CN107669273B - Magnetic resonance equipment scanning system, scanning method and computer readable storage medium - Google Patents
Magnetic resonance equipment scanning system, scanning method and computer readable storage medium Download PDFInfo
- Publication number
- CN107669273B CN107669273B CN201711085890.0A CN201711085890A CN107669273B CN 107669273 B CN107669273 B CN 107669273B CN 201711085890 A CN201711085890 A CN 201711085890A CN 107669273 B CN107669273 B CN 107669273B
- Authority
- CN
- China
- Prior art keywords
- scanning
- training data
- information
- positioning
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Informatics (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- General Physics & Mathematics (AREA)
- Veterinary Medicine (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Animal Behavior & Ethology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Surgery (AREA)
- Radiology & Medical Imaging (AREA)
- Heart & Thoracic Surgery (AREA)
- Data Mining & Analysis (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Fuzzy Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- High Energy & Nuclear Physics (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention relates to a magnetic resonance equipment scanning method. The method comprises the steps of obtaining a scout image of a scanning object through scout scanning, and obtaining coil information and scanning sequence information of clinical scanning to be executed. And determining the part to be scanned of the scanning object by taking the positioning image, the coil information and the scanning sequence information as the input of a part identification module. And optimizing subsequent parameters of the scanning object based on the part to be scanned, so that the whole clinical scanning is more accurate and more targeted. The magnetic resonance equipment scanning method can reduce the influence caused by the difference of the space sensitivities of different tissues and different scanning coils of a human body. The scanning method of the magnetic resonance equipment can save a large amount of workload caused by the fact that automatic identification of the position information cannot be automatically completed. The part identification module is adopted in the magnetic resonance equipment scanning method, so that the part to be scanned can be identified quickly, accurately and automatically.
Description
Technical Field
The present invention relates to the field of medical images, and in particular, to a magnetic resonance apparatus scanning system, a magnetic resonance apparatus scanning method, and a computer-readable storage medium.
Background
When scanning, a magnetic resonance apparatus usually acquires a scout image (possibly a scout image of the chest and abdomen of a human body), and then medical staff plans a clinical scan on the scout image to determine corresponding scan parameters of a scan part (possibly the heart, the liver, the intestinal tract, etc.) needing to be specifically obtained from the clinical scan. Due to differences between a human body and magnetic resonance equipment hardware, in order to obtain a better imaging effect, medical staff often needs to manually adjust system parameters, sequence parameters, reconstruction or image processing according to specific clinical scanned part information in a scanning process of the magnetic resonance equipment. Conventional magnetic resonance apparatus scanning systems require a user to manually plan a clinical scan through scout images.
Disclosure of Invention
The invention aims to provide a magnetic resonance equipment scanning system, a scanning method and a computer readable storage medium, and solves the problems that the traditional magnetic resonance equipment scanning system can only manually carry out clinical scanning planning on a scout image and can not automatically identify a formal scanning part so as to intelligently plan clinical scanning parameters.
In order to solve the above problems, the present invention provides a magnetic resonance apparatus scanning method,
the method comprises the following steps:
acquiring a positioning image of a scanning object through positioning scanning;
acquiring coil information and scanning sequence information of the positioning scanning;
based on the trained part recognition module, the positioning image, the coil information and the scanning sequence information are used as the input of the part recognition module to determine the part to be scanned of the scanning object;
and optimizing the parameters of the scanning object in the subsequent scanning based on the part to be scanned.
In one embodiment, the training method of the part recognition module specifically includes:
acquiring positioning images of a plurality of groups of training data;
acquiring coil information of a plurality of groups of training data, acquiring scanning sequence information of the plurality of groups of training data, and extracting image characteristic information of positioning images of the plurality of groups of training data;
establishing a multi-layer neural network unit according to the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
and setting function corresponding relations for the multi-layer neural network units, and adjusting parameters in each layer of neural network units according to multiple groups of training data to form the part identification module.
In an embodiment, the step of establishing a multi-layer neural network unit according to the coil information of the multiple sets of training data, the scanning sequence information of the multiple sets of training data, and the image feature information of the scout images of the multiple sets of training data specifically includes:
performing first-stage classification storage on training data according to the image characteristic information of the positioning image to obtain training data after preliminary classification and at least one layer of neural network units;
and performing secondary classification storage on the coil information and the scanning sequence information in the obtained training data after the preliminary classification, establishing a multilayer neural network unit, and respectively storing the training data after the preliminary classification in the multilayer neural network unit.
In one embodiment, the site identification module comprises:
the image acquisition device is used for acquiring a plurality of groups of training data positioning images;
the characteristic generator is used for acquiring coil information of a plurality of groups of training data, acquiring scanning sequence information of the plurality of groups of training data and extracting image characteristic information of positioning images of the plurality of groups of training data;
the multi-layer neural network unit is used for classifying and storing the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
and the synthesizer is used for setting function corresponding relation for the multilayer neural network units and adjusting parameters in each layer of the neural network units according to training data to form the part identification module.
In one embodiment, the training method of the part recognition module comprises the following steps:
acquiring positioning images of a plurality of groups of training data;
acquiring coil information of a plurality of groups of training data, acquiring scanning sequence information of the plurality of groups of training data, and extracting image characteristic information of positioning images of the plurality of groups of training data;
establishing different classification rules aiming at the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
setting different weight ratios among the different classification rules to form the part recognition module.
