CN113591860A - Image classification method and device and storage medium - Google Patents
Image classification method and device and storage medium Download PDFInfo
- Publication number
- CN113591860A CN113591860A CN202110705690.0A CN202110705690A CN113591860A CN 113591860 A CN113591860 A CN 113591860A CN 202110705690 A CN202110705690 A CN 202110705690A CN 113591860 A CN113591860 A CN 113591860A
- Authority
- CN
- China
- Prior art keywords
- matrix
- feature
- brain
- magnetic resonance
- matrixes
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
- G06F18/2115—Selection of the most significant subset of features by evaluating different subsets according to an optimisation criterion, e.g. class separability, forward selection or backward elimination
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses an image classification method, an image classification device and a storage medium, wherein the method comprises the following steps: acquiring magnetic resonance imaging and a plurality of preset brain area templates, wherein the division rules of human brain areas corresponding to the brain area templates are different; determining a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, wherein each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates; screening the feature vectors in the feature matrixes to obtain a first optimized feature matrix; and determining an image prediction category corresponding to the magnetic resonance imaging according to the first optimization feature matrix. Can solve the problem that doctors in the prior art mainly rely on the clinical score and the movement problem of people to diagnose the neurodegenerative diseases and are difficult to distinguish the neurodegenerative disease people from normal people.
Description
Technical Field
The present invention relates to the field of image processing, and in particular, to an image classification method, apparatus, and storage medium.
Background
Currently, as the global population ages, the number of global populations of neurodegenerative diseases is continuously increasing, and the diseases are receiving global attention. Physicians now rely primarily on the clinical scores and motor problems of the population to diagnose neurodegenerative diseases, while neurons of the brain have degenerated before cognitive impairment and motor problems occur. Therefore, in the early stage of neurodegenerative disease, i.e., when typical clinical symptoms have not appeared yet, it is difficult for a doctor to see the degeneration of cerebral neurons from the brain images of the population, so that it is difficult for the doctor to distinguish the neurodegenerative disease population from the normal population according to the clinical symptoms of the population.
Thus, there is still a need for improvement and development of the prior art.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide an image classification method, device and storage medium for solving the above-mentioned drawbacks of the prior art, and to solve the problem in the prior art that it is difficult for a doctor to distinguish a neurodegenerative disease population from a normal population by mainly relying on clinical scores and motion problems of the population to diagnose the neurodegenerative disease.
The technical scheme adopted by the invention for solving the problems is as follows:
in a first aspect, an embodiment of the present invention provides an image classification method, where the method includes:
acquiring magnetic resonance imaging and a plurality of preset brain area templates, wherein the division rules of human brain areas corresponding to the brain area templates are different;
determining a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, wherein each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates;
screening the feature vectors in the feature matrixes to obtain a first optimized feature matrix;
and determining an image prediction category corresponding to the magnetic resonance imaging according to the first optimization feature matrix.
In one embodiment, the determining a plurality of feature matrices from the plurality of brain region templates and the magnetic resonance imaging comprises:
performing image segmentation on the magnetic resonance imaging to obtain a plurality of first partial images, wherein the brain tissue types corresponding to the plurality of first partial images are different;
respectively carrying out region division on the plurality of first local images according to each brain area template to obtain a plurality of first divided images corresponding to each brain area template;
and determining a feature matrix corresponding to each brain area template according to the first divided images corresponding to each brain area template to obtain a plurality of feature matrices.
In one embodiment, the determining a feature matrix corresponding to each of the brain region templates according to a plurality of first divided images corresponding to each of the brain region templates includes:
respectively extracting features of the first divided images corresponding to each brain area template to obtain a plurality of feature vectors corresponding to each brain area template;
and determining a feature matrix corresponding to each brain area template according to a plurality of feature vectors corresponding to each brain area template.
In an embodiment, the screening the feature vectors in the feature matrices to obtain a first optimized feature matrix includes:
acquiring a plurality of relation mapping matrixes and a weight matrix, wherein the relation mapping matrixes correspond to the brain area templates one by one;
carrying out sparse constraint on the plurality of relational mapping matrixes to obtain a plurality of sparse relational mapping matrixes;
mapping the feature matrixes into a public relation pool according to the sparse relation mapping matrixes to obtain a public feature matrix, wherein the sparse relation mapping matrixes correspond to the feature matrixes one by one, and one sparse relation mapping matrix and one feature matrix which have corresponding relations correspond to the same brain region template;
carrying out sparse constraint on the weight matrix to obtain a sparse weight matrix;
and multiplying the sparse weight matrix and the public feature matrix to obtain the first optimized feature matrix.
In one embodiment, the determining the image prediction category corresponding to the magnetic resonance imaging according to the first optimized feature matrix includes:
acquiring first clinical information corresponding to the magnetic resonance imaging, and constructing a first newly-added feature matrix according to the first clinical information;
constructing a first target feature matrix according to the first newly-added feature matrix and the first optimized feature matrix;
and inputting the first target characteristic matrix into a multi-classification model to obtain the image prediction category.
In one embodiment, the relational mapping matrices and the weight matrices are pre-trained matrices, and the training process of the relational mapping matrices and the weight matrices is as follows:
acquiring a nuclear magnetic resonance training image, and determining a plurality of training feature matrixes which correspond to the brain area templates one by one according to the brain area templates and the nuclear magnetic resonance training image;
acquiring a plurality of initial relation mapping matrixes, wherein the plurality of initial relation mapping matrixes correspond to the plurality of brain area templates one by one;
carrying out sparse constraint on the plurality of initial relation mapping matrixes to obtain a plurality of initial sparse relation mapping matrixes;
mapping the training feature matrixes into the public relation pool according to the initial sparse relation mapping matrixes to obtain a public training feature matrix;
acquiring an initial weight matrix and a real label corresponding to the nuclear magnetic resonance training image;
obtaining a preset function, determining variables in the preset function according to the initial weight matrix, the real label, the public training matrix and the initial relation mapping matrices to obtain a target function, and determining the relation mapping matrices and the weight matrix according to the target function.
In an embodiment, the determining variables in the preset function according to the initial weight matrix, the real label, the common training matrix, and the initial relationship mapping matrices to obtain an objective function, and determining the relationship mapping matrices and the weight matrix according to the objective function includes:
obtaining the relation between matrixes in the public training characteristic matrix to obtain first relation information;
carrying out sparse constraint on the initial weight matrix to obtain an initial sparse weight matrix;
multiplying the initial sparse weight matrix and the public training feature matrix to obtain a sparse training feature matrix;
obtaining the relation between vectors in the sparse training feature matrix to obtain second relation information;
taking the initial relationship mapping matrixes, the public relationship pool, the first relationship information, the second relationship information, the real label and the initial weight matrix as variables of the preset function to obtain the target function;
and deriving the objective function to obtain a plurality of relation mapping matrixes and the weight matrix, wherein the function value of the objective function corresponding to the relation mapping matrixes and the weight matrix is minimum.
In one embodiment, the multi-classification model is a model that is trained in advance, and the training process of the multi-classification model is as follows:
determining a second optimized feature matrix corresponding to the nuclear magnetic resonance training image according to the relation mapping matrix and the weight matrix;
acquiring second clinical information corresponding to the nuclear magnetic resonance training image, and constructing a second newly added feature matrix according to the second clinical information;
constructing a second target feature matrix according to the second optimized feature matrix and the second newly added feature matrix;
and acquiring an initial multi-classification model, and training the initial multi-classification model according to the second target feature matrix and the real label to obtain the multi-classification model.
In a second aspect, an embodiment of the present invention further provides an image classification apparatus, where the apparatus includes:
the acquisition module is used for acquiring magnetic resonance imaging and a plurality of preset brain area templates, and the division rules of human brain areas corresponding to the brain area templates are different;
a determining module, configured to determine a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, where each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates;
the screening module is used for screening the eigenvectors in the plurality of characteristic matrixes to obtain a first optimized characteristic matrix;
and the prediction module is used for determining the image prediction category corresponding to the magnetic resonance imaging according to the first optimization characteristic matrix.
In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a plurality of instructions are stored, where the instructions are adapted to be loaded and executed by a processor to implement any of the steps of the image classification method described above.
The invention has the beneficial effects that: the embodiment of the invention obtains magnetic resonance imaging and a plurality of preset brain area templates, wherein the division rules of human brain areas respectively corresponding to the brain area templates are different; determining a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, wherein each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates; screening the feature vectors in the feature matrixes to obtain a first optimized feature matrix; and determining an image prediction category corresponding to the magnetic resonance imaging according to the first optimization feature matrix. The invention predicts the categories of the magnetic resonance imaging shot by different people by a method of extracting image characteristics, and doctors can more accurately distinguish normal people and neurodegenerative disease people by predicting the categories of the images corresponding to the magnetic resonance imaging and combining self medical experience. Therefore, the invention can solve the problem that doctors in the prior art mainly rely on the clinical score and the movement problem of people to diagnose the neurodegenerative diseases and are difficult to distinguish the neurodegenerative disease people from normal people.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flowchart of an image classification method according to an embodiment of the present invention.
Fig. 2 is an overall technical flowchart of an image classification method according to an embodiment of the present invention.
Fig. 3 is a connection diagram of internal modules of the image classification apparatus according to the embodiment of the present invention.
Fig. 4 is a schematic block diagram of a terminal according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
It should be noted that, if directional indications (such as up, down, left, right, front, and back … …) are involved in the embodiment of the present invention, the directional indications are only used to explain the relative positional relationship between the components, the movement situation, and the like in a specific posture (as shown in the drawing), and if the specific posture is changed, the directional indications are changed accordingly.
Neurodegenerative diseases are a group of diseases caused by gradual impairment of the function or structure of neurons, including alzheimer's disease, parkinson's disease, and the like. Alzheimer's disease is the most common neurodegenerative disease in the world, the major pathological changes of which are synaptic and neuronal damage to the subcortical structures and cortex of the brain. The most common early symptom of alzheimer's disease is short-term memory loss. As the disease progresses, symptoms may develop, including language disorders, disorientation, emotional instability, and loss of motivation. Parkinson's disease is the second most common neurodegenerative disease in the world. The main pathological changes of parkinson's disease are degenerative death of the mesencephalic substantia nigra dopaminergic neurons, a significant reduction of the striatal dopamine content in the brain. The main clinical manifestations of parkinson's disease include motor symptoms (tremor, bradykinesia, myotonia, gait instability) and non-motor symptoms (depression, sleep disorders, olfactory failure, cognitive disorders). Both neurodegenerative diseases are irreversible and occur primarily in the elderly, and as the disease progresses, the symptoms become more severe.
Currently, as the global population ages, the number of global populations of neurodegenerative diseases is continuously increasing, and the diseases are receiving global attention. Physicians now rely primarily on the clinical scores and motor problems of the population to diagnose neurodegenerative diseases, while neurons of the brain have degenerated before cognitive impairment and motor problems occur. Therefore, in the early stage of neurodegenerative disease, i.e., when typical clinical symptoms have not appeared yet, it is difficult for a doctor to see the degeneration of cerebral neurons from the brain images of the population, so that it is difficult for the doctor to distinguish the neurodegenerative disease population from the normal population according to the clinical symptoms of the population.
Aiming at the defects in the prior art, the invention provides an image classification method, which comprises the steps of obtaining magnetic resonance imaging and a plurality of preset brain area templates, wherein the brain area templates respectively correspond to different human brain area division rules; determining a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, wherein each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates; screening the feature vectors in the feature matrixes to obtain a first optimized feature matrix; and determining an image prediction category corresponding to the magnetic resonance imaging according to the first optimization feature matrix. The invention predicts the categories of the magnetic resonance imaging shot by different people by a method of extracting image characteristics, and doctors can more accurately distinguish normal people and neurodegenerative disease people by predicting the categories of the images corresponding to the magnetic resonance imaging and combining self medical experience. Therefore, the invention can solve the problem that doctors in the prior art mainly rely on the clinical score and the movement problem of people to diagnose the neurodegenerative diseases and are difficult to distinguish the neurodegenerative disease people from normal people.
As shown in fig. 1, the method comprises the steps of:
s100, acquiring magnetic resonance imaging and a plurality of preset brain area templates, wherein the division rules of human brain areas corresponding to the brain area templates are different.
With the development of imaging technology and machine learning, machine learning is widely used for analysis of neuroimages. The most widely used brain image data at this stage is Magnetic Resonance Imaging (MRI), because it can effectively reveal abnormalities in the structure of the human brain. In addition, because the brain has complex and various structures, it is difficult to accurately extract the image features of the magnetic resonance imaging of each user by dividing the magnetic resonance imaging by using a single brain region template, thereby reducing the accuracy of classification. Therefore, in the embodiment, a plurality of brain region templates are preset, and the brain region templates respectively correspond to different human brain region division rules.
In one implementation, the method further comprises the steps of:
step S200, determining a plurality of feature matrices according to the plurality of brain area templates and the magnetic resonance imaging, wherein each feature matrix in the plurality of feature matrices is determined by one brain area template in the plurality of brain area templates.
Specifically, because the brain region templates respectively correspond to different human brain region division rules, different feature matrices corresponding to magnetic resonance imaging can be extracted according to the brain region templates. Compared with the method that a single brain region template is adopted to divide a single brain tissue image to determine the image prediction category corresponding to the magnetic resonance imaging, the method that each brain region template is adopted to extract a feature matrix corresponding to the magnetic resonance imaging to obtain a plurality of feature matrices and perform comprehensive analysis can determine the image prediction category corresponding to the magnetic resonance imaging more accurately. For example, as shown in fig. 2, the plurality of brain region templates may be a 90 brain region template, a 116 brain region template, and a 200 brain region template, and the sizes and the numbers of the human brain regions respectively divided based on the three brain region templates are all different, so the dimensions of the image features of the magnetic resonance imaging extracted based on the three brain region templates are also different.
In one implementation, the determining a plurality of feature matrices from the plurality of brain region templates and the magnetic resonance imaging includes:
step S201, performing image segmentation on the magnetic resonance imaging to obtain a plurality of first partial images, wherein the brain tissue types corresponding to the plurality of first partial images are different;
step S202, respectively carrying out region division on the plurality of first local images according to each brain area template to obtain a plurality of first divided images corresponding to each brain area template;
step S203, determining a feature matrix corresponding to each brain area template according to a plurality of first divided images corresponding to each brain area template to obtain a plurality of feature matrices.
Specifically, since different tissues of the human brain change differently when the neurodegenerative disease occurs, in order to accurately determine the image prediction type corresponding to the magnetic resonance imaging, the present embodiment first needs to perform image segmentation on the magnetic resonance imaging to obtain the first partial images, i.e., the first partial images, corresponding to different brain tissue types. Then, for each first local image, each brain area template in the plurality of brain area templates is adopted to perform region division on the first local image, so that a plurality of first divided images corresponding to each brain area template are obtained. In addition, for each brain region template, a corresponding feature matrix is determined comprehensively according to a plurality of corresponding first divided images, so that a feature matrix corresponding to each brain region template, namely the plurality of feature matrices, is obtained.
For example, as shown in fig. 2, grey and white matter are important in assessing various stages of neurodegenerative disease because they are vital components of the human brain and are fragile. While the cerebrospinal fluid may indicate a pathological state of the human brain as a fluid around the spinal cord and brain. Therefore, in the present embodiment, a Statistical Parameter Mapping (SPM) toolbox is used to divide the magnetic resonance imaging into three first partial images, and the corresponding brain tissue types are gray matter, white matter and cerebrospinal respectively, wherein for the first partial image corresponding to the gray matter, a 90 brain region template is used to perform region division on the first partial image to obtain a first divided image a; carrying out region division on the image by adopting a 116 brain region template to obtain a first division image B; and performing region division on the image by adopting a 200-brain region template to obtain a first division image C. Aiming at a first local image corresponding to white matter, carrying out region division on the first local image by adopting a 90 brain region template to obtain a first division image D; carrying out region division on the image by adopting a 116 brain region template to obtain a first division image E; and performing region division on the image by adopting a 200-brain region template to obtain a first division image F. Aiming at a first local image corresponding to the cerebrospinal, performing region division on the first local image by adopting a 90 brain region template to obtain a first division image G; carrying out region division on the image by adopting a 116 brain region template to obtain a first division image H; and performing region division on the image by adopting a 200-brain region template to obtain a first division image I. The first division image A, the first division image D and the first division image G correspond to the 90 brain area template, so that a feature matrix 1 corresponding to the 90 brain area template is determined according to A, D, G; the first division image B, the first division image E and the first division image H correspond to the 116 brain area template, so that the feature matrix 2 corresponding to the 116 brain area template is determined according to B, E, H; the first divided image C, the first divided image F, and the first divided image I correspond to the 200 brain region template, and therefore the feature matrix 3 corresponding to the 200 brain region template is determined from C, F, I.
In one implementation, the determining, according to a plurality of first divided images corresponding to each brain region template, a feature matrix corresponding to each brain region template includes: respectively extracting features of the first divided images corresponding to each brain area template to obtain a plurality of feature vectors corresponding to each brain area template; and determining a feature matrix corresponding to each brain area template according to a plurality of feature vectors corresponding to each brain area template.
Specifically, in order to obtain a feature matrix corresponding to each brain region template in the plurality of brain region templates, for each brain region template, in this embodiment, feature extraction needs to be performed on a plurality of first divided images corresponding to the brain region template to obtain a plurality of feature vectors, where each feature vector in the plurality of feature vectors is determined by one first divided image in the plurality of first divided images. And then, comprehensively determining a feature matrix corresponding to the brain area template according to the feature vectors corresponding to the first divided images.
For example, for a 90 brain region template, feature extraction is performed on a first divided image a, a first divided image D, and a first divided image G corresponding to the 90 brain region template, so as to obtain feature vectors a, D, and G corresponding to the first divided image a, the first divided image D, and the first divided image G, respectively. For example, for the first divided image a, the average tissue density value of each region in the first divided image a may be obtained, and the feature vector a corresponding to the first divided image a may be determined according to the average tissue density value of each region.
As shown in fig. 1, the method further comprises the steps of:
and S300, screening the eigenvectors in the plurality of characteristic matrixes to obtain a first optimized characteristic matrix.
Specifically, the number of feature vectors contained in the feature matrices is large, wherein the features indicated by some feature vectors are discriminant, that is, the feature vectors are significantly different in the normal population and in the neurodegenerative disease attack population; there is also the additional presence that a part of the feature vectors indicate features that are not discriminative, i.e. that the feature vectors differ only little in the normal population and in the neurodegenerative disease-affected population. Therefore, in order to reduce the consumption of computing resources and accurately determine the image prediction category corresponding to the magnetic resonance imaging, in this embodiment, feature vectors in a plurality of feature matrices need to be screened to obtain a first optimized feature matrix, and all feature vectors existing in the first optimized feature matrix basically have discriminability.
In one implementation, the step S300 specifically includes the following steps:
s301, obtaining a plurality of relation mapping matrixes and a plurality of weight matrixes, wherein the relation mapping matrixes correspond to the brain area templates one by one;
step S302, carrying out sparse constraint on the plurality of relational mapping matrixes to obtain a plurality of sparse relational mapping matrixes;
step S303, mapping the plurality of feature matrices into a public relation pool according to the plurality of sparse relation mapping matrices to obtain a public feature matrix, wherein the plurality of sparse relation mapping matrices correspond to the plurality of feature matrices one to one, and one sparse relation mapping matrix and one feature matrix with corresponding relations correspond to the same brain region template;
s304, carrying out sparse constraint on the weight matrix to obtain a sparse weight matrix;
step S305, multiplying the sparse weight matrix and the public feature matrix to obtain the first optimized feature matrix.
Specifically, in the present embodiment, a plurality of relational mapping matrices are trained in advance, and these relational mapping matrices correspond to a plurality of brain region templates one to one, that is, each brain region template has a trained relational mapping matrix corresponding thereto. And aiming at each characteristic matrix in the plurality of characteristic matrices, acquiring a relational mapping matrix corresponding to the same brain area template with the characteristic matrix. In this embodiment, sparse constraint needs to be performed on the plurality of relational mapping matrices to obtain a plurality of sparse relational mapping matrices. Since the sparse constraint can change the dimensionality of part of the feature vectors into 0, for each feature matrix, the feature matrix is mapped into the public relation pool according to the sparse relation mapping matrix corresponding to the feature matrix, so that the mapping matrix corresponding to the feature matrix can be obtained, and the number of the feature vectors of the matrix is smaller than that of the feature matrix, thereby realizing the first screening. And when the plurality of feature matrices are mapped to the public relation pool, a plurality of mapping matrices can be obtained, and the mapping matrices jointly form a public feature matrix. In this embodiment, a weight matrix is trained in advance, sparse constraint is performed on the weight matrix to obtain a sparse weight matrix, and then the sparse weight matrix is multiplied by the common feature matrix, so that feature vectors in the common feature matrix can be screened for the second time, and a first optimized feature matrix is obtained.
As shown in fig. 1, the method further comprises the steps of:
and S400, determining an image prediction category corresponding to the magnetic resonance imaging according to the first optimization feature matrix.
Since the first optimized feature matrix is the screened feature vector with discriminant, the embodiment can analyze the first optimized feature matrix and determine the image prediction category corresponding to the magnetic resonance imaging according to the analysis result.
In one implementation, the step S400 specifically includes the following steps:
step S401, acquiring first clinical information corresponding to the magnetic resonance imaging, and constructing a first newly-added feature matrix according to the first clinical information;
step S402, constructing a first target feature matrix according to the first newly-added feature matrix and the first optimized feature matrix;
and S403, inputting the first target feature matrix into a multi-classification model to obtain the image prediction category.
Specifically, the present embodiment trains a multi-classification model in advance, that is, the image class output by the model may be multiple classes. In practical applications, in order to determine the image prediction category corresponding to the magnetic resonance imaging more accurately, the present embodiment needs to combine clinical information of the patient corresponding to the magnetic resonance imaging, that is, the first clinical information. And constructing a first newly-added feature matrix according to the first clinical information, and then fusing the first newly-added feature matrix and the first optimized feature matrix to obtain a first target feature matrix. And after the trained multi-classification model acquires the first target characteristic matrix, outputting the image prediction category corresponding to the magnetic resonance image.
For example, after the first optimized feature matrix is obtained, a first newly added feature matrix is constructed by combining clinical scores (depression, sleep, smell, cognition) and age and gender information of a patient corresponding to the magnetic resonance image, then a first target feature matrix is generated according to the first newly added feature matrix and the first optimized feature matrix, the first target feature matrix is input into a multi-classification model, and an output image of the multi-classification model predicts that the category is the stage I of the neurodegenerative disease, so that a doctor is prompted that the patient corresponding to the magnetic resonance image has a disease risk of the stage I of the neurodegenerative disease, and needs to further determine whether the patient is ill according to other medical tests.
In one implementation, the training process of the relationship mapping matrices and the weight matrix is as follows:
step S1, acquiring a nuclear magnetic resonance training image, and determining a plurality of training feature matrixes corresponding to the brain area templates one by one according to the brain area templates and the nuclear magnetic resonance training image;
step S2, obtaining a plurality of initial relation mapping matrixes, wherein the plurality of initial relation mapping matrixes correspond to the plurality of brain area templates one to one;
s3, carrying out sparse constraint on the plurality of initial relation mapping matrixes to obtain a plurality of initial sparse relation mapping matrixes;
step S4, mapping the training feature matrixes to the public relation pool according to the initial sparse relation mapping matrixes to obtain a public training feature matrix;
step S5, acquiring an initial weight matrix and a real label corresponding to the nuclear magnetic resonance training image;
step S6, obtaining a preset function, determining variables in the preset function according to the initial weight matrix, the real label, the public training matrix and the initial relation mapping matrixes to obtain a target function, and determining the relation mapping matrixes and the weight matrix according to the target function.
Specifically, the respective training processes of the plurality of relational mapping matrices and the weight matrices are similar to the respective application processes, and the difference is that the training process adopts a nuclear magnetic resonance training image, and the nuclear magnetic resonance training image has a real label corresponding to the nuclear magnetic resonance training image, and the real label is used for reflecting the real image category corresponding to the nuclear magnetic resonance training image. In the training process, the embodiment provides a preset function, and substitutes the initial weight matrix, the real label, the public training matrix and the plurality of initial relation mapping matrices into the preset function to obtain the target function. The output value of the objective function can be used for guiding the adjustment of the plurality of initial relation mapping matrixes and the initial weight matrixes, and the plurality of relation mapping matrixes and the weight matrixes which can be directly applied can be obtained after the adjustment is finished.
In an implementation manner, the determining variables in the preset function according to the initial weight matrix, the real label, the common training matrix, and the plurality of initial relationship mapping matrices to obtain an objective function includes:
s501, obtaining the relation between matrixes in the public training characteristic matrix to obtain first relation information;
step S502, carrying out sparse constraint on the initial weight matrix to obtain an initial sparse weight matrix;
step S503, multiplying the initial sparse weight matrix and the public training feature matrix to obtain a sparse training feature matrix;
step S504, obtaining the relation between vectors in the sparse training feature matrix to obtain second relation information;
step S505, taking the initial relationship mapping matrixes, the public relationship pool, the first relationship information, the second relationship information, the real tags and the initial weight matrixes as variables of the preset function to obtain the target function;
step S506, deriving the objective function to obtain a plurality of relationship mapping matrices and the weight matrix, where a function value of the objective function corresponding to the plurality of relationship mapping matrices and the weight matrix is minimum.
Specifically, in order to enable the relationship mapping matrix and the weight matrix to have a better screening effect, the embodiment needs to obtain the relationship between the matrices in the public training feature matrix to obtain the first relationship information, and since the first relationship information may reflect the relationship between the matrices and each matrix corresponds to each brain area template, the first relationship information may also reflect the relationship between each brain area template. In addition, in this embodiment, it is further required to perform sparse constraint on the initial weight matrix to obtain an initial sparse weight matrix, and then multiply the initial sparse weight matrix with the public training feature matrix to realize screening of feature vectors in the public training feature matrix to obtain a sparse training feature matrix, and obtain a relationship between vectors in the sparse training feature matrix to obtain second relationship information, where the second relationship information reflects a relationship between vectors in the sparse training feature matrix, that is, is equivalent to reflecting an internal relationship of the sparse training feature matrix. And finally, taking the initial relationship mapping matrixes, the public relationship pool, the first relationship information, the second relationship information, the real label and the initial weight matrix as variables of a preset function to obtain a target function. And finally, the derivation is carried out on the objective function, and when the function value of the objective function is minimum, a plurality of relationship mapping matrixes and weight matrixes which can be directly applied are obtained.
For example, as shown in fig. 2, in the present embodiment, a plurality of magnetic resonance imaging are processed simultaneously to obtain M sets of feature matrices obtained from M templates, and the feature matrix obtained from the mth template is defined as Xm. Then obtaining a plurality of initial relation mapping matrixes PmMapping the feature space extracted from each template to a public relationship pool R through a plurality of initial relationship mapping matrixes, and applying sparse constraint to each relationship mapping matrix to ensure that the relationship between the templates mapped to the public relationship pool is discriminant, so that the public relationship pool R can be obtained through the following formula to obtain first relationship information:
wherein λ is1Is a control parameter of the regularization term,
is the Frobenius norm (F-norm) of matrix a,
is matrix A of21Norm (l)21-norm)。
After the pool of common relations R is obtained, it is used as a feature space to perform feature selection. A weight matrix W that best fits the public relationship pool R to the true label matrix Y can be solved from the basic linear regression model, while the weight matrix is subjected to adaptive sparse learning to obtain the following formula:
wherein λ2Is a control parameter of the regularization term,
is 12,pNorm (l)2,pNorm), tr (a) is the rank of matrix a.
Second, the second relationship information may be obtained using a locally preserved projection. Firstly, defining the characteristic R in the public relation pool R by adopting a graph Laplace graph methodiAnd rjSimilarity between si,jThen, obtaining the regularization term according to a local preserving projection algorithm as follows:
RL＝tr(∑i，j(WTri-WTrj)2 si，j) (3)
the final objective function is as follows, then the objective function is derived to obtain the minimum value of the function value, and a plurality of relation mapping matrixes and weight matrixes are determined according to the objective function at the time of the minimum value:
wherein λ1，λ2，λ3And λ4Is a control parameter of the regularizing term, ATRepresented as the transpose of matrix a.
In one implementation, the training process of the multi-classification model is as follows:
step S6, determining a second optimized feature matrix corresponding to the nuclear magnetic resonance training image according to the relation mapping matrix and the weight matrix;
step S7, second clinical information corresponding to the nuclear magnetic resonance training image is obtained, and a second newly added feature matrix is constructed according to the second clinical information;
step S8, constructing a second target feature matrix according to the second optimized feature matrix and the second newly added feature matrix;
and S9, obtaining an initial multi-classification model, and training the initial multi-classification model according to the second target feature matrix and the real label to obtain the multi-classification model.
Specifically, the second optimized feature matrix comprises feature vectors with discriminability, so that a second target feature matrix is comprehensively constructed according to the second optimized feature matrix and a second newly-added feature matrix formed based on second clinical information, the second target feature matrix and the real label are input into the initial multi-classification model together, the initial multi-classification model is trained, and the multi-classification model can be obtained after training.
Based on the above embodiment, the present invention further provides an image classification apparatus, as shown in fig. 3, the apparatus including:
the acquisition module 01 is used for acquiring magnetic resonance imaging and a plurality of preset brain area templates, wherein the division rules of human brain areas corresponding to the brain area templates are different;
a determining module 02, configured to determine a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, where each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates;
the screening module 03 is configured to screen eigenvectors in the plurality of feature matrices to obtain a first optimized feature matrix;
and the prediction module 04 is configured to determine an image prediction category corresponding to the magnetic resonance imaging according to the first optimized feature matrix.
Based on the above embodiments, the present invention further provides a terminal, and a schematic block diagram thereof may be as shown in fig. 4. The terminal comprises a processor, a memory, a network interface and a display screen which are connected through a system bus. Wherein the processor of the terminal is configured to provide computing and control capabilities. The memory of the terminal comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the terminal is used for connecting and communicating with an external terminal through a network. The computer program is executed by a processor to implement an image classification method. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.
It will be understood by those skilled in the art that the block diagram of fig. 4 is a block diagram of only a portion of the structure associated with the inventive arrangements and is not intended to limit the terminals to which the inventive arrangements may be applied, and that a particular terminal may include more or less components than those shown, or may have some components combined, or may have a different arrangement of components.
In one implementation, one or more programs are stored in a memory of the terminal and configured to be executed by one or more processors, including instructions for performing the image classification method.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
In summary, the present invention discloses an image classification method, an apparatus and a storage medium, wherein the method obtains magnetic resonance imaging and a plurality of preset brain region templates, and the division rules of human brain regions respectively corresponding to the brain region templates are different; determining a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, wherein each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates; screening the feature vectors in the feature matrixes to obtain a first optimized feature matrix; and determining an image prediction category corresponding to the magnetic resonance imaging according to the first optimization feature matrix. The invention predicts the categories of the magnetic resonance imaging shot by different people by a method of extracting image characteristics, and doctors can more accurately distinguish normal people and neurodegenerative disease people by predicting the categories of the images corresponding to the magnetic resonance imaging and combining self medical experience. Therefore, the invention can solve the problem that doctors in the prior art mainly rely on the clinical score and the movement problem of people to diagnose the neurodegenerative diseases and are difficult to distinguish the neurodegenerative disease people from normal people.
It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.
Claims (10)
1. A method of image classification, the method comprising:
acquiring magnetic resonance imaging and a plurality of preset brain area templates, wherein the division rules of human brain areas corresponding to the brain area templates are different;
determining a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, wherein each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates;
screening the feature vectors in the feature matrixes to obtain a first optimized feature matrix;
and determining an image prediction category corresponding to the magnetic resonance imaging according to the first optimization feature matrix.
2. The image classification method according to claim 1, wherein the determining a number of feature matrices from the number of brain region templates and the magnetic resonance imaging comprises:
performing image segmentation on the magnetic resonance imaging to obtain a plurality of first partial images, wherein the brain tissue types corresponding to the plurality of first partial images are different;
respectively carrying out region division on the plurality of first local images according to each brain area template to obtain a plurality of first divided images corresponding to each brain area template;
and determining a feature matrix corresponding to each brain area template according to the first divided images corresponding to each brain area template to obtain a plurality of feature matrices.
3. The image classification method according to claim 2, wherein the determining the feature matrix corresponding to each brain region template according to the first divided images corresponding to each brain region template includes:
respectively extracting features of the first divided images corresponding to each brain area template to obtain a plurality of feature vectors corresponding to each brain area template;
and determining a feature matrix corresponding to each brain area template according to a plurality of feature vectors corresponding to each brain area template.
4. The image classification method according to claim 1, wherein the screening the feature vectors in the feature matrices to obtain a first optimized feature matrix comprises:
acquiring a plurality of relation mapping matrixes and a weight matrix, wherein the relation mapping matrixes correspond to the brain area templates one by one;
carrying out sparse constraint on the plurality of relational mapping matrixes to obtain a plurality of sparse relational mapping matrixes;
mapping the feature matrixes into a public relation pool according to the sparse relation mapping matrixes to obtain a public feature matrix, wherein the sparse relation mapping matrixes correspond to the feature matrixes one by one, and one sparse relation mapping matrix and one feature matrix which have corresponding relations correspond to the same brain region template;
carrying out sparse constraint on the weight matrix to obtain a sparse weight matrix;
and multiplying the sparse weight matrix and the public feature matrix to obtain the first optimized feature matrix.
5. The image classification method according to claim 4, wherein the determining the image prediction class corresponding to the magnetic resonance imaging according to the first optimized feature matrix comprises:
acquiring first clinical information corresponding to the magnetic resonance imaging, and constructing a first newly-added feature matrix according to the first clinical information;
constructing a first target feature matrix according to the first newly-added feature matrix and the first optimized feature matrix;
and inputting the first target characteristic matrix into a multi-classification model to obtain the image prediction category.
6. The image classification method according to claim 5, wherein the plurality of relational mapping matrices and the weighting matrix are pre-trained matrices, and the training process of the plurality of relational mapping matrices and the weighting matrix is as follows:
acquiring a nuclear magnetic resonance training image, and determining a plurality of training feature matrixes which correspond to the brain area templates one by one according to the brain area templates and the nuclear magnetic resonance training image;
acquiring a plurality of initial relation mapping matrixes, wherein the plurality of initial relation mapping matrixes correspond to the plurality of brain area templates one by one;
carrying out sparse constraint on the plurality of initial relation mapping matrixes to obtain a plurality of initial sparse relation mapping matrixes;
mapping the training feature matrixes into the public relation pool according to the initial sparse relation mapping matrixes to obtain a public training feature matrix;
acquiring an initial weight matrix and a real label corresponding to the nuclear magnetic resonance training image;
obtaining a preset function, determining variables in the preset function according to the initial weight matrix, the real label, the public training matrix and the initial relation mapping matrices to obtain a target function, and determining the relation mapping matrices and the weight matrix according to the target function.
7. The image classification method according to claim 6, wherein the determining variables in the preset function according to the initial weight matrix, the real label, the common training matrix, and the plurality of initial relationship mapping matrices to obtain an objective function, and determining the plurality of relationship mapping matrices and the weight matrix according to the objective function comprises:
obtaining the relation between matrixes in the public training characteristic matrix to obtain first relation information;
carrying out sparse constraint on the initial weight matrix to obtain an initial sparse weight matrix;
multiplying the initial sparse weight matrix and the public training feature matrix to obtain a sparse training feature matrix;
obtaining the relation between vectors in the sparse training feature matrix to obtain second relation information;
taking the initial relationship mapping matrixes, the public relationship pool, the first relationship information, the second relationship information, the real label and the initial weight matrix as variables of the preset function to obtain the target function;
and deriving the objective function to obtain a plurality of relation mapping matrixes and the weight matrix, wherein the function value of the objective function corresponding to the relation mapping matrixes and the weight matrix is minimum.
8. The image classification method according to claim 7, wherein the multi-classification model is a model trained in advance, and the training process of the multi-classification model is as follows:
determining a second optimized feature matrix corresponding to the nuclear magnetic resonance training image according to the relation mapping matrix and the weight matrix;
acquiring second clinical information corresponding to the nuclear magnetic resonance training image, and constructing a second newly added feature matrix according to the second clinical information;
constructing a second target feature matrix according to the second optimized feature matrix and the second newly added feature matrix;
and acquiring an initial multi-classification model, and training the initial multi-classification model according to the second target feature matrix and the real label to obtain the multi-classification model.
9. An image classification apparatus, characterized in that the apparatus comprises:
the acquisition module is used for acquiring magnetic resonance imaging and a plurality of preset brain area templates, and the division rules of human brain areas corresponding to the brain area templates are different;
a determining module, configured to determine a plurality of feature matrices according to the plurality of brain region templates and the magnetic resonance imaging, where each feature matrix in the plurality of feature matrices is determined by one brain region template in the plurality of brain region templates;
the screening module is used for screening the eigenvectors in the plurality of characteristic matrixes to obtain a first optimized characteristic matrix;
and the prediction module is used for determining the image prediction category corresponding to the magnetic resonance imaging according to the first optimization characteristic matrix.
10. A computer readable storage medium having stored thereon a plurality of instructions adapted to be loaded and executed by a processor to perform the steps of the image classification method according to any of the preceding claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110705690.0A CN113591860A (en) | 2021-06-24 | 2021-06-24 | Image classification method and device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110705690.0A CN113591860A (en) | 2021-06-24 | 2021-06-24 | Image classification method and device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113591860A true CN113591860A (en) | 2021-11-02 |
Family
ID=78244547
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110705690.0A Pending CN113591860A (en) | 2021-06-24 | 2021-06-24 | Image classification method and device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113591860A (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109770903A (en) * | 2018-12-29 | 2019-05-21 | 中国科学院自动化研究所 | The classification prediction technique of functional magnetic resonance imaging, system, device |
-
2021
- 2021-06-24 CN CN202110705690.0A patent/CN113591860A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109770903A (en) * | 2018-12-29 | 2019-05-21 | 中国科学院自动化研究所 | The classification prediction technique of functional magnetic resonance imaging, system, device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ju et al. | Early diagnosis of Alzheimer's disease based on resting-state brain networks and deep learning | |
| Zhou et al. | Multi-modal latent space inducing ensemble SVM classifier for early dementia diagnosis with neuroimaging data | |
| Novikov et al. | Deep sequential segmentation of organs in volumetric medical scans | |
| CN110993094B (en) | Intelligent auxiliary diagnosis method and terminal based on medical image | |
| Guo et al. | Human age estimation using bio-inspired features | |
| Yang et al. | A study of brain networks for autism spectrum disorder classification using resting-state functional connectivity | |
| KR20210048523A (en) | Image processing method, apparatus, electronic device and computer-readable storage medium | |
| CN109344889B (en) | Brain disease classification apparatus, user terminal, and computer-readable storage medium | |
| WO2021097675A1 (en) | Smart diagnosis assistance method and terminal based on medical images | |
| CN110473172B (en) | Medical image anatomical centerline determination method, computer device and storage medium | |
| CN110210543B (en) | Image classification system, method, apparatus and storage medium | |
| Vankdothu et al. | Brain image identification and classification on Internet of Medical Things in healthcare system using support value based deep neural network | |
| Al-Adhaileh | Diagnosis and classification of Alzheimer's disease by using a convolution neural network algorithm | |
| So et al. | Deep learning for Alzheimer’s disease classification using texture features | |
| CN113240655A (en) | Method, storage medium and device for automatically detecting type of fundus image | |
| Jia et al. | Deep learning and multimodal feature fusion for the aided diagnosis of Alzheimer's disease | |
| Pan et al. | Multi-classification prediction of Alzheimer’s disease based on fusing multi-modal features | |
| CN113052800B (en) | Alzheimer disease image analysis method and device | |
| Chen et al. | Contrastive learning for prediction of Alzheimer's disease using brain 18f-fdg pet | |
| CN112036478A (en) | Identification method and device for chronic disease reimbursement medicine and computer equipment | |
| CN113591860A (en) | Image classification method and device and storage medium | |
| CN116844730A (en) | Modeling method and system of cognitive disorder recognition model | |
| CN112950654B (en) | Brain tumor image segmentation method based on multi-core learning and super-pixel nuclear low-rank representation | |
| KR102373992B1 (en) | Method and apparatut for alzheimer's disease classification using texture features | |
| Zaragoza-Jimenez et al. | Modeling face recognition in the predictive coding framework: A combined computational modeling and functional imaging study |
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 |