CN109528196A - A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics - Google Patents

A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics Download PDF

Info

Publication number
CN109528196A
CN109528196A CN201811353945.6A CN201811353945A CN109528196A CN 109528196 A CN109528196 A CN 109528196A CN 201811353945 A CN201811353945 A CN 201811353945A CN 109528196 A CN109528196 A CN 109528196A
Authority
CN
China
Prior art keywords
hvpg
heuristics
modal
image
network
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
Application number
CN201811353945.6A
Other languages
Chinese (zh)
Other versions
CN109528196B (en
Inventor
贾熹滨
刘云峰
杨正汉
杨大为
王晓培
肖玉杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Technology
Original Assignee
Beijing University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing University of Technology filed Critical Beijing University of Technology
Priority to CN201811353945.6A priority Critical patent/CN109528196B/en
Publication of CN109528196A publication Critical patent/CN109528196A/en
Application granted granted Critical
Publication of CN109528196B publication Critical patent/CN109528196B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves 
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus for radiation diagnosis, e.g. combined with radiation therapy equipment
    • A61B6/02—Devices for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
    • A61B6/03—Computerised tomographs
    • A61B6/032—Transmission computed tomography [CT]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus for radiation diagnosis, e.g. combined with radiation therapy equipment
    • A61B6/50—Clinical applications
    • A61B6/504—Clinical applications involving diagnosis of blood vessels, e.g. by angiography
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus for radiation diagnosis, e.g. 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
    • A61B6/5229—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data combining image data of a patient, e.g. combining a functional image with an anatomical image
    • A61B6/5247—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data combining image data of a patient, e.g. combining a functional image with an anatomical image combining images from an ionising-radiation diagnostic technique and a non-ionising radiation diagnostic technique, e.g. X-ray and ultrasound
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
    • A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
    • A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
    • A61B8/5238—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for combining image data of patient, e.g. merging several images from different acquisition modes into one image
    • A61B8/5261—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for combining image data of patient, e.g. merging several images from different acquisition modes into one image combining images from different diagnostic modalities, e.g. ultrasound and X-ray

Abstract

The hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics that the invention discloses a kind of, feature extraction is carried out to multi-modal medical imaging using convolutional neural networks and obtains the HVPG estimated value based on multi-modal image, regression analysis is carried out to HVPG correlation experience knowledge parameter using deep-neural-network and obtains the HVPG estimated value based on Heuristics, the HVPG estimated value based on multi-modal image and Heuristics obtained for above-mentioned steps is merged, obtain fused HVPG estimated value, joint training is carried out to aforementioned convolutional neural networks and deep-neural-network using optimization algorithm, after the completion of training, it can be used to carry out Accurate Prediction to HVPG, obtain the HVPG quantitative estimation value based on multi-modal image and Heuristics.The present invention takes into account the complementary information of multiple modalities medical image, while the supplement of feature is carried out using corresponding Heuristics, is more in line with specific aim medically.

Description

A kind of hepatic vein pressure gradient based on multi-modal image and Heuristics is non-invasive to be commented Estimate method
Technical field
The present invention relates to medical image technical fields, and in particular to a kind of liver based on multi-modal image and Heuristics is quiet Pulse pressure force gradient Noninvasive assessmet method.
Background technique
Portal hypertension is one of most common severe complication of cirrhosis, clinical manifestation be esophagus fundus ventricularis varication simultaneously Rupture haemorrhage, ascites, splenomegaly and hypersplenia etc..With the increase of portal venous pressure, esophageal varices go out The risk of blood increases, and liver after the incidence of hepatocellular carcinoma (Hepatocellular Carcinoma, HCC), HCC resection The incidence of dirty failure and the risk of associated death also increase therewith.Therefore, precise quantification and dynamic monitoring portal hypertension water The pathogenesis of flat portal hypertension, diagnosis, treatment research have vital meaning.
" goldstandard " of generally acknowledged assessment portal venous pressure is hepatic vein pressure gradient (hepatic venous at present Pressure gradient, HVPG), which is inserted to vena hepatica for wedge guide or sacculus by puncture method, measures vena hepatica Wedge Pressure and the free pressure of vena hepatica, the difference, that is, HVPG of the two.However, HVPG measurement is that a kind of technical requirements are higher invasive Check, there is certain technical difficulty and bleeding risk, and somewhat expensive, thus clinically need it is a kind of it is noninvasive, accurate door is quiet Pulse pressure appraisal procedure.
In traditional Imaging Method, including a variety of image technologies such as ultrasound, CT, magnetic resonance imaging have been used for portal vein In the research of barometric gradient assessment, but conventional image technology has the following problems:
1) variation such as secondary liver and spleen form, hardness, haemodynamics is extremely complex when due to portal hypertension, and traditional Imaging Method is mostly qualitatively diagnosis, is influenced by subjective experience, lacks more accurate quantitative evaluation；
2) the different pathological characters that the image of different modalities would generally reflect, and traditional Imaging Method is mostly from single The angle of mode image technology is studied, and can not accomplish the comprehensive assessment of various dimensions.
Deep learning (Deep learning) passes through the side of data-driven because of its special deep-neural-network structure Method can be fitted complex data and excavate the feature wherein with extremely strong generalization ability.It is assumed that being based on multi-modal medical imaging data It can learn the image characteristics and diagnostic mode that disclose reflection HVPG level out, a side by human sample in conjunction with depth learning technology Valid data collection is established by clinical acquisitions and smart standard calibration case in face, is learnt using depth model and models multi-modal image Data and HVPG quantitative estimation relationship；On the other hand, it with clinical experience empirically knowledge, establishes in multi-modal image data and shows The calculating model of neural networks for writing empirical parameter estimation HVPG realizes that fused data study and the HVPG quantization of Heuristics are estimated It surveys, and then solves the evaluation problem of portal venous pressure.
Summary of the invention
It is assessed the purpose of the present invention is to provide a kind of based on the hepatic vein pressure gradient of multi-modal image and Heuristics Method models multi-modal image data using deep learning and models with HVPG, and is aided with form in image, function, blood The empirical parameters such as hydromechanics realize the quantitative estimation to HVPG value.
To achieve the goals above, the invention adopts the following technical scheme: firstly, establishing the multi-modal image number of effective higher-dimension According to association indicate；Then, HVPG quantitative estimation regression model of the modeling based on data-driven is indicated using multi-modal association, obtain To the HVPG quantitative estimation value based on data-driven；Meanwhile HVPG in the clinic such as make full use of form, function, haemodynamics Qualitative estimation experience empirically knowledge establishes the HVPG quantitative estimation relational model based on Heuristics；Finally, using fusion Layer decision strategy, establishes the HVPG quantitative estimation fusion calculation model based on data-driven and Heuristics, realizes more accurate HVPG estimation.
A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics, including following step It is rapid:
Step 1, feature extraction is carried out to multi-modal medical imaging using convolutional neural networks and obtained based on multi-modal shadow The HVPG estimated value H of picture1, specifically comprise the following steps:
Step 1.1, resonant spring imaging MRE is acquired, DCE-MRPV is imaged in more phase dynamic contrast-enhanced magnetic resonance portal veins and more The medical imaging sequence of flip angle tri- kinds of mode of unenhanced T1mapping, and spelled after to three kinds of medical imaging series processings It connects, obtains multi-modal medical imaging；
Step 1.2, feature extraction is carried out to multi-modal medical imaging using convolutional neural networks, and obtained based on multi-modal The HVPG estimated value H of image1；
Step 2, regression analysis is carried out to HVPG correlation experience knowledge parameter using deep-neural-network and obtained based on warp Test the HVPG estimated value H of knowledge2, specifically comprise the following steps:
Step 2.1, resonance configurations relevant to HVPG, function, hemodynamic parameter are acquired and is spliced into HVPG warp Test knowledge parameter；
Step 2.2, regression analysis is carried out to HVPG Heuristics parameter using deep-neural-network, obtains knowing based on experience The HVPG estimated value H of knowledge2；
Step 3, the HVPG estimated value H based on multi-modal image and Heuristics obtained for above-mentioned steps1With H2Into Row fusion, obtains fused HVPG estimated value H ', specific formula for calculation are as follows: H '=α H1+(1-α)H2, wherein α is as fusion Weight is used to weigh the importance of multi-modal image and Heuristics for estimating HVPG, and value range is 0 < α < 1；
Step 4, convolutional neural networks involved in steps 1 and 2 and deep-neural-network are combined using optimization algorithm Training；
Step 5, after the completion of training, that is, it can be used to carry out Accurate Prediction to HVPG, obtain based on multi-modal image and experience The HVPG quantitative estimation value of knowledge.
A kind of hepatic vein pressure gradient based on multi-modal image and Heuristics according to claim 1 is non-invasive Appraisal procedure, to the method for three kinds of medical imaging sequences pretreatments and splicing in step 1.1 are as follows: using interpolation method to three kinds of mode The size of image sequence be normalized, three kinds of mode image sequence sizes are M × N × K after normalized, M, N, K is respectively the length, width and height of image, is then averaged respectively in the dimension of channel to the image sequence of three kinds of mode, size becomes M × N × 1, then spliced in the dimension of channel, just obtain the multi-modal medical imaging that size is M × N × 3.
Convolutional neural networks employed in step 1.2 are 6 layers of neural network model, include two convolutional layers, two ponds Change layer and two full articulamentums, mean value is calculated by the output to the full articulamentum of the last layer, is obtained based on multi-modal image HVPG estimated value, or use AlexNet, ResNet.
Selected Heuristics parameter relevant to HVPG and joining method in step 2.1 are as follows: resonance configurations parameter It include: liver maximum diameter S1, spleen three-dimensional maximum diameter S2, portal vein blood vessels caliber S3, vena hepatica blood vessels caliber S4 up and down； Acquiring magnetic resonance functional parameter includes: liver elasticity number F1, spleen elasticity number F2, liver parenchyma T1 value F3, spleen essence T1 value F4；Acquisition Hemodynamic parameter includes: portal venous flow speed V1, Hepatic venous flow speed V2, hepatic arterial blood flow speed V3, arteria linenalis Blood flow velocity V4.Then above-mentioned form, function, hemodynam ics param eter are built into matrixKnow as HVPG experience Know parameter.
Deep-neural-network used by step 2.2 is 5 layer networks, includes an input layer, an output layer and three Hidden layer, each hidden layer include five neurons.
Optimization algorithm described in step 4 uses the exponential decay rate α of single order moments estimation1=0.9, the finger of second order moments estimation Number attenuation rate α2The adaptive matrix of=0.999, initial learning rate e=0.09 estimates, optimization aim be adjust in network can Training parameter minimizes loss function.
The training method used in step 4 is network association training: the specific method using alternately training keeps a net The parameter constant of network only trains another network, then on the contrary；Wherein, training data be nuclear magnetic resonance image and pathological data, wherein Training sample 100,50, sample are verified, training label is the HVPG true value H that puncture method obtains；Loss function is based on most The Squared Error Loss Loss=(H-H ') of small square law2。
Detailed description of the invention
Fig. 1 is logic diagram of the invention；
Fig. 2 is flow chart of the method for the present invention；
Fig. 3 is the HVPG estimation models schematic diagram based on multi-modal image；
Fig. 4 is the HVPG estimation models schematic diagram based on Heuristics；
Beneficial effect
The present invention carries out the study of various dimensions by the medical imaging to different modalities using the method for deep learning, and auxiliary With the Heuristics diagnostic message of doctor, the available non-invasive HVPG estimated value precisely quantified.The present invention is compared with traditional Clinical puncture diagnosis method can accomplish completely noninvasive, and compared with traditional imaging diagnosis method, then eliminate subjective factor It influences and considers existing high latitude connection between multi-modal medical imaging simultaneously, so that the result made is more accurate.
Specific embodiment
Yi Xiajiehejutishishili,Bing Canzhaofutu,Dui Benfamingjinyibuxiangxishuoming.
The flow chart of the method for the invention as shown in Fig. 2, specifically includes the following steps:
Step 1, using convolutional neural networks (Convolutional Neural Network) to multi-modal medical imaging It carries out feature extraction and obtains the HVPG estimated value H based on multi-modal image1, specifically comprise the following steps:
Step 1.1, acquire resonant spring imaging (MRE), more phase dynamic contrast-enhanced magnetic resonance portal veins be imaged (DCE-MRPV) with And the medical imaging sequence of more unenhanced (T1 mapping) three kinds of mode of flip angle, and after to three kinds of medical imaging series processings Spliced, obtains multi-modal medical imaging；
The medical imaging sequence of tri- kinds of MRE, DCE-MRPV, T1 mapping mode are obtained by magnetic resonant imaging examination, Three kinds of image sequences are three-dimensional stereo data.The possible different from of the equipment used is checked due to different, firstly the need of The size of the image sequence of three kinds of mode is normalized, herein using traditional interpolation method that is, after normalized Three kinds of mode image sequence sizes are M × N × K (length, width and height that M, N, K are respectively image).Then to the image sequence of three kinds of mode It is listed in the dimension of channel and averages respectively (size becomes M × N × 1), then spliced in the dimension of channel, just having obtained size is M The multi-modal medical imaging of × N × 3.
Step 1.2, feature extraction is carried out to multi-modal medical imaging using convolutional neural networks, and obtained based on multi-modal The HVPG estimated value H of image1；
Present invention employs convolutional neural networks as shown in Figure 3 to carry out feature extraction to multi-modal medical imaging, and right The output of the full articulamentum of the last layer calculates mean value, obtains the HVPG estimated value based on multi-modal image.In network implementations, own Convolution kernel uses the random value exported from the normal distribution that the mean value of truncation is 0, standard deviation is 0.01 to be initialized；It uses Full connection of the rectifier linear unit (Rectified linear unit, ReLU) as convolutional layer and other than the last layer The nonlinear activation function of layer；The Dropout method for being 0.5 to full articulamentum use ratio prevents network over-fitting；
Step 2, HVPG correlation experience knowledge parameter is carried out using deep-neural-network (Deep Neural Network) Regression analysis simultaneously obtains the HVPG estimated value H based on Heuristics2, specifically comprise the following steps:
Step 2.1, resonance configurations relevant to HVPG, function, hemodynamic parameter are acquired and is spliced into HVPG warp Test knowledge parameter；
Acquisition resonance configurations parameter includes: liver maximum diameter S up and down1, spleen three-dimensional maximum diameter S2, portal vein Pipe caliber S3, vena hepatica blood vessels caliber S4；Acquiring magnetic resonance functional parameter includes: liver elasticity number F1, spleen elasticity number F2, liver it is real Matter T1 value F3, spleen essence T1 value F4；Acquiring hemodynamic parameter includes: portal venous flow speed V1, Hepatic venous flow speed V2, hepatic arterial blood flow speed V3, arteria linenalis blood flow velocity V4.Then above-mentioned form, function, hemodynam ics param eter are built into square Battle arrayAs HVPG Heuristics parameter.
Step 2.2, regression analysis is carried out to HVPG Heuristics parameter using deep-neural-network, obtains knowing based on experience The HVPG estimated value H of knowledge2；
The present invention includes an input layer, an output layer and three using 5 layers of deep-neural-network as shown in Figure 4 Hidden layer (each hidden layer includes five neurons), using HVPG Heuristics parameter as the input of network, the output of network HVPG estimated value of the layer output based on Heuristics.In network implementations, the initialization of network parameter uses random initializtion, and makes Use rectifier linear unit as the nonlinear activation function of hidden layer.
Step 3, the HVPG estimated value H based on multi-modal image and Heuristics obtained for above-mentioned steps1With H2Into Row fusion, obtains fused HVPG estimated value H ', specific formula for calculation are as follows: H '=α H1+(1-α)H2, wherein α is as fusion Weight, value range are (0 < α < 1)；
To make full use of the complementarity of multi-modal image data and Heuristics data in estimation HVPG, the present invention is used By the HVPG estimated value H based on multi-modal image1With the HVPG estimated value H based on Heuristics2Make further fusion to obtain more Accurately HVPG estimated value H ', is shown below:
H '=α H1+(1-α)H2
Wherein weight of the α (0 < α < 1) as fusion, for weighing multi-modal image and Heuristics for estimation HVPG's Importance.
Step 4, using the optimization algorithm of mainstream to convolutional neural networks involved in steps 1 and 2 and deep-neural-network into Row joint training；
Network association training keeps the parameter constant of a network specifically using the method for alternately training, only training is another A network, then it is on the contrary；Training data is the nuclear magnetic resonance image and pathological data of 150 patients with chronic liver, i.e. every sample standard deviation Magnetic resonance shape relevant to HVPG described in medical imaging sequence, step 2.1 comprising three kinds of mode described in step 1.1 State, function, hemodynamic parameter are simultaneously spliced into HVPG Heuristics parameter and obtaining by puncture method as training label The HVPG true value H of the patient arrived, wherein 150 samples are divided into the training sample for being trained to network 100, And 50, verifying sample for verifying network training effect；Loss function uses the Squared Error Loss based on least square method Loss=(H-H ')2；Optimization algorithm uses the exponential decay rate α of single order moments estimation1=0.9, the exponential decay rate of second order moments estimation α2The adaptive matrix of=0.999, initial learning rate e=0.09 estimate (adaptive moment estimation, Adam), Optimization aim be adjust network in can training parameter minimize loss function.
Step 5, after the completion of training, that is, it can be used to carry out Accurate Prediction to HVPG, obtain based on multi-modal image and experience The HVPG quantitative estimation value of knowledge
The performance of sample verifying network is verified using 50, and is repeated 10 times experiment, calculates the mean value of 10 experimental results And standard deviation, it is as shown in the table for experimental result:
Method 1 is that the multi-modal image data of 150 patients is trained and is verified only with the network of step 1 Result；Method 2 is to instruct only with the network of step 2 to the HPVG correlation experience knowledge supplemental characteristic of 150 patients The result practiced and verified；Method 1+2 is the converged network of the invention finally proposed to the multi-modal shadow to 150 patients As and the joint training of Heuristics supplemental characteristic and verify as a result, wherein fusion weight use 0.8, i.e., more stress multi-modal Estimation degree of the image to HVPG value.
So far, specific implementation process of the invention is just described.

Claims (7)

1. a kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics, including following step It is rapid:
Step 1, feature extraction is carried out to multi-modal medical imaging using convolutional neural networks and obtained based on multi-modal image HVPG estimated value H1, specifically comprise the following steps:
Step 1.1, resonant spring imaging MRE, more phase dynamic contrast-enhanced magnetic resonance portal veins imaging DCE-MRPV and more overturnings are acquired The medical imaging sequence of angle tri- kinds of mode of unenhanced T1mapping, and splice after to three kinds of medical imaging series processings, it obtains To multi-modal medical imaging；
Step 1.2, feature extraction is carried out to multi-modal medical imaging using convolutional neural networks, and obtained based on multi-modal image HVPG estimated value H1；
Step 2, regression analysis is carried out to HVPG correlation experience knowledge parameter using deep-neural-network and obtains knowing based on experience The HVPG estimated value H of knowledge2, specifically comprise the following steps:
Step 2.1, relevant to HVPG resonance configurations, function, hemodynamic parameter are acquired and is spliced into HVPG experience and knows Know parameter；
Step 2.2, regression analysis is carried out to HVPG Heuristics parameter using deep-neural-network, obtained based on Heuristics HVPG estimated value H2；
Step 3, the HVPG estimated value H based on multi-modal image and Heuristics obtained for above-mentioned steps1With H2Melted It closes, obtains fused HVPG estimated value H ', specific formula for calculation are as follows: H '=α H1+(1-α)H2, wherein α is as the weight merged For weighing multi-modal image and Heuristics for the importance of estimation HVPG, value range is 0 < α < 1；
Step 4, joint instruction is carried out to convolutional neural networks involved in steps 1 and 2 and deep-neural-network using optimization algorithm Practice；
Step 5, after the completion of training, that is, it can be used to carry out Accurate Prediction to HVPG, obtain based on multi-modal image and Heuristics HVPG quantitative estimation value.
It is commented 2. a kind of hepatic vein pressure gradient based on multi-modal image and Heuristics according to claim 1 is non-invasive Estimate method, it is characterised in that:
To the method for three kinds of medical imaging sequence pretreatments and splicing in step 1.1 are as follows: using interpolation method to the shadow of three kinds of mode As the size of sequence is normalized, three kinds of mode image sequence sizes are M × N × K after normalized, M, N, K points Not Wei image length, width and height, then to the image sequence of three kinds of mode channel dimension on average respectively, size become M × N × 1, then spliced in the dimension of channel, just obtain the multi-modal medical imaging that size is M × N × 3.
It is commented 3. a kind of hepatic vein pressure gradient based on multi-modal image and Heuristics according to claim 1 is non-invasive Estimate method, it is characterised in that:
Convolutional neural networks employed in step 1.2 are 6 layers of neural network model, include two convolutional layers, two pond layers And two full articulamentums, mean value is calculated by the output to the full articulamentum of the last layer, is obtained based on multi-modal image HVPG estimated value, or use AlexNet, ResNet.
It is commented 4. a kind of hepatic vein pressure gradient based on multi-modal image and Heuristics according to claim 1 is non-invasive Estimate method, it is characterised in that:
Selected Heuristics parameter relevant to HVPG and joining method in step 2.1 are as follows: resonance configurations parameter includes: Maximum diameter S1, spleen three-dimensional maximum diameter S2, portal vein blood vessels caliber S3, vena hepatica blood vessels caliber S4 above and below liver；Acquisition Magnetic resonance functional parameter includes: liver elasticity number F1, spleen elasticity number F2, liver parenchyma T1 value F3, spleen essence T1 value F4；Acquire blood flow Kinetic parameter includes: portal venous flow speed V1, Hepatic venous flow speed V2, hepatic arterial blood flow speed V3, arteria linenalis blood flow Speed V4.Then above-mentioned form, function, hemodynam ics param eter are built into matrixJoin as HVPG Heuristics Number.
It is commented 5. a kind of hepatic vein pressure gradient based on multi-modal image and Heuristics according to claim 1 is non-invasive Estimate method, it is characterised in that:
Deep-neural-network used by step 2.2 is 5 layer networks, is hidden comprising an input layer, an output layer and three Layer, each hidden layer include five neurons.
It is commented 6. a kind of hepatic vein pressure gradient based on multi-modal image and Heuristics according to claim 1 is non-invasive Estimate method, it is characterised in that:
Optimization algorithm described in step 4 uses the exponential decay rate α of single order moments estimation1=0.9, the index of second order moments estimation declines Lapse rate α2The adaptive matrix of=0.999, initial learning rate e=0.09 estimate that optimization aim is to adjust training in network Parameter minimizes loss function.
It is commented 7. a kind of hepatic vein pressure gradient based on multi-modal image and Heuristics according to claim 1 is non-invasive Estimate method, it is characterised in that:
The training method used in step 4 is network association training: the specific method using alternately training keeps a network Parameter constant only trains another network, then on the contrary；Wherein, training data is nuclear magnetic resonance image and pathological data, is divided into training Sample and verifying sample, training label are the HVPG true value H that puncture method obtains；Loss function is flat based on least square method Side's loss Loss=(H-H ')2。
CN201811353945.6A 2018-11-14 2018-11-14 Hepatic vein pressure gradient non-invasive evaluation method Active CN109528196B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811353945.6A CN109528196B (en) 2018-11-14 2018-11-14 Hepatic vein pressure gradient non-invasive evaluation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811353945.6A CN109528196B (en) 2018-11-14 2018-11-14 Hepatic vein pressure gradient non-invasive evaluation method

Publications (2)

Publication Number Publication Date
CN109528196A true CN109528196A (en) 2019-03-29
CN109528196B CN109528196B (en) 2022-07-01

Family

ID=65847272

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811353945.6A Active CN109528196B (en) 2018-11-14 2018-11-14 Hepatic vein pressure gradient non-invasive evaluation method

Country Status (1)

Country Link
CN (1) CN109528196B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110074809A (en) * 2019-06-17 2019-08-02 祁小龙 The hepatic vein pressure gradient classification method and computer equipment of CT image
CN110298383A (en) * 2019-05-28 2019-10-01 中国科学院计算技术研究所 Pathological classification method and system based on multi-modal deep learning
CN110895817A (en) * 2019-11-01 2020-03-20 复旦大学 MRI image hepatic fibrosis automatic grading method based on image omics analysis
CN111598864A (en) * 2020-05-14 2020-08-28 北京工业大学 Hepatocellular carcinoma differentiation assessment method based on multi-modal image contribution fusion
CN112216379A (en) * 2019-07-12 2021-01-12 刘璐 Disease diagnosis system based on intelligent joint learning
CN114678105A (en) * 2022-03-21 2022-06-28 南京圣德医疗科技有限公司 Method for automatically calculating balloon parameters by combining artificial intelligence technology

Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6135966A (en) * 1998-05-01 2000-10-24 Ko; Gary Kam-Yuen Method and apparatus for non-invasive diagnosis of cardiovascular and related disorders
WO2012011872A1 (en) * 2010-07-23 2012-01-26 National Cancer Centre Singapore A method and/or system for determining portal hemodynamics of a subject
US20130096394A1 (en) * 2011-02-04 2013-04-18 Analytics4Life System and method for evaluating an electrophysiological signal
US20140200824A1 (en) * 2008-09-19 2014-07-17 University Of Pittsburgh Of The Commonwealth System Of Higher Education K-partite graph based formalism for characterization of complex phenotypes in clinical data analyses and disease outcome prognosis
US20150112182A1 (en) * 2013-10-17 2015-04-23 Siemens Aktiengesellschaft Method and System for Machine Learning Based Assessment of Fractional Flow Reserve
CN106037710A (en) * 2014-11-24 2016-10-26 西门子公司 Synthetic data-driven hemodynamic determination in medical imaging
US20170027496A1 (en) * 2014-04-08 2017-02-02 Oxford University Innovation Limited Medical imaging
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
US20180078313A1 (en) * 2016-09-20 2018-03-22 Siemens Healthcare Gmbh Liver disease assessment in medical imaging
CN108109140A (en) * 2017-12-18 2018-06-01 复旦大学 Low Grade Gliomas citric dehydrogenase non-destructive prediction method and system based on deep learning
CN108095716A (en) * 2017-11-21 2018-06-01 郑州鼎创智能科技有限公司 A kind of electrocardiograph signal detection method based on confidence rule base and deep neural network
US20180160917A1 (en) * 2015-05-05 2018-06-14 The Johns Hopkins University A device and method for non-invasive left ventricular end diastolic pressure (lvedp) measurement
CN108376558A (en) * 2018-01-24 2018-08-07 复旦大学 A kind of multi-modal nuclear magnetic resonance image Case report no automatic generation method

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6135966A (en) * 1998-05-01 2000-10-24 Ko; Gary Kam-Yuen Method and apparatus for non-invasive diagnosis of cardiovascular and related disorders
US20140200824A1 (en) * 2008-09-19 2014-07-17 University Of Pittsburgh Of The Commonwealth System Of Higher Education K-partite graph based formalism for characterization of complex phenotypes in clinical data analyses and disease outcome prognosis
WO2012011872A1 (en) * 2010-07-23 2012-01-26 National Cancer Centre Singapore A method and/or system for determining portal hemodynamics of a subject
US20130096394A1 (en) * 2011-02-04 2013-04-18 Analytics4Life System and method for evaluating an electrophysiological signal
US20150112182A1 (en) * 2013-10-17 2015-04-23 Siemens Aktiengesellschaft Method and System for Machine Learning Based Assessment of Fractional Flow Reserve
CN106456078A (en) * 2013-10-17 2017-02-22 西门子保健有限责任公司 Method and system for machine learning based assessment of fractional flow reserve
US20170027496A1 (en) * 2014-04-08 2017-02-02 Oxford University Innovation Limited Medical imaging
CN106037710A (en) * 2014-11-24 2016-10-26 西门子公司 Synthetic data-driven hemodynamic determination in medical imaging
US20180160917A1 (en) * 2015-05-05 2018-06-14 The Johns Hopkins University A device and method for non-invasive left ventricular end diastolic pressure (lvedp) measurement
US20180078313A1 (en) * 2016-09-20 2018-03-22 Siemens Healthcare Gmbh Liver disease assessment in medical imaging
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
CN108095716A (en) * 2017-11-21 2018-06-01 郑州鼎创智能科技有限公司 A kind of electrocardiograph signal detection method based on confidence rule base and deep neural network
CN108109140A (en) * 2017-12-18 2018-06-01 复旦大学 Low Grade Gliomas citric dehydrogenase non-destructive prediction method and system based on deep learning
CN108376558A (en) * 2018-01-24 2018-08-07 复旦大学 A kind of multi-modal nuclear magnetic resonance image Case report no automatic generation method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
PROCOPET BOGDAN ET AL: "Serum tests, liver stiffness and artificial neural networks for diagnosing cirrhosis and portal hypertension", 《DIGESTIVE AND LIVER DISEASE》 *
龙莉玲: "计算机辅助软件及影像学在肝纤维化诊断中的应用价值", 《中华医学会第十八次全国放射学学术会议论文汇编》 *

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298383A (en) * 2019-05-28 2019-10-01 中国科学院计算技术研究所 Pathological classification method and system based on multi-modal deep learning
CN110298383B (en) * 2019-05-28 2021-07-13 中国科学院计算技术研究所 Multi-mode deep learning-based pathology classification method and system
CN110074809A (en) * 2019-06-17 2019-08-02 祁小龙 The hepatic vein pressure gradient classification method and computer equipment of CT image
CN110074809B (en) * 2019-06-17 2023-06-20 祁小龙 Hepatic vein pressure gradient classification method of CT image and computer equipment
CN112216379A (en) * 2019-07-12 2021-01-12 刘璐 Disease diagnosis system based on intelligent joint learning
CN110895817A (en) * 2019-11-01 2020-03-20 复旦大学 MRI image hepatic fibrosis automatic grading method based on image omics analysis
CN110895817B (en) * 2019-11-01 2023-06-30 复旦大学 MRI (magnetic resonance imaging) image liver fibrosis automatic grading method based on image histology analysis
CN111598864A (en) * 2020-05-14 2020-08-28 北京工业大学 Hepatocellular carcinoma differentiation assessment method based on multi-modal image contribution fusion
CN114678105A (en) * 2022-03-21 2022-06-28 南京圣德医疗科技有限公司 Method for automatically calculating balloon parameters by combining artificial intelligence technology
CN114678105B (en) * 2022-03-21 2023-10-17 南京圣德医疗科技有限公司 Method for automatically calculating balloon parameters by combining artificial intelligence technology

Also Published As

Publication number Publication date
CN109528196B (en) 2022-07-01

Similar Documents

Publication Publication Date Title
CN109528196A (en) A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics
US10580526B2 (en) System and method for calculating vessel flow parameters based on angiography
US10249048B1 (en) Method and system for predicting blood flow features based on medical images
US10872698B2 (en) Method and system for enhancing medical image-based blood flow computations using physiological measurements
EP3429465B1 (en) Systems and methods for estimating healthy lumen diameter and stenosis quantification in coronary arteries
US9629563B2 (en) Method and system for functional assessment of renal artery stenosis from medical images
US10299862B2 (en) Three-dimensional quantitative heart hemodynamics in medical imaging
KR101910233B1 (en) Systems and methods for numerically evaluating vasculature
US9386933B2 (en) Systems and methods for determination of blood flow characteristics and pathologies through modeling of myocardial blood supply
CN105326486A (en) Method and system for calculating blood vessel pressure difference and fractional flow reserve
Urbina et al. Realistic aortic phantom to study hemodynamics using MRI and cardiac catheterization in normal and aortic coarctation conditions
CN107582041A (en) Tubular structure resolver and tubular structure analytic method
CN108665449B (en) Deep learning model and device for predicting blood flow characteristics on blood flow vector path
CN109843161A (en) For determining the device of the function index for stenosis assessment
CN111317455B (en) Method, device and equipment for determining hemodynamic parameters and storage medium
CN111067494A (en) Microcirculation resistance rapid calculation method based on blood flow reserve fraction and blood flow resistance model
Reddy et al. Video-based deep learning for automated assessment of left ventricular ejection fraction in pediatric patients
Horvat et al. Aging influences pulmonary artery flow and stiffness in healthy individuals: non-invasive assessment using cardiac MRI
CN105120743A (en) Determination of physiological parameters of tissue from dynamic contrast-enhanced mr data
Babin et al. Robust segmentation methods with an application to aortic pulse wave velocity calculation
CN116486211A (en) Model training method, fractional flow reserve calculation method, device and equipment
Wong et al. Principal curves for lumen center extraction and flow channel width estimation in 3-D arterial networks: Theory, algorithm, and validation
Hentschke et al. Estimating blood flow velocity in angiographic image data
Paun et al. Inference in cardiovascular modelling subject to medical interventions
CN109411085A (en) The method and device and computer installation of perfusion analysis

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