CN106709907A - MR image processing method and device - Google Patents
MR image processing method and device Download PDFInfo
- Publication number
- CN106709907A CN106709907A CN201611122086.0A CN201611122086A CN106709907A CN 106709907 A CN106709907 A CN 106709907A CN 201611122086 A CN201611122086 A CN 201611122086A CN 106709907 A CN106709907 A CN 106709907A
- Authority
- CN
- China
- Prior art keywords
- images
- area
- interest
- gray
- sequence
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- 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
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Multimedia (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention provides an MR image processing method. MR images of a plurality of sequences are comprised. The processing method comprises the steps that regions of interest are divided from each MR image; grayscale normalization processing is carried out on the regions of interest for the MR image of each sequence; feature calculation is carried out on based on each normalized region of interest to extract a texture feature; and feature training is carried out based on the extracted texture to acquire a classifier related to the classification result of the region of interest.
Description
Technical field
The present invention relates to image processing techniques, more particularly to a kind of method and device processed MR images.
Background technology
Nuclear magnetic resonance is a kind of physical phenomenon, and being widely used in physics, chemical-biological, medical science as a kind of analysis means faces
The fields such as bed detection.In order to avoid obscuring with radiophotography in nuclear medicine, magnetic resonance imaging art (Magnetic is referred to as
Resonance, MR).
MR is imaged the rf pulse sequence by applying certain CF to the human body in magnetostatic field, makes the hydrogen in human body
Proton is activated and electromagnetic induction phenomenon occurs.After stop pulse, proton produces MR signals in relaxation process.Believe by MR
Number the processing procedure such as reception, space encoding and image reconstruction, that is, produce MR images.
MR imagings have been applied to the imaging diagnosis of each system of whole body.For soft tissue, MR images can compared with other mode
To provide more structures and texture information, but doctor directly judges that area-of-interest lesion degree is difficult from MR images, only
After MR confirms to have lesion pathology can be obtained by histological analysis.
The content of the invention
The brief overview of one or more aspects given below is providing to the basic comprehension in terms of these.This general introduction is not
The extensive overview of all aspects for contemplating, and it is also non-to be both not intended to identify the key or decisive key element of all aspects
Attempt to define the scope in terms of any or all.Its unique purpose is to provide the one of one or more aspects in simplified form
A little concepts think the sequence of more detailed description given later.
According to an aspect of the present invention, there is provided a kind of processing method of MR images, the MR images include various sequences
MR images, the processing method includes：
From every width MR image segmentation area-of-interests；
MR images for every kind of sequence perform the normalized of the gray scale between each area-of-interest；
Each area-of-interest after based on normalization performs feature calculation with texture feature extraction；And
Features training is carried out based on the texture for being extracted, the grader of the classification results on the area-of-interest is obtained.
In one example, the area-of-interest includes focal area, and the normalized includes the MR for every kind of sequence
Image performs the normalized of the gray scale between focal area.
In one example, the normalized includes each MR images for every kind of sequence, by the gray scale of each focal area
Value is divided by the gray average in the focal area multiplied by with identical preset constant.
In one example, the area-of-interest also parenchyma section including anatomical tissue, the processing method also includes will be every
The gray scale in width MR image focus region subtracts the gray average of its parenchyma section to obtain revised focal area, the normalization
Treatment includes each MR images for every kind of sequence, the gray scale of the gray value of the focal area that will be respectively corrected divided by the MR images
Average is multiplied by with identical preset constant.
In one example, the processing method also includes：Focal area after normalization is performed at local contrast enhancing
Reason, local contrast enhancing treatment includes subtracting the gray value of the focal area after each normalization the estimate of local mean value
Afterwards again divided by the estimate of Local standard deviation, this feature is calculated to be included based on each through the enhanced focal area of local contrast
Feature calculation is performed with texture feature extraction.
In one example, the MR images of various sequences include the MR images of DWI sequences and ADC sequences.
In one example, the area-of-interest includes the parenchyma section of anatomical tissue, and the normalized is included for every
The MR images for planting sequence perform the normalized of the gray scale between the parenchyma section of anatomical tissue.
In one example, the normalized includes each MR images for every kind of sequence, by the essence of each anatomical tissue
The pixel gray level that the gray value in region is in the range of the σ of μ ± 3 is normalized to identical preset range, while discarding is not at μ
Pixel in the range of ± 3 σ, wherein μ refer to the gray average of the parenchyma section of current anatomical tissue, and σ refers to standard deviation.
In one example, the normalized includes each MR images for every kind of sequence, by the solid area of anatomical tissue
The gray value in domain is divided by the gray average in the parenchyma section of anatomical tissue multiplied by with identical preset constant.
In one example, the processing method also includes：Obtained grader is used to classify target image.
In one example, the classification results include the classification of liver tumour grade malignancy, capilary infiltration degree, liver fibrosis journey
At least one in degree classification, hepatic inflammatory grading.
According to another aspect of the present invention, there is provided a kind of processing unit of MR images, the MR images include various sequences
MR images, the processing unit includes：
Image segmentation module, for from every width MR image segmentation area-of-interests；
Normalization module, the normalization of the gray scale between each area-of-interest is performed for the MR images for every kind of sequence
Treatment；
Feature calculation module, performs feature calculation to extract texture spy for each area-of-interest after based on normalization
Levy；And
Training module, for carrying out features training based on the texture for being extracted, obtains the classification on the area-of-interest
The grader of result.
In one example, the area-of-interest includes focal area, MR image of the normalization module for every kind of sequence
Perform the normalized of the gray scale between focal area.
In one example, the normalization module removes the gray value of each focal area for each MR images of every kind of sequence
With the gray average in the focal area multiplied by with identical preset constant.
In one example, the area-of-interest also parenchyma section including anatomical tissue, the processing unit also includes amendment
Module, for the gray scale in every width MR image focus region to be subtracted the gray average of its parenchyma section to obtain revised focus
Region, for each MR images of every kind of sequence, the gray value of the focal area that will be respectively corrected is divided by the MR for the normalization module
The gray average of image is multiplied by with identical preset constant.
In one example, the processing unit also includes：Local contrast strengthens module, for by the focus after each normalization
The gray value in region is subtracted after the estimate of local mean value again divided by the estimate of Local standard deviation, this feature computing module base
Feature calculation is performed with texture feature extraction through the enhanced focal area of local contrast in each.
In one example, the MR images of various sequences include the MR images of DWI sequences and ADC sequences.
In one example, the area-of-interest includes the parenchyma section of anatomical tissue, and the normalization module is directed to every kind of sequence
The MR images of row perform the normalized of the gray scale between the parenchyma section of anatomical tissue.
In one example, the normalization module is directed to each MR images of every kind of sequence, by the parenchyma section of each anatomical tissue
Gray value be in μ ± 3 σ in the range of pixel gray level be normalized to identical preset range, while discarding be not at μ ± 3 σ
In the range of pixel, wherein μ refers to the gray average of the parenchyma section of current anatomical tissue, and σ refers to standard deviation.
In one example, the normalization module is directed to each MR images of every kind of sequence, by the parenchyma section of anatomical tissue
Gray value is divided by the gray average in the parenchyma section of anatomical tissue multiplied by with identical preset constant.
In one example, the processing unit also includes：Sort module, for using obtained grader to target image
Classified.
In one example, the classification results include the classification of liver tumour grade malignancy, capilary infiltration degree, liver fibrosis journey
At least one in degree classification, hepatic inflammatory grading.
The present invention is anticipated by a series of means to MR images, and such as normalized, focal area is subtracted
Gray average treatment, topography's enhancing treatment of anatomical tissue essence etc., extract on image thus through processing
Feature significantly improves the degree of accuracy of grader in classifier training.
Brief description of the drawings
After the detailed description for reading embodiment of the disclosure in conjunction with the following drawings, better understood when of the invention
Features described above and advantage.In the accompanying drawings, each component is not necessarily drawn to scale, and with similar correlation properties or feature
Component may have same or like reference.
Fig. 1 shows the block schematic illustration of the handling process of MR images according to an aspect of the present invention；
Fig. 2 shows 7 liver MR images of sequence；
Fig. 3 shows the flow chart of the processing method 300 of MR images according to an aspect of the present invention；
Fig. 4 a, Fig. 4 b respectively illustrate the liver tumour area-of-interest and liver parenchyma area-of-interest in liver MR images；
Fig. 5 a-5f show example of two tumours of the pernicious grade of difference under T2 sequences and local Contrast enhanced result
Example；
Fig. 6 shows typical case's training framework of grader；
Fig. 7 shows the general flow of feature selecting；
Fig. 8 shows the ROC curve contrast of liver tumour grade malignancy height test；And
Fig. 9 shows the block diagram of the processing unit of MR images according to an aspect of the present invention.
Specific embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.Note, it is below in conjunction with accompanying drawing and specifically real
It is only exemplary to apply the aspects of example description, and is understood not to carry out any limitation to protection scope of the present invention.
MR is imaged the rf pulse sequence by applying certain CF to the human body in magnetostatic field, makes the hydrogen in human body
Proton is activated and electromagnetic induction phenomenon occurs.After stop pulse, proton produces MR signals in relaxation process.Believe by MR
Number the processing procedure such as reception, space encoding and image reconstruction, that is, produce MR images.
Fig. 1 shows the block schematic illustration of the handling process of MR images according to an aspect of the present invention.Such as Fig. 1 institutes
Show, whole treatment framework includes backstage and foreground two parts.The MR images that back partition is mainly from the case collected are trained
Grader, foreground partition is then that the grader obtained using training is classified to new MR images, to obtain pathological analysis knot
Really.
In the following description of the present invention, MR image procossing schemes of the invention are described with the MR images of liver as example
's.However, MR image procossings scheme of the invention can also be used for the MR graphical analyses of other position organs.
Back partition mainly includes feature calculation 101,103 3 parts of feature selecting 102 and output category device.
Feature calculation refers to go out for the helpful feature of pathological analysis from the region of interesting extraction of MR images.MR images
" area-of-interest " refer to the part for needing in image to process emphatically, such as tumour area-of-interest, or anatomical tissue
Parenchyma section etc..Area-of-interest can manually be drawn by user, or can also be obtained with automatic/semi-automatic cutting techniques
Arrive.
The MR images of multiple sequences are make use of in the present invention.The sequence of NMRS is with certain bandwidth, necessarily
The radio-frequency pulse of amplitude and the organic assembling of gradient pulse, the radio-frequency pulse combination different from gradient pulse constitute different
Sequence, the characteristics of the image that different sequence is obtained has respective.
As an example, the present invention uses DWI, ADC, T1, T1_AP, T1_PP, T1_DP, T2 this 7 sequences, to this 7 sequences
The MR images of row are processed.Fig. 2 shows 7 liver MR images of sequence.
During feature calculation 101, the area-of-interest to MR images carries out feature extraction and calculation.It is any for pathology
Analyzing significant feature can be extracted by calculating.Usually, the feature based on Intensity (intensity) in image,
The statistical value such as such as average, Entropy (entropy), Skewness (degree of skewness), Kurtosis (kurtosis), based on Gradient (ladders
Degree) feature, for example statistical value such as average, difference of area-of-interest inside or edge gradient etc. is all for pathological analysis
Significant feature.
The step of feature selecting 102 is selected the feature for extracting, and is combined using selected feature in step 103
Case sample carries out classifier training, to obtain grader.
The concept of classification is to learn a classification function on the basis of data with existing or construct a disaggregated model, i.e.,
Grader (Classifier).The function or model can be mapped to the data recording in database a certain in given classification
It is individual, such that it is able to be applied to data prediction.Grader is the general designation of the method classified to sample in data mining, comprising certainly
Plan tree, logistic regression, naive Bayesian, neutral net scheduling algorithm.
For example after the MR images that have collected patient cases, with reference to the characteristics of image and the pathological examination of case of MR images
The grader that component goes out is established and established between feature and pathological examination a kind of mapping.So as to should can be with by the grader
MR images to new patient carry out classification prediction, to judge the pathological examination of the patient.This utilizes the assorting process of grader
Be 201 the step of foreground in carry out.The result for being exported is the pathology made based on MR images and judged.
In the present invention, with liver MR images as example, can be to the pernicious grade of sample, capilary Infiltrating, or fiber
Change grade, inflammatory grade to be classified.
Fig. 3 shows the flow chart of the processing method 300 of MR images according to an aspect of the present invention.
Here MR images may include the MR images of various sequences, such as DWI, ADC, T1, T1_AP, T1_PP, T1_DP,
T2 sequences.MR images can be the image of above-mentioned one or more sequence.These MR images are that patient shoots in hospital diagnosis
Image, can be from infection from hospital.These MR images as processing method of the invention basis.
In step 301：From every width MR image segmentation area-of-interests.
Area-of-interest can be the focal area in image, such as tumor region, or anatomical tissue parenchyma section.
Fig. 4 a show the liver tumour area-of-interest in liver MR images, and Fig. 4 b show the liver parenchyma region of interest in liver MR images
Domain.
The determination of area-of-interest as shown in Fig. 4 a, 4b, or can use automatic/semi-automatic skill by user's hand drawn
Art determines area-of-interest.
As described above, MR images can include one or more MR image of sequence.Assuming that having 50 cases, each case
There are 7 MR images of sequence, then determine area-of-interest from the MR images of each sequence of each case, then be therefrom partitioned into
Area-of-interest is analyzed for follow-up calculating.
Image segmentation can be performed using any of cutting techniques, be repeated no more again.
In step 302：MR images for every kind of sequence perform the normalized of the gray scale between each area-of-interest.
For each sequence, there is the difference in brightness and contrast in the image of different patients.In the present invention, it is right
In each sequence, the gray scale normalization between area-of-interest is performed.It ensure that data are special between different patient images
The uniformity levied, improves the accuracy of the grader of follow-up training.
In the present invention, for the normalization of area-of-interest, some special normalization computational methods are especially employed, from
From the point of view of experimental result, the accuracy of grader is substantially increased, as described in detail.
On the one hand, area-of-interest can be focal area.Now, the MR images that can be directed to every kind of sequence perform focal zone
The normalized of the gray scale between domain.
In one example, it is normalized (herein referred to as " based on sense by the average of each area-of-interest of unification
The identical average in interest region ").That is, for each MR images of every kind of sequence, by the gray value of each focal area divided by the focus
Gray average in region is multiplied by with identical preset constant.As an example, the preset constant can be set to such as 500.
This normalized has obvious help for the classifier training degree of accuracy of some illnesss.For example, for liver MR
The capilary of the liver tumour of image is impregnated with the classifier training of the judgement of nothing, significantly improves accuracy.
In another example, before normalization, first pre-processed by the parenchyma section of anatomical tissue.It is specific and
Speech, area-of-interest in addition to focal area, the also parenchyma section including anatomical tissue.Now, before normalization, for every
One width MR images, can subtract the gray average of its parenchyma section to be corrected by the gray value in every width MR image focus region
Focal area, normalization is then performed as object with the focal area that this is corrected.
Then, it is normalized (herein referred to as " identical average ") by the average of each image of unification.It is directed to
Each MR images of every kind of series, will in the above described manner pass through the gray scale of each focal area gray value divided by the MR images of amendment
Average is multiplied by with identical preset constant.As an example, the preset constant can be set to such as 500.
This normalized has obvious help for the classifier training degree of accuracy of some illnesss.For example, for liver MR
The classifier training of the judgement of the grade malignancy height of the liver tumour of image, significantly improves accuracy, because subtracting liver
Between the operation of parenchyma section gray average can be removed the different patients of same sequence, not a pair of the liver parenchyma brightness of Relative tumor
The influence of texture analysis.
On the other hand, area-of-interest can be the parenchyma section of anatomical tissue.By taking liver parenchyma region as an example, liver parenchyma sense
Interest region is referred on MR images in the range of liver, exclude blood vessel, bile duct, focus and motion artifacts, brightness and texture it is equal
Even hepatic tissue VOI.The analysis of the parenchyma section of anatomical tissue is meaningful for illness, by taking liver parenchyma region as an example, can
To obtain the information of liver fibrosis and inflammatory degree.
In the case of parenchyma section of the area-of-interest for anatomical tissue, the MR images that can be directed to every kind of sequence perform solution
Cut open the normalized of the gray scale between the parenchyma section of tissue.
In one example, normalized by the way that the value of image is limited in the range of 6 standard deviations of average (referred to herein as
It is " 3 times of standard deviations of mean value ± ").Specifically, for each MR images of every kind of sequence, by the ash of the parenchyma section of each anatomical tissue
The pixel gray level that angle value is in the range of the σ of μ ± 3 is normalized to identical preset range, for example, normalize in [0, VAL],
The pixel being not in the range of the σ of μ ± 3 is abandoned simultaneously, and wherein μ refers to the gray average of the parenchyma section of current anatomical tissue, σ
Standard deviation is referred to, the wherein VAL is a constant, such as can be 1000 as an example.
This normalization mode can remove the skew of average and multiplication sex differernce between different images.This normalization is right
There is obvious help in the classifier training degree of accuracy of some illnesss.For example, for liver MR images degree of hepatic fibrosis just
Judgement classifier training, significantly improve accuracy.
In another example, it is normalized yet by the average of each area-of-interest of unification.That is, for every kind of sequence
Each MR images of row, the gray average in parenchyma section by the gray value of the parenchyma section of anatomical tissue divided by anatomical tissue is again
It is multiplied by identical preset constant.
This normalization has obvious help for the classifier training degree of accuracy of some illnesss.For example, for liver MR images
Hepatic inflammatory degree height judgement classifier training, significantly improve accuracy.
In step 303：Each area-of-interest after based on normalization performs feature calculation with texture feature extraction.
As described above, any can be extracted for the significant feature of pathological analysis by calculating.Such as base in image
Feature in Intensity (intensity), the feature based on Gradient (gradient).As example, can calculate in the present invention
253 features of each area-of-interest, as shown in the following chart：
Table 1
In table 1, LBP-HF (Local Binary Pattern Histogram Fourier) refers to local binary patterns
Histogram Fourier's feature；GLCM (Grey-level co-occurrence matrix) refers to gray level co-occurrence matrixes feature.This
The calculating of a little features is all known technology, be will not be repeated here.
For 7 images of series, then can altogether to calculate 253x7=1771 texture special for the MR images of each case
Levy.Certainly, this is only example, it is also possible to calculate more or less textural characteristics.
Specifically, in the present invention, when calculating feature, the feature based on LCE images is especially calculated.
LCE (Local Contrast Enhancement) local contrast enhancing technology be a kind of enhancing image detail and
In the prior art of local message, it is suggested and for the feature calculation of Lung neoplasm detection.The LCE of two dimension is calculated, and is by image
The value of each pixel is subtracted after the estimate of local mean value, then divided by the estimate of Local standard deviation, is shown below：
The value of the pixel of coordinate x, y is located at wherein in i (x, y) representing input images, o (x, y) is output image.It is local equal
The calculating formula of value is：
μ (x, y)=i (x, y) * h (x, y)
The calculating formula of the estimate of Local standard deviation is：
Wherein * represents two-dimensional convolution, and the type and size of the selection of kernel function h (x, y) and target in pending image
It is relevant.For example, in order to process different size of, liver tumour or hepatic parenchymal area-of-interest, from 11*11 and two kinds of 51*51
The kernel function of the Gauss of size.The example of two pernicious grade tumors of difference under T2 sequences as shown in Fig. 5 a-f, corresponding LCE
Image is calculated by two kinds of different size of kernel functions respectively.Fig. 5 a are the example T2 images of pernicious II grades of liver tumour, Fig. 5 b and
Fig. 5 c are two kinds of LCE images of Fig. 5 a, and Fig. 5 d are the example T2 images of pernicious III level liver tumour, and Fig. 5 e and Fig. 5 f are Fig. 5 d's
Two kinds of LCE images.
In the present invention, portion of the region of interest area image executive board enhancing after to normalization, is then based on each warp
The enhanced focal area of local contrast performs feature calculation with texture feature extraction.Local enhancement technology is for some illnesss
The classifier training degree of accuracy has obvious help.For example, for the judgement of the grade malignancy height of the liver tumour of liver MR images
Classifier training, significantly improves accuracy.More preferably, can be using the image of DWI and ADC sequences, now classifier training is accurate
Exactness is also significantly increased.
In step 304：Features training is carried out based on the texture for being extracted, the classification knot on the area-of-interest is obtained
The grader of fruit.
After textural characteristics are extracted, can carry out features training to obtain grader based on these features.Grader is several
According to the general designation of the method classified to sample in excavation, comprising decision tree, logistic regression, naive Bayesian, neutral net etc.
Classic algorithm.The training of grader can use any algorithm known, and the training framework of grader is shown in Fig. 6.
By taking the pernicious classification of liver tumour as an example.86 cases are obtained from hospital, altogether 107 tumours.From at the doctor of cooperation
Obtain tumour and hepatic parenchymal area-of-interest in each sequence, and tumour pernicious classification pathological examination.Each tumour can
To calculate 1771 features.107 tumours are divided into 2 parts, used as training set, 35 tumours are used as test set for 72 tumours.
And training set is randomly divided into k parts, system carries out feature selecting based on training set, and the character subset that will be selected is rolled over K and intersected
Verify the formal verification classifying quality of (k-fold Cross Validation).For example it is bent with the Receiver Operating Characteristics for obtaining
The TG-AUC (AUC) of line (ROC), or the error rate classified, as the judge criterion of feature selecting.With what is finally selected
All cases in character subset, and training set, train a grader, and test this grader with the case of test set,
Obtain ROC curve and AUC.Grader described in this example is LDA.
As described above, in the big frame of Fig. 6, the training set of 72 tumours is randomly divided into k parts, the final test for obtaining test set
As a result this process of (AUC) can repeat n times, obtain AUC1, AUC2,…,AUCN, finally therefrom select the AUC pair of maximum
The character subset answered, and corresponding grader, tumour VOI being used for system as last output, to user input enter
The grader of the pernicious classification of row.
K folding cross validations are a conventional cross validation methods, will the data of training set be divided into k parts again, make
With wherein k-1 numbers according to as training data, training obtains grader and tests the number that remaining that portion is not engaged in training
According to repetition k times.As a result, all k numbers are gone over according to by wheel current test, after the test result of all these data collects
It is used to evaluate.In experiment of the invention, k takes 4.By taking the pernicious classification of above-mentioned liver tumour as an example, the training set of division includes 72
Individual tumour, after being divided into 4 parts, every part is 18 tumours.In cross validation being every time 3 parts is used to train, remaining 1 part
For testing.
Feature selecting flow in Fig. 6 is as shown in Figure 7.The search strategy of feature selecting is selected first.Common search plan
Slightly include forward, backward or the mode such as random.For example, according to SFS (to selection before order) strategy, character subset X is opened from empty set
Begin, one feature fx of selection adds character subset X every time so that evaluation function J (X) is optimal.It is exactly briefly to select every time
One so that the feature that the value of evaluation function is optimal is added.According to SFFS (being selected to trim before order) strategy, in SFS
On the basis of, to the feature in the character subset X that has selected, the evaluation function J (X) after removing certain feature fx is calculated one by one,
If J (X) is lifted on the contrary, the removal of feature fx is performed, that is, the existing features that increase of character subset X also have the behaviour for removing feature
Make.It is right using the feature selection approach based on mutual information according to the combined method of mRMR (minimal redundancy maximum correlation)+SFS
Feature is ranked up, and selects candidate characteristic set X ', then carries out the feature selection approach of the combining classification device such as SFS, selects final
Character subset X.In experiment of the invention, it is SFS methods to have used the feature selection approach based on mutual information.
After obtaining character subset, it is estimated with selected evaluation criterion, this feature is exported if stop condition is met
Collection, otherwise continues search for feature.In experiment of the invention, classifying quality of this feature subset in training set is employed to comment
Price card is accurate, i.e., training set data carries out the AUC obtained after cross validation.Stopping criterion is that new character subset cannot carry AUC
Rise and then stop search.
The relevant knowledge of classifier training and some means used in experiment of the invention have been briefly described above,
In view of the process of classifier training is the highly developed technology in this area, will not be repeated here.
By above-mentioned training, the grader of the classification results on area-of-interest can be obtained.
Although for make explanation simplify the above method is illustrated and is described as a series of actions, it should be understood that and understand,
The order that these methods are not acted is limited, because according to one or more embodiments, some actions can occur in different order
And/or with from it is depicted and described herein or herein it is not shown and describe but it will be appreciated by those skilled in the art that other
Action concomitantly occurs.
Such scheme of the invention, can train the grader on various classification results.For example, for liver MR
For image, classification results may include the classification of liver tumour grade malignancy, the classification of capilary infiltration degree, degree of hepatic fibrosis, liver
Inflammatory grading etc..
The present invention is anticipated by a series of means to MR images, and such as normalized, focal area is subtracted
Gray average treatment, topography's enhancing treatment of anatomical tissue essence etc., extract on image thus through processing
Feature significantly improves the degree of accuracy of grader in classifier training.
By taking liver MR images as an example, (liver tumour area-of-interest), capilary infiltration degree are classified in liver tumour grade malignancy
(liver tumour region of interest), degree of hepatic fibrosis are classified (liver parenchyma area-of-interest), (liver parenchyma sense is emerging for hepatic inflammatory grading
Interesting region) classifier training result such as table 2 below：
Table 2
In order to contrast, also list the instruction of DWI and ADC series is not used using the feature based on LCE and herein
Practice the contrast of result, as shown in Figure 8.It is to be compared with liver tumour grade malignancy height in Fig. 8, as can be seen from Figure 8, does not make
Do not use with the feature based on LCE and the result of DWI and ADC sequences clearly worse compared to the solution of the present invention.
Fig. 9 shows the block diagram of the processing unit 900 of MR images according to an aspect of the present invention.As shown in figure 9, place
Reason device 900 may include image segmentation module 901, normalization module 902, feature calculation module 903, training module 904.
Image segmentation module 901 is used for from every width MR image segmentation area-of-interests.As it was previously stated, area-of-interest can be wrapped
Include the parenchyma section of focal area or anatomical tissue.The solution of the present invention is mainly at the area-of-interest to MR images
Reason.
Normalization module 902 is used to be performed for the MR images of every kind of sequence the normalizing of the gray scale between each area-of-interest
Change is processed.Because the image of different patients has the difference in brightness and contrast, by normalized, it is ensured that different
The uniformity of data characteristics between patient image, improves the accuracy of the grader of follow-up training.
Especially, normalization module employs a series of special normalization algorithms, significantly there is provided the accurate of grader
Property.
On the one hand, area-of-interest is focal area, and the MR images that normalization module 902 can be directed to every kind of sequence perform disease
The normalized of the gray scale between stove region.
In one example, normalization module 902 can be by the gray value of each focal area divided by the gray scale in the focal area
Average is multiplied by with identical preset constant.This normalized has obvious side for the classifier training degree of accuracy of some illnesss
Help.For example, being impregnated with the classifier training of the judgement of nothing for the capilary of the liver tumour of liver MR images, it is accurate to significantly improve
Property.
In another example, processing unit 900 may also include correcting module 905, and normalization is performed in normalization module 902
Before, the gray scale in every width MR image focus region can be subtracted the gray average of its parenchyma section to be corrected by correcting module 905
Focal area afterwards, then normalizes each MR image of the module 902 for every kind of sequence, the ash of the focal area that will be respectively corrected
Angle value divided by the MR images gray average multiplied by with identical preset constant.Classification of this normalized for some illnesss
The device training degree of accuracy has obvious help.For example, the classification of the judgement for the grade malignancy height of the liver tumour of liver MR images
Device is trained, and significantly improves accuracy.
More preferably, processing unit 900 may also include local contrast enhancing module 906.Local contrast strengthens module 906
The gray value of the focal area after each normalization can be subtracted after the estimate of local mean value estimating divided by Local standard deviation again
Evaluation.Thus obtained image carries out follow-up feature calculation again, can further improve the classifier training degree of accuracy, such as right
In the classifier training of the judgement of the grade malignancy height of the liver tumour of liver MR images.
On the other hand, area-of-interest is the parenchyma section of anatomical tissue, and normalization module 902 can be directed to every kind of sequence
MR images perform the normalized of the gray scale between the parenchyma section of anatomical tissue.
In one example, normalization module 902 can be directed to each MR images of every kind of sequence, by the solid area of each anatomical tissue
The pixel gray level that the gray value in domain is in the range of the σ of μ ± 3 is normalized to identical preset range, for example, normalize to [0,
VAL] in, while abandoning the pixel being not in the range of the σ of μ ± 3, wherein μ refers to the ash of the parenchyma section of current anatomical tissue
Degree average, σ refers to standard deviation, and the wherein VAL is a constant, such as can be 1000 as an example.This can be removed
The skew of average and multiplication sex differernce between different images.The classifier training degree of accuracy of this normalization for some illnesss
There is obvious help.For example, the classifier training of the judgement for the degree of hepatic fibrosis height of liver MR images, significantly improves
Accuracy.
In another example, normalization module 902 can be directed to each MR images of every kind of sequence, by the solid area of anatomical tissue
The gray value in domain is divided by the gray average in the parenchyma section of anatomical tissue multiplied by with identical preset constant.This normalization for
The classifier training degree of accuracy of some illnesss has obvious help.For example, sentencing for the hepatic inflammatory degree height of liver MR images
Disconnected classifier training, significantly improves accuracy.
After the normalization for carrying out area-of-interest, feature calculation module 903 is based on each region of interest after normalization
Domain performs feature calculation with texture feature extraction, then carries out features training based on extracted texture by training module 904, obtains
Take the grader of the classification results on the area-of-interest.Feature calculation module 903 and training module 904 can use any
Conventional algorithm performs, will not be repeated here.
Finally, processing unit 900 may also include sort module 905, and sort module 905 can be used obtained grader pair
Target image is classified.For example, for the MR images of new patient, classification can be carried out by using the grader of foregoing acquisition and sentenced
It is disconnected.
Those skilled in the art will further appreciate that, the various illustratives described with reference to the embodiments described herein
Logic plate, module, circuit and algorithm steps can realize being electronic hardware, computer software or combination of the two.For clear
Chu ground explains this interchangeability of hardware and software, various illustrative components, frame, module, circuit and step be above with
Its functional form makees vague generalization description.Such feature be implemented as hardware or software depend on concrete application and
Put on the design constraint of total system.Technical staff can be realized described for every kind of application-specific with different modes
Feature, but such realize that decision-making should not be interpreted to cause departing from the scope of the present invention.
With reference to presently disclosed embodiment describe various illustrative logic modules and circuit can with general processor,
Digital signal processor (DSP), application specific integrated circuit (ASIC), field programmable gate array (FPGA) or other FPGAs
Device, discrete door or transistor logic, discrete nextport hardware component NextPort or its be designed to carry out function described herein any group
Close to realize or perform.General processor can be microprocessor, but in alternative, the processor can be any routine
Processor, controller, microcontroller or state machine.Processor is also implemented as the combination of computing device, such as DSP
One or more microprocessors that combination, multi-microprocessor with microprocessor cooperate with DSP core or any other this
Class is configured.
The step of method or algorithm for being described with reference to embodiment disclosed herein, can be embodied directly in hardware, in by processor
Embodied in the software module of execution or in combination of the two.Software module can reside in RAM memory, flash memory, ROM and deposit
Reservoir, eprom memory, eeprom memory, register, hard disk, removable disk, CD-ROM or known in the art appoint
In the storage medium of what other forms.Exemplary storage medium is coupled to processor to enable the processor from/to the storage
Medium reads and write-in information.In alternative, storage medium can be integrated into processor.Processor and storage medium can
In residing in ASIC.ASIC can reside in user terminal.In alternative, processor and storage medium can be used as discrete sets
Part is resident in the user terminal.
In one or more exemplary embodiments, described function can be in hardware, software, firmware or its any combinations
Middle realization.If being embodied as computer program product in software, each function can be as the instruction of one or more bars or generation
Code storage is transmitted on a computer-readable medium or by it.Computer-readable medium includes computer-readable storage medium and communication
Both media, it includes any medium for facilitating computer program to shift from one place to another.Storage medium can be can quilt
Any usable medium that computer is accessed.It is non-limiting as an example, such computer-readable medium may include RAM, ROM,
EEPROM, CD-ROM or other optical disc storages, disk storage or other magnetic storage apparatus can be used to carry or store instruction
Or the desirable program code and any other medium that can be accessed by a computer of data structure form.Any connection is also by by rights
Referred to as computer-readable medium.If for example, software is to use coaxial cable, fiber optic cables, twisted-pair feeder, digital subscriber line
Or the wireless technology of such as infrared, radio and microwave etc is passed from web site, server or other remote sources (DSL)
Send, then the coaxial cable, fiber optic cables, twisted-pair feeder, DSL or such as infrared, radio and microwave etc is wireless
Technology is just included among the definition of medium.Disk (disk) as used herein and dish (disc) include compact disc
(CD), laser disc, laser disc, digital versatile disc (DVD), floppy disk and blu-ray disc, which disk (disk) are often reproduced in the way of magnetic
Data, and dish (disc) laser reproduce data optically.Combinations of the above should also be included in computer-readable medium
In the range of.
It is for so that any person skilled in the art can all make or use this public affairs to provide of this disclosure being previously described
Open.Various modifications of this disclosure all will be apparent for a person skilled in the art, and as defined herein general
Suitable principle can be applied to spirit or scope of other variants without departing from the disclosure.Thus, the disclosure is not intended to be limited
Due to example described herein and design, but should be awarded and principle disclosed herein and novel features phase one
The widest scope of cause.
Claims (22)
1. a kind of processing method of MR images, MR image of the MR images including various sequences, the treating method comprises：
From every width MR image segmentation area-of-interests；
MR images for every kind of sequence perform the normalized of the gray scale between each area-of-interest；
Each area-of-interest after based on normalization performs feature calculation with texture feature extraction；And
Features training is carried out based on the texture for being extracted, the grader of the classification results on the area-of-interest is obtained.
2. processing method as claimed in claim 1, it is characterised in that the area-of-interest includes focal area, described to return
One changes treatment includes being performed for the MR images of every kind of sequence the normalized of the gray scale between focal area.
3. processing method as claimed in claim 2, it is characterised in that the normalized is included for each of every kind of sequence
MR images, by the gray value of each focal area divided by the gray average in the focal area multiplied by with identical preset constant.
4. processing method as claimed in claim 2, it is characterised in that the area-of-interest also essence including anatomical tissue
Region, the processing method also include by the gray value in every width MR image focus region subtract the gray average of its parenchyma section with
The focal area that acquisition is corrected, the normalized includes each MR images for every kind of sequence, the disease that will be respectively corrected
The gray value in stove region divided by the MR images gray average multiplied by with identical preset constant.
5. processing method as claimed in claim 4, it is characterised in that also include：
Local contrast enhancement processing is performed to the focal area after normalization, the local contrast enhancing treatment includes will be each
The gray value of the focal area after normalization is subtracted after the estimate of local mean value again divided by the estimate of Local standard deviation,
The feature calculation includes performing feature calculation to extract texture through the enhanced focal area of local contrast based on each
Feature.
6. processing method as claimed in claim 5, it is characterised in that the MR images of various sequences include DWI sequences and
The MR images of ADC sequences.
7. processing method as claimed in claim 1, it is characterised in that the area-of-interest includes the solid area of anatomical tissue
Domain, the normalized includes being performed for the MR images of every kind of sequence returning for the gray scale between the parenchyma section of anatomical tissue
One change is processed.
8. processing method as claimed in claim 7, it is characterised in that the normalized is included for each of every kind of sequence
MR images, the pixel gray level that the gray value of the parenchyma section of each anatomical tissue is in the range of the σ of μ ± 3 are normalized to identical
Preset range, while abandoning the pixel for being not in the range of the σ of μ ± 3, wherein μ refers to the parenchyma section of current anatomical tissue
Gray average, σ refer to standard deviation.
9. processing method as claimed in claim 7, it is characterised in that the normalized is included for each of every kind of sequence
MR images, the gray average in parenchyma section by the gray value of the parenchyma section of anatomical tissue divided by anatomical tissue is multiplied by with phase
Same preset constant.
10. processing method as claimed in claim 1, it is characterised in that also include：
Obtained grader is used to classify target image.
11. processing methods as claimed in claim 1, it is characterised in that the classification results include liver tumour grade malignancy point
At least one in level, the classification of capilary infiltration degree, degree of hepatic fibrosis, hepatic inflammatory grading.
A kind of 12. processing units of MR images, the MR images include the MR images of various sequences, and the processing unit includes：
Image segmentation module, for from every width MR image segmentation area-of-interests；
Normalization module, at the normalization that the gray scale between each area-of-interest is performed for the MR images of every kind of sequence
Reason；
Feature calculation module, feature calculation is performed with texture feature extraction for each area-of-interest after based on normalization；With
And
Training module, for carrying out features training based on the texture for being extracted, obtains the classification knot on the area-of-interest
The grader of fruit.
13. processing units as claimed in claim 12, it is characterised in that the area-of-interest includes focal area, described
Normalization module performs the normalized of the gray scale between focal area for the MR images of every kind of sequence.
14. processing units as claimed in claim 13, it is characterised in that each MR of the normalization module for every kind of sequence
Image, by the gray value of each focal area divided by the gray average in the focal area multiplied by with identical preset constant.
15. processing units as claimed in claim 13, it is characterised in that the area-of-interest also reality including anatomical tissue
Matter region, the processing unit also includes：
Correcting module, for the gray scale in every width MR image focus region to be subtracted into the gray average of its parenchyma section to be corrected
Focal area afterwards,
, for each MR images of every kind of sequence, the gray value of the focal area that will be respectively corrected is divided by the MR for the normalization module
The gray average of image is multiplied by with identical preset constant.
16. processing units as claimed in claim 15, it is characterised in that also include：
Local contrast strengthens module, the estimate for the gray value of the focal area after each normalization to be subtracted local mean value
Afterwards again divided by the estimate of Local standard deviation,
The feature calculation module is based on each and performs feature calculation to extract texture through the enhanced focal area of local contrast
Feature.
17. processing units as claimed in claim 16, it is characterised in that the MR images of various sequences include DWI sequences
With the MR images of ADC sequences.
18. processing units as claimed in claim 12, it is characterised in that the area-of-interest includes the essence of anatomical tissue
Region, the normalization module performs the normalizing of the gray scale between the parenchyma section of anatomical tissue for the MR images of every kind of sequence
Change is processed.
19. processing units as claimed in claim 18, it is characterised in that each MR of the normalization module for every kind of sequence
Image, identical is normalized to by the pixel gray level that the gray value of the parenchyma section of each anatomical tissue is in the range of the σ of μ ± 3
Preset range, while abandoning the pixel being not in the range of the σ of μ ± 3, wherein μ refers to the parenchyma section of current anatomical tissue
Gray average, σ refers to standard deviation.
20. processing units as claimed in claim 18, it is characterised in that each MR of the normalization module for every kind of sequence
Image, the gray average in parenchyma section by the gray value of the parenchyma section of anatomical tissue divided by anatomical tissue is multiplied by with identical
Preset constant.
21. processing units as claimed in claim 12, it is characterised in that also include：
Sort module, for using obtained grader to classify target image.
22. processing units as claimed in claim 12, it is characterised in that the classification results include liver tumour grade malignancy point
At least one in level, the classification of capilary infiltration degree, degree of hepatic fibrosis, hepatic inflammatory grading.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611122086.0A CN106709907A (en) | 2016-12-08 | 2016-12-08 | MR image processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611122086.0A CN106709907A (en) | 2016-12-08 | 2016-12-08 | MR image processing method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106709907A true CN106709907A (en) | 2017-05-24 |
Family
ID=58936487
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611122086.0A Pending CN106709907A (en) | 2016-12-08 | 2016-12-08 | MR image processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106709907A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107357842A (en) * | 2017-06-23 | 2017-11-17 | 华中师范大学 | A kind of multi-screen Knowledge Visualization method of knowledge based classification |
| CN108154509A (en) * | 2018-01-12 | 2018-06-12 | 平安科技（深圳）有限公司 | Cancer recognition methods, device and storage medium |
| CN109410289A (en) * | 2018-11-09 | 2019-03-01 | 中国科学院武汉物理与数学研究所 | A kind of high lack sampling hyperpolarized gas lung MRI method for reconstructing of deep learning |
| CN109840554A (en) * | 2018-12-26 | 2019-06-04 | 贵州联科卫信科技有限公司 | A kind of Alzheimer's disease MRI image classification method based on SVM-RFE-MRMR algorithm |
| CN110148192A (en) * | 2019-04-18 | 2019-08-20 | 上海联影智能医疗科技有限公司 | Medical image imaging method, device, computer equipment and storage medium |
| CN111033523A (en) * | 2017-09-21 | 2020-04-17 | 国际商业机器公司 | Data enhancement for image classification tasks |
| CN111553352A (en) * | 2020-04-27 | 2020-08-18 | 上海工程技术大学 | DICOM image processing method and system |
| CN112508953A (en) * | 2021-02-05 | 2021-03-16 | 四川大学 | Meningioma rapid segmentation qualitative method based on deep neural network |
| CN112862797A (en) * | 2021-02-23 | 2021-05-28 | 复旦大学附属华山医院 | Liver fibrosis nondestructive prediction method and system |
| CN113077417A (en) * | 2021-03-16 | 2021-07-06 | 中国人民解放军总医院第一医学中心 | Method and system for predicting hepatocellular carcinoma microvascular invasion |
| US11529054B2 (en) * | 2018-07-25 | 2022-12-20 | The Charles Stark Draper Laboratory, Inc. | Method and system for post-traumatic stress disorder (PTSD) and mild traumatic brain injury (mTBI) diagnosis using magnetic resonance spectroscopy |
| CN118037583A (en) * | 2024-04-11 | 2024-05-14 | 灏麟(天津)生物科技有限公司 | Nuclear magnetic data optimization method for sarcopenia diagnosis and evaluation |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800089A (en) * | 2012-06-28 | 2012-11-28 | 华中科技大学 | Main carotid artery blood vessel extraction and thickness measuring method based on neck ultrasound images |
| CN103699904A (en) * | 2013-12-25 | 2014-04-02 | 大连理工大学 | Image computer-aided diagnosis method for multi-sequence nuclear magnetic resonance images |
| CN104000619A (en) * | 2014-06-16 | 2014-08-27 | 彭文献 | Thyroid CT image computer-aided diagnosis system and method |
| CN104217437A (en) * | 2014-09-19 | 2014-12-17 | 西安电子科技大学 | Lesion area partitioning method for prostate KVCT image |
| CN105590324A (en) * | 2016-02-03 | 2016-05-18 | 上海联影医疗科技有限公司 | Segmentation method and device of medical images |
-
2016
- 2016-12-08 CN CN201611122086.0A patent/CN106709907A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800089A (en) * | 2012-06-28 | 2012-11-28 | 华中科技大学 | Main carotid artery blood vessel extraction and thickness measuring method based on neck ultrasound images |
| CN103699904A (en) * | 2013-12-25 | 2014-04-02 | 大连理工大学 | Image computer-aided diagnosis method for multi-sequence nuclear magnetic resonance images |
| CN104000619A (en) * | 2014-06-16 | 2014-08-27 | 彭文献 | Thyroid CT image computer-aided diagnosis system and method |
| CN104217437A (en) * | 2014-09-19 | 2014-12-17 | 西安电子科技大学 | Lesion area partitioning method for prostate KVCT image |
| CN105590324A (en) * | 2016-02-03 | 2016-05-18 | 上海联影医疗科技有限公司 | Segmentation method and device of medical images |
Non-Patent Citations (2)
| Title |
|---|
| 朱浩栋等: "医学图像灰度归一化方法研究", 《国际生物医学工程杂志》 * |
| 郭冬梅: "肝脏病变MRI诊断与计算机辅助诊断", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107357842A (en) * | 2017-06-23 | 2017-11-17 | 华中师范大学 | A kind of multi-screen Knowledge Visualization method of knowledge based classification |
| CN111033523A (en) * | 2017-09-21 | 2020-04-17 | 国际商业机器公司 | Data enhancement for image classification tasks |
| CN111033523B (en) * | 2017-09-21 | 2023-12-29 | 国际商业机器公司 | Data enhancement for image classification tasks |
| CN108154509A (en) * | 2018-01-12 | 2018-06-12 | 平安科技（深圳）有限公司 | Cancer recognition methods, device and storage medium |
| US11529054B2 (en) * | 2018-07-25 | 2022-12-20 | The Charles Stark Draper Laboratory, Inc. | Method and system for post-traumatic stress disorder (PTSD) and mild traumatic brain injury (mTBI) diagnosis using magnetic resonance spectroscopy |
| CN109410289B (en) * | 2018-11-09 | 2021-11-12 | 中国科学院精密测量科学与技术创新研究院 | Deep learning high undersampling hyperpolarized gas lung MRI reconstruction method |
| CN109410289A (en) * | 2018-11-09 | 2019-03-01 | 中国科学院武汉物理与数学研究所 | A kind of high lack sampling hyperpolarized gas lung MRI method for reconstructing of deep learning |
| CN109840554A (en) * | 2018-12-26 | 2019-06-04 | 贵州联科卫信科技有限公司 | A kind of Alzheimer's disease MRI image classification method based on SVM-RFE-MRMR algorithm |
| CN109840554B (en) * | 2018-12-26 | 2023-05-02 | 贵州联科卫信科技有限公司 | Alzheimer's disease MRI image classification method based on SVM-RFE-MRMR algorithm |
| CN110148192A (en) * | 2019-04-18 | 2019-08-20 | 上海联影智能医疗科技有限公司 | Medical image imaging method, device, computer equipment and storage medium |
| CN111553352A (en) * | 2020-04-27 | 2020-08-18 | 上海工程技术大学 | DICOM image processing method and system |
| CN112508953A (en) * | 2021-02-05 | 2021-03-16 | 四川大学 | Meningioma rapid segmentation qualitative method based on deep neural network |
| CN112862797A (en) * | 2021-02-23 | 2021-05-28 | 复旦大学附属华山医院 | Liver fibrosis nondestructive prediction method and system |
| CN112862797B (en) * | 2021-02-23 | 2024-03-19 | 复旦大学附属华山医院 | Liver fibrosis nondestructive prediction method and system |
| CN113077417A (en) * | 2021-03-16 | 2021-07-06 | 中国人民解放军总医院第一医学中心 | Method and system for predicting hepatocellular carcinoma microvascular invasion |
| CN118037583A (en) * | 2024-04-11 | 2024-05-14 | 灏麟(天津)生物科技有限公司 | Nuclear magnetic data optimization method for sarcopenia diagnosis and evaluation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709907A (en) | MR image processing method and device | |
| Soltaninejad et al. | Automated brain tumour detection and segmentation using superpixel-based extremely randomized trees in FLAIR MRI | |
| Dalmış et al. | Using deep learning to segment breast and fibroglandular tissue in MRI volumes | |
| Zhuang et al. | Breast ultrasound tumor image classification using image decomposition and fusion based on adaptive multi-model spatial feature fusion | |
| Nasor et al. | Detection and Localization of Early‐Stage Multiple Brain Tumors Using a Hybrid Technique of Patch‐Based Processing, k‐means Clustering and Object Counting | |
| Kavur et al. | Basic ensembles of vanilla-style deep learning models improve liver segmentation from ct images | |
| Shahajad et al. | Features extraction for classification of brain tumor MRI images using support vector machine | |
| Das et al. | A fast and automated segmentation method for detection of masses using folded kernel based fuzzy c-means clustering algorithm | |
| Dillman et al. | Current and emerging artificial intelligence applications for pediatric abdominal imaging | |
| Süleyman Yıldırım et al. | Automatic detection of multiple sclerosis lesions using Mask R‐CNN on magnetic resonance scans | |
| Affane et al. | Robust deep 3-D architectures based on vascular patterns for liver vessel segmentation | |
| Lo et al. | Voxel classification based airway tree segmentation | |
| Jucevičius et al. | Investigation of MRI prostate localization using different MRI modality scans | |
| Benrabha et al. | Automatic ROI detection and classification of the achilles tendon ultrasound images | |
| Mourya et al. | CT guided diagnosis: Cascaded U-Net for 3D segmentation of liver and tumor | |
| CN111971751A (en) | System and method for evaluating dynamic data | |
| Saad et al. | A review on image segmentation techniques for MRI brain stroke lesion | |
| Çetingül et al. | Estimation of local orientations in fibrous structures with applications to the Purkinje system | |
| Braik et al. | Exploration of genetic algorithms and particle swarm optimization in improving the quality of medical images | |
| Judice et al. | A novel assessment of various bio-imaging methods for lung tumor detection and treatment by using 4-D and 2-D CT images | |
| Saumiya et al. | Unified automated deep learning framework for segmentation and classification of liver tumors | |
| Huang et al. | Extraction of media adventitia and luminal intima borders by reconstructing intravascular ultrasound image sequences with vascular structural continuity | |
| Amritha et al. | Liver tumor segmentation and classification using deep learning | |
| Sari et al. | Review of brain MRI image segmentation based on deep learning method | |
| Wu et al. | Deep artificial neural network approach to automated lesion segmentation in breast |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170524 |
|
| RJ01 | Rejection of invention patent application after publication |