CN110074809A - The hepatic vein pressure gradient classification method and computer equipment of CT image - Google Patents
The hepatic vein pressure gradient classification method and computer equipment of CT image Download PDFInfo
- Publication number
- CN110074809A CN110074809A CN201910522932.5A CN201910522932A CN110074809A CN 110074809 A CN110074809 A CN 110074809A CN 201910522932 A CN201910522932 A CN 201910522932A CN 110074809 A CN110074809 A CN 110074809A
- Authority
- CN
- China
- Prior art keywords
- pressure gradient
- classification
- liver
- hepatic vein
- vein pressure
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 210000002989 hepatic vein Anatomy 0.000 title claims abstract description 125
- 238000000034 method Methods 0.000 title claims abstract description 52
- 210000004185 liver Anatomy 0.000 claims abstract description 142
- 238000012360 testing method Methods 0.000 claims abstract description 53
- 238000005520 cutting process Methods 0.000 claims abstract description 30
- 230000003187 abdominal effect Effects 0.000 claims abstract description 29
- 210000000952 spleen Anatomy 0.000 claims description 102
- 238000012549 training Methods 0.000 claims description 68
- 238000013527 convolutional neural network Methods 0.000 claims description 50
- 230000006870 function Effects 0.000 claims description 44
- 238000012545 processing Methods 0.000 claims description 19
- 238000009499 grossing Methods 0.000 claims description 14
- 241000415078 Anemone hepatica Species 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 230000015654 memory Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 210000001015 abdomen Anatomy 0.000 claims description 5
- 230000002440 hepatic effect Effects 0.000 claims description 5
- 210000004218 nerve net Anatomy 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 2
- 208000007232 portal hypertension Diseases 0.000 description 13
- 238000005516 engineering process Methods 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 7
- 208000019425 cirrhosis of liver Diseases 0.000 description 7
- 238000000605 extraction Methods 0.000 description 7
- 230000008520 organization Effects 0.000 description 7
- 206010016654 Fibrosis Diseases 0.000 description 6
- 230000007882 cirrhosis Effects 0.000 description 6
- 238000003745 diagnosis Methods 0.000 description 6
- 238000003384 imaging method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000003860 storage Methods 0.000 description 5
- 238000003759 clinical diagnosis Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 239000003814 drug Substances 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 4
- 230000035485 pulse pressure Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 208000032843 Hemorrhage Diseases 0.000 description 2
- 206010061218 Inflammation Diseases 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 208000034158 bleeding Diseases 0.000 description 2
- 230000000740 bleeding effect Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 230000001684 chronic effect Effects 0.000 description 2
- 208000035850 clinical syndrome Diseases 0.000 description 2
- 238000002591 computed tomography Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 229940079593 drug Drugs 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 230000004054 inflammatory process Effects 0.000 description 2
- 201000007270 liver cancer Diseases 0.000 description 2
- 208000019423 liver disease Diseases 0.000 description 2
- 208000014018 liver neoplasm Diseases 0.000 description 2
- 239000003550 marker Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 210000004258 portal system Anatomy 0.000 description 2
- 210000003240 portal vein Anatomy 0.000 description 2
- 238000010008 shearing Methods 0.000 description 2
- 210000001519 tissue Anatomy 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000012752 Hepatectomy Methods 0.000 description 1
- 206010019728 Hepatitis alcoholic Diseases 0.000 description 1
- 102000000589 Interleukin-1 Human genes 0.000 description 1
- 108010002352 Interleukin-1 Proteins 0.000 description 1
- 206010066901 Treatment failure Diseases 0.000 description 1
- 108010000134 Vascular Cell Adhesion Molecule-1 Proteins 0.000 description 1
- 102100023543 Vascular cell adhesion protein 1 Human genes 0.000 description 1
- 230000001154 acute effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 208000002353 alcoholic hepatitis Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000017531 blood circulation Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000001447 compensatory effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012631 diagnostic technique Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000002439 hemostatic effect Effects 0.000 description 1
- MOFVSTNWEDAEEK-UHFFFAOYSA-M indocyanine green Chemical compound [Na+].[O-]S(=O)(=O)CCCCN1C2=CC=C3C=CC=CC3=C2C(C)(C)C1=CC=CC=CC=CC1=[N+](CCCCS([O-])(=O)=O)C2=CC=C(C=CC=C3)C3=C2C1(C)C MOFVSTNWEDAEEK-UHFFFAOYSA-M 0.000 description 1
- 229960004657 indocyanine green Drugs 0.000 description 1
- 238000012045 magnetic resonance elastography Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000005180 public health Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 210000002966 serum Anatomy 0.000 description 1
- 210000000955 splenic vein Anatomy 0.000 description 1
- 238000003325 tomography Methods 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
- 210000003462 vein Anatomy 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
- A61B6/03—Computed tomography [CT]
- A61B6/032—Transmission computed tomography [CT]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/50—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5205—Devices using data or image processing specially adapted for radiation diagnosis involving processing of raw data to produce diagnostic data
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
- A61B6/5217—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data extracting a diagnostic or physiological parameter from medical diagnostic data
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/30056—Liver; Hepatic
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- High Energy & Nuclear Physics (AREA)
- Data Mining & Analysis (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Optics & Photonics (AREA)
- Pathology (AREA)
- Radiology & Medical Imaging (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Dentistry (AREA)
- Physiology (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Pulmonology (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The present embodiments relate to the hepatic vein pressure gradient classification methods and computer equipment of CT image to obtain multiple liver segments to be measured this method comprises: the abdominal CT images for the tester that will acquire carry out liver image cutting pretreatment;Multiple liver segments to be measured are input in trained liver prediction submodel in advance and carry out hepatic vein pressure gradient class test, obtain multiple predictions classification of each liver segment to be measured, and predict corresponding class probability of classifying;Wherein, the hepatic vein pressure gradient of tester is same prediction classification within the scope of same default hepatic vein pressure gradient;Calculate the average value for belonging to the class probability of same prediction classification, and by each average value maximum value and the corresponding prediction classification of the maximum value be determined as the hepatic vein pressure gradient class test result of tester.Various embodiments of the present invention realize the noninvasive test of more classification to hepatic vein pressure gradient, and testing classification result precision is higher, and generalization ability is stronger.
Description
Technical field
The present invention relates to field of artificial intelligence more particularly to a kind of hepatic vein pressure gradient classification methods of CT image
And computer equipment.
Background technique
Portal hypertension refers to that one group is persistently increased caused clinical syndrome by portalsystem pressure, is common in chronic
Hepatopathy development enters the cirrhosis stage, and poor clinical diagnosis causes a degree of public health problem.Traditional technology
In, common portal hypertension non-invasive diagnosis and detection technique are mainly used for the identification of clinically significant conspicuousness portal hypertension,
And it is easy to be interfered by different factors.Simultaneously in traditional technology, clinical diagnosis is carried out by building prediction model although having, its
The generalization ability of model is relatively limited, is not suitable for the preferable feature of selection for image adaptive, and major part is only used for
The diagnostic classification of clinical significance portal hypertension, limitation are larger.
Summary of the invention
It is an object of the invention to be directed to the deficiency of traditional technology, a kind of hepatic vein pressure gradient classification of CT image is provided
Method and computer equipment.
In one embodiment, the present invention provides a kind of hepatic vein pressure gradient classification methods of CT image, comprising:
The abdominal CT images for the tester that will acquire carry out liver image cutting pretreatment, obtain multiple liver figures to be measured
Block;
Multiple liver segments to be measured are input in trained liver prediction submodel in advance and carry out vena hepatica pressure ladder
Class test is spent, multiple predictions classification of each liver segment to be measured, and the corresponding class probability of prediction classification are obtained;Its
In, the hepatic vein pressure gradient of the tester is same prediction classification within the scope of same default hepatic vein pressure gradient;
Calculate belong to it is same prediction classification class probability average value, and by each average value maximum value and this most
It is worth the hepatic vein pressure gradient class test result that corresponding prediction classification is determined as tester greatly.
The abdominal CT images for the tester that will acquire in one of the embodiments, carry out the pre- place of spleen image cutting
Reason, obtains multiple spleen segments to be measured;
Multiple spleen segments to be measured are input in trained spleen prediction submodel in advance and carry out vena hepatica pressure ladder
Class test is spent, multiple predictions classification of each spleen segment to be measured, and the corresponding class probability of prediction classification are obtained;
Hepatic vein pressure gradient class test, which is carried out, for liver segment to be measured and spleen segment to be measured obtains all classification
The class probability for belonging to same prediction classification in probability is averaged, and the average value for accordingly predicting the class probability of classification is obtained,
And by each average value maximum value and the corresponding prediction classification of the maximum value be determined as the hepatic vein pressure gradient of tester
Class test result.
Abdominal CT images are subjected to liver image cutting pretreatment in one of the embodiments, comprising:
Liver area image is extracted from the hepatic portal figure layer of abdominal CT images, and liver area image is smoothly located
Reason;
Liver area image after smoothing processing is cut according to the first default segment size, obtains multiple livers to be measured
Dirty segment.
Abdominal CT images are subjected to the cutting pretreatment of spleen image in one of the embodiments, comprising:
Spleen area image is extracted from the hilus lienis figure layer of abdominal CT images, and spleen area image is smoothly located
Reason;
Spleen area image after smoothing processing is cut according to the second default segment size, obtains multiple spleens to be measured
Dirty segment.
The training step of liver prediction submodel in one of the embodiments, comprising:
It is random within the scope of default weight coefficient by all weight coefficient initializations of the first initial convolutional neural networks
It is uniformly distributed;
According to default hepatic vein pressure gradient range, it will acquire and cut pretreated each liver by liver image
Training segment is grouped, and the liver of each group training segment is input to the first initial convolution mind after weight coefficient initialization
Through in network;
Obtain each liver training segment by multiple predictions classification of the corresponding output of the first initial convolutional neural networks and
The corresponding class probability of prediction classification is based on the corresponding first function value of preset first-loss function to calculate;
Using the average value of each first function value as first object value, and back-propagation algorithm is utilized based on first object value
The weight coefficient of the first initial convolutional neural networks is updated, until the prediction of the updated first initial convolutional neural networks is classified
As a result first object value is made to be less than or equal to the first predetermined target value, then by the first initial convolutional neural networks after final updated
Submodel is predicted as liver;First object value indicate to each liver training segment prediction classification results and actual packet it is inclined
Difference.
The training step of spleen prediction submodel in one of the embodiments, comprising:
The weight coefficient of second initial convolutional neural networks is initialized as random uniform within the scope of default weight coefficient
Distribution;
According to default hepatic vein pressure gradient range, it will acquire and cut pretreated each spleen by spleen image
Training segment is grouped, and the spleen of each group training segment is input to the second initial convolution mind after weight coefficient initialization
Through network;
Obtain each spleen training segment by multiple predictions classification of the corresponding output of the second initial convolutional neural networks and
The corresponding class probability value of prediction classification is based on the corresponding second function value of preset second loss function to calculate;
Using the average value of each second function value as the second target value, and back-propagation algorithm is utilized based on the second target value
The weight coefficient of the second initial convolutional neural networks is updated, until the prediction of the updated second initial convolutional neural networks is classified
As a result so that the second target value is less than or equal to the second predetermined target value, then make the second initial convolutional neural networks after final updated
Submodel is predicted for spleen;Second target value indicate to each spleen training segment prediction classification results and actual packet it is inclined
Difference.
In one of the embodiments, further include:
At least one preset hepatic vein pressure gradient key value is obtained, and is drawn according to each hepatic vein pressure gradient key value
Divide corresponding default hepatic vein pressure gradient range.
Preset hepatic vein pressure gradient key value is five in one of the embodiments,;
Each preset hepatic vein pressure gradient key value be respectively 5mmHg, 10mmHg, 12mmHg, 16mmHg, 20mmHg with
And 22mmHg.
Loss function is obtained based on following formula in one of the embodiments:
Wherein, c indicates prediction classification;M indicates the number of prediction classification;O indicates the category set of prediction classification;yo,cTable
Show the corresponding label value of prediction classification;Po,cIndicate the corresponding class probability of prediction classification.
On the other hand, the embodiment of the invention also provides a kind of computer equipment, including memory and processor, memories
It is stored with computer program, processor realizes hepatic vein pressure gradient classification method when executing computer program.
The hepatic vein pressure gradient classification method and computer equipment of CT image of the invention, by the abdominal CT of tester
Image by liver image cutting pretreatment after, obtain multiple liver images to be measured be input to liver prediction submodel in carry out liver
Venous pressure gradient test, being divided into multiple liver images to be measured can analyze about the pylic more details of tester, so that
Test result is more accurate.Further, each liver image to be measured predicts that submodel exports multiple predictions and classifies by liver
With corresponding probability value, belong to same prediction by calculating and classify the first average value of corresponding class probability, and by maximum value
The first average value and corresponding hepatic vein pressure gradient classification results of the prediction classification as tester.Various embodiments of the present invention
The highest prediction classification of class probability value can be regard as test result, realize the noninvasive survey of more classification to hepatic vein pressure gradient
Examination, testing classification result precision is higher, and generalization ability is stronger.
Detailed description of the invention
In order to illustrate more clearly of technical solution of the present invention, letter will be made to attached drawing needed in the embodiment below
It singly introduces, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as to the present invention
The restriction of protection scope.In various figures, part is similarly comprised using similar number.
Fig. 1 shows the flow diagram of the hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention;
Fig. 2 shows the signals of another process of the hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention
Figure;
Fig. 3 shows training liver prediction in the hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention
The flow diagram of model;
Fig. 4 shows training spleen prediction in the hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention
The flow diagram of model;
Fig. 5 shows the structural schematic diagram of the hepatic vein pressure gradient sorter of the CT image of the embodiment of the present invention.
Specific embodiment
Below in conjunction with attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.
The component of embodiments of the present invention, which are generally described and illustrated herein in the accompanying drawings can be come with a variety of different configurations
Arrangement and design.Therefore, requirement is not intended to limit to the detailed description of the embodiment of the present invention provided in the accompanying drawings below
The scope of the present invention of protection, but it is merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, this field skill
Art personnel every other embodiment obtained without making creative work belongs to the model that the present invention protects
It encloses.
Hereinafter, term " includes ", " having " and its cognate that can be used in various embodiments of the present invention are only
It is intended to mean that special characteristic, number, step, operation, the combination of element, component or aforementioned item, and is understood not to first
Exclude the combined presence or increase by one of one or more other features, number, step, operation, element, component or aforementioned item
A or more feature, number, step, operation, element, component or aforementioned item combination a possibility that.
In addition, term " first ", " second ", " third " etc. are only used for distinguishing description, it is not understood to indicate or imply
Relative importance.
Unless otherwise defined, otherwise all terms (including technical terms and scientific terms) used herein have and this
The identical meaning of the various normally understood meanings of embodiment one skilled in the art of invention.The term (such as exists
The term limited in the dictionary generally used) it is to be interpreted as that there is contain identical with situational meaning in the related technical field
Justice and Utopian meaning or meaning too formal will be interpreted as having, unless in various embodiments of the present invention
It is clearly defined.
Portal hypertension refers to that one group is persistently increased caused clinical syndrome by portalsystem pressure, is common in chronic
Hepatopathy development enters the cirrhosis stage, and high incidence and poor Clinical Outcome cause serious in global range public defend
Raw problem.Currently, hepatic vein pressure gradient (hepatic venous pressure gradient, HVPG) is international recommends
Portal hypertension diagnoses goldstandard, but this be detected as it is invasive and relatively expensive, in facing for China or even European and American developed countries
It is very limited in bed application.The non-invasive diagnosis and monitoring technology for establishing a kind of cirrhotic patients with portal hypertension have great
Clinical meaning.
Portal hypertension non-invasive diagnosis and detection technique common at present is mainly used for clinical significance portal hypertension
The identification of (clinically significant portal hypertension, CSPH, HVPG >=10mmHg).Conventional
Laboratory checks (such as platelet count reduction) and iconography means [as ultrasound, x-ray computerized tomography (CT) and magnetic are total
Vibration imaging pylephlebectasis or Doppler flow mapping, portal vein flows decrease even adverse current] there is certain diagnostic value to CSPH, but
Sensibility is poor.The model pair constructed based on the detected liver elasticity combination platelet count of FibroScan and spleen major diameter
CSPH has good diagnostic value.Point shearing wave elastogram, Real-time Two-dimensional shearing wave elastogram and magnetic resonance elastography
The interpretation that can also be used for liver elasticity and spleen elasticity carries out CSPH diagnosis.In addition, being obtained by quantitative MR imaging technology
Liver T1 relaxation time and arteria linenalis blood flow velocity calculate the ratio between liver and spleen volume based on CT image and combine liver week abdomen
Water, indocyanine green 15min retention rate and combination serum of cirrhosis patients marker of inflammation interleukin-1 ' beta ', leucocyte are situated between
The various methods such as element -1R- α, Fas-R and vascular cell adhesion molecule-1 are used equally for the non-invasive diagnosis of CSPH.
There is radiation group model based on ct images to carry out the prediction of CSPH, the key step packet of the technology in the prior art
It includes: using specific software to the horizontal bitmap layer of porta hepatis and the horizontal bitmap layer progress of hilus lienis in Portal venous phase CT image sequence
The region for determining liver organization and spleen tissue respectively is delineated manually;Liver organization and spleen tissue region foundation are set
11 textures and non-grain feature carry out feature extraction;The feature extracted is input to the calculation formula set, according to meter
Patient is divided into CSPH (HVPG > 10mmHg) or non-CSPH by the result of calculation.
Common portal hypertension non-invasive diagnostic techniques above-mentioned are easy to be interfered by different factors, such as liver elastomeric check pair
It is very big in the fat or more serious patient's error of inflammation;Serologic marker object is influenced vulnerable to other diseases factor;Spleen diameter
Subjective, the poor repeatabilities etc. such as long or stereometry, the clinical diagnosis for causing these technology practical manifestations to go out still have
Larger dispute.And omics technology is radiated although by using the CT image and automatic feature extraction side that process obtains is standardized
Method reduces subjectivity error as far as possible, improves the effect of clinical diagnosis, but this method needs artificial definition to be used for
The characteristics of image of prediction model is constructed, the generalization ability of model is relatively limited, and the selection that can not be directed to image adaptive is optimal
Feature.The model is only used for the diagnostic classification (HVPG > 10mmHg) of CSPH simultaneously, is not suitable for based on other HVPG threshold values
Hepatic vein pressure gradient classification.
Referring to Fig. 1, in one embodiment, the embodiment of the invention provides the hepatic vein pressure gradient of a kind of CT image point
Class method, comprising:
Step S110: the abdominal CT images for the tester that will acquire carry out liver image cutting pretreatment, obtain multiple
Liver segment to be measured.
Liver image cutting pretreatment includes importeding into the abdominal CT images of tester to show and delineate DICOM
The software of (Digital Imaging and Communications in Medicine, digital imaging and communications in medicine),
The figure layer of porta hepatis is selected in the horizontal bit image of vena hepatica figure layer sequence, and the profile for extracting liver organization is hooked
It draws, the tissue regions delineated is smoothed.And the liver organization extracted region after smoothing processing is gone out according to predetermined
Segment size is cut into a certain size non-overlap segment.
The abdominal CT images of tester are carried out liver image cutting pretreatment by the embodiment of the present invention, convenient for more being closed
In pylic minutia, the precision of hepatic vein pressure gradient class test is helped to improve.
Step S120: it is quiet that multiple liver segments to be measured are input to progress liver in trained liver prediction submodel in advance
The test of pulse pressure gradient classifications obtains multiple predictions classification of each liver segment to be measured, and the corresponding classification of prediction classification
Probability;Wherein, the hepatic vein pressure gradient of tester is same prediction classification within the scope of same default hepatic vein pressure gradient.
Liver prediction submodel is to cross multiple groups based on the first initial convolution neural network to belong to different default vena hepatica pressures
The liver training image training of force gradient range and generate, wherein the first convolution neural network model is more points built
Connectionist model.Specifically, choose clinical practice use in important several HVPG key values as class object, with
These HVPG key values divide default hepatic vein pressure gradient range, and each default hepatic vein pressure gradient range belongs to one
Prediction classification, is indicated with tag along sort, and in more Classification Neural models, which is that (N is class to 0,1,2 ... N-1
Number) label value.
Step S130: the average value for belonging to the class probability of same prediction classification is calculated, and by the maximum in each average value
Value and the corresponding prediction classification of the maximum value are determined as the hepatic vein pressure gradient class test result of tester.
Each liver segment to be measured is input in liver prediction submodel, obtains multiple predictions point of the liver image to be measured
Class and the corresponding class probability of each prediction classification.Complete prediction result in order to obtain, in the pre- of all liver images to be measured
It surveys in result, the corresponding class probability averaged of same prediction classification will be belonged to, to obtain each prediction classification most
Whole class probability, which corresponding final classification probability of prediction classification is high, that is, it is quiet to show which liver is hepatic vein pressure gradient belong to
Pulse pressure gradient scope.As a result, by each average value maximum value and the maximum value corresponding prediction classification be determined as testing
The hepatic vein pressure gradient class test result of person.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention passes through the abdominal CT images of tester
After liver image cutting pretreatment, obtains multiple liver images to be measured and be input to progress vena hepatica pressure in liver prediction submodel
Gradient test, being divided into multiple liver images to be measured can analyze about the pylic more details of tester, so that test result
It is more accurate.Further, each liver image to be measured by liver predict submodel export multiple predictions classification with it is corresponding
Probability value belongs to same prediction and classifies the average value of corresponding class probability by calculating, and by the average value of maximum value and right
Hepatic vein pressure gradient classification results of the prediction classification answered as tester.Various embodiments of the present invention can be by class probability value most
High prediction classification is used as test result, realizes the noninvasive test of more classification to hepatic vein pressure gradient, testing classification result
Precision is higher, and generalization ability is stronger.
Referring to fig. 2, in a specific embodiment, further includes:
Step S210: the abdominal CT images for the tester that will acquire carry out the cutting pretreatment of spleen image, obtain multiple
Spleen segment to be measured.
The cutting pretreatment of spleen image includes importeding into the abdominal CT images of tester to show and delineate DICOM
The software of (Digital Imaging and Communications in Medicine, digital imaging and communications in medicine),
The figure layer of the first hilus lienis is selected in the horizontal bit image of splenic vein figure layer sequence, and the profile for extracting spleen tissue is hooked
It draws, the tissue regions delineated is smoothed.And the spleen tissue extracted region after smoothing processing is gone out according to predetermined
Segment size is cut into a certain size non-overlap segment.
The abdominal CT images of tester are carried out the cutting pretreatment of spleen image by the embodiment of the present invention, convenient for more being closed
In pylic minutia, the precision of hepatic vein pressure gradient class test is helped to improve.
Step S220: it is quiet that multiple spleen segments to be measured are input to progress liver in trained spleen prediction submodel in advance
The test of pulse pressure gradient classifications obtains multiple predictions classification of each spleen segment to be measured, and the corresponding classification of prediction classification
Probability.
Spleen prediction submodel is to cross multiple groups based on the second initial convolution neural network to belong to different default vena hepatica pressures
The spleen training image training of force gradient range and generate, wherein the second convolution neural network model is more points built
Connectionist model.
Step S230: hepatic vein pressure gradient class test is carried out for liver segment to be measured and spleen segment to be measured and is obtained
All class probabilities in belong to it is same prediction classification class probability be averaged, obtain accordingly predict classify class probability
Average value, and by each average value maximum value and the corresponding prediction classification of the maximum value be determined as the vena hepatica of tester
Barometric gradient class test result.
Classify for each prediction, by all trained segments (including liver training segment and spleen training segment) for this
The class probability of class prediction classification is averagely obtained average value, and by the maximum value and the maximum value pair in each average value
The prediction classification answered is determined as the hepatic vein pressure gradient class test result of tester.
Further, it in averaged, can be sought by average weighted mode, wherein in weighted average, liver
The weight of the prediction result of the weight and spleen prediction submodel of the prediction result of dirty prediction submodel can pass through greedy search
Method determines.
The hepatic vein pressure gradient class test of the CT image of the embodiment of the present invention, has merged spleen information, thus into one
Step improves the precision of prediction for noninvasive hepatic vein pressure gradient, helps to realize more classification to hepatic vein pressure gradient
Noninvasive test.
In a specific embodiment, abdominal CT images are subjected to liver image cutting pretreatment, comprising:
Liver area image is extracted from the hepatic portal figure layer of abdominal CT images, and liver area image is smoothly located
Reason.
Extracting liver area image for example can be by that can show and delineate software (such as ITK- of DICOM image
SNAP3.6 software) carry out delineate extraction manually, can also image segmentation algorithm automatically extracted.Liver area image is carried out
Smoothing processing is, for example, the average value (M) and standard deviation (θ) for calculating all pixels point gray value in liver area image, into one
The pixel value linear transformation of all pixels point in liver area image to section [M-3 × θ] is arrived the section [M+3 × θ] by step ground
It is interior.
Liver area image after smoothing processing is cut according to the first default segment size, obtains multiple livers to be measured
Dirty segment.
First default segment size can be 64*64 pixel, 32*33 pixel or 128*128 pixel, it is ensured that in each segment
Contain liver organization at least 50% region.To according to the first default segment size by the liver area image after smoothing processing
It is cut into a certain size such as square segment of non-overlap segment.If containing non-liver organization region in segment, by these
The pixel value of pixel in region is set as 0, the pixel value of the pixel in liver organization region is set as non-zero, so as not to shadow
Ring subsequent analysis.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention will carry out liver figure in abdominal CT images
As cutting pretreatment, liver area image is divided into multiple segments, to be analyzed for each segment, so as to analyze more
Mostly pylic detailed information, improves the precision of the noninvasive prediction of hepatic vein pressure gradient, so that Accurate Prediction goes out vena hepatica pressure
The affiliated range of force gradient.
In a specific embodiment, abdominal CT images are subjected to the cutting pretreatment of spleen image, comprising:
Spleen area image is extracted from the hilus lienis figure layer of abdominal CT images, and spleen area image is smoothly located
Reason.
Extracting spleen area image for example can be by that can show and delineate software (such as ITK- of DICOM image
SNAP3.6 software) carry out delineate extraction manually, can also image segmentation algorithm automatically extracted.Spleen area image is carried out
Smoothing processing is, for example, the average value (N) and standard deviation (β) for calculating all pixels point gray value in spleen area image, into one
The pixel value linear transformation of all pixels point in spleen area image to section [N-3 × β] is arrived the section [N+3 × β] by step ground
It is interior.
Spleen area image after smoothing processing is cut according to the second default segment size, obtains multiple spleens to be measured
Dirty segment.
Second default segment size can be 64*64 pixel, 32*33 pixel or 128*128 pixel, it is ensured that in each segment
Contain spleen tissue at least 50% region.To according to the second default segment size by the spleen area image after smoothing processing
It is cut into a certain size such as square segment of non-overlap segment.If containing non-spleen tissue region in segment, by these areas
The pixel value of pixel in domain is set as 0, the pixel value of the pixel in spleen tissue region is set as non-zero, so as not to influence
Subsequent analysis.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention will carry out spleen figure in abdominal CT images
As cutting pretreatment, spleen area image is divided into multiple segments, to be analyzed for each segment, so as to analyze more
Mostly pylic detailed information, improves the precision of the noninvasive prediction of hepatic vein pressure gradient, so that more to go out liver quiet for Accurate Prediction
The affiliated range of pulse pressure force gradient.
Referring to Fig. 3, in a specific embodiment, liver predicts the training step of submodel, comprising:
Step S310: being default weight coefficient range by all weight coefficient initializations of the first initial convolutional neural networks
Interior is uniformly distributed at random.
Convolutional neural networks are integrated with the function of coding and classification, mainly include input layer, hidden layer and output layer,
In, hidden layer includes several convolutional layers, pond layer and full articulamentum.Convolutional layer carries out feature extraction to liver training segment automatically
Afterwards, pond layer can carry out selection and information filtering to feature, and finally all extraction features are integrated in full articulamentum, then are input to
It is exported after being classified in classifier, that is, output layer.Further, the first convolution neural network structure that the present embodiment is built can
With but be not limited to include an input layer, seven convolutional layers, three pond layers, two full articulamentums and an output layer,
To meet the prediction classification demand to hepatic vein pressure gradient.Further, default weight coefficient may range from -0.5 to
0.5.Further, also the deviation of first initial each neuron of convolutional neural networks can be initialized as 0.
Step S320: it according to default hepatic vein pressure gradient range, will acquire and by liver image cutting pretreatment
Each liver training segment afterwards is grouped, and the liver of each group is trained first that segment is input to after weight coefficient initializes
In initial convolutional neural networks.
Pretreated liver training segment is cut according to default hepatic vein pressure gradient model by liver image by all
It encloses and is divided into multiple groups, the number of group is identical as the number of prediction classification.
Step S330: it obtains each liver training segment and passes through the multiple pre- of the corresponding output of the first initial convolutional neural networks
Classification and the corresponding class probability of prediction classification are surveyed, is based on the corresponding first function value of preset first-loss function to calculate.
Preset first-loss function is to minimize classification to intersect entropy function.Each liver training segment obtains one and is based on
The first function value that preset first-loss function is calculated.
Step S340: using the average value of each first function value as first object value, and based on first object value using instead
The weight coefficient of the first initial convolutional neural networks is updated to propagation algorithm, until the updated first initial convolutional neural networks
Prediction classification results make first object value be less than or equal to the first predetermined target value, then by the first initial volume after final updated
Product neural network predicts submodel as liver;First object value indicates the prediction classification results and reality to each liver training segment
The deviation of border grouping.
The corresponding first function value of each liver training segment is added averaged, obtains first object value.To
The first initial convolution is updated from the direction of output layer, hidden layer to input layer using back-propagation algorithm based on the first object value
The weight coefficient of neural network, so that prediction classification results and reality of the first initial convolutional neural networks to each liver training segment
Border grouping is consistent or deviation is smaller.Further, each liver training segment is re-entered into updated first initial volume
Product neural network in be iterated training, until first object value be less than or equal to the first predetermined target value, i.e., preset first
Loss function convergence, then it represents that the deviation of prediction classification results and actual packet to each liver training segment is smaller or one
It causes, then predicts submodel for the first initial convolutional neural networks after final updated as liver.Wherein, Adam optimization can be used
Device is trained, and learning rate can be 0.0012.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention can will be instructed by pretreated each liver
Practice image to be grouped according to default hepatic vein pressure gradient range, then is input in the first initial convolutional neural networks.As a result,
First initial convolutional neural networks can be corresponding more by automatically extracting characteristics of image and carrying out classification output liver training image
A prediction classification and corresponding class probability obtain first object value based on each prediction classification and each class probability, so as to root
Weight coefficient is adjusted according to first object value, by successive ignition training up to first-loss function convergence, and then it is pre- to complete liver
Survey the training of submodel.The embodiment of the present invention extracts characteristics of image without Manual definition, can carry out feature extraction automatically for door
Vein classification model construction improves the efficiency of model training, can be realized the classification of the prediction to different hepatic vein pressure gradient.
Referring to fig. 4, in a specific embodiment, the training step of spleen prediction submodel, comprising:
Step S410: the weight coefficient of the second initial convolutional neural networks is initialized as within the scope of default weight coefficient
It is uniformly distributed at random.
Step S420: it according to default hepatic vein pressure gradient range, will acquire and by the cutting pretreatment of spleen image
Each spleen training segment afterwards is grouped, and the spleen of each group is trained second that segment is input to after weight coefficient initializes
Initial convolutional neural networks.
Step S430: it obtains each spleen training segment and passes through the multiple pre- of the corresponding output of the second initial convolutional neural networks
Classification and the corresponding class probability value of prediction classification are surveyed, is based on the corresponding second function of preset second loss function to calculate
Value.
Step S440: using the average value of each second function value as the second target value, and based on the second target value using instead
The weight coefficient of the second initial convolutional neural networks is updated to propagation algorithm, until the updated second initial convolutional neural networks
Prediction classification results make the second target value be less than or equal to the second predetermined target value, then by the second initial convolution after final updated
Neural network predicts submodel as spleen;Second target value indicates the prediction classification results and reality to each spleen training segment
The deviation of grouping.
It should be noted that the present embodiment predicts submodule to liver to the training of spleen prediction submodel and above-described embodiment
The training method process of type is identical, and details are not described herein.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention, can further improve to vena hepatica pressure
The precision of force gradient non-invasive diagnosis prediction, to facilitate the affiliated range that more Accurate Prediction goes out hepatic vein pressure gradient.
In a specific embodiment, at least one preset hepatic vein pressure gradient key value is obtained, and according to each
The hepatic vein pressure gradient key value divides the corresponding preset pressure gradient scope.
In a specific embodiment, preset hepatic vein pressure gradient key value is five;
Each preset hepatic vein pressure gradient key value be respectively 5mmHg, 10mmHg, 12mmHg, 16mmHg, 20mmHg with
And 22mmHg.
Important in clinical practice is several key values, such as 5mmHg, 10mmHg, 12mmHg,
16mmHg, 20mmHg, 22mmHg etc..HVPG > 5mmHg prompt has portal hypertension;> 10mmHg prompts clinical significance portal vein high
It presses (CSPH), the compensatory phase patient of cirrhosis occurs varication, decompensation event and liver cancer risk and increases, Post hepatectomy of liver cancer hair
The risk of raw decompensation event increases;> 12mmHg prompt decompensated liver cirrhosis, be occur variceal bleeding it is high-risk because
Element;> 16mmHg prompts the mortality risk of cirrhotic patients with portal hypertension patient to increase;> 20mmHg prompts cirrhosis Acute Venous bent
The hemostatic treatment failure rate and mortality risk for opening bleeding patients increase;The death of > 22mmHg prompt acute alcoholic hepatitis patient
Risk increases.
Therefore, it can be, but not limited to be confined to be divided into multiple livers according to 5 preset hepatic vein pressure gradient key values
Venous pressure gradient range, each hepatic vein pressure gradient range belong to a prediction classification.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention, realizes to hepatic vein pressure gradient
Classify noninvasive test, testing classification result precision is higher, and generalization ability is stronger more.
In a specific embodiment, loss function is obtained based on following formula:
Wherein, c indicates prediction classification;M indicates the number of prediction classification;O indicates the category set of prediction classification;yo,cTable
Show the corresponding label value of prediction classification;Po,cIndicate the corresponding class probability of prediction classification.
For example, the hepatic vein pressure gradient key value chosen is respectively 5mmHg, 10mmHg, 12mmHg, 16mmHg,
20mmHg, 22mmHg, the then range divided are that HVPG≤5mmHg is that a range belongs to a prediction classification, corresponding label value
It is 1.5mmHg≤HVPG≤10mmHg is that a range belongs to a prediction classification, and corresponding label value is 2.10mmHg≤HVPG
≤ 12mmHg is that a range belongs to a prediction classification, and corresponding label value is 3.12mmHg≤HVPG≤20mmHg is a model
It encloses and belongs to a prediction classification, corresponding label value is 4.20mmHg≤HVPG≤22mmHg is that a range belongs to a prediction point
Class, corresponding label value are 5.In turn, hepatic vein pressure gradient range 5 classifications, the corresponding moulds of classifying of each classification have been divided into more
Label value in type.Therefore, each trained segment exports multiple prediction classification and prediction point after corresponding prediction submodel
After the corresponding class probability of class, the corresponding functional value of every trained segment can be calculated based on above-mentioned loss function, further
Obtain overall first object value or the second target value.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention can be based on first-loss function and second
Loss function obtains first object value and the second target value, by the first initial convolutional neural networks and the second initial convolution nerve net
The prediction classification results of network and the deviation of actual packet quantify, and help fast and effeciently to train more classification vena hepatica pressure ladders
The liver prediction submodel and spleen of degree predict submodel.
As a preferred embodiment, for training the data source of liver prediction submodel and spleen prediction submodel such as
Under: cirrhosis person, HVPG measurement and abdomen Contrast enhanced CT scan patient and possess the strong of abdomen Contrast enhanced CT scan
Kang personage.Further, according to a certain percentage by all data, such as 3:1:1 is divided into training set, verifying collection and test
Collection.
Further, training set for liver prediction submodel and spleen prediction submodel training, verifying collection for
The classification performance that model corresponds to the data not occurred in training set is verified in training process, so that model is according to the property on verifying collection
Energy situation adjusting parameter, test set are used to after model determines preferably to test and measure the performance of model.It is testing
On collection, the classification accuracy tested (is indicated, AUC is area under ROC curve, the value between 0-1, bigger expression with AUC
It is more accurate to classify) it is significantly higher than other methods, the AUC of classification can achieve 0.946, and the AUC of other comparative approach is respectively
0.778,0.549,0.505,0.473,0.558,0.483.
The hepatic vein pressure gradient classification method of the CT image of the embodiment of the present invention does not need manually to set characteristics of image,
By automatically extracting the feature in CT image by convolutional neural networks, therefore it is more conducive to extract suitable hepatic vein pressure gradient
The more advanced CT characteristics of image of classification;And the hepatic vein pressure gradient disaggregated model classification results that the embodiment of the present invention obtains are more
Accurately, and generalization ability is stronger, is applicable to the classification of different hepatic vein pressure gradient key values.
Referring to Fig. 5, in one embodiment, the present invention also provides a kind of classification of the hepatic vein pressure gradient of CT image to fill
It sets, comprising:
First test preprocessing module 510, the abdominal CT images of the tester for will acquire carry out liver image and cut
Pretreatment is cut, multiple liver segments to be measured are obtained.
First categorization module 520 predicts submodule for multiple liver segments to be measured to be input to trained liver in advance
Hepatic vein pressure gradient class test is carried out in type, obtains multiple predictions classification of each liver segment to be measured, and prediction point
The corresponding class probability of class;Wherein, the hepatic vein pressure gradient of tester is within the scope of same default hepatic vein pressure gradient
Same prediction classification.
First prediction module 530, for calculating the average value for belonging to the class probability of same prediction classification, and will be each average
The corresponding prediction classification of maximum value and the maximum value in value is determined as the hepatic vein pressure gradient class test knot of tester
Fruit.
In a specific embodiment, further includes:
Second test preprocessing module, it is pre- that the abdominal CT images of the tester for will acquire carry out the cutting of spleen image
Processing, obtains multiple spleen segments to be measured.
Second categorization module, for multiple spleen segments to be measured to be input in trained spleen prediction submodel in advance
Hepatic vein pressure gradient class test is carried out, multiple predictions classification of each spleen segment to be measured, and prediction classification pair are obtained
The class probability answered.
Second prediction module, for hepatic vein pressure gradient point will to be carried out for liver segment to be measured and spleen segment to be measured
The class probability for belonging to same prediction classification in all class probabilities that class testing obtains is averaged, and obtains accordingly predicting classification
Class probability average value, and by each average value maximum value and the maximum value corresponding prediction classification be determined as testing
The hepatic vein pressure gradient class test result of person.
In a specific embodiment, the first test preprocessing module includes:
First image processing unit, for extracting liver area image from the hepatic portal figure layer of abdominal CT images, and it is right
Liver area image is smoothed.
First image cutting, for by the liver area image after smoothing processing according to the first default segment size into
Row cutting, obtains multiple liver segments to be measured.
In a specific embodiment, the second test preprocessing module includes:
Second image processing unit, for extracting spleen area image from the hilus lienis figure layer of abdominal CT images, and it is right
Spleen area image is smoothed.
Second image cutting, for by the spleen area image after smoothing processing according to the second default segment size into
Row cutting, obtains multiple spleen segments to be measured.
It in a specific embodiment, further include liver prediction submodel training module, liver predicts submodel training
Module includes:
First initial cell, for being default weight by all weight coefficient initializations of the first initial convolutional neural networks
Being uniformly distributed in coefficient range at random.
First grouped element, for will acquire and cut by liver image according to hepatic vein pressure gradient range is preset
It cuts pretreated each liver training segment to be grouped, and the liver of each group training segment is input to weight coefficient initialization
In the initial convolutional neural networks of first afterwards.
First computing unit passes through the corresponding output of the first initial convolutional neural networks for obtaining each liver training segment
Multiple predictions classification and prediction classify corresponding class probability, to calculate based on preset first-loss function corresponding the
One functional value.
First model modification unit for using the average value of each first function value as first object value, and is based on first
Target value updates the weight coefficient of the first initial convolutional neural networks using back-propagation algorithm, until updated first is initial
The prediction classification results of convolutional neural networks make first object value be less than or equal to the first predetermined target value, then will be after final updated
The first initial convolutional neural networks as liver predict submodel;First object value indicates the prediction to each liver training segment
The deviation of classification results and actual packet.
It in a specific embodiment, further include spleen prediction submodel training module, spleen predicts submodel training
Module includes:
Second initial cell, for the weight coefficient of the second initial convolutional neural networks to be initialized as default weight coefficient
Being uniformly distributed in range at random.
Second packet unit, for will acquire and cut by spleen image according to hepatic vein pressure gradient range is preset
It cuts pretreated each spleen training segment to be grouped, and the spleen of each group training segment is input to weight coefficient initialization
The initial convolutional neural networks of second afterwards.
Second computing unit passes through the corresponding output of the second initial convolutional neural networks for obtaining each spleen training segment
Multiple predictions classification and prediction classify corresponding class probability value, it is corresponding based on preset second loss function to calculate
Second function value.
Second model modification unit for using the average value of each second function value as the second target value, and is based on second
Target value updates the weight coefficient of the second initial convolutional neural networks using back-propagation algorithm, until updated second is initial
The prediction classification results of convolutional neural networks make the second target value be less than or equal to the second predetermined target value, then will be after final updated
Second initial convolutional neural networks predict submodel as spleen;Second target value indicates the prediction point to each spleen training segment
The deviation of class result and actual packet.
In a specific embodiment, further includes:
Hepatic vein pressure gradient range division module, it is crucial for obtaining at least one preset hepatic vein pressure gradient
Value, and corresponding default hepatic vein pressure gradient range is divided according to each hepatic vein pressure gradient key value.
In a specific embodiment, preset hepatic vein pressure gradient key value is five;
Each preset hepatic vein pressure gradient key value be respectively 5mmHg, 10mmHg, 12mmHg, 16mmHg, 20mmHg with
And 22mmHg.
In a specific embodiment, loss function is obtained based on following formula:
Wherein, c indicates prediction classification;M indicates the number of prediction classification;O indicates the category set of prediction classification;yo,cTable
Show the corresponding label value of prediction classification;Po,cIndicate the corresponding class probability of prediction classification.
The specific restriction of hepatic vein pressure gradient sorter about CT image may refer to above for CT image
Hepatic vein pressure gradient classification method restriction, details are not described herein.The hepatic vein pressure gradient of above-mentioned CT image, which is classified, to be filled
Modules in setting can be realized fully or partially through software, hardware and combinations thereof.Above-mentioned each module can be in the form of hardware
It is embedded in or independently of the storage that in the processor in computer equipment, can also be stored in a software form in computer equipment
In device, the corresponding operation of the above modules is executed in order to which processor calls.
In one embodiment, the present invention also provides a kind of computer equipments, including memory and processor, memory to deposit
Computer program is contained, processor realizes the hepatic vein pressure gradient classification method of CT image when executing computer program.
It should be noted that the detailed process and reality of the hepatic vein pressure gradient classification method of the CT image in the present embodiment
Existing principle is identical as the hepatic vein pressure gradient classification method of CT image of the various embodiments described above, and details are not described herein.
The computer equipment of the embodiment of the present invention realizes the noninvasive test of more classification to hepatic vein pressure gradient, test
Classification results precision is higher, and generalization ability is stronger.
In one embodiment, the present invention also provides a kind of computer readable storage mediums, are stored thereon with computer journey
Sequence realizes each step in the hepatic vein pressure gradient classification method of CT image when the computer program is executed by processor.
In several embodiments provided herein, it should be understood that disclosed device and method can also pass through
Other modes are realized.The apparatus embodiments described above are merely exemplary, for example, flow chart and structure in attached drawing
Figure shows the system frame in the cards of the device of multiple embodiments according to the present invention, method and computer program product
Structure, function and operation.In this regard, each box in flowchart or block diagram can represent a module, section or code
A part, a part of the module, section or code includes one or more for implementing the specified logical function
Executable instruction.It should also be noted that function marked in the box can also be to be different from the implementation as replacement
The sequence marked in attached drawing occurs.For example, two continuous boxes can actually be basically executed in parallel, they are sometimes
It can execute in the opposite order, this depends on the function involved.It is also noted that in structure chart and/or flow chart
The combination of each box and the box in structure chart and/or flow chart, can function or movement as defined in executing it is dedicated
Hardware based system realize, or can realize using a combination of dedicated hardware and computer instructions.
In addition, each functional module or unit in each embodiment of the present invention can integrate one independence of formation together
Part, be also possible to modules individualism, an independent part can also be integrated to form with two or more modules.
It, can be with if the function is realized and when sold or used as an independent product in the form of software function module
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be intelligence
Can mobile phone, personal computer, server or network equipment etc.) execute each embodiment the method for the present invention whole or
Part steps.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory),
Random access memory (RAM, Random Access Memory), magnetic or disk etc. be various to can store program code
Medium.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.
Claims (10)
1. a kind of hepatic vein pressure gradient classification method of CT image characterized by comprising
The abdominal CT images for the tester that will acquire carry out liver image cutting pretreatment, obtain multiple liver segments to be measured;
Multiple described liver segments to be measured are input in trained liver prediction submodel in advance and carry out vena hepatica pressure ladder
Class test is spent, the multiple predictions classification and the corresponding classification of prediction classification of each liver segment to be measured are obtained
Probability;Wherein, the hepatic vein pressure gradient of the tester is same prediction within the scope of same default hepatic vein pressure gradient
Classification;
Calculate belong to it is same it is described prediction classification class probability average value, and by each average value maximum value and this most
It is worth the hepatic vein pressure gradient class test result that the corresponding prediction classification is determined as the tester greatly.
2. the hepatic vein pressure gradient classification method of CT image according to claim 1, which is characterized in that further include:
The abdominal CT images for the tester that will acquire carry out the cutting pretreatment of spleen image, obtain multiple spleen figures to be measured
Block;
Multiple described spleen segments to be measured are input in trained spleen prediction submodel in advance and carry out vena hepatica pressure ladder
Class test is spent, multiple prediction classification and the prediction classification for obtaining each spleen segment to be measured are corresponding
Class probability;
It will carry out what hepatic vein pressure gradient class test obtained for the liver segment to be measured and the spleen segment to be measured
The class probability for belonging to same prediction classification in all class probabilities is averaged, and the class probability for accordingly predicting classification is obtained
Average value, and by each average value maximum value and the corresponding prediction classification of the maximum value be determined as the tester's
Hepatic vein pressure gradient class test result.
3. the hepatic vein pressure gradient classification method of CT image according to claim 1, which is characterized in that by the abdomen
CT image carries out liver image cutting pretreatment, comprising:
Liver area image is extracted from the hepatic portal figure layer of the abdominal CT images, and the liver area image is carried out flat
Sliding processing;
Liver area image after smoothing processing is cut according to the first default segment size, obtains multiple described livers to be measured
Dirty segment.
4. the hepatic vein pressure gradient classification method of CT image according to claim 2, which is characterized in that by the abdomen
CT image carries out the cutting pretreatment of spleen image, comprising:
Spleen area image is extracted from the hilus lienis figure layer of the abdominal CT images, and the spleen area image is carried out flat
Sliding processing;
Spleen area image after smoothing processing is cut according to the second default segment size, obtains multiple described spleens to be measured
Dirty segment.
5. the hepatic vein pressure gradient classification method of CT image according to claim 1, which is characterized in that the liver is pre-
Survey the training step of submodel, comprising:
It is random uniform within the scope of default weight coefficient by all weight coefficient initializations of the first initial convolutional neural networks
Distribution;
According to the default hepatic vein pressure gradient range, will acquire and pretreated each by liver image cutting
Liver training segment be grouped, and by the liver of each group training segment be input to weight coefficient initialization after described first at the beginning of
In beginning convolutional neural networks;
Obtain multiple predictions point that each liver training segment passes through the corresponding output of the described first initial convolutional neural networks
Class and the corresponding class probability of prediction classification, are based on the corresponding first function value of preset first-loss function to calculate;
Using the average value of each first function value as first object value, and backpropagation is utilized based on the first object value
Algorithm updates the weight coefficient of the described first initial convolutional neural networks, until the updated first initial convolution nerve net
The prediction classification results of network make the first object value be less than or equal to the first predetermined target value, then will be described in after final updated
First initial convolutional neural networks predict submodel as the liver;The first object value is indicated to each liver training
The deviation of the prediction classification results and actual packet of segment.
6. the hepatic vein pressure gradient classification method of CT image according to claim 2, which is characterized in that the spleen is pre-
Survey the training step of submodel, comprising:
The weight coefficient of second initial convolutional neural networks is initialized as being uniformly distributed within the scope of default weight coefficient at random;
According to the default hepatic vein pressure gradient range, will acquire and pretreated each by spleen image cutting
Spleen training segment be grouped, and by the spleen of each group training segment be input to weight coefficient initialization after described second at the beginning of
Beginning convolutional neural networks;
Obtain multiple predictions point that each spleen training segment passes through the corresponding output of the described second initial convolutional neural networks
Class and the corresponding class probability value of prediction classification, are based on the corresponding second function value of preset second loss function to calculate;
Using the average value of each second function value as the second target value, and backpropagation is utilized based on second target value
Algorithm updates the weight coefficient of the described second initial convolutional neural networks, until the updated second initial convolution nerve net
The prediction classification results of network make second target value be less than or equal to the second predetermined target value, then by described the after final updated
Two initial convolutional neural networks predict submodel as the spleen;Second target value is indicated to each spleen training figure
The deviation of the prediction classification results and actual packet of block.
7. the hepatic vein pressure gradient classification method of CT image according to claim 1, which is characterized in that further include:
At least one preset hepatic vein pressure gradient key value is obtained, and is drawn according to each hepatic vein pressure gradient key value
Divide the corresponding default hepatic vein pressure gradient range.
8. the hepatic vein pressure gradient classification method of CT image according to claim 7, which is characterized in that described preset
Hepatic vein pressure gradient key value is five;
Each preset hepatic vein pressure gradient key value be respectively 5mmHg, 10mmHg, 12mmHg, 16mmHg, 20mmHg with
And 22mmHg.
9. the hepatic vein pressure gradient classification method of CT image according to claim 5 or 6, which is characterized in that loss letter
Base is obtained in following formula:
Wherein, c indicates the prediction classification;M indicates the number of the prediction classification;O indicates the classification collection of the prediction classification
It closes;yo,cIndicate the corresponding label value of the prediction classification;Po,cIndicate the corresponding class probability of the prediction classification.
10. a kind of computer equipment, which is characterized in that including memory and processor, the memory is stored with computer journey
Sequence, the processor realize the vena hepatica of CT image described in any one of claims 1 to 9 when executing the computer program
Barometric gradient classification method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910522932.5A CN110074809B (en) | 2019-06-17 | 2019-06-17 | Hepatic vein pressure gradient classification method of CT image and computer equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910522932.5A CN110074809B (en) | 2019-06-17 | 2019-06-17 | Hepatic vein pressure gradient classification method of CT image and computer equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110074809A true CN110074809A (en) | 2019-08-02 |
CN110074809B CN110074809B (en) | 2023-06-20 |
Family
ID=67424427
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910522932.5A Active CN110074809B (en) | 2019-06-17 | 2019-06-17 | Hepatic vein pressure gradient classification method of CT image and computer equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110074809B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110755104A (en) * | 2019-11-05 | 2020-02-07 | 四川大学华西医院 | Liver cirrhosis portal hypertension noninvasive evaluation method based on 3D multichannel convolutional neural network |
CN112057751A (en) * | 2020-08-14 | 2020-12-11 | 中南大学湘雅医院 | Automatic delineation method for organs endangered in pelvic cavity radiotherapy |
CN112733708A (en) * | 2021-01-08 | 2021-04-30 | 山东交通学院 | Hepatic portal vein detection positioning method and system based on semi-supervised learning |
CN113658700A (en) * | 2021-08-16 | 2021-11-16 | 复旦大学附属中山医院 | Gate pulse high pressure non-invasive assessment method and system based on machine learning |
CN115019082A (en) * | 2022-04-06 | 2022-09-06 | 江苏科技大学 | Non-invasive liver function grading system |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014190170A1 (en) * | 2013-05-22 | 2014-11-27 | The Regents Of The University Of California | Biomarker composite test for hepatic vein pressure gradient and cirrhosis treatment |
WO2015155521A1 (en) * | 2014-04-08 | 2015-10-15 | Isis Innovation Limited | Medical imaging |
RU2014118282A (en) * | 2014-05-06 | 2015-11-27 | Федеральное государственное бюджетное образовательное учреждение высшего профессионального образования "Петрозаводский государственный университет" | METHOD FOR NON-INVASIVE DIAGNOSTICS OF PORTAL HYPERTENSION IN CHRONIC DIFFUSAL LIVER DISEASES |
CN105825070A (en) * | 2016-04-06 | 2016-08-03 | 南方医科大学南方医院 | Measurement method of virtual hepatic vein pressure gradient |
CN107480675A (en) * | 2017-07-28 | 2017-12-15 | 祁小龙 | A kind of construction method of the HVPG computation model based on radiation group |
CN107945878A (en) * | 2017-11-16 | 2018-04-20 | 祁小龙 | A kind of construction method of the hepatic vein pressure gradient computation model based on radiation group |
CN109528196A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics |
CN109758136A (en) * | 2019-02-28 | 2019-05-17 | 四川大学华西医院 | Hepatic vein pressure gradient measurement method based on portal vein blood vessel feature |
-
2019
- 2019-06-17 CN CN201910522932.5A patent/CN110074809B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014190170A1 (en) * | 2013-05-22 | 2014-11-27 | The Regents Of The University Of California | Biomarker composite test for hepatic vein pressure gradient and cirrhosis treatment |
WO2015155521A1 (en) * | 2014-04-08 | 2015-10-15 | Isis Innovation Limited | Medical imaging |
RU2014118282A (en) * | 2014-05-06 | 2015-11-27 | Федеральное государственное бюджетное образовательное учреждение высшего профессионального образования "Петрозаводский государственный университет" | METHOD FOR NON-INVASIVE DIAGNOSTICS OF PORTAL HYPERTENSION IN CHRONIC DIFFUSAL LIVER DISEASES |
CN105825070A (en) * | 2016-04-06 | 2016-08-03 | 南方医科大学南方医院 | Measurement method of virtual hepatic vein pressure gradient |
CN107480675A (en) * | 2017-07-28 | 2017-12-15 | 祁小龙 | A kind of construction method of the HVPG computation model based on radiation group |
CN107945878A (en) * | 2017-11-16 | 2018-04-20 | 祁小龙 | A kind of construction method of the hepatic vein pressure gradient computation model based on radiation group |
CN109528196A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics |
CN109758136A (en) * | 2019-02-28 | 2019-05-17 | 四川大学华西医院 | Hepatic vein pressure gradient measurement method based on portal vein blood vessel feature |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110755104A (en) * | 2019-11-05 | 2020-02-07 | 四川大学华西医院 | Liver cirrhosis portal hypertension noninvasive evaluation method based on 3D multichannel convolutional neural network |
CN112057751A (en) * | 2020-08-14 | 2020-12-11 | 中南大学湘雅医院 | Automatic delineation method for organs endangered in pelvic cavity radiotherapy |
CN112733708A (en) * | 2021-01-08 | 2021-04-30 | 山东交通学院 | Hepatic portal vein detection positioning method and system based on semi-supervised learning |
CN113658700A (en) * | 2021-08-16 | 2021-11-16 | 复旦大学附属中山医院 | Gate pulse high pressure non-invasive assessment method and system based on machine learning |
CN113658700B (en) * | 2021-08-16 | 2023-10-27 | 复旦大学附属中山医院 | Gate pulse high-pressure noninvasive evaluation method and system based on machine learning |
CN115019082A (en) * | 2022-04-06 | 2022-09-06 | 江苏科技大学 | Non-invasive liver function grading system |
CN115019082B (en) * | 2022-04-06 | 2024-05-07 | 江苏科技大学 | Non-invasive liver function grading system |
Also Published As
Publication number | Publication date |
---|---|
CN110074809B (en) | 2023-06-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109584254B (en) | Heart left ventricle segmentation method based on deep full convolution neural network | |
CN110074809A (en) | The hepatic vein pressure gradient classification method and computer equipment of CT image | |
Chouhan et al. | Soft computing approaches for image segmentation: a survey | |
Wu et al. | Cascaded fully convolutional networks for automatic prenatal ultrasound image segmentation | |
CN109472263B (en) | Global and local information combined brain magnetic resonance image segmentation method | |
CN105825509A (en) | Cerebral vessel segmentation method based on 3D convolutional neural network | |
CN111539947A (en) | Image detection method, training method of related model, related device and equipment | |
CN103514597A (en) | Image processing device | |
CN101667289B (en) | Retinal image segmentation method based on NSCT feature extraction and supervised classification | |
CN105488781A (en) | Dividing method based on CT image liver tumor focus | |
CN110363772B (en) | Heart MRI segmentation method and system based on antagonistic network | |
CN110211139A (en) | Automatic segmentation Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ | |
CN108256453A (en) | A kind of method based on one-dimensional ECG signal extraction two dimension CNN features | |
CN110991254B (en) | Ultrasonic image video classification prediction method and system | |
Omar et al. | Automated myocardial wall motion classification using handcrafted features vs a deep cnn-based mapping | |
CN111297399A (en) | Fetal heart positioning and fetal heart rate extraction method based on ultrasonic video | |
CN111275706A (en) | Shear wave elastic imaging-based ultrasound omics depth analysis method and system | |
Banerjee et al. | A CADe system for gliomas in brain MRI using convolutional neural networks | |
Fathima et al. | A deep learning approach on segmentation of bone for bmd measurement from dexa scan images | |
CN114862865B (en) | Vessel segmentation method and system based on multi-view coronary angiography sequence image | |
Chouat et al. | Lung disease detection in chest x-ray images using transfer learning | |
CN110084800A (en) | A kind of Lung metastases prediction technique for four limbs soft tissue sarcoma patient | |
Chevaillier et al. | Functional segmentation of renal DCE-MRI sequences using vector quantization algorithms | |
CN114864095A (en) | Analysis method for blood circulation change of narrow coronary artery under combination of multiple exercise strengths | |
Baracho et al. | A hybrid neural system for the automatic segmentation of the interventricular septum in echocardiographic images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |