CN109222972A - A kind of full brain data classification method of fMRI based on deep learning - Google Patents
A kind of full brain data classification method of fMRI based on deep learning Download PDFInfo
- Publication number
- CN109222972A CN109222972A CN201811054390.5A CN201811054390A CN109222972A CN 109222972 A CN109222972 A CN 109222972A CN 201811054390 A CN201811054390 A CN 201811054390A CN 109222972 A CN109222972 A CN 109222972A
- Authority
- CN
- China
- Prior art keywords
- fmri
- layer
- image
- dimensional
- full
- 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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
- A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
- A61B5/0042—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part for the brain
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2576/00—Medical imaging apparatus involving image processing or analysis
- A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
- A61B2576/026—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part for the brain
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Radiology & Medical Imaging (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Veterinary Medicine (AREA)
- High Energy & Nuclear Physics (AREA)
- Neurology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of full brain data classification methods of fMRI based on deep learning, comprising: (1) obtains fMRI test data, pre-processed, obtain corresponding label；(2) the full brain data of fMRI are polymerize；(3) average 3-D image is sliced with orthogonal x, y, z axis direction respectively；(4) three groups of two dimensional images are respectively converted into a frame multichannel two dimensional image；(5) building is used for the mixing multichannel convolutive neural network model of the full brain data classification of fMRI；(6) fMRI data are handled, is trained using obtained tag along sort as input data, obtained parameter is used for the mixing convolutional neural networks model of the full brain data classification of fMRI；(7) fMRI data are handled, three obtained frame multichannel two dimensional images is input in the mixing convolutional neural networks model after training and are classified.The present invention can effectively improve the accuracy rate of the full brain data classification of fMRI, while reduce the calculation amount of fMRI full brain data classification model training and classification.
Description
Technical field
The present invention relates to data classification field more particularly to a kind of full brain data classification sides fMRI based on deep learning
Method.
Background technique
Functional mri (fMRI) is a kind of noninvasive brain function activity measurement means, and fMRI data reflect the mankind
The oxygen content of blood situation of brain, FMRI has been widely used in the fields such as cognitive science, development science, mental disease at present.
Deep learning is the method that a kind of pair of data carry out representative learning in machine learning, deep neural network (DNN), volume
Product neural network (CNN) and recurrent neural network (RNN) even deep learning model has been successfully applied to computer vision, voice is known
Not, the fields such as natural language processing.Deep learning model has been used for the classification to the full brain data of fMRI, but for complicated dynamic
The full brain data of fMRI, how in the case where keeping the lesser situation of calculation amount using deep learning improve classification accuracy, still
It is a problem to be solved.
Summary of the invention
The purpose of the present invention is to provide a kind of full brain data classification methods of fMRI based on deep learning.The present invention compares
It in the prior art, can preferably learn the full brain characteristic information of fMRI, and carry out model training using lesser calculation amount simultaneously.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of full brain data classification method of fMRI based on deep learning, specific steps include:
(1) the fMRI test data for obtaining test participant, pre-processes fMRI test data, while obtaining fMRI
The corresponding label of data；
(2) the full brain data of fMRI of each test participant are polymerize；
(3) the average 3-D image obtained after polymerization is sliced, obtains three with orthogonal x, y, z axis direction respectively
Group two dimensional image；
(4) obtain three groups of two dimensional images are respectively converted into a frame multichannel two dimensional image；
(5) building is used for the mixing multichannel convolutive neural network model of the full brain data classification of fMRI；
(6) the fMRI data that will be used for the participant of model training part pass through the processing of step (1)-(4), by what is obtained
Three frame multichannel two dimensional images and its tag along sort are input in mixing convolutional neural networks as input data and carry out model instruction
Practice, obtains the parameter of mixing convolutional neural networks, the mixing convolutional neural networks model for the full brain data classification of fMRI；
(7) step (1)-(4) processing is successively carried out to the fMRI data of acquisition, the three frame multichannel X-Y schemes that will be obtained
Classify as being input in the mixing convolutional neural networks model after training.
Specifically, the pretreatment in the step (1) include head shift calibrating, time horizon correction, Spatial normalization and
Space smoothing etc.；The label is the attribute (such as certain movement of test participant) of finger to finger test participant, or test participates in
The behavior property of person during the test (such as certain movement of test participant).
