CN110097128A - Medical Images Classification apparatus and system - Google Patents
Medical Images Classification apparatus and system Download PDFInfo
- Publication number
- CN110097128A CN110097128A CN201910376120.4A CN201910376120A CN110097128A CN 110097128 A CN110097128 A CN 110097128A CN 201910376120 A CN201910376120 A CN 201910376120A CN 110097128 A CN110097128 A CN 110097128A
- Authority
- CN
- China
- Prior art keywords
- image
- classifier
- medical
- module
- roi region
- 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.)
- Granted
Links
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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Genetics & Genomics (AREA)
- Physiology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The embodiment of the invention discloses a kind of Medical Images Classification apparatus and systems.Wherein, device includes classifier prebuild module and image classification module.Then classifier prebuild module is respectively trained every class ROI region image using Three dimensional convolution neural network for extracting multiple ROI region images that training sample concentrates each medicine sample image, obtains corresponding base classifier；Based on preset fitness value, optimal base classifier is selected from multiple base classifiers using genetic algorithm, and Image Classifier is finally generated based on each optimum classifier.Image classification module obtains the classification results of medical image to be identified for medical image to be identified to be input in the Image Classifier constructed in advance.The application improves the accuracy of Medical Images Classification, is conducive to precisely identify early stage Alzheimer disease medical image, provides auxiliary for the prevention and diagnosis of early stage Alzheimer Disease patient.
Description
Technical field
The present embodiments relate to technical field of medical image processing, more particularly to a kind of Medical Images Classification device and
System.
Background technique
Alzheimer disease is a kind of nervous system degenerative disease of the progress sexual development of onset concealment, can medically be led to
Cross structure image or functional nerve image discovery Alzheimer disease.
Structure image such as Cranial Computed Tomography (Computer Tomography, computed tomography) and MRI (Magnetic
Resonance Imaging, Magnetic resonance imaging) it checks, it can show that Cerebral cortex atrophy is obvious, especially hippocampus and inside temporo
Leaf supports the clinical diagnosis of Alzheimer.Functional nerve image can be as disconnected such as positron scanning and single photon emission computed
Diagnosis of dementias confidence level can be improved in layer scanning.
The clinical manifestation of Alzheimer different times is different, if advanced stage visible frontal lobe metabolism is lowered, how to identify A Er
Ci Haimo disease early stage crowd, provides auxiliary for the prevention and diagnosis of early stage patients with Alzheimer disease, is those skilled in the art
Urgent problem to be solved.
The relevant technologies generally realize patients with Alzheimer disease medical image by carrying out analysis to the medical image of acquisition
Identification still during medical image analysis, since the Alzheimer disease cause of disease is unknown so far, choose medical image
((region of interest, area-of-interest) performance is bad, causes last image classification effect accuracy not high by ROI.
Summary of the invention
The embodiment of the present disclosure provides a kind of Medical Images Classification apparatus and system, improves the accurate of Medical Images Classification
Degree, is conducive to precisely identify early stage Alzheimer disease medical image, is the prevention and diagnosis of early stage Alzheimer Disease patient
Auxiliary is provided.
In order to solve the above technical problems, the embodiment of the present invention the following technical schemes are provided:
On the one hand the embodiment of the present invention provides a kind of Medical Images Classification device, including classifier prebuild module and figure
As categorization module；
The classifier prebuild module includes ROI region extracting sub-module, model training submodule, optimum combination screening
Submodule and model integrated submodule；
Described image categorization module is obtained for medical image to be identified to be input in the Image Classifier constructed in advance
The classification results of the medical image to be identified；
Wherein, the ROI region extracting sub-module concentrates the multiple of each medicine sample image for extracting training sample
ROI region image, the training sample set include the brain area figure of healthy population with alzheimer's disease difference illness Phase patient
Picture；The model training submodule is obtained for being trained respectively to every class ROI region image using Three dimensional convolution neural network
To corresponding base classifier；The optimum combination screening submodule is used to be based on preset fitness value, is calculated using heredity
Method selects optimal base classifier from multiple base classifiers；The model integrated submodule is used to generate based on each optimum classifier
Described image classifier.
Optionally, the model integrated submodule includes:
Target ROI region determination unit, for being to determine the corresponding target ROI region of each optimal base classifier；
Target ROI region combines image extraction unit, for each medicine sample image in the trained set of stereotypes,
Multiple target ROI regions are extracted simultaneously, to constitute target ROI region combination image；
Initial pictures classifier training unit, for utilizing the Three dimensional convolution neural network to each target ROI region group
It closes image to be trained, obtains initial pictures classifier；
Integrated unit, for integrating the initial pictures classifier and each optimal base classifier, described in generating
Image Classifier.
Optionally, the classifier prebuild module further includes that base classifier selects submodule, and the base classifier is selected
Submodule includes:
Sequencing unit, for being ranked up according to the classification accuracy of base classifier to each base classifier,
Unit is deleted, the base classifier for being not more than preset threshold for deleting classification accuracy.
Optionally, the ROI region extracting sub-module is to utilize the 116 of each medicine sample image of ALL116 template extraction
The module of a ROI region image.
Optionally, the Three dimensional convolution neural network is the two-dimensional network level and three-dimensional network knot using VGG16 model
Structure, and the network structure that number of filters is 8.
Optionally, the classifier prebuild module further includes image preprocessing submodule, and described image pre-processes submodule
Block includes:
First denoising unit, for concentrating each medicine sample image to carry out the dynamic correction of head and stripping head the training sample
Bone operation is influenced with removing noise and non-brain tissue structure；
Registration process unit, for treated each medicine sample image to be carried out Spatial normalization, with registration to system
One coordinate space.
Optionally, described image pre-processes submodule further include:
Gaussian smoothing unit, for carrying out Gaussian smoothing to each medicine sample image by Spatial normalization processing
Processing.
Optionally, described image pre-processes submodule further include:
Normalization unit, for carrying out gray scale normalization to by each medicine sample image of Gaussian smoothing.
On the other hand the embodiment of the present invention provides a kind of Medical Images Classification system, including cure described in any one as above
Learn image classification device.
It optionally, further include the Medical imaging module being connected with the Medical Images Classification device；
The Medical imaging module is for storing current medical image to be identified, the medical image letter
Breath include medical image to be identified and label information, the label information be the Medical Images Classification device output it is described to
Identify the classification results of medical image.
The advantages of technical solution provided by the present application, is, is constructed in conjunction with 3DCNN for medicine by genetic algorithm
The Image Classifier that image is identified, 3DCNN can not only learn the key feature of image, can also reduce over-fitting
Risk；Based on genetic algorithm can solve optimization problem globally optimal solution and optimum results it is unrelated with primary condition, may search for
There is the ROI image region of remarkable result to combine sick people's classification, the final image for determining which region of brain is to classification
There is positive influence, it is ensured that the ROI brain zone function of selection has certain identical property with the clinical manifestation of the disease, finally will be hereditary
Algorithm is selected different ROI models and is integrated, and an efficient strong classifier is ultimately formed, to obtain more acurrate, stable
With strong as a result, improve the accuracy of Medical Images Classification, be conducive to precisely identify early stage Alzheimer disease medicine figure
Picture provides auxiliary for the prevention and diagnosis of early stage Alzheimer Disease patient.
In addition, the embodiment of the present invention provides corresponding system also directed to classification method of medical image, further such that institute
Method is stated with more practicability, the system has the advantages that corresponding.
It should be understood that the above general description and the following detailed description are merely exemplary, this can not be limited
It is open.
Detailed description of the invention
It, below will be to embodiment or correlation for the clearer technical solution for illustrating the embodiment of the present invention or the relevant technologies
Attached drawing needed in technical description is briefly described, it should be apparent that, the accompanying drawings in the following description is only this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is a kind of specific embodiment structure chart of Medical Images Classification device provided in an embodiment of the present invention；
Fig. 2 is a kind of flow diagram of Image Classifier building process provided in an embodiment of the present invention；
Fig. 3 provides for the embodiment of the present invention according to an exemplary CNN operation principle schematic diagram；
Fig. 4 provides for the embodiment of the present invention according to Three dimensional convolution neural network structure schematic diagram；
Fig. 5 provides for the embodiment of the present invention according to image preprocessing result schematic diagram；
Fig. 6 is a kind of flow diagram of image classification method provided in an embodiment of the present invention.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.Obviously, described embodiments are only a part of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
The description and claims of this application and term " first ", " second ", " third " " in above-mentioned attached drawing
Four " etc. be for distinguishing different objects, rather than for describing specific sequence.Furthermore term " includes " and " having " and
Their any deformations, it is intended that cover and non-exclusive include.Such as contain a series of steps or units process, method,
System, product or equipment are not limited to listed step or unit, but may include the step of not listing or unit.
After describing the technical solution of the embodiment of the present invention, the various non-limiting realities of detailed description below the application
Apply mode.
Referring first to Fig. 1, Fig. 1 is a kind of Medical Images Classification device provided in an embodiment of the present invention in a kind of specific implementation
Block schematic illustration under mode, the embodiment of the present invention may include the following contents:
Medical Images Classification device may include classifier prebuild module 1 and image classification module 2.
In the present embodiment, classifier prebuild module 1 for constructing the module of medical image for identification, building in advance
Process is seen shown in Fig. 2, specifically may include ROI region extracting sub-module, model training submodule, optimum combination screening submodule
Block and model integrated submodule.
Wherein, ROI region extracting sub-module is for extracting multiple areas ROI that training sample concentrates each medicine sample image
Area image, that is to say, that number, the classification for the ROI region that each medicine sample image extracts are all the same.Every medicine sample graph
The number and classification of the ROI region of picture, those skilled in the art can choose according to the actual situation.Training sample set may include
The brain area image of healthy population and the brain area image of alzheimer's disease difference illness Phase patient, and every brain area image is all
A label information is preset, which identifies the disease condition for identifying corresponding brain area image, such as
One sample image is the sample image of healthy population, and label may be configured as y=0, and second sample image is early stage A Er
The sample image of Zi Haimo disease patient, label may be configured as y=1, and third sample image is advanced stage alzheimer's disease trouble
The sample image of person, label may be configured as y=2, and brain area image may be, for example, brain CT image or MRI image.Due to medicine sample
This image is brain area image, can be using each subregion of brain area image as ROI region, such as precentral gyrus ROI region, back outside
Superior frontal gyrus ROI region, middle frontal gyrus ROI region, parahippocampal gyrus ROI region.Using ROI image rather than full brain image is as model
Input reduces network query function amount, improves convolutional neural networks training speed.
It is understood that taking off a ROI region from primitive medicine sample image, it is corresponding just to generate the ROI region
ROI region image is taken off the ROI region image and primitive medicine sample image label information having the same of processing, be can be used
Any one takes off the method that ROI region generates ROI region image from primitive medicine sample image, and the application does not appoint this
What is limited.Optionally, 116 ROI region images of each medicine sample image of ALL116 template extraction, AAL can be used
(Anatomical Automatic Labeling, anatomical automatic label), for by Montreal Neurological
What Institute (MNI) mechanism provided, which shares 116 regions, wherein 90 regions belong to brain, it is 26 remaining
Belong to cerebellum structure.It can extract the image in this 116 regions using ALL116 template, so that 116 ROI region images are generated,
The image number that every one kind ROI region image includes is the total number of samples that training sample set includes.
In this application, model training submodule is used to utilize Three dimensional convolution neural network (3DConvolutional
Neural Networks, 3DCNN) every class ROI region image is trained respectively, corresponding base classifier is obtained, that is,
Say, every class ROI region image one base classifier of training, the base classifier can be used as an Individual classifier to medical image into
Row identification classification, according to classification accuracy ROI region to the influence degree of alzheimer's disease.
Fig. 3 is the working principle diagram of CNN, and CNN includes convolutional layer and pond layer.Convolutional layer is also referred to as feature extraction layer, right
Input image data applies several filters, and an input parameter is used to do the feature extraction of multiple types.Convolutional layer is used
In the feature for extracting input image data.The image data feature that each different convolution kernel extracts would also vary from,
Convolution nuclear volume is more, and the image data feature extracted is also more.To image using the knot obtained after a filter
Fruit is known as characteristic spectrum, and the quantity of characteristic spectrum and the quantity of filter are consistent.In network training process, if only led to
The feature that convolutional layer extracts image is crossed, can be very huge when encountering larger-size image hour operation quantity, the speed of network training
Also relatively slow.In order to reduce operand, net training time is reduced, convolutional neural networks are connected to one layer of pond behind convolutional layer
Layer, it can reduce the resolution ratio of image, reduce data operation quantity, while network can be enhanced to the adaptability of image change.
In one embodiment, the structure of Three dimensional convolution neural network is seen shown in Fig. 4, Three dimensional convolution neural network
Structure can be the two-dimensional network level and three-dimensional net structure using VGG16 model, and number of filters is 8 network knots
Structure.Such as may include containing the first convolutional layer, the second convolutional layer, the first module of the first pond layer, containing third convolutional layer, the
Four convolutional layers, the second pond layer the second module, contain the 5th convolutional layer, the 6th convolutional layer, the 7th convolutional layer, third pond layer
Third module, the first full articulamentum FC1, the second full articulamentum FC2, full articulamentum FC3, Softmax layers of third.Input picture
It is successively transmitted to the second module after the first resume module, processing result is transmitted to third mould after the second resume module
Block is exported after third resume module by full articulamentum.As to how model is obtained using convolutional neural networks training sample, it can
It is just repeated no more herein refering to the realization process of any description of Related Art.
In the present embodiment, optimum combination screens submodule and is used to be based on preset fitness value, utilizes genetic algorithm
Optimal base classifier is selected from multiple base classifiers.Healthy brain area image and suffer from alzheimer's disease brain area image in different brains
The clinical presentation that region is presented is different, and the clinical presentation that same a part of the brain area image in the different illness stages is presented can
It can be identical, it is also possible to different, this is just needed according to image classification demand, such as identification early stage alzheimer's disease brain area image,
The brain area region being affected to the classification results is determined, as optimal ROI region.Since genetic algorithm (GA) can solve to optimize
The globally optimal solution of problem, and optimum results are unrelated with primary condition has stronger robustness, can be used genetic algorithm from
Optimal base classifiers combination is selected in multiple base classifiers, the corresponding ROI region of each optimal base classifier is the optimal area ROI
Domain that is to say the brain area region being affected to classification results.In genetic algorithm searching process, fitness value can be preset
For as judgment criteria, it can also to be F value, the application does not do this any that fitness value, which can be the classification accuracy of base classifier,
It limits.As to how seeing any description of Related Art using the optimum combination of genetic algorithm optimizing base classifier
Realization process just repeats no more herein.
It is understood that model integrated submodule is used to generate Image Classifier based on each optimum classifier, it will be different
ROI network model integrated to improve the classifying quality of sick people.Any integrated study model can be used will be more
A single classifier obtains unified integrated if base classifier or initial pictures classifier organically combine
Model is practised, to can get more acurrate, stable and strong classification results.
Image classification module 2 for medical image to be identified to be input in the Image Classifier constructed in advance, obtain to
Identify the classification results of medical image.Classify if Image Classifier is used for the identification to Alzheimer's brain area image,
The classification results of medical image so to be identified are illness or health；If Image Classifier is further used for alzheimer's disease
The identification of patient's brain area image in illness stage is classified, then the classification results of medical image to be identified are early stage illness, mid-term
Illness, advanced stage illness or health.
In technical solution provided in an embodiment of the present invention, constructed in conjunction with 3DCNN for medicine by genetic algorithm
The Image Classifier that image is identified, 3DCNN can not only learn the key feature of image, can also reduce over-fitting
Risk；Based on genetic algorithm can solve optimization problem globally optimal solution and optimum results it is unrelated with primary condition, may search for
There is the ROI image region of remarkable result to combine sick people's classification, the final image for determining which region of brain is to classification
There is positive influence, it is ensured that the ROI brain zone function of selection has certain identical property with the clinical manifestation of the disease, finally will be hereditary
Algorithm is selected different ROI models and is integrated, and an efficient strong classifier is ultimately formed, to obtain more acurrate, stable
With strong as a result, improve the accuracy of Medical Images Classification, be conducive to precisely identify early stage Alzheimer disease medicine figure
Picture provides auxiliary for the prevention and diagnosis of early stage Alzheimer Disease patient.
In this application, it integrates, generates final between each optimal base classifier that GA algorithm optimizing can be obtained
Image Classifier；In addition, can also be trained for the corresponding ROI region image of optimal base classifier, other parts are excluded
The influence of area image, the model and other optimum classifiers that training obtains are integrated, in this embodiment, model integrated
Submodule may include target ROI region determination unit, target ROI region combination image extraction unit, initial pictures classifier instruction
Practice unit and integrated unit.
Wherein, target ROI region determination unit is used for determine the corresponding target ROI region of each optimal base classifier,
Since the training of every class ROI region image obtains a base classifier, therefore such ROI region can be locked according to base classifier, as
Optimal ROI region, that is, target ROI region.Target ROI region combines image extraction unit and is used for in training set of stereotypes
Each medicine sample image, while multiple target ROI regions are extracted, to constitute target ROI region combination image；For example,
Optimal ROI region after GA algorithm optimizing is Medial Temporal Lobe, precentral gyrus, carries on the back outside superior frontal gyrus, middle frontal gyrus, parahippocampal gyrus, that
Will be for each medicine sample image, while extracting the Medial Temporal Lobe of each brain area image, precentral gyrus, carrying on the back on the volume of outside
It returns, these regions of middle frontal gyrus, parahippocampal gyrus and processing group are combined into target ROI region combination image, target ROI region combination image
Quantity can with training set of stereotypes in comprising the total number of medicine sample image it is identical.Initial pictures classifier training unit is used for
Each target ROI region combination image is trained using Three dimensional convolution neural network, obtains initial pictures classifier；As for such as
What obtains model using convolutional neural networks training sample, sees the realization process of any description of Related Art, herein,
Just it repeats no more.Integrated unit, for integrating initial pictures classifier and each optimal base classifier, to generate image point
Class device.
The optimal ROI combination that the embodiment of the present invention is selected by GA algorithm can also be conducive to extraction figure by 3DCNN training
The deep layer key feature of picture, excludes the influence of other local image regions, is conducive to the performance for promoting classifier.
In view of there may be the regions little to Classification and Identification influential effect in the ROI region selected, in order to reduce
The entire model training time, classifier training efficiency is improved, classifier prebuild module 1 may also include base classifier and select submodule
Block, it includes sequencing unit and deletion unit that base classifier, which selects submodule,.
Wherein, sequencing unit is for being ranked up each base classifier according to the classification accuracy of base classifier；It deletes single
Member, the base classifier for being not more than preset threshold for deleting classification accuracy.It can be concentrated from training sample and choose a part of sample
Image verifies classification accuracy of each base classifier to verifying sample image, by each base classifier as verifying sample
Classification accuracy is as its percentage contribution to classification recognition result of measurement.Threshold value can be determined according to the actual situation, this Shen
Any restriction is not done to this please.
It, can also be to instruction for the ease of the accuracy and efficiency of subsequent Medical Images Classification in another embodiment
Practice each medicine sample image in sample set and carry out image preprocessing, original image is carried out standardization and drop appropriate by purpose
Low noise sonication.Based on this, classifier prebuild module 1 may also include image preprocessing submodule, and utilization can be soft using SPM12
Part carries out image preprocessing, and primitive medicine sample image is seen shown in Fig. 5 through image preprocessing post-processing result, and image is located in advance
Managing submodule specifically may include the first denoising unit and registration process unit.
First denoising unit, for concentrating each medicine sample image to carry out the dynamic correction of head and stripping skull behaviour training sample
Make, is influenced with removing noise and non-brain tissue structure.
Registration process unit, for treated each medicine sample image to be carried out Spatial normalization, with registration to system
One coordinate space, to eliminate the difference between the corresponding individual of each medicine sample image.
In addition, in order to remove the influence of noise on image, so that data are joined closer to being similar to just be distributed very much with this to increase
The validity that number is examined, image preprocessing submodule may also include Gaussian smoothing unit, for by Spatial normalization
Each medicine sample image of processing carries out Gaussian smoothing.
After Gauss denoising, for subsequent convenience of calculation, image can be also normalized namely image is located in advance
Reason submodule may also include normalization unit, for carrying out gray scale normalizing to by each medicine sample image of Gaussian smoothing
Change.
It should be noted that medical image to be identified is being input to figure in order to improve the classifying quality of images to be recognized
As that can also carry out image preprocessing to the image before classifier, namely respectively, progress head moves correction, the operation of stripping skull, Gauss
Smoothing processing and gray scale normalization processing.
The embodiment of the present invention provides corresponding implementation method also directed to Medical Images Classification device, below to of the invention real
The classification method of medical image for applying example offer is introduced, classification method of medical image described below and above-described medicine
Image classification device can correspond to each other reference.
Fig. 6 is referred to, Fig. 6 is the flow diagram of another classification method of medical image provided in an embodiment of the present invention,
The embodiment of the present invention for example can be applied in the diagnostic system of alzheimer's disease, specifically may include the following contents:
S601: medical image to be identified is obtained.
S602: medical image to be identified is input in the Image Classifier constructed in advance, obtains medical image to be identified
Classification results.
Wherein, the building process of Image Classifier includes:
Extract multiple ROI region images that training sample concentrates each medicine sample image；Training sample set includes health
Crowd and brain area image in different alzheimer's disease illness Phase patients；
Every class ROI region image is trained respectively using Three dimensional convolution neural network, obtains corresponding base classifier；
Based on preset fitness value, optimal base classifier is selected from multiple base classifiers using genetic algorithm,
Image Classifier is generated based on each optimum classifier.
Optionally, Image Classifier is generated based on each optimum classifier can include:
Determine the corresponding target ROI region of each optimal base classifier；
To each medicine sample image in training set of stereotypes, while multiple target ROI regions are extracted, to constitute target
ROI region combines image；
Each target ROI region combination image is trained using Three dimensional convolution neural network, obtains initial pictures classification
Device；
Initial pictures classifier and each optimal base classifier are integrated, to generate Image Classifier.
In another embodiment, every class ROI region image is being instructed respectively using Three dimensional convolution neural network
Practice, after obtaining corresponding base classifier, using genetic algorithm before selecting optimal base classifier in multiple base classifiers, also
Can include:
Each base classifier is ranked up according to the classification accuracy of base classifier；
Delete the base classifier that classification accuracy is not more than preset threshold.
In some other embodiment, multiple ROI region figures of each medicine sample image are concentrated in extraction training sample
Before picture, it may also include that
Each medicine sample image is concentrated to carry out the dynamic correction of head and stripping skull operation training sample, to remove noise and non-
Brain tissue structure influences；
Treated each medicine sample image is subjected to Spatial normalization, with registration to uniform coordinate space；
Gaussian smoothing is carried out to each medicine sample image by Spatial normalization processing, and carries out gray scale normalizing
Change.
From the foregoing, it will be observed that the embodiment of the present invention improves the accuracy of Medical Images Classification, be conducive to precisely to identify early stage Ah
Alzheimer's disease medical image provides auxiliary for the prevention and diagnosis of early stage Alzheimer Disease patient.
The embodiment of the invention also provides a kind of Medical Images Classification systems, it may include described in any one embodiment as above
Medical Images Classification device.
Training sample number and type directly determine the classifying quality of Image Classifier, in order to improve point of Image Classifier
Class accuracy rate can expand training sample real-time perfoming.Optionally, it may also include the medicine being connected with Medical Images Classification device
Image data library module.It include multiple medicine sample images, each doctor that training sample is concentrated in Medical imaging module
Learning sample image can directly obtain from Medical imaging module.Medical imaging module is for storing currently wait know
Other medical image, medical image include medical image and label information to be identified, and label information is medical image
The classification results of the medical image to be identified of sorter output.
That is, after Medical Images Classification device carries out image classification identification to the medical image to be identified of input,
The identification classification results of output Medical imaging module can be sent to store, it can be straight as medical image to be identified
It connects and is obtained from system, may also set up Medical Images Classification device when sending classification results while sending the medicine figure to be identified
Picture.
The function of each functional module of Medical Images Classification device described in the embodiment of the present invention can be implemented according to above-mentioned apparatus
Specific implementation in example, specific implementation process are referred to the associated description of above-mentioned apparatus embodiment, and details are not described herein again.
From the foregoing, it will be observed that the embodiment of the present invention improves the accuracy of Medical Images Classification, be conducive to precisely to identify early stage Ah
Alzheimer's disease medical image provides auxiliary for the prevention and diagnosis of early stage Alzheimer Disease patient.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with it is other
The difference of embodiment, same or similar part may refer to each other between each embodiment.For being filled disclosed in embodiment
For setting, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is referring to method part
Explanation.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
A kind of Medical Images Classification apparatus and system provided by the present invention is described in detail above.It answers herein
With a specific example illustrates the principle and implementation of the invention, the explanation of above example is only intended to help to manage
Solve method and its core concept of the invention.It should be pointed out that for those skilled in the art, not departing from
, can be with several improvements and modifications are made to the present invention under the premise of the principle of the invention, these improvement and modification also fall into this hair
In bright scope of protection of the claims.
Claims (10)
1. a kind of Medical Images Classification device, which is characterized in that including classifier prebuild module and image classification module；
The classifier prebuild module includes ROI region extracting sub-module, model training submodule, optimum combination screening submodule
Block and model integrated submodule；
Described image categorization module obtains described for medical image to be identified to be input in the Image Classifier constructed in advance
The classification results of medical image to be identified；
Wherein, the ROI region extracting sub-module is for extracting multiple areas ROI that training sample concentrates each medicine sample image
Area image, the training sample set include the brain area image of healthy population with alzheimer's disease difference illness Phase patient；Institute
Model training submodule is stated for being trained respectively to every class ROI region image using Three dimensional convolution neural network, obtains phase
The base classifier answered；Optimum combination screening submodule is used to be based on preset fitness value, using genetic algorithm from
Optimal base classifier is selected in multiple base classifiers；Described in the model integrated submodule is used to generate based on each optimum classifier
Image Classifier.
2. Medical Images Classification device according to claim 1, which is characterized in that the model integrated submodule includes:
Target ROI region determination unit, for being to determine the corresponding target ROI region of each optimal base classifier；
Target ROI region combines image extraction unit, for each medicine sample image in the trained set of stereotypes, simultaneously
Multiple target ROI regions are extracted, to constitute target ROI region combination image；
Initial pictures classifier training unit, for utilizing the Three dimensional convolution neural network to each target ROI region constitutional diagram
As being trained, initial pictures classifier is obtained；
Integrated unit, for integrating the initial pictures classifier and each optimal base classifier, to generate described image
Classifier.
3. Medical Images Classification device according to claim 2, which is characterized in that the classifier prebuild module is also wrapped
It includes base classifier and selects submodule, the base classifier selects submodule and includes:
Sequencing unit, for being ranked up according to the classification accuracy of base classifier to each base classifier,
Unit is deleted, the base classifier for being not more than preset threshold for deleting classification accuracy.
4. Medical Images Classification device according to claim 2, which is characterized in that the ROI region extracting sub-module is
Utilize the module of 116 ROI region images of each medicine sample image of ALL116 template extraction.
5. Medical Images Classification device according to claim 1, which is characterized in that the Three dimensional convolution neural network is to adopt
With the two-dimensional network level and three-dimensional net structure of VGG16 model, and the network structure that number of filters is 8.
6. according to claim 1 to Medical Images Classification device described in 5 any one, which is characterized in that the classifier is pre-
Building module further includes image preprocessing submodule, and described image pretreatment submodule includes:
First denoising unit, for concentrating each medicine sample image to carry out the dynamic correction of head and stripping skull behaviour the training sample
Make, is influenced with removing noise and non-brain tissue structure；
Registration process unit is sat with registration to unified for treated each medicine sample image to be carried out Spatial normalization
Mark space.
7. Medical Images Classification device according to claim 6, which is characterized in that described image pretreatment submodule also wraps
It includes:
Gaussian smoothing unit, for being carried out at Gaussian smoothing to each medicine sample image by Spatial normalization processing
Reason.
8. Medical Images Classification device according to claim 7, which is characterized in that described image pretreatment submodule also wraps
It includes:
Normalization unit, for carrying out gray scale normalization to by each medicine sample image of Gaussian smoothing.
9. a kind of Medical Images Classification system, which is characterized in that including the medical image point as described in claim 1-8 any one
Class device.
10. Medical Images Classification device according to claim 9, which is characterized in that further include and the medical image point
The connected Medical imaging module of class device；
The Medical imaging module is for storing current medical image to be identified, the medical image packet
Medical image and label information to be identified are included, the label information is the described to be identified of Medical Images Classification device output
The classification results of medical image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910376120.4A CN110097128B (en) | 2019-05-07 | 2019-05-07 | Medical image classification device and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910376120.4A CN110097128B (en) | 2019-05-07 | 2019-05-07 | Medical image classification device and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110097128A true CN110097128A (en) | 2019-08-06 |
| CN110097128B CN110097128B (en) | 2023-02-07 |
Family
ID=67447072
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910376120.4A Active CN110097128B (en) | 2019-05-07 | 2019-05-07 | Medical image classification device and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110097128B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674773A (en) * | 2019-09-29 | 2020-01-10 | 燧人(上海)医疗科技有限公司 | Dementia recognition system, device and storage medium |
| CN112364892A (en) * | 2020-10-21 | 2021-02-12 | 北京麟卓信息科技有限公司 | Image identification method and device based on dynamic model |
| CN112561935A (en) * | 2020-12-26 | 2021-03-26 | 广东工业大学 | Method, device and equipment for identifying Alzheimer's disease |
| CN112884754A (en) * | 2021-03-11 | 2021-06-01 | 广东工业大学 | Multi-modal Alzheimer's disease medical image recognition and classification method and system |
| CN113344198A (en) * | 2021-06-09 | 2021-09-03 | 北京三快在线科技有限公司 | Model training method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107131883A (en) * | 2017-04-26 | 2017-09-05 | 中山大学 | The full-automatic mobile terminal indoor locating system of view-based access control model |
| CN107480677A (en) * | 2017-08-07 | 2017-12-15 | 北京深睿博联科技有限责任公司 | The method and device of area-of-interest in a kind of identification three-dimensional CT image |
| US20180276813A1 (en) * | 2017-03-23 | 2018-09-27 | International Business Machines Corporation | Weakly supervised probabilistic atlas generation through multi-atlas label fusion |
| CN109589092A (en) * | 2018-10-08 | 2019-04-09 | 广州市本真网络科技有限公司 | Method and system are determined based on the Alzheimer's disease of integrated study |
-
2019
- 2019-05-07 CN CN201910376120.4A patent/CN110097128B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180276813A1 (en) * | 2017-03-23 | 2018-09-27 | International Business Machines Corporation | Weakly supervised probabilistic atlas generation through multi-atlas label fusion |
| CN107131883A (en) * | 2017-04-26 | 2017-09-05 | 中山大学 | The full-automatic mobile terminal indoor locating system of view-based access control model |
| CN107480677A (en) * | 2017-08-07 | 2017-12-15 | 北京深睿博联科技有限责任公司 | The method and device of area-of-interest in a kind of identification three-dimensional CT image |
| CN109589092A (en) * | 2018-10-08 | 2019-04-09 | 广州市本真网络科技有限公司 | Method and system are determined based on the Alzheimer's disease of integrated study |
Non-Patent Citations (2)
| Title |
|---|
| MOMOYO ITO等: "Extraction of brain regions for image diagnosis of Alzheimer-type Dementia based on atrophy progress speeds", 《IEEE NUCLEAR SCIENCE SYMPOSUIM & MEDICAL IMAGING CONFERENCE》 * |
| 任湘: "基于深度学习的DCE-MRI乳腺癌诊断及分子分型预测", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674773A (en) * | 2019-09-29 | 2020-01-10 | 燧人(上海)医疗科技有限公司 | Dementia recognition system, device and storage medium |
| CN112364892A (en) * | 2020-10-21 | 2021-02-12 | 北京麟卓信息科技有限公司 | Image identification method and device based on dynamic model |
| CN112364892B (en) * | 2020-10-21 | 2021-11-23 | 北京麟卓信息科技有限公司 | Image identification method and device based on dynamic model |
| CN112561935A (en) * | 2020-12-26 | 2021-03-26 | 广东工业大学 | Method, device and equipment for identifying Alzheimer's disease |
| CN112561935B (en) * | 2020-12-26 | 2022-09-16 | 广东工业大学 | Intelligent classification method, device and equipment for brain images |
| CN112884754A (en) * | 2021-03-11 | 2021-06-01 | 广东工业大学 | Multi-modal Alzheimer's disease medical image recognition and classification method and system |
| CN113344198A (en) * | 2021-06-09 | 2021-09-03 | 北京三快在线科技有限公司 | Model training method and device |
| CN113344198B (en) * | 2021-06-09 | 2022-08-26 | 北京三快在线科技有限公司 | Model training method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110097128B (en) | 2023-02-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097128A (en) | Medical Images Classification apparatus and system | |
| CN110070935B (en) | Medical image synthesis method, classification method and device based on antagonistic neural network | |
| US10997466B2 (en) | Method and system for image segmentation and identification | |
| CN110121749A (en) | Deep learning medical system and method for Image Acquisition | |
| CN108230296A (en) | The recognition methods of characteristics of image and device, storage medium, electronic device | |
| CN106339571A (en) | Artificial Neural Network And A Method For The Classification Of Medical Image Data Records | |
| CN107624193A (en) | The method and system of semantic segmentation in laparoscope and endoscope 2D/2.5D view data | |
| CN1934589A (en) | Systems and methods providing automated decision support for medical imaging | |
| US20220254022A1 (en) | Method and system for automatic multiple lesion annotation of medical images | |
| US11693071B2 (en) | Systems and methods for mapping neuronal circuitry and clinical applications thereof | |
| CN110363760A (en) | The computer system of medical image for identification | |
| Whitehead et al. | A deep learning approach to spine segmentation using a feed-forward chain of pixel-wise convolutional networks | |
| Fashandi et al. | An investigation of the effect of fat suppression and dimensionality on the accuracy of breast MRI segmentation using U‐nets | |
| CN113159040B (en) | Method, device and system for generating medical image segmentation model | |
| Chuang et al. | Efficient triple output network for vertebral segmentation and identification | |
| CN110910342A (en) | Analyzing bone trauma by using deep learning | |
| AU2020223750B2 (en) | Method and System for Image Annotation | |
| Li et al. | Anatomy completor: A multi-class completion framework for 3d anatomy reconstruction | |
| Boorboor et al. | NeuRegenerate: A framework for visualizing neurodegeneration | |
| CN109124635A (en) | Model generating method, MRI scan method and system | |
| Braik et al. | Exploration of genetic algorithms and particle swarm optimization in improving the quality of medical images | |
| Paul et al. | Computer-Aided Diagnosis Using Hybrid Technique for Fastened and Accurate Analysis of Tuberculosis Detection with Adaboost and Learning Vector Quantization | |
| Gómez López | Soft computing y visión por ordenador para la identificación forense mediante comparación de radiografías | |
| Kamber | Automated detection of multiple sclerosis lesions in magnetic resonance images of the human brain | |
| EP4152248A1 (en) | Medical image analysis system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |