on any embodiment of the above-mentioned method 300, and the medical diagnostic image is obtained based on the formal scan, the orientation information of each medical diagnostic image can be determined based on the identified body position information.

[0050] By automatically identifying the body position of the subject, even if the doctor or technician is no longer required to manually set the body position information, the position information of each reconstructed image can be determined as in the traditional scanning process, making the scanning process smoother and the scanning results more robust. For example, the medical imaging system can determine which position the corresponding pre-scan image is based on the automatically identified body position information HFS (head first, supine) and the adopted scanning sequence.

[0051] The above method can be used in magnetic resonance imaging or other medical imaging scanning techniques. Figure 5 A structural diagram of an example of a magnetic resonance imaging scanning system is shown below. Figure 5 An example of applying the above-mentioned medical imaging scanning method to magnetic resonance imaging scanning technology is described.

[0052] like Figure 5As shown, the magnetic resonance imaging system 500 includes a scanner 510 , a table 520 , a controller unit 530 , a data processing unit 540 , an operation console 550 and a display unit 560 .

[0053] In one example, the scanner 510 may include a main magnet assembly 511, which generally includes an annular superconducting magnet defined in a housing, and the annular superconducting magnet is installed in an annular vacuum container. The annular superconducting magnet and its housing define a cylindrical space surrounding the detection object 56, that is, Figure 5 A scanning bore 512 is shown. The scanning bore 512 defines an imaging region or at least a portion of an imaging region of the magnetic resonance imaging system.

[0054] The bed 520 is configured to be connected to the patient entrance of the scanning cavity 512, and is used to carry the test object 56, so that the test object 56 can be moved to the scanning area in a specific posture (such as one of the above-mentioned multiple posture types) to receive imaging scanning.

[0055] The scanner 510 further includes an RF transmitting coil 516, a radio frequency generator 513, a gradient coil assembly 514, a gradient coil driver 515, an RF receiving coil 570, and a data acquisition unit 517. When performing an imaging scanning process on the detection object 56, the scanner 510 is used to obtain image data of the detection object 56.

[0056] The image data may be processed by a data processing unit 540 such as calculation and reconstruction. The data processing unit 540 may include a computer and a storage medium on which a program for predetermined data processing to be executed by the computer is recorded.

[0057] The controller unit 530 is coupled to the scanner 510 , the bed 520 , and the data processing unit 540 to control these components to perform a scanning procedure of magnetic resonance imaging, which may specifically include a pre-scanning procedure and a formal scanning procedure.

[0058] The controller unit 530 may include a computer and a storage medium, wherein the storage medium is used to store a program that can be executed by the computer. When the computer executes the program, the scanner 510, the bed 520, the display unit 560 and other components may perform corresponding operations in the pre-scanning process and the scanning process. The data processing unit 540 may also be enabled to perform predetermined data processing.

[0059] The storage medium of the controller unit 530 and the data processing unit 540 may include, for example, a ROM, a floppy disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, or a nonvolatile memory card.

[0060] The operation console 550 may include user input devices, such as a keyboard and a mouse, and the operator can input control signals or parameter settings to the controller unit 530 through the user input devices. These parameter settings generally include setting the body position information of the detection object. However, this step can be omitted in the present embodiment, and the body position information of the detection object can be automatically identified through a deep learning network.

[0061] The display unit 560 can be connected to the operation console 550 to display an operation interface, including a parameter setting interface, and can also be connected to the data processing unit 540 to display a reconstructed image or various images captured by a camera coupled to the magnetic resonance imaging system.

[0062] When performing magnetic resonance imaging scanning, doctors or technicians usually need to help or guide the subject to be positioned on the bed 520 in a desired position. The subject's position information is then set on the operation interface before performing a pre-scan. If the position information is not consistent with the subject's actual positioning posture, it takes time to correct such an error, or the scan may be directly re-performed.

[0063] After adopting the method of the present invention, the body position setting is not performed. Instead, a pre-scan is performed by a scanner to obtain a set of pre-scan images, and then the pre-scan images are used to identify the body position of the detection object. After the region of interest is determined based on the pre-scan images, a formal scan can be performed on the region of interest. The data processing unit 540 reconstructs a medical diagnostic image for diagnosis based on the data obtained by the formal scan, and when the medical diagnostic image is presented through the display unit, the orientation information of the medical diagnostic image can be annotated at the same time. Therefore, even if the body position setting is not performed, it will not cause confusion among doctors about the orientation of the image.

[0064] In one embodiment, the above-mentioned trained deep learning network is trained based on a training system on an external carrier (e.g., a device other than a medical imaging system). In some embodiments, the training system may include a first module for storing a training data set, a second module for training and / or updating based on a model, and a network for connecting the first module and the second module. In some embodiments, the first module includes a first processing unit and a first storage unit, wherein the first storage unit is used to store a training data set, and the first processing unit is used to receive relevant instructions (e.g., obtain a training data set) and send the training data set to the second module according to the instructions. The second module includes a second processing unit and a second storage unit, wherein the second storage unit is used to store a training model, and the second processing unit is used to receive relevant instructions, train and / or update the learning network, etc. In some embodiments, the network may include various connection types, such as wired, wireless communication links or fiber optic cables, etc.

[0065] Once the deep learning network is generated and / or configured, its data can be copied and / or loaded into the magnetic resonance imaging system 500 described above, which can be done in different ways. For example, the model can be loaded via a directed connection or link between the controller unit 530 and the second storage unit of the magnetic resonance imaging system 500. In this regard, the communication between the different elements can be completed using available wired and / or wireless connections and / or according to any suitable communication (and / or network) standards or protocols. Alternatively, the data can be loaded into the magnetic resonance imaging system 500 indirectly. For example, the data can be stored in a suitable machine-readable medium (e.g., a flash memory card, etc.), which is then used to load the data into the magnetic resonance imaging system 500 (on-site, such as by a user or authorized personnel of the system), or the data can be downloaded to an electronic device capable of local communication (e.g., a laptop, etc.), which is then used on-site (e.g., by a user or authorized personnel of the system) to upload the data to the magnetic resonance imaging system 500 via a direct connection (e.g., a USB connector, etc.).

[0066] Based on the above description, an embodiment of the present invention can also provide a medical imaging system, which includes a controller unit, which is used to control a trained deep learning network to receive an image group and receive the body position information of the detection object output by the deep learning network. The image group includes multiple pre-scan images in multiple orientations obtained by performing a pre-scan on the detection object through the medical imaging system.

[0067] The deep learning network is specifically used to determine a body position information of the detection object from a plurality of predetermined body position types.

[0068] The deep learning network may include a first learning network 402 and a second learning network 404. The first learning network 402 is used to receive the image group and output an image category of each pre-scan image, wherein the image category includes a combination of at least one of the multiple orientations and at least one of the multiple body position types. The second learning network 404 is used to receive the image category of each pre-scan image and output body position information of the detected object.

[0069] The second learning network 404 is also used to extract a smaller number of image categories from the received image categories, and correct the extracted image categories to be consistent with the larger number of image categories from the received image categories, wherein the smaller number of image categories and the larger number of image categories include the same orientation.

[0070] An embodiment of the present invention may also provide a computer-readable storage medium for storing an instruction set and / or a computer program. When the instruction set and / or the computer program are executed by a computer, the computer executes the method for identifying the body position of the detection object in any of the above embodiments. The computer that executes the instruction set and / or the computer program may be a computer of the medical imaging system or other devices / modules of the medical imaging system. In one embodiment, the instruction set and / or the computer program may be compiled in a processor / controller of the computer, such as the above-mentioned controller unit 530.

[0071] Specifically, the set of instructions and / or computer program, when executed by a computer, causes the computer to:

[0072] Identifying image quality types of medical images based on the trained learning network; and

[0073] Based on the identified image quality type, a corresponding control signal for controlling the medical imaging system is generated.

[0074] The instructions described above may be combined into one instruction for execution, and any instruction may be split into multiple instructions for execution. In addition, the execution order of the instructions described above is not limited to the above order.

[0075] In some embodiments, before identifying the type of image quality in the medical image, the method further includes receiving the medical image generated by the medical imaging system based on an instruction of a user.

[0076] As used herein, the term "computer" may include any processor-based or microprocessor-based system, including systems using microcontrollers, reduced instruction set computers (RISC), application specific integrated circuits (ASICs), logic circuits, and any other circuits or processors capable of performing the functions described herein. The above examples are exemplary only, and are thus not intended to limit the definition and / or meaning of the term "computer" in any way.

[0077] The instruction set may include various commands that instruct a computer or processor as a processing machine to perform specific operations, such as the methods and processes of various embodiments. The instruction set may take the form of a software program that may form part of one or more tangible non-transitory computer-readable media. The software may take various forms such as system software or application software. In addition, the software may take the form of a collection of independent programs or modules, a program module within a larger program, or a portion of a program module. The software may also include modular programming in the form of object-oriented programming. The processing of input data by the processing machine may be in response to operator commands, or in response to previous processing results, or in response to a request made by another processing machine.

[0078] Some exemplary embodiments have been described above, however, it should be understood that various modifications may be made. For example, if the described techniques are performed in a different order and / or if the components in the described systems, architectures, devices, or circuits are combined in different ways and / or replaced or supplemented by other components or their equivalents, suitable results may be achieved. Accordingly, other implementations also fall within the scope of protection of the claims.

Claims

1. A method for identifying the body position of a detection object in medical imaging, comprising: Receiving an image group through a first learning network in the trained deep learning network, and outputting an image category of each pre-scan image, wherein the image group includes a plurality of pre-scan images in a plurality of orientations obtained by performing a pre-scan on a detection object, and the image category includes a combination of at least one of the plurality of orientations and at least one of a plurality of predetermined body position types; as well as, The image category of each pre-scanned image is received via a second learning network in the trained deep learning network, and the body position information of the detection object is output, which includes: determining one of a plurality of predetermined body position types as the body position information of the detection object via the second learning network.

2. The method of claim 1, wherein: The multiple orientations include coronal, sagittal and transverse orientations, and each body orientation type includes a combination of the orientation of the subject's head or feet and one of supine, prone, left side or right side.

3. The method of claim 1, wherein: The second learning network extracts a smaller number of image categories from the received image categories, and corrects the extracted image categories to be consistent with the larger number of image categories from the received image categories.

4. The method of claim 3, wherein: The smaller number of image categories and the larger number of image categories include the same orientation.

5. The method according to claim 1, wherein the deep learning network comprises a VGG (Visual Geometry Group) convolutional neural network.

6. A computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a computer, the computer executes the method according to any one of claims 1 to 5.

7. A medical imaging system comprising a controller unit, the controller unit being configured to: Receiving an image group via a first learning network in a trained deep learning network, and outputting an image category of each pre-scan image, wherein the image group includes a plurality of pre-scan images in a plurality of orientations obtained by performing a pre-scan on a detection object, and the image category includes a combination of at least one of the plurality of orientations and at least one of a plurality of predetermined body position types; and, The image category of each pre-scanned image is received via a second learning network in the trained deep learning network, and the body position information of the detected object is output, including: The second learning network determines one of a plurality of predetermined body position types as body position information of the detection object.

8. The system of claim 7, wherein: The multiple orientations include coronal, sagittal and transverse orientations, and each body orientation type includes a combination of the orientation of the subject's head or feet and one of supine, prone, left side or right side.

9. The system of claim 7, wherein: The second learning network is used to extract a smaller number of image categories from the received image categories, and correct the extracted image categories to be consistent with the larger number of image categories from the received image categories.

10. The system of claim 9, wherein: The smaller number of image categories and the larger number of image categories include the same orientation.

11. The system of claim 7, wherein the deep learning network comprises a VGG (Visual Geometry Group) convolutional neural network.

Images