Specifically, three-dimensional to the N frame of acquisition if the full brain data of fMRI are tranquillization state fMRI data in step (2)
The tissue points of image (dimX × dimY × dimZ) corresponding position carry out arithmetic average, obtain a frame and are averaged 3-D image.
Specifically, in step (2), if the full brain data of fMRI are task state fMRI data, to the N in test process
Frame 3-D image uses signal intensity percentage (PSC) method, come calculate each tissue points during the test opposite tranquillization when
The average change value at quarter is converted into a frame and is averaged 3-D image.
Further, the average PSC calculation formula of each tissue points are as follows:
Wherein, N indicates the frame number of 3-D image during test, yiIndicate tissue points the i-th frame image value,It indicates
Average value of the tissue points at the tranquillization moment, tranquillization moment Selection experiment person calculate in the rest period of no trial stimulus, p expression
To the average change value of the tissue points.
It is dimX, y-axis dimY, z-axis dimZ that wherein the size of the 3-D image, which is x-axis,；During the test
N frame 3-D image label having the same.
Specifically, the concrete operations average 3-D image being sliced in step (3) are as follows: along the x-axis direction in x-axis
Each unit length is sliced, and dimX two dimensional images on the y-z plane are obtained, and every size is dimY × dimZ；
Unit length each in y-axis is sliced along the y-axis direction, obtains the two dimensional image of dimY on x-z plane, every big
Small is dimX × dimZ；Unit length each in z-axis is sliced along the z-axis direction, obtains dimZ two on the x-y plane
Image is tieed up, every size is dimX × dimY.With the two dimensional image on same level for one group, final one is obtained three group two
Tie up image.
Further, the step (4) specifically: according to the concept in channel in convolutional neural networks, for dimX y-z
Two dimensional image in plane, by the two dimensional image of each slice position as a channel, be converted into a frame can input into
The two dimensional image for having dimX channel of convolutional neural networks；For the two dimensional image on dimY x-z-planes, each is cut
For the two dimensional image of piece position as a channel, the dimY channel that have into convolutional neural networks can be inputted by being converted into a frame
Two dimensional image；For the two dimensional image in dimZ x-y planes, equally by the two dimensional image of each slice position as one
Channel, the two dimensional image for having dimZ channel into convolutional neural networks can be inputted by being converted into a frame.
Specifically, the mixing multichannel convolutive neural network model successively includes three in parallel from output is input to
Multichannel two-dimensional convolution neural network and a full Connection Neural Network.Wherein the input of each two-dimensional convolution neural network is corresponding
The output of a kind of multichannel two dimensional image, three multichannel two-dimensional convolution neural networks is spliced into one-dimensional characteristic with cascade,
It is input to full Connection Neural Network, the probability value of every kind of tag along sort of finally output prediction.
Further, the multichannel two-dimensional convolution neural network successively includes input layer (Input), the first convolutional layer
(Conv2d_1), the first pond layer (MaxPooling2d_1), the first Dropout layers, the second convolutional layer (Conv2d_2), second
Pond layer (MaxPooling2d_2), the 2nd Dropout layers and flattening layer (Flatten).The wherein convolution of the first convolutional layer
Nuclear volume is 32, and convolution kernel size is 3 × 3；The convolution nuclear volume of second convolutional layer is 64, and convolution kernel size is 3 × 3.It is described
First convolutional layer and the second convolutional layer are all made of LeakyReLU function as activation primitive.First pond layer and the second pond
Change layer and be all made of maximum pondization operation, pond window size is 2 × 2.Described first Dropout layers and the 2nd Dropout layer it is equal
Retain upper one layer of result passed over 0.25 probability.The flattening layer is by the result flattening output of convolutional layer at a pile knot
Fruit.The one-dimensional result of three multichannel two-dimensional convolution neural networks output is spliced into one-dimensional characteristic by fused layer (Merge), defeated
Enter to full Connection Neural Network.
Further, the full Connection Neural Network successively includes the first full articulamentum (Dense_1), specification layer
(BatchNormalization), Dropout layers, the second full articulamentum (Dense_2).The wherein neuron of the first full articulamentum
Quantity is 625；The neuronal quantity of second full articulamentum is determined according to the categorical measure of classification task.The first full articulamentum
Using LeakyReLU function as activation primitive；Second full articulamentum is using Softmax function as activation primitive.The rule
Model layer is standardized upper one layer of transmitting result again, makes the mean value of its result close to 0, and standard deviation is close to 1.Dropout
Layer with 0.5 probability retain upper one layer pass over record a demerit.The output of full Connection Neural Network is multiple probability values, indicates pre-
Survey the probability value that result is every kind of tag along sort.
The present invention compared to the prior art, have it is below the utility model has the advantages that
The present invention extracts feature in three orthogonal planes using multichannel two-dimensional convolution.For fMRI high dimensional data
Feature enables to model learning to abundant feature in three orthogonal planes with the multichannel two-dimensional convolution of fast speed,
It avoids simultaneously using the biggish Three dimensional convolution of calculation amount, reduces calculation amount, improve the standard of the classification to the full brain data of fMRI
True rate and classification speed.
Detailed description of the invention
Fig. 1 is a kind of specific flow chart of the full brain data classification method of fMRI based on deep learning；
Fig. 2 is the structural schematic diagram for mixing convolutional neural networks.
Specific embodiment
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited
In this.
Embodiment
In the present embodiment, the psychomotor task for choosing Task fMRI acts five kinds --- mobile right finger, movement
Left-hand finger squeezes right crus of diaphragm toe, squeezes left foot toe, mobile tongue, fMRI data classify.
It is as shown in Figure 1 a kind of flow chart of the full brain data classification method of fMRI based on deep learning, specific steps
Include:
(1) the FMRI test data for obtaining test participant, pre-processes fMRI test data, while obtaining fMRI
The corresponding label of data；
The pretreatment includes head shift calibrating, time horizon correction, Spatial normalization and space smoothing etc.；
The label refers to the corresponding action classification of fMRI data, be respectively as follows: mobile right finger, mobile left-hand finger,
It squeezes right crus of diaphragm toe, squeeze left foot toe, mobile tongue.
(2) the full brain data of fMRI of each test participant are polymerize；
The full brain data of fMRI in the present embodiment are task state fMRI data, therefore, to the N frame three-dimensional figure in test process
As using signal intensity percentage (PSC) method, to calculate each tissue points being averaged with respect to the tranquillization moment during the test
Changing value is converted into a frame and is averaged 3-D image.
The average PSC calculation formula of each tissue points are as follows:
Wherein, N indicates the frame number of 3-D image during test, yiIndicate tissue points the i-th frame image value,It indicates
Average value of the tissue points at the tranquillization moment, tranquillization moment Selection experiment person calculate in the rest period of no trial stimulus, p expression
To the average change value of the tissue points.
The wherein size of the 3-D image are as follows: x-axis 91, y-axis 109, z-axis 91；N frame in the action process
3-D image action classification label having the same.
(3) the average 3-D image obtained after polymerization is sliced, obtains three with orthogonal x, y, z axis direction respectively
Group two dimensional image；
The detailed process that average 3-D image is sliced are as follows: be sliced along the x-axis direction, obtain 91 on the y-z plane
Two dimensional image, every size is 109 × 91；It is sliced along the y-axis direction, obtains 109 two dimensional images on x-z plane,
Every size is 91 × 91；It is sliced along the z-axis direction, obtains 91 two dimensional images on the x-y plane, every size is 91
×109.Final one is obtained three groups of two dimensional images.
(4) obtain three groups of two dimensional images are respectively converted into a frame multichannel two dimensional image；
Specific conversion process are as follows: by the two dimensional image on 91 y-z planes, be converted into the X-Y scheme that a frame has 91 channels
Picture；By the two dimensional image on 109 x-z-planes, it is converted into the two dimensional image that a frame there are 109 channels；It will be in 91 x-y planes
Two dimensional image, be converted into the two dimensional image that a frame has 91 channels.
(5) building is used for the mixing multichannel convolutive neural network model of the full brain data classification of fMRI；
Specifically, the mixing multichannel convolutive neural network model structure as shown in Fig. 2, specifically: from being input to
Output successively includes that three multichannel two dimensions in parallel involve in neural network and a full Connection Neural Network.Wherein each two
The input of dimension convolutional neural networks corresponds to a kind of multichannel two dimensional image, the output of three multichannel two-dimensional convolution neural networks with
Cascade is spliced into one-dimensional characteristic, is input to full Connection Neural Network, the probability value of every kind of tag along sort of finally output prediction.
The multichannel two-dimensional convolution neural network successively include input layer (Input), the first convolutional layer (Conv2d_1),
First pond layer (MaxPooling2d_1), the first Dropout layers, the second convolutional layer (Conv2d_2), the second pond layer
(MaxPooling2d_2), the 2nd Dropout layers and flattening layer (Flatten).The wherein convolution nuclear volume of the first convolutional layer
It is 32, convolution kernel size is 3 × 3；The convolution nuclear volume of second convolutional layer is 64, and convolution kernel size is 3 × 3.The first volume
Lamination and the second convolutional layer are all made of LeakyReLU function as activation primitive.First pond layer and the second pond layer are equal
It is operated using maximum pondization, pond window size is 2 × 2.Described first Dropout layers and the 2nd Dropout layers with 0.25
Probability retain one layer of result passed over.The flattening layer is by the result flattening output of convolutional layer at a pile result.Three
The one-dimensional result of a multichannel two-dimensional convolution neural network output is spliced into one-dimensional characteristic by fused layer (Merge), is input to
Full Connection Neural Network.
The full Connection Neural Network successively includes the first full articulamentum (Dense_1), specification layer
(BatchNormalization), Dropout layers, the second full articulamentum (Dense_2).The wherein neuron of the first full articulamentum
Quantity is 625；The neuronal quantity of second full articulamentum is determined according to the categorical measure of classification task.The first full articulamentum
Using LeakyReLU function as activation primitive；Second full articulamentum is using Softmax function as activation primitive.The rule
Model layer is standardized upper one layer of transmitting result again, makes the mean value of its result close to 0, and standard deviation is close to 1.Dropout
Layer with 0.5 probability retain upper one layer pass over record a demerit.The output of full Connection Neural Network is multiple probability values, indicates pre-
Survey the probability value that result is every kind of tag along sort.
(6) the fMRI data that will be used for the participant of model training part pass through the processing of step (1)-(4), by what is obtained
Three frame multichannel two dimensional image machine sort labels are input in mixing convolutional neural networks as input data and carry out model instruction
Practice, obtains the parameter of mixing convolutional neural networks, the mixing convolutional neural networks model for the full brain data classification of fMRI；
(7) step (1)-(4) processing is successively carried out to the fMRI data of acquisition, the three frame multichannel X-Y schemes that will be obtained
Classify as being input in the mixing convolutional neural networks model after training.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (9)
1. a kind of full brain data classification method of fMRI based on deep learning, which is characterized in that specific steps include:
(1) the fMRI test data for obtaining test participant, pre-processes fMRI test data, while obtaining fMRI data
Corresponding label；
(2) the full brain data of fMRI of each test participant are polymerize；
(3) the average 3-D image obtained after polymerization is sliced, obtains three group two with orthogonal x, y, z axis direction respectively
Tie up image；
(4) obtain three groups of two dimensional images are respectively converted into a frame multichannel two dimensional image；
(5) building is used for the mixing multichannel convolutive neural network model of the full brain data classification of fMRI；
(6) the fMRI data that will be used for the participant of model training part pass through the processing of step (1)-(4), three frames that will be obtained
Multichannel two dimensional image and its tag along sort are input in mixing convolutional neural networks as input data and carry out model training,
Obtain the parameter of mixing convolutional neural networks, the mixing convolutional neural networks model for the full brain data classification of fMRI；
(7) step (1)-(4) processing is successively carried out to the fMRI data of acquisition, three obtained frame multichannel two dimensional images are defeated
Enter and classifies in the mixing convolutional neural networks model to after training.
2. the full brain data classification method of a kind of fMRI based on deep learning according to claim 1, which is characterized in that institute
Stating the pretreatment in step (1) includes head shift calibrating, time horizon correction, Spatial normalization and space smoothing；The label
It is the behavior property of the attribute or test participant of finger to finger test participant during the test.
3. the full brain data classification method of a kind of fMRI based on deep learning according to claim 1, which is characterized in that
In step (2), if the full brain data of fMRI are tranquillization state fMRI data, to the N frame 3-D image of acquisition (dimX × dimY ×
DimZ) tissue points of corresponding position carry out arithmetic average, obtain a frame and are averaged 3-D image；
If the full brain data of fMRI are task state fMRI data, signal intensity is used to the N frame 3-D image in test process
It is flat to be converted into a frame to calculate each tissue points during the test with respect to the average change value at tranquillization moment for percentage ratio method
Equal 3-D image.
4. the full brain data classification method of a kind of fMRI based on deep learning according to claim 3, which is characterized in that every
The calculation formula of the average signal variation percentage of a tissue points are as follows:
Wherein, N indicates the frame number of 3-D image during test, yiIndicate tissue points the i-th frame image value,Indicate voxel
This is calculated in the rest period of no trial stimulus, p expression in average value of the point at the tranquillization moment, tranquillization moment Selection experiment person
The average change value of tissue points；
It is dimX, y-axis dimY, z-axis dimZ that wherein the size of the 3-D image, which is x-axis,；N frame during the test
3-D image label having the same.
5. the full brain data classification method of a kind of fMRI based on deep learning according to claim 1, which is characterized in that
The concrete operations that average 3-D image is sliced in step (3) are as follows: unit length each in x-axis is carried out along the x-axis direction
Slice, obtains dimX two dimensional images on the y-z plane, and every size is dimY × dimZ；Along the y-axis direction in y-axis
Each unit length is sliced, and dimY two dimensional images on x-z plane are obtained, and every size is dimX × dimZ；
Unit length each in z-axis is sliced along the z-axis direction, obtains the two dimensional image of dimZ on the x-y plane, every big
Small is dimX × dimY；With the two dimensional image on same level for one group, final one is obtained three groups of two dimensional images.
6. the full brain data classification method of a kind of fMRI based on deep learning according to claim 1, which is characterized in that institute
State step (4) specifically: will for the two dimensional image on dimX y-z planes according to the concept in channel in convolutional neural networks
The two dimensional image of each slice position is converted into a frame and can input into convolutional neural networks have dimX as a channel
The two dimensional image in a channel；For the two dimensional image on dimY x-z-planes, by the two dimensional image of each slice position as
One channel, the two dimensional image for having dimY channel into convolutional neural networks can be inputted by being converted into a frame；For dimZ
Two dimensional image in x-y plane, equally by the two dimensional image of each slice position as a channel, being converted into a frame can
Input the two dimensional image for having dimZ channel into convolutional neural networks.
7. the full brain data classification method of a kind of fMRI based on deep learning according to claim 1, which is characterized in that institute
Mixing multichannel convolutive neural network model is stated from output is input to, successively includes three multichannel two-dimensional convolution nerves in parallel
Network and a full Connection Neural Network；Wherein the input of each two-dimensional convolution neural network corresponds to a kind of multichannel X-Y scheme
The output of picture, three multichannel two-dimensional convolution neural networks is spliced into one-dimensional characteristic with cascade, is input to full connection nerve
Network, the probability value of every kind of tag along sort of finally output prediction.
8. the full brain data classification method of a kind of fMRI based on deep learning according to claim 7, which is characterized in that institute
Stating multichannel two-dimensional convolution neural network successively includes input layer (Input), the first convolutional layer (Conv2d_1), the first pond layer
(MaxPooling2d_1), the first Dropout layers, the second convolutional layer (Conv2d_2), the second pond layer (MaxPooling2d_
2), the 2nd Dropout layers and flattening layer (Flatten).Wherein the convolution nuclear volume of the first convolutional layer is 32, convolution kernel size
It is 3 × 3；The convolution nuclear volume of second convolutional layer is 64, and convolution kernel size is 3 × 3；First convolutional layer and the second convolutional layer
LeakyReLU function is all made of as activation primitive；First pond layer and the second pond layer are all made of maximum pondization operation,
Pond window size is 2 × 2；Described first Dropout layers and the 2nd Dropout layer with 0.25 probability retain it is upper one layer pass
The result passed；The flattening layer is by the result flattening output of convolutional layer at a pile result；Three multichannel two-dimensional convolution minds
One-dimensional result through network output is spliced into one-dimensional characteristic by fused layer (Merge), is input to full Connection Neural Network.
9. the full brain data classification method of a kind of fMRI based on deep learning according to claim 7, which is characterized in that institute
State full Connection Neural Network successively and include the first full articulamentum (Dense_1), specification layer (BatchNormalization),
Dropout layers, the second full articulamentum (Dense_2)；Wherein the neuronal quantity of the first full articulamentum is 625；Second full connection
The neuronal quantity of layer is determined according to the categorical measure of classification task；The first full articulamentum is made using LeakyReLU function
For activation primitive；Second full articulamentum is using Softmax function as activation primitive；The specification layer is by upper one layer of transmitting knot
Fruit is standardized again, makes the mean value of its result close to 0, and standard deviation is close to 1；Dropout layers retain upper one with 0.5 probability
What layer passed over records a demerit；The output of full Connection Neural Network is multiple probability values, and expression prediction result is every kind of tag along sort
Probability value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811054390.5A CN109222972B (en) | 2018-09-11 | 2018-09-11 | fMRI whole brain data classification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811054390.5A CN109222972B (en) | 2018-09-11 | 2018-09-11 | fMRI whole brain data classification method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109222972A true CN109222972A (en) | 2019-01-18 |
| CN109222972B CN109222972B (en) | 2020-09-22 |
Family
ID=65067767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811054390.5A Active CN109222972B (en) | 2018-09-11 | 2018-09-11 | fMRI whole brain data classification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109222972B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816037A (en) * | 2019-01-31 | 2019-05-28 | 北京字节跳动网络技术有限公司 | The method and apparatus for extracting the characteristic pattern of image |
| CN110197729A (en) * | 2019-05-20 | 2019-09-03 | 华南理工大学 | Tranquillization state fMRI data classification method and device based on deep learning |
| CN110192860A (en) * | 2019-05-06 | 2019-09-03 | 复旦大学 | A kind of the Brian Imaging intelligent test analyzing method and system of network-oriented information cognition |
| CN110246566A (en) * | 2019-04-24 | 2019-09-17 | 中南大学湘雅二医院 | Method, system and storage medium are determined based on the conduct disorder of convolutional neural networks |
| CN110322969A (en) * | 2019-07-03 | 2019-10-11 | 北京工业大学 | A kind of fMRI data classification method based on width study |
| CN110604572A (en) * | 2019-10-08 | 2019-12-24 | 江苏海洋大学 | Brain activity state identification method based on human brain characteristic map |
| CN110870770A (en) * | 2019-11-21 | 2020-03-10 | 大连理工大学 | ICA-CNN classified fMRI space activation map smoothing and broadening method |
| CN110916661A (en) * | 2019-11-21 | 2020-03-27 | 大连理工大学 | ICA-CNN classified fMRI intracerebral data time pre-filtering and amplifying method |
| CN110992351A (en) * | 2019-12-12 | 2020-04-10 | 南京邮电大学 | sMRI image classification method and device based on multi-input convolutional neural network |
| CN111046918A (en) * | 2019-11-21 | 2020-04-21 | 大连理工大学 | ICA-CNN classified fMRI data space pre-smoothing and broadening method |
| CN111709787A (en) * | 2020-06-18 | 2020-09-25 | 北京字节跳动网络技术有限公司 | Method, apparatus, electronic device, and medium for generating user retention time |
| CN111728590A (en) * | 2020-06-30 | 2020-10-02 | 中国人民解放军国防科技大学 | Individual cognitive ability prediction method and system based on dynamic function connection |
| US20210174939A1 (en) * | 2019-12-09 | 2021-06-10 | Tencent America LLC | Deep learning system for detecting acute intracranial hemorrhage in non-contrast head ct images |
| CN113096096A (en) * | 2021-04-13 | 2021-07-09 | 中山市华南理工大学现代产业技术研究院 | Microscopic image bone marrow cell counting method and system fusing morphological characteristics |
| CN113313232A (en) * | 2021-05-19 | 2021-08-27 | 华南理工大学 | Functional brain network classification method based on pre-training and graph neural network |
| CN113313673A (en) * | 2021-05-08 | 2021-08-27 | 华中科技大学 | TB-level cranial nerve fiber data reduction method and system based on deep learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107067395A (en) * | 2017-04-26 | 2017-08-18 | 中国人民解放军总医院 | A kind of nuclear magnetic resonance image processing unit and method based on convolutional neural networks |
| CN107145727A (en) * | 2017-04-26 | 2017-09-08 | 中国人民解放军总医院 | The medical image processing devices and method of a kind of utilization convolutional neural networks |
| CN107424145A (en) * | 2017-06-08 | 2017-12-01 | 广州中国科学院软件应用技术研究所 | The dividing method of nuclear magnetic resonance image based on three-dimensional full convolutional neural networks |
| CN107563434A (en) * | 2017-08-30 | 2018-01-09 | 山东大学 | A kind of brain MRI image sorting technique based on Three dimensional convolution neutral net, device |
| CN107767378A (en) * | 2017-11-13 | 2018-03-06 | 浙江中医药大学 | The multi-modal Magnetic Resonance Image Segmentation methods of GBM based on deep neural network |
-
2018
- 2018-09-11 CN CN201811054390.5A patent/CN109222972B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107067395A (en) * | 2017-04-26 | 2017-08-18 | 中国人民解放军总医院 | A kind of nuclear magnetic resonance image processing unit and method based on convolutional neural networks |
| CN107145727A (en) * | 2017-04-26 | 2017-09-08 | 中国人民解放军总医院 | The medical image processing devices and method of a kind of utilization convolutional neural networks |
| CN107424145A (en) * | 2017-06-08 | 2017-12-01 | 广州中国科学院软件应用技术研究所 | The dividing method of nuclear magnetic resonance image based on three-dimensional full convolutional neural networks |
| CN107563434A (en) * | 2017-08-30 | 2018-01-09 | 山东大学 | A kind of brain MRI image sorting technique based on Three dimensional convolution neutral net, device |
| CN107767378A (en) * | 2017-11-13 | 2018-03-06 | 浙江中医药大学 | The multi-modal Magnetic Resonance Image Segmentation methods of GBM based on deep neural network |
Non-Patent Citations (3)
| Title |
|---|
| HENG HUANG ET AL.: "Modeling Task fMRI Data Via Deep Convolutional Autoencoder", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
| HEUNG-IL SUK ET AL.: "State-spacemodel with deep learning for functional dynamics estimation in resting-state fMRI", 《NEUROIMAGE》 * |
| 张兆晨，冀俊忠: "基于卷积神经网络的fMRI数据分类方法", 《模式识别与人工智能》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816037A (en) * | 2019-01-31 | 2019-05-28 | 北京字节跳动网络技术有限公司 | The method and apparatus for extracting the characteristic pattern of image |
| CN109816037B (en) * | 2019-01-31 | 2021-05-25 | 北京字节跳动网络技术有限公司 | Method and device for extracting feature map of image |
| CN110246566A (en) * | 2019-04-24 | 2019-09-17 | 中南大学湘雅二医院 | Method, system and storage medium are determined based on the conduct disorder of convolutional neural networks |
| CN110192860A (en) * | 2019-05-06 | 2019-09-03 | 复旦大学 | A kind of the Brian Imaging intelligent test analyzing method and system of network-oriented information cognition |
| CN110192860B (en) * | 2019-05-06 | 2022-10-11 | 复旦大学 | Brain imaging intelligent test analysis method and system for network information cognition |
| CN110197729A (en) * | 2019-05-20 | 2019-09-03 | 华南理工大学 | Tranquillization state fMRI data classification method and device based on deep learning |
| CN110322969A (en) * | 2019-07-03 | 2019-10-11 | 北京工业大学 | A kind of fMRI data classification method based on width study |
| CN110604572A (en) * | 2019-10-08 | 2019-12-24 | 江苏海洋大学 | Brain activity state identification method based on human brain characteristic map |
| CN110870770B (en) * | 2019-11-21 | 2021-05-11 | 大连理工大学 | ICA-CNN classified fMRI space activation map smoothing and broadening method |
| CN111046918A (en) * | 2019-11-21 | 2020-04-21 | 大连理工大学 | ICA-CNN classified fMRI data space pre-smoothing and broadening method |
| CN110916661A (en) * | 2019-11-21 | 2020-03-27 | 大连理工大学 | ICA-CNN classified fMRI intracerebral data time pre-filtering and amplifying method |
| CN110870770A (en) * | 2019-11-21 | 2020-03-10 | 大连理工大学 | ICA-CNN classified fMRI space activation map smoothing and broadening method |
| CN111046918B (en) * | 2019-11-21 | 2022-09-20 | 大连理工大学 | ICA-CNN classified fMRI data space pre-smoothing and broadening method |
| US20210174939A1 (en) * | 2019-12-09 | 2021-06-10 | Tencent America LLC | Deep learning system for detecting acute intracranial hemorrhage in non-contrast head ct images |
| CN110992351A (en) * | 2019-12-12 | 2020-04-10 | 南京邮电大学 | sMRI image classification method and device based on multi-input convolutional neural network |
| CN110992351B (en) * | 2019-12-12 | 2022-08-16 | 南京邮电大学 | sMRI image classification method and device based on multi-input convolution neural network |
| CN111709787A (en) * | 2020-06-18 | 2020-09-25 | 北京字节跳动网络技术有限公司 | Method, apparatus, electronic device, and medium for generating user retention time |
| CN111709787B (en) * | 2020-06-18 | 2023-08-22 | 抖音视界有限公司 | Method, device, electronic equipment and medium for generating user retention time |
| CN111728590A (en) * | 2020-06-30 | 2020-10-02 | 中国人民解放军国防科技大学 | Individual cognitive ability prediction method and system based on dynamic function connection |
| CN113096096A (en) * | 2021-04-13 | 2021-07-09 | 中山市华南理工大学现代产业技术研究院 | Microscopic image bone marrow cell counting method and system fusing morphological characteristics |
| CN113313673A (en) * | 2021-05-08 | 2021-08-27 | 华中科技大学 | TB-level cranial nerve fiber data reduction method and system based on deep learning |
| CN113313673B (en) * | 2021-05-08 | 2022-05-20 | 华中科技大学 | TB-level cranial nerve fiber data reduction method and system based on deep learning |
| CN113313232A (en) * | 2021-05-19 | 2021-08-27 | 华南理工大学 | Functional brain network classification method based on pre-training and graph neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109222972B (en) | 2020-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109222972A (en) | A kind of full brain data classification method of fMRI based on deep learning | |
| CN106650806B (en) | A kind of cooperating type depth net model methodology for pedestrian detection | |
| KR102433971B1 (en) | Systems and Methods for Facial Expression Recognition and Annotation | |
| CN109685819B (en) | Three-dimensional medical image segmentation method based on feature enhancement | |
| CN110298262A (en) | Object identification method and device | |
| CN108734290B (en) | Convolutional neural network construction method based on attention mechanism and application | |
| CN114220035A (en) | Rapid pest detection method based on improved YOLO V4 | |
| CN110070107A (en) | Object identification method and device | |
| CN108710906B (en) | Real-time point cloud model classification method based on lightweight network LightPointNet | |
| CN108596039A (en) | A kind of bimodal emotion recognition method and system based on 3D convolutional neural networks | |
| CN109255375A (en) | Panoramic picture method for checking object based on deep learning | |
| CN106407889A (en) | Video human body interaction motion identification method based on optical flow graph depth learning model | |
| CN110414349A (en) | Introduce the twin convolutional neural networks face recognition algorithms of sensor model | |
| CN107871106A (en) | Face detection method and device | |
| CN107609638A (en) | A kind of method based on line decoder and interpolation sampling optimization convolutional neural networks | |
| CN106203487A (en) | A kind of image classification method based on Multiple Kernel Learning Multiple Classifier Fusion and device | |
| CN110197729A (en) | Tranquillization state fMRI data classification method and device based on deep learning | |
| CN110084285A (en) | Fish fine grit classification method based on deep learning | |
| CN107679543A (en) | Sparse autocoder and extreme learning machine stereo image quality evaluation method | |
| CN110222717A (en) | Image processing method and device | |
| CN110399821A (en) | Customer satisfaction acquisition methods based on facial expression recognition | |
| CN111488827A (en) | Crowd counting method and system based on multi-scale feature information | |
| CN110674774A (en) | Improved deep learning facial expression recognition method and system | |
| EP3557487A1 (en) | Generation of validation data with generative contradictory networks | |
| CN110059593A (en) | A kind of human facial expression recognition method based on feedback convolutional neural networks |
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 |