CN112465905A - Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning - Google Patents
Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning Download PDFInfo
- Publication number
- CN112465905A CN112465905A CN201910843425.1A CN201910843425A CN112465905A CN 112465905 A CN112465905 A CN 112465905A CN 201910843425 A CN201910843425 A CN 201910843425A CN 112465905 A CN112465905 A CN 112465905A
- Authority
- CN
- China
- Prior art keywords
- data
- deep learning
- brain region
- magnetic resonance
- mri
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Abstract
The invention discloses a method for positioning a characteristic brain region of magnetic resonance imaging data based on deep learning, which comprises the following steps: a. preprocessing original MRI image data to obtain preprocessed MRI data; b. performing ALFF mapping extraction on the preprocessed MRI data to obtain image data after mapping extraction; c. taking the image after mapping extraction as the input quantity of a convolutional neural network model, and carrying out deep learning, wherein the convolutional neural network model is of an inclusion structure, and obtaining the classification result of the MRI image data; d. and determining the characteristic brain area location according to the classification result. The invention can realize the characteristic brain region positioning obtained by the resting MRI image by adopting the related algorithm in the deep learning, and has high accuracy.
Description
Technical Field
The invention relates to the technical field of medical treatment, in particular to a method for positioning a characteristic brain region of magnetic resonance imaging data based on deep learning.
Background
In recent years, various neurological diseases are emerging, which not only seriously affect the health and life quality of patients, but also bring great economic burden to families and society of patients. The resting state magnetic resonance imaging is a basic research and clinical research method widely applied to various neurologic disease imaging, and is mainly used for measuring intrinsic or spontaneous brain activity according to low-frequency fluctuation of a blood oxygen level dependent signal. The traditional method of impact data analysis is hypothesis testing of voxels in the brain alone, with inferences being possible only at the group level.
Clinically, certain diseases can cause changes of brain structures, certain brain areas of the brains of patients with the diseases are changed on MRI medical images, doctors can preliminarily judge whether the patients suffer from the diseases according to the characteristics of the brain areas, so that the basis is provided for detecting the diseases, the workload of the doctors is reduced, and therefore, the characteristic brain areas for distinguishing the patients with the diseases from the patients without the diseases play a key role in detecting the diseases.
Deep learning is a branch of machine learning, and is mainly a learning method based on unsupervised feature learning and a feature hierarchical structure. The identification of the medical image is a branch of image identification, compared with other digital images, the medical image contains a large amount of information reflecting the health level of a human body, but at present, the information is mainly analyzed manually, and the medical image is not only easily influenced by the subjective factors of a clinician, but also easily causes data waste. Therefore, in the medical image big data era, it becomes important how to extract important information from medical images of patients with neurological and mental diseases to help clinicians accurately analyze the diseases of patients. With the continuous and deep application of the deep learning algorithm, researchers begin to pay more attention to the application of the deep learning algorithm in the medical field, and there are also many researchers applying the deep learning algorithm in the analysis processing of the magnetic resonance image.
The deep learning learns abstract or high-level characteristics of data from a large amount of data, so that the classification or prediction accuracy of a model on new data is improved, and compared with shallow machine learning, the deep learning has the following two characteristics:
1. and (3) feature learning: the deep learning method can automatically learn high-level characteristic representation of data from mass data according to different tasks, and can express intrinsic information of the data;
2. deep layer structure: the structure of the deep learning model generally has 5 or more layers of hidden layer nodes, and more nonlinear transformation units are included, so that the generalization capability of the model can be enhanced;
the common deep learning model comprises an automatic encoder model, a sparse coding model, a limiting Boltzmann machine model, a deep belief network model, a convolutional neural network model CNN and the like.
Disclosure of Invention
In order to better apply the characteristic information found in neuroimaging to clinical diagnosis, the method is realized by means of a related algorithm in deep learning, and the more complex images can be better processed by the powerful automatic characteristic extraction and high-efficiency characteristic expression capability of the deep learning.
The invention aims to provide a method for positioning a characteristic brain region of magnetic resonance imaging data based on deep learning, which can obtain the characteristic brain region positioning through a resting state MRI mapping image.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
the invention discloses a method for positioning a characteristic brain region of magnetic resonance imaging data based on deep learning, which comprises the following steps:
a. preprocessing original MRI image data to obtain preprocessed MRI data;
b. performing ALFF mapping extraction on the preprocessed MRI data to obtain image data after mapping extraction;
c. taking the image after mapping extraction as the input quantity of a convolutional neural network model, and carrying out deep learning, wherein the convolutional neural network model is of an inclusion structure, and obtaining the classification result of the MRI image data;
d. and determining the characteristic brain area location according to the classification result.
Preferably, the method for determining the characteristic brain region location according to the classification result comprises the following steps:
replacing the last full-connection layer by the global average pooling layer to obtain the average value of each feature map of the last convolution layer, and obtaining output through weighted sum; for each class c, the mean value of k of each feature map has a corresponding W, which is recorded as
The CAM model obtained through training is used for corresponding classes
Extracting and calculating the weighted sum of the characteristic graphs corresponding to the characteristic graphs, wherein the classification task of the CAM is as the following formula (1):
wherein
The widths and heights of the kth feature map are shown as u and v, then ycRepresents the calculated classification result, wherein
A weight result for each class is represented, obtained through the full connection layer;
operating the weights according to the formula (2) to obtain an output graph of the CAM,
therein
The weighted sum output result of the feature map is shown, and normalized to 0 to 1 to obtain a thermodynamic map corresponding to the MRI image.
As another preferred method, the method for determining the feature brain region location according to the classification result is as follows:
deriving the output characteristic diagram of the convolutional layer according to the classification result to obtain the weight
The weight is
Obtained by using formula (3):
will be provided with
Accumulating the output characteristic diagram to obtain a weight-weighted result characteristic diagram;
the weight-weighted resultant feature map removes the influence of negative values by equation (4):
preferably, the following substeps are included in step a:
a1, removing the data of the previous time points in the original MRI image data;
a2, time-layer correction;
a3, head movement correction: deleting data with head translation exceeding 2mm or head rotation exceeding 3 degrees in any direction of X, Y, Z;
a4, space normalization: scaling the MRI image data according to a standard brain template;
a5, smoothing, and performing null smoothing by convolving the MRI image data with a 3D gaussian kernel.
Preferably, the number of sub-steps a1 is 10.
Preferably, the ALFF map extraction on the preprocessed MRI data includes the following sub-steps:
b1, performing band-pass filtering on the signals among each time sequence, and selecting the signals with the frequency within the range of 0.01-0.08 Hz;
b2, performing fast Fourier transform on the signal obtained in b1, calculating a frequency power spectrum of the signal, and squaring the power spectrum to obtain low-frequency amplitude;
b3, calculating the mean value of the power spectrum of each voxel in the frequency range of 0.01-0.08 Hz, namely the ALFF value;
b4, ALFF normalization: and dividing the average value of the low-frequency amplitude and the average value of the low-frequency amplitude of the whole brain voxel to obtain a standard ALFF value.
Preferably, the raw MRI image data is brain MRI image data in a resting state.
Preferably, in step c, after the input quantity is the ALFF map extraction, the 3D data slice is converted into the PNG format data.
The invention can adopt a related algorithm in deep learning to realize the characteristic brain region positioning obtained by the resting state MRI mapping image, and has high accuracy.
The model of the invention allows to locate brain regions associated with diseases in MRI images without any prior knowledge. Not only can the diseased brain regions be reflected at the group level, but the diseased brain regions associated with each patient can be found for the individual level.
Drawings
FIG. 1 is a flow chart of pre-processing raw MRI image data;
FIG. 2 is an image of an ALFF map extraction;
FIG. 3 is a schematic diagram of a convolutional neural network;
FIG. 4 is a schematic diagram of the inclusion structure;
FIG. 5 is a block diagram of the present invention;
FIG. 6 is a schematic representation of a ROC curve;
figure 7 is a schematic diagram of a confusion matrix.
FIG. 8 is a schematic diagram of an experimental framework;
fig. 9 is a schematic diagram of the experimental results.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings.
The invention discloses a method for positioning a characteristic brain region of magnetic resonance imaging data based on deep learning, which actually processes the following steps:
the MRI data is pre-processed prior to analysis thereof, the main purpose of which is to detect and correct for differences in the acquisition of the subject data.
The flow of preprocessing is shown in fig. 1, and the specific steps of MRI data preprocessing are as follows:
1. top 10 time-point data for each subject MRI data were culled: when MRI data are collected, because initial MRI signals are possibly unstable and other factors, the accuracy of the MRI data included in analysis is ensured by eliminating data of the previous 10 time points in consideration of the problem of magnetic field uniformity;
2. and (3) correcting the time layer: because MRI data is composed of several "2D" layers (slices), the acquisition time of each layer is different, but different layers are acquired in the same layer by default in the subsequent analysis process, and therefore, the correction of the time layer is required, and the main purpose of the correction of the time layer is to avoid the mutual influence between adjacent brain regions when the brain scans;
3. head movement correction: the examinee inevitably has autonomous or involuntary head movement in the scanning process, small head movement can cause large change of BOLD signals, therefore, the head movement correction mainly has the function of removing the influence caused by head movement in the scanning process, so that the acquired experimental data approaches to reality, the preprocessing process takes translation of 2mm or rotation of 3 degrees in the direction of X, Y, Z axes as a standard, and the tested data exceeding the two specified ranges are eliminated;
4. spatial normalization: because the brain of each person has obvious differences in shape, size and the like, all data need to be standardized, specifically, MRI data of different subjects are standardized according to the proportional size according to a standard brain template;
5. smoothing treatment: since the signal-to-noise ratio of the acquired MRI image is not high, it is necessary to convolve the data with a 3D gaussian kernel for null smoothing.
MRI feature extraction
The ALFF method reflects the spontaneous activity of brain neurons in a resting state directly through the amplitude of the change of the blood oxygen dependent horizontal signal baseline of the brain functional activity, converts the signal into a frequency domain for analysis, and calculates the power spectrum of functional magnetic resonance data in frequency. The method comprises the following specific steps:
(1) performing band-pass filtering on the signals between each time sequence, and selecting the signals with the frequency within the range of 0.01-0.08 Hz;
(2) performing fast Fourier transform on the signal obtained in the step (1), calculating a frequency power spectrum of the signal, and squaring the power spectrum to obtain low-frequency amplitude;
(3) calculating the mean value of the power spectrum of each voxel in the frequency range of 0.01-0.08 Hz, namely ALFF; (4) ALFF normalization: and dividing the average value of the low-frequency amplitude and the average value of the low-frequency amplitude of the whole brain voxel to obtain a standard ALFF value.
The image extracted by the ALFF map is shown in fig. 2.
The image subjected to feature mapping by using a low-frequency amplitude method is used as the input of the classification model, so that a better classification result can be obtained.
Convolutional neural network based classification recognition
As shown in fig. 3, the convolutional neural network generally includes a convolutional layer, a pooling layer, and a fully-connected layer, and finally, classification and identification are realized by using a classifier in the fully-connected layer, and an output layer is generally composed of a Softmax classifier.
Convolutional neural networks use mathematical convolution operations instead of traditional matrix operations, with two advantages: firstly, the neurons are connected in a local mode, so that the training time is reduced; and secondly, the number of weight parameters is reduced among the neuron nodes of the same layer of the convolutional neural network through a weight sharing mechanism, and the purpose of reducing the complexity of the convolutional neural network can be achieved.
Constructing convolutional neural networks
The convolutional neural network is inspired by a biological receptive field mechanism, and is finally developed into a deep learning model particularly suitable for image processing through continuous improvement of researchers. For the identification of MRI images, our framework is inspired by GoogleNet, and the mainstream method for improving CNN identification effect is to make the network deeper and wider before GoogleNet comes out, but this often results in the increase of overfitting and calculation amount. The inclusion structure proposed by GoogleNet can not only keep the sparsity of the network structure, but also utilize the high computational performance of the dense matrix, and the structure is shown in fig. 4.
Convolution kernels with different sizes are used, which means that the receptive fields with different sizes are obtained, and finally splicing means fusion of features with different scales; to reduce the amount of computation, a1 × 1 convolution kernel is used for dimensionality reduction. The final frame diagram is shown in fig. 5.
In the specific example, after the resting state MRI images of patients with migraine aura, patients without migraine aura and normal people are used as examples, the ALFF characteristics are extracted, 3D data slices are converted into PNG format and input into the CNN framework, and the final average classification accuracy reaches about 90%.
The ROC curve of the test results is shown in FIG. 6.
The confusion matrix of the test results is shown in figure 7.
Grad-CAM technology-based characteristic brain area positioning
At present, most neuroimaging studies find abnormal brain regions by comparing MRI data of patient groups and healthy groups through a method of principal variable analysis, and the obtained results are limited to statistics of group level and are difficult to be accurate to individual level. The use of machine learning-based medical image data analysis has been an encouraging advancement in the identification of brain diseases over the past few years. Machine learning enables localization of diseased brain regions at the individual level, as compared to being based on a large number of univariate analyses. In recent years, with the development of deep learning, the deep learning model has a powerful automatic feature extraction function, and high-level features of input data are extracted to create the deep learning model. Today, deep learning models are widely used to improve the accuracy of classification models, which can mine more disease-related information of MRI images.
Few studies have used deep learning techniques to obtain a classification result of an MRI image, and then use a classification model to find a characteristic brain region related to a disease, and for a black box model such as a deep model, how to identify the model, i.e., find which regions in a picture contain classification information related to the disease, is helpful to enhance understanding of pathogenesis of the disease. To achieve this goal, the cam (class activation mapping) technique can be used to explain the deep learning model of the present invention to locate the most classified characteristic region in the picture, which shows its decision basis in the form of thermodynamic diagram, as telling us which heat-generating object is in the night.
For a deep convolutional neural network, after convolution and pooling for many times, the last convolutional layer of the convolutional neural network contains the most abundant spatial and semantic information, and then the last convolutional layer is a full-link layer and a softmax layer, wherein the contained information is difficult to understand by human beings and is difficult to display in a visual mode. Therefore, to make the convolutional neural network give a reasonable interpretation to its classification result, the last convolutional layer must be fully utilized. The CAM replaces the last fully-connected layer with a Global Average Pooling layer (GAP), which is as large as the entire feature map. I.e. averaging all pixels of each feature map. After GAP, we get the mean of each feature map of the last convolution layer, and get the output by weighted sumAnd (6) discharging. For each class c, the mean value of k of each feature map has a corresponding W, which is recorded as
The CAM model obtained through training is used for corresponding classes
Extracted and the weighted sum of their feature maps corresponding to them is obtained.
The classification task of the CAM can be expressed as:
wherein
The widths and heights of the kth feature map are shown as u and v, then ycRepresents the calculated classification result, wherein
One weight result for each class is represented, obtained through the fully connected layer. The output map of the CAM is obtained by applying this weight as follows.
Therein
The weighted and output result of the feature map is shown, and a thermodynamic diagram can be drawn corresponding to the category interested region in the image by normalizing the result to be between 0 and 1.
Since the CAM technology requires modification of the structure of the original model, the model needs to be retrained, which greatly limits its usage scenarios. In our study we used the Grad-CAM proposed later, the basic idea of Grad-CAMThe way and the CAM are consistent, and a weighted sum is finally obtained by obtaining the corresponding weight of each pair of feature maps. But it differs from CAM mainly in weighting
The process of (1). The CAM is retrained to obtain weights by replacing the fully-connected layer with the GAP layer, and the Grad-CAM uses the global average of the gradients to calculate the weights. In fact, the Grad-CAM and the weights computed by the CAM are equivalent through strict mathematical derivation. First, the class output result of the training model is derived from the output characteristic diagram of the convolutional layer, e.g.
Weights similar to those found in GAP can be obtained by calculation
The formula is as follows:
will be provided with
The visualization result of the CAM can be obtained similarly to the feature map accumulation, and the weight represents the importance of the target class c to the feature mapping. Since we only focus on the influence of the most final classification result of positive values in the feature map, a ReLU function is needed to be applied to the weight-weighted result feature map to remove the influence of negative values. The results were:
the present study initially utilized convolutional neural networks to obtain patient MRI images.
The experimental frame is shown in FIG. 8
In this framework, 45 migraine patient data were used to obtain activation zones as shown in fig. 9 compared to normal:
in conclusion, the model of the present invention can locate brain regions associated with disease in MRI images without any a priori knowledge. The method can not only reflect the brain areas with diseases on a group level, but also find the brain areas with diseases related to each patient aiming at an individual level, and in the migraine experiment, the result basically accords with the brain areas obtained by a double-sample T test, but the accuracy of the method is still required to be proved by more researches.
The present invention is capable of other embodiments, and various changes and modifications may be made by one skilled in the art without departing from the spirit and scope of the invention.
Claims (8)
1. The method for locating the characteristic brain region of the magnetic resonance imaging data based on deep learning is characterized by comprising the following steps of:
a. preprocessing original MRI image data to obtain preprocessed MRI data;
b. performing mapping extraction such as ALFF, REHO, RCFS and the like on the preprocessed MRI data to obtain image data after mapping extraction;
c. taking the image after mapping extraction as the input quantity of a convolutional neural network model, and carrying out deep learning, wherein the convolutional neural network model is of an inclusion structure, and obtaining the classification result of the MRI image data;
d. and determining the characteristic brain area location according to the classification result.
2. The method of deep learning based magnetic resonance imaging data feature brain region localization according to claim 1, wherein the method of determining feature brain region localization according to the classification result is as follows:
replacing the last full-connection layer by the global average pooling layer to obtain the average value of each feature map of the last convolution layer, and obtaining output through weighted sum; corresponding to each class c, each featureThe mean values of k in the graph all have a corresponding W, noted as
The CAM model obtained through training is used for corresponding classes
Extracting and calculating the weighted sum of the characteristic graphs corresponding to the characteristic graphs, wherein the classification task of the CAM is as the following formula (1):
wherein
The widths and heights of the kth feature map are shown as u and v, then ycRepresents the calculated classification result, wherein
A weight result for each class is represented, obtained through the full connection layer;
operating the weights according to the formula (2) to obtain an output graph of the CAM,
therein
The weighted sum output result of the feature map is shown, and normalized to 0 to 1 to obtain a thermodynamic map corresponding to the MRI image.
3. The method of deep learning based magnetic resonance imaging data feature brain region localization according to claim 2, wherein the method of determining feature brain region localization according to the classification result is as follows:
deriving the output characteristic diagram of the convolutional layer according to the classification result to obtain the weight
The weight is
Obtained by using formula (3):
will be provided with
Accumulating the output characteristic diagram to obtain a weight-weighted result characteristic diagram;
the weight-weighted resultant feature map removes the influence of negative values by equation (4):
4. a method for feature brain region localization of magnetic resonance imaging data based on deep learning according to any one of claims 1-3, characterized in that in step a comprises the following sub-steps:
a1, removing the data of the previous time points in the original MRI image data;
a2, time-layer correction;
a3, head movement correction: deleting data with head translation exceeding 2mm or head rotation exceeding 3 degrees in any direction of X, Y, Z;
a4, space normalization: scaling the MRI image data according to a standard brain template;
a5, smoothing, and performing null smoothing by convolving the MRI image data with a 3D gaussian kernel.
5. The method for feature brain region localization based on deep learning magnetic resonance imaging data according to claim 4, wherein the number of sub-steps a1 is 10.
6. The method for feature brain region localization of deep learning based magnetic resonance imaging data according to claim 54, wherein the ALFF map extraction of the pre-processed MRI data comprises the sub-steps of:
b1, performing band-pass filtering on the signals among each time sequence, and selecting the signals with the frequency within the range of 0.01-0.08 Hz;
b2, performing fast Fourier transform on the signal obtained in b1, calculating a frequency power spectrum of the signal, and squaring the power spectrum to obtain low-frequency amplitude;
b3, calculating the mean value of the power spectrum of each voxel in the frequency range of 0.01-0.08 Hz, namely the ALFF value;
b4, ALFF normalization: and dividing the average value of the low-frequency amplitude and the average value of the low-frequency amplitude of the whole brain voxel to obtain a standard ALFF value.
7. The method for characteristic brain region localization of magnetic resonance imaging data based on deep learning of claim 1, wherein the raw MRI image data is resting brain MRI image data.
8. The method for locating the characteristic brain region based on the deep learning magnetic resonance imaging data as claimed in claim 6, wherein in step c, the input quantity is the data converted from the 3D data slice into the PNG format after the ALFF mapping is extracted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910843425.1A CN112465905A (en) | 2019-09-06 | 2019-09-06 | Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910843425.1A CN112465905A (en) | 2019-09-06 | 2019-09-06 | Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112465905A true CN112465905A (en) | 2021-03-09 |
Family
ID=74807221
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910843425.1A Pending CN112465905A (en) | 2019-09-06 | 2019-09-06 | Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112465905A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113052898A (en) * | 2021-04-08 | 2021-06-29 | 四川大学华西医院 | Point cloud and strong-reflection target real-time positioning method based on active binocular camera |
| CN113705670A (en) * | 2021-08-27 | 2021-11-26 | 上海交通大学医学院附属仁济医院 | Brain image classification method and device based on magnetic resonance imaging and deep learning |
| CN113925606A (en) * | 2021-10-25 | 2022-01-14 | 四川大学华西医院 | Nerve regulation and control navigation positioning method and device and nerve regulation and control treatment system |
| CN114419309A (en) * | 2022-01-07 | 2022-04-29 | 福州大学 | High-dimensional feature automatic extraction method based on brain T1-w magnetic resonance image |
| CN114595748A (en) * | 2022-02-21 | 2022-06-07 | 南昌大学 | Data segmentation method for fall protection system |
| CN115005798A (en) * | 2022-06-02 | 2022-09-06 | 四川大学 | Brain image feature extraction method based on edge connection function connection |
| CN116312971A (en) * | 2023-05-15 | 2023-06-23 | 之江实验室 | Cognitive training material generation method, cognitive training method, device and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096616A (en) * | 2016-06-08 | 2016-11-09 | 四川大学华西医院 | A kind of nuclear magnetic resonance image feature extraction based on degree of depth study and sorting technique |
| CN110168573A (en) * | 2016-11-18 | 2019-08-23 | 易享信息技术有限公司 | Spatial attention model for image labeling |
| CN110176001A (en) * | 2019-06-03 | 2019-08-27 | 浙江大学 | A kind of high iron catenary insulator breakage accurate positioning method based on Grad-CAM algorithm |
-
2019
- 2019-09-06 CN CN201910843425.1A patent/CN112465905A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096616A (en) * | 2016-06-08 | 2016-11-09 | 四川大学华西医院 | A kind of nuclear magnetic resonance image feature extraction based on degree of depth study and sorting technique |
| CN110168573A (en) * | 2016-11-18 | 2019-08-23 | 易享信息技术有限公司 | Spatial attention model for image labeling |
| CN110176001A (en) * | 2019-06-03 | 2019-08-27 | 浙江大学 | A kind of high iron catenary insulator breakage accurate positioning method based on Grad-CAM algorithm |
Non-Patent Citations (3)
| Title |
|---|
| HAO YANG 等: "Multimodal MRI-based classification of migraine: using deep learning convolutional neural network", 《BIOMEDICAL ENGINEERING ONLINE 2018》 * |
| R.R.SELVARAJU 等: "Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| Y. LIU 等: "A Pilot Study of Diabetes Mellitus Classification from rs-fMRI Data Using Convolutional Neural Networks", 《MATHEMATICAL PROBLEMS IN ENGINEERING》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113052898A (en) * | 2021-04-08 | 2021-06-29 | 四川大学华西医院 | Point cloud and strong-reflection target real-time positioning method based on active binocular camera |
| CN113052898B (en) * | 2021-04-08 | 2022-07-12 | 四川大学华西医院 | Point cloud and strong-reflection target real-time positioning method based on active binocular camera |
| CN113705670A (en) * | 2021-08-27 | 2021-11-26 | 上海交通大学医学院附属仁济医院 | Brain image classification method and device based on magnetic resonance imaging and deep learning |
| CN113925606A (en) * | 2021-10-25 | 2022-01-14 | 四川大学华西医院 | Nerve regulation and control navigation positioning method and device and nerve regulation and control treatment system |
| CN113925606B (en) * | 2021-10-25 | 2023-07-07 | 四川大学华西医院 | Nerve regulation navigation positioning method and device and nerve regulation treatment system |
| CN114419309A (en) * | 2022-01-07 | 2022-04-29 | 福州大学 | High-dimensional feature automatic extraction method based on brain T1-w magnetic resonance image |
| CN114595748A (en) * | 2022-02-21 | 2022-06-07 | 南昌大学 | Data segmentation method for fall protection system |
| CN114595748B (en) * | 2022-02-21 | 2024-02-13 | 南昌大学 | Data segmentation method for fall protection system |
| CN115005798A (en) * | 2022-06-02 | 2022-09-06 | 四川大学 | Brain image feature extraction method based on edge connection function connection |
| CN116312971A (en) * | 2023-05-15 | 2023-06-23 | 之江实验室 | Cognitive training material generation method, cognitive training method, device and medium |
| CN116312971B (en) * | 2023-05-15 | 2023-08-29 | 之江实验室 | Cognitive training material generation method, cognitive training method, device and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112465905A (en) | Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning | |
| Zhao et al. | Supervised segmentation of un-annotated retinal fundus images by synthesis | |
| Almotiri et al. | Retinal vessels segmentation techniques and algorithms: a survey | |
| CN110503630B (en) | Cerebral hemorrhage classifying, positioning and predicting method based on three-dimensional deep learning model | |
| Bourbakis | Detecting abnormal patterns in WCE images | |
| Arif et al. | Automated detection of nonmelanoma skin cancer based on deep convolutional neural network | |
| CN112348059A (en) | Deep learning-based method and system for classifying multiple dyeing pathological images | |
| CN109118487A (en) | Bone age assessment method based on non-down sampling contourlet transform and convolutional neural networks | |
| CN115147600A (en) | GBM multi-mode MR image segmentation method based on classifier weight converter | |
| Dresp-Langley et al. | Using the quantization error from Self‐Organizing Map (SOM) output for fast detection of critical variations in image time series | |
| CN114998651A (en) | Skin lesion image classification and identification method, system and medium based on transfer learning | |
| Alkabawi et al. | Computer-aided classification of multi-types of dementia via convolutional neural networks | |
| Chen et al. | Skin lesion segmentation using recurrent attentional convolutional networks | |
| Javidi et al. | Retinal image assessment using bi-level adaptive morphological component analysis | |
| CN115496720A (en) | Gastrointestinal cancer pathological image segmentation method based on ViT mechanism model and related equipment | |
| CN116579975A (en) | Brain age prediction method and system of convolutional neural network | |
| Asyhar et al. | Implementation LSTM Algorithm for Cervical Cancer using Colposcopy Data | |
| Bindhu et al. | Segmentation of skin cancer using Fuzzy U-network via deep learning | |
| Liu et al. | Weakly-supervised localization and classification of biomarkers in OCT images with integrated reconstruction and attention | |
| CN112861881A (en) | Honeycomb lung recognition method based on improved MobileNet model | |
| CN116030063B (en) | Classification diagnosis system, method, electronic device and medium for MRI image | |
| Kanse et al. | HG-SVNN: harmonic genetic-based support vector neural network classifier for the glaucoma detection | |
| Li et al. | MC-UNet: Multimodule Concatenation Based on U-Shape Network for Retinal Blood Vessels Segmentation | |
| Pallawi et al. | Study of Alzheimer’s disease brain impairment and methods for its early diagnosis: a comprehensive survey | |
| Bellary | The MRI knee pain classification using CNN algorithm and segmentation using clustering algorithm |
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 |