CN110120051A - A kind of right ventricle automatic division method based on deep learning - Google Patents

A kind of right ventricle automatic division method based on deep learning Download PDF

Info

Publication number
CN110120051A
CN110120051A CN201910388790.8A CN201910388790A CN110120051A CN 110120051 A CN110120051 A CN 110120051A CN 201910388790 A CN201910388790 A CN 201910388790A CN 110120051 A CN110120051 A CN 110120051A
Authority
CN
China
Prior art keywords
right ventricle
deep learning
network model
layer
method based
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201910388790.8A
Other languages
Chinese (zh)
Inventor
刘鹏
王丽嘉
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.)
University of Shanghai for Science and Technology
Original Assignee
University of Shanghai for Science and 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 University of Shanghai for Science and Technology filed Critical University of Shanghai for Science and Technology
Priority to CN201910388790.8A priority Critical patent/CN110120051A/en
Publication of CN110120051A publication Critical patent/CN110120051A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20048—Transform domain processing
    • G06T2207/20061—Hough transform
    • 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
    • G06T2207/20076—Probabilistic image processing
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/20212—Image combination
    • G06T2207/20224—Image subtraction
    • 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
    • G06T2207/30048—Heart; Cardiac

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Geometry (AREA)
  • Probability & Statistics with Applications (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)
  • Image Processing (AREA)

Abstract

The present invention provides a kind of right ventricle automatic division method based on deep learning, which comprises the following steps: all cardiac magnetic resonance film short axis images of acquisition are pre-processed；Extract region of interest ROI；EDS extended data set；U-shaped network model is constructed in the library keras of deep learning, and is trained；Prediction of result is carried out using trained U-shaped network model.The present invention provides a kind of right ventricle full-automatic partition method based on deep learning, influence of the surrounding tissue to segmentation result is reduced by automatic identification ventricular area first, original U-shaped network is improved, to realize the accurate full-automatic dividing to right ventricle, provide the foundation for the further functional analysis and medical diagnosis on disease of heart.

Description

A kind of right ventricle automatic division method based on deep learning
Technical field
The present invention relates to field of medical image processing, specially a kind of right ventricle based on deep learning side of segmentation automatically Method.
Background technique
With being constantly progressive for Medical Image Processing, the segmentation of medical image segmentation, especially vitals is meter The basis of calculation machine auxiliary diagnosis and treatment.Magnetic resonance imaging (Magnetic Resonance Imaging) is a kind of extensive use In the standard technique of medical diagnosis.In cardiac magnetic resonance film image, the accurate segmentation of right ventricle can help people effective Calculating ventricle end-systole and the parameters such as end-diastolic volume, stroke output and ejection fraction, to be carried out to heart Further functional analysis and medical diagnosis on disease.
Left ventricular shape rule is surrounded by thicker cardiac muscle, and segmentation task is relatively easy, is studied also relatively deeply.Than Left ventricle is played, right ventricle shape is in irregular semilune and myocardial wall is thin, adjoins with fat, makes accurate point of right ventricle It is cut into for a difficult point.Common right ventricle dividing method has active contour model, level set, figure to cut, cluster and multichannel chromatogram Segmentation etc., these methods show good effect on limited data set, but the often data except training data It performs poor in library, needs manual intervention, cause sliced time to increase and the subjective impact being subject to is big.
Summary of the invention
The purpose of the present invention is: the segmentation precision of right ventricle is improved, the cost manually participated in is reduced.
In order to achieve the above object, the technical solution of the present invention is to provide a kind of right ventricle based on deep learning is automatic Dividing method, which comprises the following steps:
Step 1, the pretreatment that all cardiac magnetic resonance film short axis images of acquisition are carried out with gray scale normalization, by pixel Primary system one arrives [0,255] section；
Step 2 extracts region of interest ROI: seeking pretreated cardiac magnetic resonance film short axis images every two in step 1 The sum of the absolute difference image of a continuous phase, determines left ventricle center using hough-circle transform on error image, The part for containing Ventricular is intercepted simultaneously as region of interest ROI image according to the positional relationship of Ventricular；
The data set of the image data of all region of interest ROI images obtained in step 3, expansion step 2, will own Image data is adjusted to onesize and the data set after expansion is divided into training set, verifying collection and test set three parts；
Step 4 constructs U-shaped network model in the library keras of deep learning；
Step 5, the input by training set obtained in step 3 and verifying collection as U-shaped network model, to the U built Type network model is trained and parameter adjustment, obtains trained U-shaped network model；
Step 6 predicts test set using U-shaped network model trained in step 5, carries out to the result of prediction Post-processing, i.e., carry out binary conversion treatment to the probability graph that U-shaped neural network forecast obtains, and the binary map divided is tied by segmentation Fruit calculates the different physical signs of right ventricle, and carries out analysis and assessment to result, wherein carrying out analysis and assessment to result includes essence Exactness, correlation, consistency analysis, the evaluation index of accuracy be Dice coefficient (Dice Metric, DM) and Hao Siduofu away from From (HausdorffDistance, HD), Dice coefficient is used to measure the similarity of automatic segmentation result and goldstandard, using public affairs Formula:
In formula, DM (A, B) indicates that the Dice coefficient of A and B, A are the resulting size of automatic segmentation result, and B is expert The size of the goldstandard marked manually；
Hausdorff distance is used to measure the asymmetric difference of maximum of two images, using formula:
HD (A, B)=max (maxa∈A(minb∈BD (a, b)), maxb∈B(mina∈AD (a, b)))
In formula, HD (A, B) indicates that the Hausdorff distance of A and B, A indicate that the profile divided automatically, B indicate that expert is manual The profile of the goldstandard of mark, a are the point in profile A, and b is the point in profile B, d (a, b) be a point and b point Euclid away from From.
Preferably, in step 3, region of interest ROI image obtained in step 2 is translated, rotated, overturn, is drawn It stretches and achievees the purpose that EDS extended data set.
Preferably, in step 4, the U-shaped network model includes: the convolutional layer that convolution kernel is 3 × 3, batch normalized Layer, the maximum pond layer that LeakyRelu activation primitive layer, core size are 2 × 2, extension convolutional layer, up-sampling layer, warp lamination And softmax activation primitive layer.
Preferably, in step 6, the physical signs include end-diastolic volume (End Diastolic Volume, EDV), end-systolic volume (End Systolic Volume, ESV), stroke output (StrokeVolume, SV), penetrate blood system Number (EjectionFraction, EF).
Compared with prior art, the beneficial effects of the present invention are:
The present invention provides a kind of right ventricle full-automatic partition method based on deep learning passes through the automatic identification heart first Chamber region reduces influence of the surrounding tissue to segmentation result, improves to original U-shaped network, to realize to right ventricle Accurate full-automatic dividing provides the foundation for the further functional analysis and medical diagnosis on disease of heart.
Detailed description of the invention
Fig. 1 is the flow diagram of the method for the present invention；
Fig. 2 is the schematic network structure of the method for the present invention；
Fig. 3 is cardiac short axis image diastasis bottom segmentation result；
Fig. 4 is segmentation result in the middle part of cardiac short axis image diastasis；
Fig. 5 is segmentation result at the top of cardiac short axis image diastasis；
Fig. 6 is cardiac short axis image end-systole bottom segmentation result；
Fig. 7 is segmentation result in the middle part of cardiac short axis image end-systole；
Fig. 8 is to divide at the top of cardiac short axis image end-systole as a result, Fig. 3 into Fig. 8, is followed successively by interested from left to right Region, segmentation result of the present invention and expert divide goldstandard.
Specific embodiment
With reference to the accompanying drawing, the present invention is further explained.It should be understood that these embodiments are merely to illustrate the present invention and do not have to In limiting the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, those skilled in the art can be with The present invention is made various changes or modifications, such equivalent forms equally fall within model defined by the application the appended claims It encloses.
A kind of right ventricle automatic division method based on deep learning disclosed in the present embodiment including the following steps:
(1) data prediction
The present invention carries out retrospective analysis (1.5T GE magnetic resonance imaging system) to heart film magnetic resonance image, this reality Apply in example 844 width MRI image datas altogether including 61 patients, wherein male 22, women 39, the age covers 23~ 93 years old, specific imaging parameters are as follows: image size 256 × 256, thickness 6-8mm, interlamellar spacing 2-4mm, every number of cases evidence include 6-10 Layer, every layer of 20-28 phase.Every number of cases is felt according to according to the absolute difference image of end-systole and the image of diastasis The extraction in interest region, area-of-interest includes left and right ventricles, and area-of-interest is uniformly carried out gray scale normalization and is located in advance Reason, makes image grayscale size value in [0,255] section, handles for follow-up data.41 number of cases are schemed according to 569 width MRI are amounted to Parameters as being used to train and adjust network, wherein 80% image is as training set, 20% image is as verifying collection； 20 number of cases are split according to 275 width MRI images are amounted to as test set.For lack of training samples problem, the present invention is using flat It moves, rotation, overturning, the method exptended sample amount stretched.
(2) network establishment and training
The network architecture of the present embodiment uses python language and keras deep learning library, uses NVIDIAGeForceGTX 1050Ti is accelerated.
Improved U-shaped network model is built under the library keras of deep learning, will be adjusted to same 120 × 120 size Training set and verifying collection input network are trained and adjust with parameter, and the output of network is that each pixel belongs to the general of right ventricle Rate.
Network architecture includes: the convolutional layer that convolution kernel is 3 × 3, and batch normalized layer, LeakyRelu activate letter Several layers, the maximum pond layer that core size is 2 × 2, extension convolutional layer, up-sampling layer, warp lamination and softmax activation primitive Layer, network structure is as shown in Fig. 2, its parameter setting is as shown in table 1.Down-sampling is reduced on the basis of original U-shaped network Number expands receptive field using convolution is extended, and batch normalized is added, changes activation primitive and loss function raising precision.
Table 1
The parameter of convolutional layer (Convolutional layer) is optimized by back-propagation algorithm.Convolution The purpose of operation is to extract the different characteristic of input, first layer convolutional layer may can only extract some rudimentary feature such as edges, Lines, the network of more layers can from low-level features the more complicated feature of iterative extraction.The output size of each layer of convolutional layer with The relationship of input is indicated using following formula:
In formula, heightinput、heightoutput、heightkernelIt respectively inputs, the length of output and convolution kernel； widthinput、widthoutput、widthkernelIt respectively inputs, the width of output and convolution kernel；Padding is 0 filling size, Stride is step-length, and in the present embodiment, padding=1, stride=1 guarantee that each layer of output size is identical as input.
Normalized layer is criticized for handling the problem of data distribution of middle layer in training process changes.This implementation Example passes through preprocessing process for the data normalization of input layer, but each layer of input data distribution of network is to become always Change, the update of front layer training parameter will lead to the variation of back layer input data distribution, therefore will necessarily cause every below The change of one layer of input data distribution, reducing data distribution difference bring using batch normalized influences.
Pond layer is used to reduce the output parameter of network, selects validity feature output, uses convolution kernel for 2 in the present embodiment × 2 maximum pond layer takes the maximum value in four values for including in window as output, residual value is ignored.Activation primitive Using LeakyRelu, a non-zero slope is assigned to all negative values on the basis of correcting linear unit R elu, in the present embodiment Slope is set as 10.
Since data set picture size is smaller, therefore the down-sampling process of U-shaped network is reduced to 3 times, can effectively be extracted Characteristics of image.According to can only there is the fact that a heart in piece image, influence of the surrounding tissue to feature extraction, benefit are reduced Convolutional layer is replaced with extension convolutional layer, neural network can be reduced to similar features with the receptive field for increasing each layer of index times Error extraction.Picture can be restored to size identical with input picture with deconvolution using up-sampling.
The loss function L (w) of the present embodiment uses Dice coefficient, is defined as:
In formula, ynIndicate n-th of sample training label, for right ventricle bianry image, i.e., 0 represents background, and 1 represents the right side Ventricle；Indicate the prediction result of network, the process that training process, that is, undated parameter of network minimizes L (w).
(3) test data is split
The diastasis of each of test set patient and the image of end-systole are carried out using trained network Prediction, the output result of neural network forecast are the probability that each pixel belongs to right ventricle, probability think greater than 0.5 be Right ventricle carries out binary conversion treatment to probabilistic image by setting threshold value and obtains segmentation result to the end.Obtained by the present embodiment most Whole segmentation result is shown in Fig. 5.

Claims (4)

1. a kind of right ventricle automatic division method based on deep learning, which comprises the following steps:
Step 1, the pretreatment that all cardiac magnetic resonance film short axis images of acquisition are carried out with gray scale normalization, by pixel primary system One arrives [0,255] section；
Step 2 extracts region of interest ROI: pretreated cardiac magnetic resonance film short axis images every two in step 1 being asked to connect The sum of the absolute difference image of continuous phase, determines left ventricle center using hough-circle transform on error image, according to The positional relationship of Ventricular intercepts the part for containing Ventricular as region of interest ROI image simultaneously；
The data set of the image data of all region of interest ROI images obtained in step 3, expansion step 2, by all images Data point reuse is onesize and the data set after expansion is divided into training set, verifying collection and test set three parts；
Step 4 constructs U-shaped network model in the library keras of deep learning；
Step 5, the input by training set obtained in step 3 and verifying collection as U-shaped network model, to the U-shaped net built Network model is trained and parameter adjustment, obtains trained U-shaped network model；
Step 6 predicts test set using U-shaped network model trained in step 5, after carrying out to the result of prediction Reason carries out binary conversion treatment to the probability graph that U-shaped neural network forecast obtains, the binary map divided passes through segmentation result meter The different physical signs of right ventricle are calculated, and analysis and assessment are carried out to result, wherein it includes accurate for carrying out analysis and assessment to result Degree, correlation, consistency analysis, the evaluation index of accuracy are Dice coefficient and Hausdorff distance, and Dice coefficient is for spending The similarity for measuring automatic segmentation result and goldstandard, using formula:
In formula, DM (A, B) indicates that the Dice coefficient of A and B, A are the resulting size of automatic segmentation result, and B is that expert is manual The size of the goldstandard of mark；
Hausdorff distance is used to measure the asymmetric difference of maximum of two images, using formula:
HD (A, B)=max (maxa∈A(minb∈BD (a, b)), maxb∈B(mina∈AD (a, b)))
In formula, HD (A, B) indicates that the Hausdorff distance of A and B, A indicate that the profile divided automatically, B indicate that expert marks manually Goldstandard profile, a be profile A in point, b be profile B in point, d (a, b) be a point and b point Euclidean distance.
2. a kind of right ventricle automatic division method based on deep learning as described in claim 1, which is characterized in that step 3 In, region of interest ROI image obtained in step 2 is translated, is rotated, is overturn, stretches the mesh for reaching EDS extended data set 's.
3. a kind of right ventricle automatic division method based on deep learning as described in claim 1, which is characterized in that step 4 In, the U-shaped network model includes: the convolutional layer that convolution kernel is 3 × 3, batch normalized layer, LeakyRelu activation primitive Layer, the maximum pond layer that core size is 2 × 2, extension convolutional layer, up-sampling layer, warp lamination and softmax activation primitive Layer.
4. a kind of right ventricle automatic division method based on deep learning as described in claim 1, which is characterized in that step 6 In, the physical signs includes end-diastolic volume, end-systolic volume, stroke output, ejection fraction.
CN201910388790.8A 2019-05-10 2019-05-10 A kind of right ventricle automatic division method based on deep learning Pending CN110120051A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910388790.8A CN110120051A (en) 2019-05-10 2019-05-10 A kind of right ventricle automatic division method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910388790.8A CN110120051A (en) 2019-05-10 2019-05-10 A kind of right ventricle automatic division method based on deep learning

Publications (1)

Publication Number Publication Date
CN110120051A true CN110120051A (en) 2019-08-13

Family

ID=67522008

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910388790.8A Pending CN110120051A (en) 2019-05-10 2019-05-10 A kind of right ventricle automatic division method based on deep learning

Country Status (1)

Country Link
CN (1) CN110120051A (en)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110599499A (en) * 2019-08-22 2019-12-20 四川大学 MRI image heart structure segmentation method based on multipath convolutional neural network
CN110991408A (en) * 2019-12-19 2020-04-10 北京航空航天大学 Method and device for segmenting white matter high signal based on deep learning method
CN111105424A (en) * 2019-12-19 2020-05-05 广州柏视医疗科技有限公司 Lymph node automatic delineation method and device
CN111415361A (en) * 2020-03-31 2020-07-14 浙江大学 Method and device for estimating brain age of fetus and detecting abnormality based on deep learning
CN111466894A (en) * 2020-04-07 2020-07-31 上海尽星生物科技有限责任公司 Ejection fraction calculation method and system based on deep learning
CN111583271A (en) * 2020-04-13 2020-08-25 华东师范大学 Method for automatically predicting gene expression categories based on cancer CT images
CN111598838A (en) * 2020-04-22 2020-08-28 中南民族大学 Automatic heart MR image segmentation method and device, electronic equipment and storage medium
CN111784696A (en) * 2020-06-28 2020-10-16 深圳大学 Method and system for training right ventricle segmentation model and right ventricle segmentation
CN111784732A (en) * 2020-06-28 2020-10-16 深圳大学 Method and system for training cardiac motion field estimation model and cardiac motion field estimation
CN113159040A (en) * 2021-03-11 2021-07-23 福建自贸试验区厦门片区Manteia数据科技有限公司 Method, device and system for generating medical image segmentation model
CN113240659A (en) * 2021-05-26 2021-08-10 广州天鹏计算机科技有限公司 Image feature extraction method based on deep learning
CN113838068A (en) * 2021-09-27 2021-12-24 深圳科亚医疗科技有限公司 Method, apparatus and storage medium for automatic segmentation of myocardial segments
WO2022161192A1 (en) * 2021-02-01 2022-08-04 之江实验室 Method for automatically segmenting left ventricle of spect three-dimensional reconstruction image
WO2023193290A1 (en) * 2022-04-08 2023-10-12 胡冠彤 Medical imaging system and method for in-vitro heart simulator

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107578416A (en) * 2017-09-11 2018-01-12 武汉大学 It is a kind of by slightly to heart left ventricle's full-automatic partition method of smart cascade deep network
CN108564590A (en) * 2018-04-20 2018-09-21 上海理工大学 A kind of right ventricle multichannel chromatogram dividing method based on cardiac magnetic resonance film short axis images
CN109191476A (en) * 2018-09-10 2019-01-11 重庆邮电大学 The automatic segmentation of Biomedical Image based on U-net network structure
CN109584254A (en) * 2019-01-07 2019-04-05 浙江大学 A kind of heart left ventricle's dividing method based on the full convolutional neural networks of deep layer

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107578416A (en) * 2017-09-11 2018-01-12 武汉大学 It is a kind of by slightly to heart left ventricle's full-automatic partition method of smart cascade deep network
CN108564590A (en) * 2018-04-20 2018-09-21 上海理工大学 A kind of right ventricle multichannel chromatogram dividing method based on cardiac magnetic resonance film short axis images
CN109191476A (en) * 2018-09-10 2019-01-11 重庆邮电大学 The automatic segmentation of Biomedical Image based on U-net network structure
CN109584254A (en) * 2019-01-07 2019-04-05 浙江大学 A kind of heart left ventricle's dividing method based on the full convolutional neural networks of deep layer

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
苏新宇 等: "基于心脏磁共振短轴电影图像的右心室分割新进展", 《波谱学杂志》 *

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110599499B (en) * 2019-08-22 2022-04-19 四川大学 MRI image heart structure segmentation method based on multipath convolutional neural network
CN110599499A (en) * 2019-08-22 2019-12-20 四川大学 MRI image heart structure segmentation method based on multipath convolutional neural network
CN110991408A (en) * 2019-12-19 2020-04-10 北京航空航天大学 Method and device for segmenting white matter high signal based on deep learning method
CN111105424A (en) * 2019-12-19 2020-05-05 广州柏视医疗科技有限公司 Lymph node automatic delineation method and device
CN110991408B (en) * 2019-12-19 2022-09-06 北京航空航天大学 Method and device for segmenting white matter high signal based on deep learning method
CN111415361A (en) * 2020-03-31 2020-07-14 浙江大学 Method and device for estimating brain age of fetus and detecting abnormality based on deep learning
CN111466894A (en) * 2020-04-07 2020-07-31 上海尽星生物科技有限责任公司 Ejection fraction calculation method and system based on deep learning
CN111583271A (en) * 2020-04-13 2020-08-25 华东师范大学 Method for automatically predicting gene expression categories based on cancer CT images
CN111598838B (en) * 2020-04-22 2023-04-07 中南民族大学 Automatic heart MR image segmentation method and device, electronic equipment and storage medium
CN111598838A (en) * 2020-04-22 2020-08-28 中南民族大学 Automatic heart MR image segmentation method and device, electronic equipment and storage medium
CN111784732A (en) * 2020-06-28 2020-10-16 深圳大学 Method and system for training cardiac motion field estimation model and cardiac motion field estimation
CN111784696A (en) * 2020-06-28 2020-10-16 深圳大学 Method and system for training right ventricle segmentation model and right ventricle segmentation
CN111784732B (en) * 2020-06-28 2023-07-28 深圳大学 Method and system for training heart motion field estimation model and heart motion field estimation
CN111784696B (en) * 2020-06-28 2023-08-01 深圳大学 Right ventricle segmentation model training method and system
WO2022161192A1 (en) * 2021-02-01 2022-08-04 之江实验室 Method for automatically segmenting left ventricle of spect three-dimensional reconstruction image
CN113159040A (en) * 2021-03-11 2021-07-23 福建自贸试验区厦门片区Manteia数据科技有限公司 Method, device and system for generating medical image segmentation model
CN113159040B (en) * 2021-03-11 2024-01-23 福建自贸试验区厦门片区Manteia数据科技有限公司 Method, device and system for generating medical image segmentation model
CN113240659A (en) * 2021-05-26 2021-08-10 广州天鹏计算机科技有限公司 Image feature extraction method based on deep learning
CN113838068A (en) * 2021-09-27 2021-12-24 深圳科亚医疗科技有限公司 Method, apparatus and storage medium for automatic segmentation of myocardial segments
WO2023193290A1 (en) * 2022-04-08 2023-10-12 胡冠彤 Medical imaging system and method for in-vitro heart simulator

Similar Documents

Publication Publication Date Title
CN110120051A (en) A kind of right ventricle automatic division method based on deep learning
CN107369160B (en) Choroid neogenesis blood vessel segmentation algorithm in OCT image
CN108830155B (en) Heart coronary artery segmentation and identification method based on deep learning
CN109146872B (en) Heart coronary artery image segmentation and identification method based on deep learning and optical flow method
CN110930416B (en) MRI image prostate segmentation method based on U-shaped network
CN107578416B (en) Full-automatic heart left ventricle segmentation method for coarse-to-fine cascade deep network
CN109389584A (en) Multiple dimensioned rhinopharyngeal neoplasm dividing method based on CNN
CN109584254A (en) A kind of heart left ventricle's dividing method based on the full convolutional neural networks of deep layer
CN108257135A (en) The assistant diagnosis system of medical image features is understood based on deep learning method
CN111105424A (en) Lymph node automatic delineation method and device
CN108664976B (en) Super-pixel-based fuzzy spectral clustering brain tumor image automatic segmentation method
Liu et al. Automatic whole heart segmentation using a two-stage u-net framework and an adaptive threshold window
CN109242860A (en) Based on the brain tumor image partition method that deep learning and weight space are integrated
CN107146228A (en) A kind of super voxel generation method of brain magnetic resonance image based on priori
CN108805923B (en) Physiological parameter acquisition method and device
CN109102512B (en) DBN neural network-based MRI brain tumor image segmentation method
CN112336381B (en) Echocardiogram end systole/diastole frame automatic identification method based on deep learning
CN112508902B (en) White matter high signal grading method, electronic equipment and storage medium
CN110853048A (en) MRI image segmentation method, device and storage medium based on rough training and fine training
CN113643353B (en) Measurement method for enhancing resolution of vascular caliber of fundus image
CN113902738A (en) Heart MRI segmentation method and system
CN112750131A (en) Pelvis nuclear magnetic resonance image musculoskeletal segmentation method based on scale and sequence relation
CN112950555A (en) Deep learning-based type 2 diabetes cardiovascular disease image classification method
CN112950611A (en) Liver blood vessel segmentation method based on CT image
CN115471512A (en) Medical image segmentation method based on self-supervision contrast learning

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20190813