In one embodiment, in the step of establishing different classification rules for the coil information of the multiple sets of training data, the scanning sequence information of the multiple sets of training data, and the image feature information of the scout images of the multiple sets of training data, the classification rules are kernel functions with different corresponding relations;
and setting different weight ratios among the different classification rules to form the part identification module, wherein the weight ratios are used for setting different parameters in the kernel function to adjust the classification rules.
In one embodiment, in the step of optimizing subsequent parameters of the scanning object based on the part to be scanned, the subsequent parameters include any one of the following parameters: scan sequence parameters, image reconstruction parameters, and image processing parameters.
In one embodiment, the step of optimizing subsequent parameters of the scan object based on the part to be scanned further comprises:
and finishing the action of automatically optimizing the image or the action of automatically adjusting the window width and the window level according to the optimized subsequent parameters of the scanning object.
In order to solve the above problem, the present invention further provides a magnetic resonance apparatus scanning system, including a magnetic resonance scanning apparatus and a computer, wherein the computer includes a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor when executing the program can be used to execute a magnetic resonance apparatus scanning method, the method includes:
acquiring a positioning image of a scanning object through positioning scanning;
acquiring coil information and scanning sequence information of the positioning scanning;
based on the trained part recognition module, the positioning image, the coil information and the scanning sequence information are used as the input of the part recognition module to determine the part to be scanned of the scanning object;
and optimizing the parameters of the scanning object in the subsequent scanning based on the part to be scanned.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, is adapted to carry out the steps of the method of any of the preceding claims.
The invention provides a magnetic resonance equipment scanning method, which obtains a positioning image of a scanning object through positioning scanning. And acquiring coil information and scanning sequence information of the positioning scanning. And determining the part to be scanned by taking the positioning image, the coil information and the scanning sequence information as the input of the part recognition module based on the trained part recognition module. And optimizing the parameters of the subsequent scanning based on the part to be scanned. The scanning method of the magnetic resonance equipment provided by the invention can accurately, quickly and automatically identify the part to be scanned. And optimizing parameters of subsequent scanning based on the part to be scanned, so that the whole clinical scanning is more intelligent, more accurate and more targeted. The magnetic resonance equipment scanning method can reduce the influence caused by the difference of the space sensitivities of different tissues and different scanning coils of a human body. The magnetic resonance equipment scanning method can accurately judge and identify the part to be scanned.
Drawings
Figure 1 is a flow chart of a magnetic resonance apparatus scanning method of some embodiments of the present invention;
figure 2 is a flow chart of a magnetic resonance apparatus scanning method of some embodiments of the present invention;
figure 3 is a flow chart of a magnetic resonance apparatus scanning method of some embodiments of the present invention;
figure 4 is a flow chart of a magnetic resonance apparatus scanning method according to some embodiments of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the magnetic resonance apparatus scanning system, the scanning method and the computer readable storage medium of the present invention are further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Referring to fig. 1, a scanning method of a magnetic resonance apparatus is provided, which includes the following steps:
s100, acquiring a scout image of a scanning object through scout scanning.
Scout scan refers to a global scan of a scanned object by a magnetic resonance apparatus to obtain an image of a larger area of the scanned object (e.g., head to foot, neck to thigh) to provide a location for a subsequent clinical scan. Here, the obtained scout image is not limited, and may be image information of an arbitrary body area of the scanning subject. The scout image is usually vector bit information of the scanned object, and may be a coronal image or a transverse image. Such as: the positioning image can be vector bit image information of the human body thorax and abdomen, and the positions of the heart, the liver, the intestinal tract and the like of the human body are covered in the positioning image of the human body thorax and abdomen acquired by the magnetic resonance equipment. The scan sequence of the scout image may employ a 2D GRE sequence (2D gradient echo sequence).
In some cases, the scout image may not be limited to a magnetic resonance image acquired by a magnetic resonance apparatus, but may also be acquired by other medical apparatuses such as: CT images, PET images, X-ray images, and the like. The following steps of the present embodiment are mainly illustrated by magnetic resonance images, which are known to those skilled in the art and can be applied to other medical images.
And S200, acquiring coil information and scanning sequence information of positioning scanning.
In some cases, after a patient (a scanning object) enters a magnetic resonance apparatus during a magnetic resonance scan, medical staff determines a part to be examined according to registration of the patient in a scanning room, and can wear a radio frequency receiving coil for the patient. After the patient is in place, the computer in the operating room controls the magnetic resonance scanner by medical personnel to perform a scout scan as described in step S100. The computer of the operation room determines the corresponding radio frequency coil and the positioning scanning protocol according to the positioning scanning requirement by medical staff, and then the computer of the operation room sends the coil information and the scanning sequence information of the selected positioning scanning to the magnetic resonance scanner. The magnetic resonance scanner carries out positioning scanning on the patient according to the scanned coil information and the scanning sequence information, and transmits the magnetic resonance echo signals received according to the positioning scanning to a computer of an operation room for image reconstruction to obtain a positioning image. And finally, the computer of the operation room displays the obtained positioning image on the computer of the operation room, and simultaneously, the positioning image can be transmitted to the magnetic resonance scanner of the scanning room through the communication connecting device and displayed on a liquid crystal touch screen on the surface of the magnetic resonance scanner.
It can be understood that the specific scanning part (heart, liver, intestinal tract, etc.) of the scanning can not be judged only according to the positioning image, and the technical scheme of the invention provides integration of image characteristic information of the positioning image, scanning coil information of the positioning scanning and scanning sequence information of the positioning scanning, and the specific part which a doctor wants to scan can be intelligently judged without manual setting by the doctor, so that the setting time of the doctor is saved, and the scanning process is optimized.
Here, the coil information and the scan sequence information of the scout scan may be manually input by a doctor, for example, one to ten rf coils of the existing mr scanner, and the scan range of the scout scan desired by the doctor is the thoracoabdominal region, then three to six rf coils corresponding to the thoracoabdominal region may be manually selected, and simultaneously one scan sequence (for example, 2D GRE sequence) may be selected.
In some embodiments, the coil information and scan sequence information of the scout scan may also be automatically obtained and generated by the computer based on the patient's relevant data.
And S300, based on the trained part identification module, determining the part to be scanned of the scanning object by taking the positioning image, the coil information and the scanning sequence information as the input of the part identification module.
The part identification module can identify the part information of the scanning object according to the positioning image, the scanned coil information and the scanning sequence information. The part recognition module can be obtained by learning and training based on a certain algorithm and a certain process. Due to different algorithms, the specific training process will be different, and will not be described in detail herein.
The part recognition module may be an algorithm based on a neural network, and specifically may be a multi-layer neural network composed of a plurality of feature quantities as neural nodes (or neural network units). And the neural network can be trained according to some sample data, so that the part to be scanned of the scanning object is accurately identified according to the trained part identification module.
In the process of training the part recognition module, a large amount of sample data (scout images, coil information and scanning sequence information corresponding to known scanning part information) is used as the input of the part recognition module, and the algorithm parameters (neural network unit) of the part recognition module are optimized.
In the process of using the part recognition module, the positioning image, the coil information and the scanning sequence information are input into the part recognition module, and the corresponding part to be scanned is obtained according to the trained part recognition module.
S400, optimizing the parameters of the scanning object in the subsequent scanning based on the part to be scanned.
It can be understood that the parameters of the scanning object in the subsequent scanning are optimized again according to the information of the part to be scanned, which is identified by the part identification module. Such as: in some cases, after the scanning object enters the scanning area of the magnetic resonance equipment during the magnetic resonance scanning, medical staff determines the part needing to be checked according to the registration of the scanning object in a scanning room, and selects a radio frequency receiving coil. Medical personnel automatically select a positioning protocol based on the radio frequency coil or selected examination region via the computer of the operating room and then select the scanned coil information and scan sequence information to the magnetic resonance scanner. The magnetic resonance scanner performs a scout scan as described in step S100 based on the input information. The scout image, coil information and scan sequence information of the first scan are recorded and input to the part recognition module. And the part identification module determines the part to be scanned of the scanned object according to the training and learning data. And performing secondary scanning according to the determined part to be scanned of the scanning object. The clinical scan to be performed is optimized based on the site to be scanned. The clinical scan to be performed optimized here may include a reselection of the localization protocol, a re-determination of the scan coil information, a re-determination of the scan sequence information, and the like. It can be understood that the clinical scan is performed again on the part to be scanned after re-optimization, so that the scanning result is more accurate.
For example: in a conventional MR scanning process, it is often necessary to set relevant system parameters or scanning protocol parameters, such as a radio frequency transmission mode, a transmission voltage, a form and intensity of a K-space wave filter, image filter parameters, an automatic window adjusting strategy, etc., according to a specific scanned part.
For example, for the purpose of optimal imaging, the radio frequency emission modes of the head, abdomen and knee are TXMode1, TXMode2 and TXMode3, respectively. The set transmit voltages were 280, 420, and 320. According to different signal characteristics of the head, the abdomen and the knee, the shape and the strength of the K-space filter need to be set in a targeted mode, and according to the difference of attention points of the three parts in clinical diagnosis, the parameters of the image filter also need to be set in a targeted mode. Such as: the liver image is relatively flat, and can properly adopt slightly strong denoising parameters, while the knee joint image often needs to observe linear structures such as meniscus and the like, and the parameters of enhancing and smoothing the linear structures of the image filter are needed. For the default window width of the image, because the focus of clinical diagnosis is different, the default window of the image needs to be processed according to the position, for example, when abdomen is enhanced to scan, obvious signal change needs to be seen, when head is scanned, gray-white contrast needs to be good, when joint is scanned and fat is pressed, the joint surface needs to be clear, and the bone has low brightness.
In the embodiment, a magnetic resonance apparatus scanning method is provided. The part to be scanned can be rapidly and automatically identified by adopting the part identification module in the magnetic resonance equipment scanning method. The magnetic resonance equipment scanning method firstly obtains the positioning image, the coil information and the scanning sequence information of a scanning object. The part to be scanned is then determined based on the part identification module. And the parameters of the subsequent scanning are optimized based on the part to be scanned, so that the whole clinical scanning is more intelligent, more accurate and more targeted. The magnetic resonance equipment scanning method can reduce the influence caused by the difference of the space sensitivities of different tissues and different scanning coils of a human body. The magnetic resonance equipment scanning method can accurately judge and identify the part to be scanned. In this embodiment, a scout image, scanning coil information, and scanning sequence information of the magnetic resonance apparatus are used. By adopting the trained part recognition module, the part recognition module can provide accurate scanned part information in real time when scanning the positioning image every time. The scanning method of the magnetic resonance equipment can save a large amount of workload caused by the fact that automatic identification of the position information cannot be automatically completed.
In one embodiment, the part recognition module may employ a variety of training methods, such as: a training method using a neural network or a training method using a support vector machine, and the like. It is to be understood that the training method of the part recognition module is not limited to the training method described in the present application, and may be applied in combination with the method as long as the learning of the training data is completed. More specifically, a self-organizing feature mapping based neural network training method, a radial basis function based neural network training method, a dual-mode linear neural network training method, or an inverse error based neural network training method may be selected.
Referring to fig. 2, in an embodiment, the training method of the portion identification module specifically includes:
s510, positioning images of multiple groups of training data are obtained.
It will be appreciated that the scout images of the sets of training data acquired herein may be scout images of known location information.
S520, coil information of a plurality of groups of training data is obtained, scanning sequence information of the plurality of groups of training data is obtained, and image characteristic information of positioning images of the plurality of groups of training data is extracted.
It can be understood that the specific part of the scan cannot be determined only according to the scout image, and the specific part of the scan can be determined by integrating the image feature information of the scout image, the scan coil information and the scan sequence information in combination with the coil information and the scan sequence information. The coil information and the scanning sequence information of the multiple sets of training data can be acquired in real time in the clinical scanning process, and can also be manually input after the scanning is finished. The coil information and the scanning sequence information of the multiple groups of training data are acquired to complete the accurate classification of the multiple groups of training data by matching with the image characteristic information of the multiple groups of training data positioning images. A complete data information base is formed by storing a plurality of groups of training data information. In the complete data information base, each set of training data stores image characteristic signals, scanning coil information and scanning sequence information of a part to be scanned and a positioning image corresponding to the part to be scanned.
S530, establishing a multi-layer neural network unit according to the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data.
The neural network unit is established based on the corresponding relation between the position information of a plurality of groups of training data and the coil information, the scanning sequence information and the image characteristic information of the positioning image. Such as: a coil information layer may be provided for distinguishing coil information of different scan targets. And establishing a scanning sequence information layer for recording the scanning sequence information of different scanning objects. And establishing an image characteristic information layer of the positioning image, which is used for recording the image characteristic information of the positioning images of different types of scanning objects. The multi-layer neural network unit can be set for the same part, and the multi-layer neural network unit can be set for the same characteristic information (coil information, scanning sequence information and image characteristic information of a scout image). In the whole neural network unit, the corresponding relation between the neural network units in different layers and the scanned position information is obtained through integration, analysis and calculation.
S540, setting function corresponding relations for the multi-layer neural network units, and adjusting parameters in each layer of neural network units according to multiple groups of training data to form the part recognition module.
It can be understood that the parameters in the function corresponding relation of the neural network unit need to be continuously updated in the process of establishing the part identification module, so that the identification process of the part identification module is simpler and more efficient. It can be understood that the more complete the training data set is used for training, the higher the positioning accuracy of the part recognition module is. After training of the neural network is completed, when a new positioning image of clinical scanning (scanning coil information and scanning sequence information) is input, the part recognition module automatically recognizes and processes the input positioning image according to the learned data, and finally provides corresponding part information.
The embodiment provides a training method based on a neural network for completing the training of a part recognition module. The training of the part identification module is completed through a plurality of groups of training data (including part information, positioning images of corresponding parts, coil information of corresponding parts and scanning sequence information of corresponding parts). In this embodiment, the training process of the neural network is implemented by using a computer program.
Referring to fig. 3, in an embodiment, the step of establishing a multi-layer neural network unit according to the coil information of the multiple sets of training data, the scanning sequence information of the multiple sets of training data, and the image feature information of the scout images of the multiple sets of training data specifically includes:
and S531, performing first-stage classification storage on the training data according to the image feature information of the positioning image to obtain preliminarily classified training data and at least one layer of neural network unit.
The first stage of classification storage may be understood as a preliminary partitioning of the scout image of the training data. For example, the human body feature information can be roughly divided into: head, cervical, spinal, chest, lung, heart, legs (thigh, calf), feet (ankle, toe), etc. When the part recognition module receives a plurality of groups of training data, the training data is subjected to first-stage classification storage. At the first stage of classified storage, at least one layer of neural network units can be arranged according to the complexity of the characteristic part.
And S532, performing secondary classification storage aiming at the coil information and the scanning sequence information in the obtained training data after the preliminary classification, establishing a multilayer neural network unit, and respectively storing the training data after the preliminary classification in the multilayer neural network unit.
In some cases, after the scout image is classified, the training data needs to be classified and stored for the second time according to the coil information and the scanning sequence information. In the second classification storage process, a plurality of layers of different neural network elements can also be arranged. It will be appreciated that some scan information parameters other than the coil information and the scan sequence information may also be stored as reference values for classification. It will also be appreciated that a three priority neural network classification scheme may be provided. The method comprises the steps of firstly setting a first-stage neural network unit for classified storage of a scout image, then setting a second-stage neural network unit for classified storage of scanning coil information, and finally establishing a third-stage neural network unit for classified storage of scanning sequence information. The priority of the neural network can be set according to the needs of the user, that is, the priority sequence can be set according to the needs of the user.
In this embodiment, training data is stored in a hierarchical manner, and a neural network unit with a priority is established. The multiple groups of training data are classified and stored, so that the part identification module can identify the part to be scanned more quickly and accurately. The learning and training process of the part identification module is safer and more reliable, the structural design is simpler, the identification speed is higher, and the identification efficiency is higher.
In one embodiment, the site identification module comprises:
and the image acquisition device is used for acquiring positioning images of a plurality of groups of training data.
The characteristic generator is used for acquiring coil information of a plurality of groups of training data, acquiring scanning sequence information of the plurality of groups of training data and extracting image characteristic information of positioning images of the plurality of groups of training data;
the multi-layer neural network unit is used for classifying and storing the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
and the synthesizer is used for setting function corresponding relation for the multilayer neural network units and adjusting parameters in each layer of the neural network units according to training data to form the part identification module.
It can be understood that the part identification module may be configured with different actual execution units or devices for completing different steps, implementing classification storage of multiple sets of training data, and establishing different function corresponding relationships. A part identification module is integrally established to realize the function of determining the part to be scanned of the scanning object by inputting information such as a positioning image, coil information, scanning sequence information and the like.
In this embodiment, the structures of the image acquisition device, the feature generator, the multilayer neural network unit, and the synthesizer are not limited, and the corresponding functions may be implemented. The specific image acquiring device is not limited to acquiring the positioning image, and may acquire other image information. The characteristic generator is used for generating some characteristic information capable of distinguishing the part to be scanned according to the coil information, the scanning sequence information, the positioning image and other parameter information of a plurality of groups of training data. The multi-layer neural network unit is used for classifying and storing the characteristic information generated by the characteristic generator. The multi-layer neural network unit can also establish different priorities according to different requirements. The synthesizer is used for establishing function corresponding relations among the formed multilayer neural network units and adjusting parameters among the function corresponding relations according to training data. Finally, the whole neural network determined according to the multiple groups of training data can quickly identify the part to be scanned.
Referring to fig. 4, in an embodiment, the training method of the portion identification module includes:
s510', positioning images of multiple groups of training data are obtained.
S520', coil information of a plurality of groups of training data is obtained, scanning sequence information of the plurality of groups of training data is obtained, and image characteristic information of positioning images of the plurality of groups of training data is extracted.
S530', different classification rules are established for the coil information of the multiple sets of training data, the scan sequence information of the multiple sets of training data, and the image feature information of the scout images of the multiple sets of training data.
And S540', setting different weight ratios among the different classification rules to form the part identification module.
In this embodiment, the training of the part identification module is completed by inputting multiple sets of training data by using a machine learning method of a support vector machine. A Support Vector Machine (SVM) establishes a specific Machine learning model (a part recognition module) according to limited sample information (multiple sets of training data). An optimal compromise is sought between the complexity of the model (i.e., the learning accuracy over sets of training data) and the learning ability (i.e., the ability to identify arbitrary samples without error) in anticipation of the best application and deployment.
The learning method of the support vector machine sets some feature information as a vector. The vectors are mapped into a higher dimensional space in which a maximally spaced hyperplane is established. Two hyperplanes are built parallel to each other on both sides of the hyperplane separating the data. Establishing a suitably oriented separation hyperplane maximizes the distance between two hyperplanes parallel thereto. It is assumed that the larger the distance or difference between the parallel hyperplanes is, the smaller the total error of the classifier is, and the higher the accuracy of the part identification module is.
Specifically, in the training process of the part recognition module, steps S510 'and S520' are similar to steps S510 and S520 in the training process using the neural network. Feature information including coil information, scan sequence information, and image feature information of a scout image is generated for a plurality of sets of training data.
In step S530', different classification rules are established for different coil information, scan sequence information, and image feature information of the scout image. Different classification rules may be established for different scan site information. Classification rules may be established for different parameters at different locations. The learned classification rules can be corrected to a certain degree in the process of machine learning according to a plurality of groups of training data, and the recognition and identification degrees of the part recognition module are higher due to continuous learning and training.
In step S540', different weight ratios are set between different classification rules to form a part recognition module. Different weight ratios are set among different classification rules, so that the part identification module can have pertinence when identifying the part to be scanned. The part identification module identifies the output information of the part to be scanned more accurately.
In one embodiment, in the step of establishing different classification rules for the coil information of the multiple sets of training data, the scanning sequence information of the multiple sets of training data, and the image feature information of the scout images of the multiple sets of training data, the classification rules are kernel functions with different corresponding relationships.
And setting different weight ratios among the different classification rules to form the part identification module, wherein the weight ratios are used for setting different parameters in the kernel function to adjust the classification rules.
After the kernel function is determined, because the known data for determining the kernel function has a certain error, two parameters, namely a relaxation coefficient and a penalty coefficient, are introduced to correct the error in consideration of the generalization problem. On the basis of determining the kernel function, the two coefficients are determined through a large number of comparison experiments and the like, the research is basically completed, and the method is suitable for application in related subjects or businesses and has certain popularization capability. The error is absolute, and the requirements of different disciplines and different specialties are different.
In this embodiment, a learning method of a support vector machine is used to train the part recognition module. In the specific training process, different kernel functions are adopted by the classification rules, and different parameters can be set in the different kernel functions according to the weight ratio of the parameters.
In one embodiment, the learning method using the support vector machine may also include:
and the image acquisition device is used for acquiring positioning images of a plurality of groups of training data.
And the characteristic generator is used for acquiring coil information and scanning sequence information of a plurality of groups of training data and extracting image characteristic information of a positioning image of the training data.
And the classification memory is used for establishing different classification rules for the different coil information, the scanning sequence information and the image characteristic information of the positioning image.
And the synthesizer is used for setting different weight ratios among different classification memories to form the part identification module.
It is understood that both machine learning methods using neural networks and machine learning methods using support vector machines can build the part recognition module. The structures of the image acquisition device, the feature generator, the classification memory, and the synthesizer are not limited, and the corresponding functions can be realized.
In one embodiment, in the step of optimizing subsequent parameters of the scanning object based on the part to be scanned, the subsequent parameters include any one of the following parameters: scan sequence parameters, image reconstruction parameters, and image processing parameters.
During the magnetic resonance scanning, the image is reconstructed and processed after the image scanning is completed. The parameters of the magnetic resonance equipment can be reset according to the scanning position information given by the position identification module so as to complete the whole clinical scanning process. It will be appreciated that parameters specifically optimizing subsequent scans include: optimization of system parameters of the magnetic resonance device, optimization of protocol parameters of the scout scan, optimization of scan sequence parameters of the magnetic resonance device, optimization of image reconstruction parameters of the scout scan, or optimization adjustment for image processing parameters. For example, when the protocol parameters of the positioning scan are optimized: the transmit voltage in the protocol parameters can be automatically adjusted using the site information or the K-space filter parameters to achieve optimal imaging quality. Specifically, among the subsequent parameters, the scan sequence parameter may be echo time (echo) and repetition time (repetition time) of the scan sequence. Among the subsequent parameters, the image reconstruction parameter may be a filling order, a filling trajectory, which fills the K-space data. Among the subsequent parameters, the image processing parameter may be a parameter for removing image noise and filtering.
In one embodiment, the step of optimizing subsequent parameters of the scan object based on the part to be scanned further comprises: and finishing the action of automatically optimizing the image or the action of automatically adjusting the window width and the window level according to the optimized subsequent parameters of the scanning object.
And finishing the action of automatically optimizing the image or the action of automatically adjusting the window width and the window level according to the optimized parameters of the subsequent scanning of the scanning object. All these requirements, which require special adjustments for different parts, are currently fulfilled without manual adjustment of protocol parameters. The automatic optimization image filter can be adjusted according to the detected parameters, the automatic part identification function is realized, and the manual workload can be greatly reduced. The output image with the automatically adjusted window width and level can also automatically adjust the window width and level of the output image according to the optimized parameters in the subsequent scanning process based on the scanning object.
To solve the above problem, there is also provided a magnetic resonance apparatus scanning system comprising a magnetic resonance scanning apparatus and a computer, wherein the computer comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor when executing the program is operable to perform a magnetic resonance apparatus scanning method, the method comprising:
acquiring a positioning image of a scanning object through positioning scanning;
acquiring coil information and scanning sequence information of clinical scanning to be executed;
based on the trained part recognition module, the positioning image, the coil information and the scanning sequence information are used as the input of the part recognition module to determine the part to be scanned of the scanning object;
optimizing the clinical scan to be performed based on the site to be scanned.
To solve the above problem, there is also provided a computer readable storage medium having stored thereon a computer program which, when being executed by a processor, is adapted to carry out the steps of the method of any of the above.
The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (11)
1. A magnetic resonance apparatus scanning method, characterized by comprising the steps of: acquiring a positioning image of a scanning object through positioning scanning;
acquiring coil information of the positioning scanning and scanning sequence information of the positioning scanning;
the positioning image, the coil information and the scanning sequence information are used as the input of a part identification module to determine the part to be scanned of the scanning object;
optimizing subsequent parameters of the scanning object based on the part to be scanned;
the part recognition module is trained in advance by the following method: acquiring positioning images of a plurality of groups of training data;
acquiring coil information of a plurality of groups of training data, acquiring scanning sequence information of the plurality of groups of training data, and extracting image characteristic information of positioning images of the plurality of groups of training data;
establishing a multi-layer neural network unit according to the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
and setting function corresponding relations for the multi-layer neural network units, and adjusting parameters in each layer of neural network units according to multiple groups of training data to form the part identification module.
2. The method according to claim 1, wherein the step of building a multi-layer neural network unit according to the coil information of the plurality of sets of training data, the scanning sequence information of the plurality of sets of training data, and the image feature information of the scout images of the plurality of sets of training data specifically comprises: performing first-stage classification storage on training data according to image characteristic information of the positioning image to obtain training data after preliminary classification and at least one layer of neural network units;
and performing secondary classification storage on the coil information and the scanning sequence information in the obtained training data after the preliminary classification, establishing a multilayer neural network unit, and respectively storing the training data after the preliminary classification in the multilayer neural network unit.
3. The method as claimed in claim 1, wherein the coil information of the scout scan and the scan sequence information of the scout scan are manually input information or automatically generated information by the magnetic resonance apparatus according to the related data of the scanned object.
4. The magnetic resonance apparatus scanning method as set forth in claim 1, wherein the site identification module includes: the image acquisition device is used for acquiring positioning images of a plurality of groups of training data;
the characteristic generator is used for acquiring coil information of a plurality of groups of training data, acquiring scanning sequence information of the plurality of groups of training data and extracting image characteristic information of positioning images of the plurality of groups of training data;
the multi-layer neural network unit is used for classifying and storing the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
and the synthesizer is used for setting function corresponding relation for the multilayer neural network units and adjusting parameters in each layer of the neural network units according to training data to form the part identification module.
5. A magnetic resonance apparatus scanning method, characterized by comprising the steps of: acquiring a positioning image of a scanning object through positioning scanning;
acquiring coil information of the positioning scanning and scanning sequence information of the positioning scanning;
the positioning image, the coil information and the scanning sequence information are used as the input of a part identification module to determine the part to be scanned of the scanning object;
optimizing subsequent parameters of the scanning object based on the part to be scanned;
the training method of the part recognition module comprises the following steps: acquiring positioning images of a plurality of groups of training data;
acquiring coil information of a plurality of groups of training data, acquiring scanning sequence information of the plurality of groups of training data, and extracting image characteristic information of positioning images of the plurality of groups of training data;
establishing different classification rules aiming at the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
setting different weight ratios among the different classification rules to form the part recognition module.
6. The magnetic resonance apparatus scanning method as set forth in claim 5, wherein in the step of establishing different classification rules for the coil information of the plurality of sets of training data, the scanning sequence information of the plurality of sets of training data, and the image feature information of the scout images of the plurality of sets of training data, the classification rules are kernel functions having different correspondences;
and setting different weight ratios among the different classification rules to form the part identification module, wherein the weight ratios are used for setting different parameters in the kernel function to adjust the classification rules.
7. The magnetic resonance apparatus scanning method as set forth in claim 5, wherein the site identification module includes:
the image acquisition device is used for acquiring positioning images of a plurality of groups of training data;
the characteristic generator is used for acquiring coil information of a plurality of groups of training data and scanning sequence information of a plurality of groups of training data and extracting image characteristic information of a positioning image of the training data;
the classification memory is used for establishing different classification rules for the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
and the synthesizer is used for setting different weight ratios among different classification memories to form the part identification module.
8. The method as claimed in claim 1, wherein the step of optimizing subsequent parameters of the scanned object based on the portion to be scanned comprises any one of the following parameters: scanning sequence parameters, filling sequence for filling K space data, parameters for removing image noise and parameters for filtering.
9. The method as set forth in claim 8, wherein the step of optimizing subsequent parameters of the scanned object based on the region to be scanned further comprises: and finishing the action of automatically optimizing the image according to the optimized subsequent parameters of the scanning object.
10. A magnetic resonance apparatus scanning system comprising a magnetic resonance scanning apparatus and a computer, wherein the computer comprises a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the program when executed by the processor is operable to perform a magnetic resonance apparatus scanning method, the method comprising: acquiring a positioning image of a scanning object through positioning scanning;
acquiring coil information of the positioning scanning and scanning sequence information of the positioning scanning;
based on the trained part recognition module, the positioning image, the coil information and the scanning sequence information are used as the input of the part recognition module to determine the part to be scanned of the scanning object;
optimizing parameters of subsequent scanning of the scanning object based on the part to be scanned;
the part recognition module includes:
an image acquisition unit for acquiring positioning images of a plurality of groups of training data;
an information acquisition unit for acquiring coil information of a plurality of groups of training data and acquiring scanning sequence information of the plurality of groups of training data;
an image characteristic information extraction unit for extracting the image characteristic information of the scout images of the plurality of groups of training data;
the multilayer neural network unit is established according to the coil information of the multiple groups of training data, the scanning sequence information of the multiple groups of training data and the image characteristic information of the positioning images of the multiple groups of training data;
a function operation unit for setting function corresponding relation to the multilayer neural network unit; and the function operation unit is used for adjusting parameters in each layer of the neural network unit according to a plurality of groups of training data to form the part identification module.
11. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, is adapted to carry out the steps of the method of any one of claims 1 to 9.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711085890.0A CN107669273B (en) | 2017-11-07 | 2017-11-07 | Magnetic resonance equipment scanning system, scanning method and computer readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711085890.0A CN107669273B (en) | 2017-11-07 | 2017-11-07 | Magnetic resonance equipment scanning system, scanning method and computer readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107669273A CN107669273A (en) | 2018-02-09 |
CN107669273B true CN107669273B (en) | 2021-02-19 |
Family
ID=61146003
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711085890.0A Active CN107669273B (en) | 2017-11-07 | 2017-11-07 | Magnetic resonance equipment scanning system, scanning method and computer readable storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107669273B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108968960A (en) * | 2018-06-27 | 2018-12-11 | 上海联影医疗科技有限公司 | Localization method and magnetic resonance system for magnetic resonance system |
CN109124635B (en) * | 2018-09-25 | 2022-09-02 | 上海联影医疗科技股份有限公司 | Model generation method, magnetic resonance imaging scanning method and system |
CN111374690A (en) * | 2018-12-28 | 2020-07-07 | 通用电气公司 | Medical imaging method and system |
CN109709503B (en) * | 2019-02-13 | 2021-03-23 | 上海联影医疗科技股份有限公司 | Magnetic resonance system control method, magnetic resonance system and computer equipment |
CN112820383B (en) * | 2019-11-15 | 2023-04-28 | 上海联影医疗科技股份有限公司 | Medical imaging method, system and storage medium |
CN112823741A (en) * | 2019-11-20 | 2021-05-21 | 上海联影医疗科技股份有限公司 | Magnetic resonance scanning method and magnetic resonance system |
CN110960241A (en) * | 2019-12-09 | 2020-04-07 | 上海联影医疗科技有限公司 | Method and device for determining scanning parameters of medical image scanning and computer equipment |
CN113075599B (en) * | 2020-01-03 | 2023-05-16 | 上海联影医疗科技股份有限公司 | Magnetic resonance signal acquisition method, magnetic resonance system and medium |
CN115951282A (en) * | 2020-02-28 | 2023-04-11 | 上海联影医疗科技股份有限公司 | Magnetic resonance system |
CN113842222B (en) * | 2021-09-24 | 2024-02-27 | 深圳市联影高端医疗装备创新研究院 | Storage cabinet structure for medical device |
CN114913383B (en) * | 2022-06-24 | 2023-06-30 | 北京赛迈特锐医疗科技有限公司 | Model training method for identifying image sequence type and method for configuring image equipment |
CN115919285A (en) * | 2023-02-28 | 2023-04-07 | 山东奥新医疗科技有限公司 | Nuclear magnetic resonance positioning method, device, equipment and storage medium |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102004026616B4 (en) * | 2004-06-01 | 2007-09-20 | Siemens Ag | Method for measuring an examination area with a magnetic resonance apparatus |
CN100587512C (en) * | 2006-08-03 | 2010-02-03 | 西门子(中国)有限公司 | Method and device for recognizing radiofrequency signal receiver coil in MRI system |
CN102440777B (en) * | 2010-10-12 | 2013-09-04 | 深圳迈瑞生物医疗电子股份有限公司 | Method and equipment for scanning control in magnetic resonance scanning room |
DE102013210855A1 (en) * | 2013-06-11 | 2014-12-11 | Siemens Aktiengesellschaft | A method for adjusting a slice position within a slice protocol for a magnetic resonance examination and a magnetic resonance apparatus for carrying out the method |
CN107292257A (en) * | 2017-06-14 | 2017-10-24 | 深圳先进技术研究院 | Body part automatic identification magnetic resonance scanning method and device based on deep learning |
CN107273885A (en) * | 2017-06-30 | 2017-10-20 | 上海联影医疗科技有限公司 | A kind of method that scanning area is automatically determined based on positioning image |
-
2017
- 2017-11-07 CN CN201711085890.0A patent/CN107669273B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN107669273A (en) | 2018-02-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107669273B (en) | Magnetic resonance equipment scanning system, scanning method and computer readable storage medium | |
US10417762B2 (en) | Matching patient images and images of an anatomical atlas | |
Onofrey et al. | Generalizable multi-site training and testing of deep neural networks using image normalization | |
JP6567179B2 (en) | Pseudo CT generation from MR data using feature regression model | |
KR101883258B1 (en) | Detection of anatomical landmarks | |
JP2018535732A (en) | Pseudo CT generation from MR data using tissue parameter estimation | |
US9974490B2 (en) | Method and device for segmenting a medical examination object with quantitative magnetic resonance imaging | |
US9406130B2 (en) | Determining an anatomical atlas | |
CN110310723A (en) | Bone image processing method, electronic equipment and storage medium | |
Mukherjee et al. | Automatic segmentation of spinal cord MRI using symmetric boundary tracing | |
CN109350059A (en) | For ancon self-aligning combined steering engine and boundary mark engine | |
CN106557767A (en) | A kind of method of ROI region in determination interventional imaging | |
Kumar et al. | Medical image fusion based on type-2 fuzzy sets with teaching learning based optimization. | |
CN110276772B (en) | Automatic positioning method and system for structural elements in muscle tissue | |
CN108877922A (en) | Lesion degree judges system and method | |
CN111640127A (en) | Accurate clinical diagnosis navigation method for orthopedics department | |
KR20230050253A (en) | Method for detecting pleurl effusion and the apparatus for therof | |
Cheng et al. | Low rank self-calibrated brain network estimation and autoweighted centralized multi-task learning for early mild cognitive impairment diagnosis | |
US20240122565A1 (en) | Determination system, method of controlling determination system, and control program | |
Mejía-Rodríguez et al. | 3D Kidney Reconstruction from 2D Ultrasound Images | |
CN117115288A (en) | Magnetic resonance scanning scheme generation method, device, electronic device and storage medium | |
EP2912632A1 (en) | Determining an anatomical atlas | |
CN115917349A (en) | Automatic detection of critical stations in multi-station magnetic resonance imaging | |
KR20230049937A (en) | Method for detecting pleurl effusion and the apparatus for therof | |
CN114998297A (en) | Vertebra numbering method and device, electronic device and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information |
Address after: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant after: Shanghai Lianying Medical Technology Co., Ltd Address before: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant before: SHANGHAI UNITED IMAGING HEALTHCARE Co.,Ltd. |
|
CB02 | Change of applicant information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |