WO2021051555A1 - 基于图像识别的左心室测量方法、装置以及计算机设备 - Google Patents

基于图像识别的左心室测量方法、装置以及计算机设备 Download PDF

Info

Publication number
WO2021051555A1
WO2021051555A1 PCT/CN2019/117552 CN2019117552W WO2021051555A1 WO 2021051555 A1 WO2021051555 A1 WO 2021051555A1 CN 2019117552 W CN2019117552 W CN 2019117552W WO 2021051555 A1 WO2021051555 A1 WO 2021051555A1
Authority
WO
WIPO (PCT)
Prior art keywords
layer
sampling
image
left ventricle
data
Prior art date
Application number
PCT/CN2019/117552
Other languages
English (en)
French (fr)
Inventor
刘莉红
王健宗
Original Assignee
平安科技（深圳）有限公司
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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021051555A1 publication Critical patent/WO2021051555A1/zh

Links

Images

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Definitions

  • This application relates to the technical field of left ventricular measurement based on image recognition, and in particular to a method, device, computer equipment, and non-volatile computer-readable storage medium for left ventricle measurement based on image recognition.
  • Heart-related diseases are one of the diseases with the highest mortality rate in today's society. Early prevention and timely detection are of great significance to heart-related diseases, and left ventricular measurement is very important for the diagnosis of heart-related diseases.
  • MRI Magnetic Resonance Imaging
  • MRI Magnetic Resonance Imaging
  • this application proposes a left ventricular measurement method, device, computer equipment, and non-volatile computer-readable storage medium based on image recognition, which can receive the left ventricle image to be recognized and then input it into a pre-established left ventricle image.
  • the measurement is performed using a ventricular measurement model.
  • the left ventricle measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification layer, which are sequentially connected; then, the M directions of the left ventricle wall and N values extracted from the circulation layer are obtained.
  • the case data table is set to query the measurement result of the left ventricular image.
  • the present application provides a method for measuring the left ventricle based on image recognition, the method including:
  • the present application also provides a left ventricular measurement device based on image recognition, the device including:
  • the receiving module is used to receive the left ventricle image to be recognized; the input module is used to input the left ventricle image to a pre-established left ventricular measurement model, wherein the left ventricular measurement model includes image segmentation layers connected in sequence, Convolutional layer, circulation layer, and classification layer; an acquisition module for acquiring M-directional wall, N-directional diameter and inner/outer cavity area data of the left ventricle extracted by the circulation layer, and the classification
  • a preset case data table queries the measurement result of the left ventricle image, wherein the case data query table includes the correspondence relationship between the L item data and the measurement result.
  • this application also proposes a computer device, the computer device includes a memory and a processor, the memory stores computer-readable instructions that can run on the processor, and the computer-readable instructions are The implementation steps when the processor is executed:
  • the present application also provides a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be At least one processor executes, so that the at least one processor executes the steps:
  • the image recognition-based left ventricular measurement method, device, computer equipment, and non-volatile computer-readable storage medium proposed in this application can receive the left ventricular image to be recognized and then input it into a pre-established left ventricular measurement model for evaluation.
  • the left ventricle measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification layer that are sequentially connected; then, the left ventricle's M azimuth wall and N azimuth diameters extracted by the circulatory layer are obtained And internal/external cavity area data, as well as the cardiac phase data extracted by the classification layer; and then according to M azimuth wall, N azimuth diameter, internal/external cavity area and cardiac phase data and preset case data
  • the table queries the measurement results of the left ventricle image. Through the above method, more accurate left ventricular measurement can be achieved based on image recognition.
  • Fig. 1 is a schematic diagram of an optional hardware architecture of the computer equipment of the present application
  • FIG. 2 is a schematic diagram of program modules of an embodiment of a left ventricular measurement device based on image recognition according to the present application;
  • FIG. 3 is a schematic flowchart of an embodiment of a method for measuring a left ventricle based on image recognition according to the present application.
  • FIG. 1 is a schematic diagram of an optional hardware architecture of the computer device 1 of the present application.
  • the computer device 1 may include, but is not limited to, a memory 11, a processor 12, and a network interface 13 that can communicate with each other through a system bus.
  • the computer device 1 is connected to the network through the network interface 13 (not shown in FIG. 1), and connected to other terminal devices such as mobile terminal (Mobile Terminal), user equipment (UE), and portable equipment (portable equipment) through the network, PC side and so on.
  • the network may be Intranet, Internet, Global System of Mobile Communication (GSM), Wideband Code Division Multiple Access (WCDMA), 4G network, 5G Network, Bluetooth (Bluetooth), Wi-Fi, call network and other wireless or wired networks.
  • FIG. 1 only shows the computer device 1 with components 11-13, but it should be understood that it is not required to implement all the illustrated components, and more or fewer components may be implemented instead.
  • the memory 11 includes at least one type of non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX). Memory, etc.), random access memory (RAM), static random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory , Disks, CDs, etc.
  • the memory 11 may be an internal storage unit of the computer device 1, for example, a hard disk or a memory of the computer device 1.
  • the memory 11 may also be an external storage device of the computer device 1, such as a plug-in hard disk, a smart media card (SMC), and a secure digital ( Secure Digital, SD card, Flash Card, etc.
  • the memory 11 may also include both an internal storage unit of the computer device 1 and an external storage device thereof.
  • the memory 11 is generally used to store the operating system and various application software installed in the computer device 1, such as the program code of the left ventricular measurement device 200 based on image recognition.
  • the memory 11 can also be used to temporarily store various types of data that have been output or will be output.
  • the memory 11 stores computer readable instructions, and the computer readable instructions can be executed by at least one processor, so that the at least one processor executes the steps:
  • the processor 12 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips.
  • the processor 12 is generally used to control the overall operation of the computer device 1, such as performing data interaction or communication-related control and processing.
  • the processor 12 is used to run the program code or process data stored in the memory 11, for example, to run the left ventricular measurement device 200 based on image recognition.
  • the network interface 13 may include a wireless network interface or a wired network interface.
  • the network interface 13 is usually used to establish a communication connection between the computer device 1 and other terminal devices such as mobile terminals, user equipment and portable devices, PC terminals, etc. .
  • the left ventricular measurement device 200 based on image recognition when the left ventricle measurement device 200 based on image recognition is installed and running in the computer device 1, when the left ventricle measurement device 200 based on image recognition is running, it can receive the left ventricle image to be recognized. Input it into a pre-established left ventricular measurement model for measurement, the left ventricular measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification layer that are sequentially connected; then the left ventricle extracted from the circulation layer is obtained.
  • the cavity area and the cardiac phase data and the preset case data table query the measurement result of the left ventricle image.
  • this application proposes a left ventricular measurement device 200 based on image recognition.
  • FIG. 2 is a program module diagram of an embodiment of a left ventricular measurement device 200 based on image recognition in the present application.
  • the image recognition-based left ventricular measurement device 200 includes a series of computer-readable instructions stored on the memory 11, and when the computer-readable instructions are executed by the processor 12, various implementations of the present application can be implemented.
  • the left ventricular measurement device 200 based on image recognition may be divided into one or more modules based on specific operations implemented by various parts of the computer-readable instructions.
  • the left ventricular measurement device 200 based on image recognition can be divided into a receiving module 201, an input module 202, an acquiring module 203 and a query module 204. among them:
  • the receiving module 201 is used to receive the left ventricle image to be recognized.
  • the computer device 1 is connected to a user terminal, such as a mobile terminal, a PC terminal, and other devices, and then the receiving module 201 can receive the user's left ventricle image to be recognized through the user terminal.
  • the computer device 1 may also directly provide a data interface to receive the user's left ventricular image to be identified, or connect to a database system, and then receive the left ventricular image to be identified sent from the database system.
  • the left ventricular image to be recognized may be MRI, or other types of images such as CT (Computed Tomography, computer tomography) images.
  • the input module 202 is configured to input the left ventricle image into a pre-established left ventricular measurement model for measurement, wherein the left ventricle measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification that are sequentially connected Floor.
  • the left ventricle measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification that are sequentially connected Floor.
  • the input module 202 further inputs the left ventricle image into a preset left ventricular measurement model to perform left ventricular measurement, wherein the left ventricle
  • the ventricular measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification layer that are sequentially connected.
  • the left ventricle image is down-sampled and up-sampled multiple times by the image segmentation layer in the left ventricle measurement model to achieve image standardization, wherein the image segmentation layer includes sequential connections The down-sampling layer and the up-sampling layer, the down-sampling layer includes a plurality of down-sampling layers of at least a first down-sampling layer and a second down-sampling layer, and the up-sampling layer includes at least a first up-sampling layer And multiple upsampling layers of the second upsampling layer.
  • the process of performing the multiple down-sampling and up-sampling by the image segmentation layer includes: inputting the left ventricle image into a first down-sampling layer and a second down-sampling layer, respectively; and performing the first down-sampling
  • the output data of the layer and the output data of the first up-sampling layer are combined as input data of the second up-sampling layer; then the output data of the second down-sampling layer and the first layer are combined.
  • the output data of the second down-sampling layer is merged into the convolutional layer as input data.
  • the image segmentation layer adopts a CNN neural network architecture, including a down-sampling layer and an up-sampling layer, the down-sampling layer is composed of four down-sampling layers, and the up-sampling layer is composed of four up-sampling layers;
  • the up-sampling layer is connected to the subsequent convolutional layer through a Dense block.
  • each down-sampling layer in the down-sampling layer includes a Dense block, an extended block, and a down-sampling block.
  • the Dense block is a basic module of a densenet block
  • the extended block is a Dense block.
  • the expansion rate in the block for example, the output of the convolution block of (1, 3, 5) is combined as the input of the down-sampling layer.
  • the expansion rate is a parameter of the expansion convolution and is used to indicate the size of the expansion of the convolution kernel.
  • the down-sampling layer adopts maximum pooling, for example, a 2 ⁇ 2 convolution kernel is selected; each up-sampling layer of the up-sampling layer includes a dense block and an up-sampling block, where the first up-sampling layer
  • the output of and the output of the Dense block in the fourth downsampling group are combined as the input of the second upsampling layer, and the output of the second upsampling layer and the output of the Dense block in the third downsampling layer are combined as the third upsampling layer
  • the input of the third up-sampling layer and the output of the Dense block in the second down-sampling layer are combined as the input of the fourth up-sampling layer, and the output of the Dense block in the fourth up-sampling layer and the first down-sampling layer
  • the merge is used as the input of the last Dense block, and the output of the last Dense block is used as the downstream node, that is, the input of the
  • the down-sampling layer of the image segmentation layer includes an extension block
  • the extension block includes a convolution kernel
  • the image segmentation layer passes through the volume of the extension block of the down-sampling layer.
  • the product kernel performs convolution processing on the input left ventricular image, thereby realizing a down-sampling process, wherein the convolution kernel performs convolution by sampling a hole convolution method.
  • the hole convolution can increase the receptive field without pooling.
  • the expansion block merges and connects the output of the convolution block with the expansion rate (1, 3, 5) as the input of the convolution in the down-sampling block, and up-sampling Path fusion downsampling the output of the path part.
  • the features of different dimensions in the left ventricular image are fused together, and the multi-dimensional features of the left ventricular image are repeatedly and fully utilized.
  • the convolutional layer further extracts task features, and uses the extracted feature maps as the input of the third layer of neural network.
  • the convolutional layer is composed of an input layer, a plurality of hidden layers, and an output layer.
  • the hidden layers include a convolutional layer, a nonlinear layer, and a pool. ⁇ The layer and so on. Further feature extraction of the feature value output by the image segmentation layer by the convolutional layer can improve the convergence of the feature value and ensure accuracy.
  • the recurrent layer further performs feature recognition and judgment on the multi-dimensional features extracted by the convolutional layer.
  • the recurrent layer includes a first sub-network and a second sub-network, and the recurrent layer fully connects the first sub-network to each neuron to determine the characteristic value of the left ventricle image. Regression; by fully connecting the second sub-network to each neuron to classify the feature value of the left ventricular image, the first sub-network and the second sub-network both use LSTM (Long short-term memory) neural network model structure.
  • LSTM Long short-term memory
  • the recurrent layer includes specifically two standard LSTM neural networks.
  • the standard LSTM neural network includes: an input layer, an LSTM cell layer, and an output layer.
  • the standard LSTM neural network can store and Use the input and output information of the system in the past. Therefore, the output of the first sub-network is fully connected with N neurons to output N feature maps for regression, and the results of various tasks are obtained after regression.
  • Each task includes M azimuth walls and N The diameter of the azimuth, the area of the inner cavity/the area of the outer cavity, etc.; the output of the second self-network is connected with neurons for classification tasks, and the classification tasks include cardiac phases and the like.
  • the cycle layer can be used to solve the sequence and time problems in the cardiac phase determination process.
  • the convolutional layer and the recurrent layer cross-use feature data by sharing a hidden layer. Therefore, the accuracy of the multi-dimensional feature extraction of the left ventricle image is improved.
  • the classification layer is connected to the loop layer, the multiple features output by the second sub-network are aggregated, and then the classification function of the classification layer is used to output the calculation of the classification probability for the multiple features , Obtain the phase feature of the left ventricle in the feature map.
  • the left ventricle measurement model performs a series of recognition operations on the left ventricle image, and then passes through the loop
  • the first sub-network in the layer outputs the extracted left ventricular wall in M directions, the diameter of the N directions and the inner/outer cavity area data; and the cardiac phase data extracted through the classification layer. Therefore,
  • the acquisition module 203 can acquire the M-direction wall of the left ventricle, the N-direction diameter and internal/external cavity area data, and cardiac phase data.
  • the left ventricular measurement model needs to be trained in advance, for example, through a large amount of relevant data as input, the weights and biases are constantly updated, and the weight is one of the neurons of the left ventricular measurement model.
  • the connection weight between the left ventricular measurement model represents the values at different positions in the convolution kernel in the image segmentation layer, convolution layer, and circulation layer in the left ventricular measurement model.
  • the bias is a correction to the connection relationship of neurons, Make the output of the neuron closer to the true value.
  • the training set used by the left ventricular measurement model is 145 MRI images published by DIG-Cardiac, covering a wide range of ages, the MRI sampling rate ranges from 0.6836 to 2.0833 mm/pixel, and the types of heart diseases are comprehensive
  • the training set is universal and can accurately train the weight value of the neural network.
  • the CT (Computed Tomography, computer tomography) image set of N heart patients can be selected as the training set, or other types of The image is used as the training set, and there is no restriction here.
  • the query module 204 is used to query the measurement result of the left ventricle image according to the ventricular wall in M directions, the diameter in N directions, the area of the inner/outer cavity, the heart phase data, and the preset case data table, where ,
  • the case data query table includes the corresponding relationship between the L item data and the measurement result.
  • the acquisition module 203 acquires the M azimuth ventricular wall, the N azimuth diameter, the inner/outer cavity area, and the cardiac phase data output by the left ventricular measurement model of the left ventricle image
  • the measurement result of the left ventricle image is queried in a preset case data table based on the data, wherein the case data table includes the correspondence relationship between the above L item data and the measurement result.
  • the computer device 1 receives a case data table created by the user in advance, and the case data table includes M positions of the ventricular wall, N positions of diameter, internal/external cavity area, and cardiac phase data at different values. Under the circumstances, it corresponds to different heart disease cases. Therefore, the query module 204 can query the measurement result corresponding to the L item data in the case data table according to the L item data acquired by the acquiring module 203.
  • the computer device 1 can receive the left ventricular image to be recognized and then input it into a pre-established left ventricular measurement model for measurement.
  • the left ventricular measurement model includes an image segmentation layer and a convolution layer connected in sequence. , Circulatory layer and classification layer; then obtain the left ventricle wall in M directions, the diameter and inner/outer cavity area data in N directions extracted by the circulatory layer, and the cardiac phase data extracted by the classification layer ; And then query the measurement results of the left ventricular image according to the ventricular wall in M directions, the diameter of the N directions, the inner/outer cavity area, and the heart phase data and the preset case data table.
  • this application also proposes a left ventricular measurement method based on image recognition, which is applied to computer equipment.
  • FIG. 3 is a schematic flowchart of an embodiment of a method for measuring a left ventricle based on image recognition according to the present application.
  • the execution order of the steps in the flowchart shown in FIG. 3 can be changed, and some steps can be omitted.
  • Step S500 Receive an image of the left ventricle to be identified.
  • the computer device is connected to a user terminal, such as a mobile terminal, a PC terminal, and other devices, and then the user's left ventricle image to be recognized is received through the user terminal.
  • a user terminal such as a mobile terminal, a PC terminal, and other devices
  • the computer device may also directly provide a data interface to receive the user's left ventricular image to be identified, or connect to a database system, and then receive the left ventricular image to be identified sent from the database system.
  • the left ventricular image to be recognized may be MRI, or other types of images such as CT (Computed Tomography, computer tomography) images.
  • Step S502 Input the left ventricle image to a pre-established left ventricular measurement model, where the left ventricular measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification layer that are sequentially connected.
  • the left ventricular measurement model includes an image segmentation layer, a convolution layer, a circulation layer, and a classification layer that are sequentially connected.
  • the computer device receives the image of the left ventricle to be recognized, it further inputs the image of the left ventricle into a preset left ventricular measurement model to perform left ventricular measurement, wherein the left ventricular measurement model includes successively Connected image segmentation layer, convolution layer, loop layer and classification layer.
  • the left ventricle image is down-sampled and up-sampled multiple times by the image segmentation layer in the left ventricle measurement model to achieve image standardization, wherein the image segmentation layer includes sequential connections The down-sampling layer and the up-sampling layer, the down-sampling layer includes a plurality of down-sampling layers of at least a first down-sampling layer and a second down-sampling layer, and the up-sampling layer includes at least a first up-sampling layer And multiple upsampling layers of the second upsampling layer.
  • the process of performing the multiple down-sampling and up-sampling by the image segmentation layer includes: inputting the left ventricle image into a first down-sampling layer and a second down-sampling layer, respectively; and performing the first down-sampling
  • the output data of the layer and the output data of the first up-sampling layer are combined as input data of the second up-sampling layer; then the output data of the second down-sampling layer and the first layer are combined.
  • the output data of the second down-sampling layer is merged into the convolutional layer as input data.
  • the image segmentation layer adopts a CNN neural network architecture, including a down-sampling layer and an up-sampling layer, the down-sampling layer is composed of four down-sampling layers, and the up-sampling layer is composed of four up-sampling layers;
  • the up-sampling layer is connected to the subsequent convolutional layer through a Dense block.
  • each down-sampling layer in the down-sampling layer includes a Dense block, an extended block, and a down-sampling block.
  • the Dense block is a basic module of a densenet block
  • the extended block is a Dense block.
  • the expansion rate in the block for example, the output of the convolution block of (1, 3, 5) is combined as the input of the down-sampling layer.
  • the expansion rate is a parameter of the expansion convolution and is used to indicate the size of the expansion of the convolution kernel.
  • the down-sampling layer adopts maximum pooling, for example, a 2 ⁇ 2 convolution kernel is selected; each up-sampling layer of the up-sampling layer includes a dense block and an up-sampling block, where the first up-sampling layer
  • the output of and the output of the Dense block in the fourth downsampling group are combined as the input of the second upsampling layer, and the output of the second upsampling layer and the output of the Dense block in the third downsampling layer are combined as the third upsampling layer
  • the input of the third up-sampling layer and the output of the Dense block in the second down-sampling layer are combined as the input of the fourth up-sampling layer, and the output of the Dense block in the fourth up-sampling layer and the first down-sampling layer
  • the merge is used as the input of the last Dense block, and the output of the last Dense block is used as the downstream node, that is, the input of the
  • the down-sampling layer of the image segmentation layer includes an extension block
  • the extension block includes a convolution kernel
  • the image segmentation layer passes through the volume of the extension block of the down-sampling layer.
  • the convolution kernel implements the down-sampling process by performing convolution processing on the input left ventricular image, wherein the convolution kernel performs convolution by sampling the hole convolution method.
  • the hole convolution can increase the receptive field without pooling.
  • the expansion block merges and connects the output of the convolution block with the expansion rate (1, 3, 5) as the input of the convolution in the down-sampling block, and up-sampling Path fusion downsampling the output of the path part.
  • the features of different dimensions in the left ventricular image are fused together, and the multi-dimensional features of the left ventricular image are repeatedly and fully utilized.
  • the convolutional layer further extracts task features, and uses the extracted feature maps as the input of the third layer of neural network.
  • the convolutional layer is composed of an input layer, a plurality of hidden layers, and an output layer.
  • the hidden layers include a convolutional layer, a nonlinear layer, and a pool. ⁇ The layer and so on. Further feature extraction of the feature value output by the image segmentation layer by the convolutional layer can improve the convergence of the feature value and ensure accuracy.
  • the recurrent layer further performs feature recognition and judgment on the multi-dimensional features extracted by the convolutional layer.
  • the recurrent layer includes a first sub-network and a second sub-network, and the recurrent layer fully connects the first sub-network to each neuron to determine the characteristic value of the left ventricle image. Perform regression; by fully connecting the second sub-network to each neuron to classify the feature value of the left ventricular diagram, the first sub-network and the second sub-network both use LSTM (Long short -term memory, long and short-term memory) neural network model structure.
  • LSTM Long short -term memory, long and short-term memory
  • the recurrent layer includes specifically two standard LSTM neural networks.
  • the standard LSTM neural network includes: an input layer, an LSTM cell layer, and an output layer.
  • the standard LSTM neural network can store and Use the input and output information of the system in the past. Therefore, the output of the first sub-network is fully connected with N neurons to output N feature maps for regression, and the results of various tasks are obtained after regression.
  • Each task includes M azimuth walls and N The diameter of the azimuth, the area of the inner cavity/the area of the outer cavity, etc.; the output of the second self-network is connected with neurons for classification tasks, and the classification tasks include cardiac phases and the like.
  • the cycle layer can be used to solve the sequence and time problems in the cardiac phase determination process.
  • the convolutional layer and the recurrent layer cross-use feature data by sharing a hidden layer. Therefore, the accuracy of the multi-dimensional feature extraction of the left ventricle image is improved.
  • the classification layer is connected to the loop layer, the multiple features output by the second sub-network are aggregated, and then the classification function of the classification layer is used to output the calculation of the classification probability for the multiple features , Obtain the phase feature of the left ventricle in the feature map.
  • the left ventricle measurement model will perform a series of recognition operations on the left ventricle image, and then pass through the circulatory layer
  • the first sub-network in the output of the extracted left ventricular wall in M directions, the diameter of the N directions and the internal/external cavity area data; and the cardiac phase data extracted through the classification layer therefore, the The computer equipment can obtain the left ventricle wall in M directions, the diameter and inner/outer cavity area data in N directions, and the heart phase data.
  • the left ventricular measurement model needs to be trained in advance, for example, through a large amount of relevant data as input, the weights and biases are constantly updated, and the weight is one of the neurons of the left ventricular measurement model.
  • the connection weight between the left ventricular measurement model represents the values at different positions in the convolution kernel in the image segmentation layer, convolution layer, and circulation layer in the left ventricular measurement model.
  • the bias is a correction to the connection relationship of neurons, Make the output of the neuron closer to the true value.
  • the training set used by the left ventricular measurement model is 145 MRI images published by DIG-Cardiac, covering a wide range of ages, the MRI sampling rate ranges from 0.6836 to 2.0833 mm/pixel, and the types of heart diseases are comprehensive
  • the training set is universal and can accurately train the weight value of the neural network.
  • the CT (Computed Tomography, computer tomography) image set of N heart patients can be selected as the training set, or other types of The image is used as the training set, and there is no restriction here.
  • Step S506 Query the measurement result of the left ventricle image according to the ventricular wall in M directions, the diameter of the N directions, the inner/outer cavity area, the heart phase data, and the preset case data table.
  • the case data query table includes the corresponding relationship between the L item data and the measurement result.
  • the computer device After the computer device obtains the M azimuth wall, the N azimuth diameter, the inner/outer cavity area, and the heart phase data of the left ventricle image output by the left ventricle measurement model , The measurement result of the left ventricle image will be queried in a preset case data table based on these data, where the case data table includes the correspondence between the L items above and the measurement result.
  • the computer device receives a case data table created by the user in advance, and the case data table includes M positions of the ventricular wall, N positions of diameter, internal/external cavity area, and heart phase data at different values. Below, corresponding to different heart disease cases. Therefore, the computer device can query the case data table for the measurement result corresponding to the L item data according to the acquired L item data.
  • the left ventricular measurement method based on image recognition proposed in this embodiment can receive a left ventricular image to be recognized and then input it into a pre-established left ventricular measurement model for measurement, and the left ventricular measurement model includes successively connected image segmentation layers. , Convolutional layer, circulation layer, and classification layer; and then obtain the left ventricle wall in M directions, the diameter and inner/outer cavity area data in N directions extracted by the circulation layer, and the classification layer extracted Heart phase data; and then query the measurement results of the left ventricular image according to the M-direction ventricular wall, the N-direction diameter, the inner/outer cavity area, and the heart phase data and the preset case data table.
  • the left ventricular measurement model includes successively connected image segmentation layers. , Convolutional layer, circulation layer, and classification layer; and then obtain the left ventricle wall in M directions, the diameter and inner/outer cavity area data in N directions extracted by the circulation layer, and the classification layer extracted Heart phase data; and then query the measurement results of the left ventricular image according to the M
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the method described in each embodiment of the present application.
  • a terminal device which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Public Health (AREA)
  • Mathematical Physics (AREA)
  • Medical Informatics (AREA)
  • Pathology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Theoretical Computer Science (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Veterinary Medicine (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Computational Linguistics (AREA)
  • Fuzzy Systems (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Radiology & Medical Imaging (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种基于图像识别的左心室测量方法，该方法包括：接收待识别的左心室图像，并将其输入到预先建立的左心室测量模型进行测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据；根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果。还提供一种基于图像识别的左心室测量装置、计算机设备以及非易失性计算机可读存储介质，能够基于图像识别实现更加精确的左心室测量。

Description

基于图像识别的左心室测量方法、装置以及计算机设备
本申请要求于2019年09月16日提交中国专利局、申请号为201910871715.7、发明名称为“基于图像识别的左心室测量方法、装置以及计算机设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及基于图像识别的左心室测量技术领域，尤其涉及一种基于图像识别的左心室测量方法、装置、计算机设备及非易失性计算机可读存储介质。
背景技术
心脏相关的疾病是当今社会导致死亡率最高的病种之一，提前的预防和及时的发现对心脏相关的疾病有重大的意义，而左心室测量对于心脏相关病的诊断非常重要。当前，主要通过对心脏的左心室的磁共振图像MRI(Magnetic Resonance Imaging，MRI)进行分割，分割出多个维度的特征，包括心脏的M个方位室壁、N个方位的直径、内/外腔面积以及心脏相位数据，然后进行测量并分析出心脏是否异常，这种方式工作量大、耗时长，且诊断过程对于医师的经验需要比较高。随着图像识别技术的发展，通过图像识别模型对左心室图像进行识别和分析也可以诊断出心脏是否异常。然而，发明人意识到，由于左心室图像的测量需要相当高的精细度，因此，现有的通过图像识别模型对左心室图像进行识别的方法应用到左心室图像识别和测量，结果的准确度并不高。
发明内容
有鉴于此，本申请提出一种基于图像识别的左心室测量方法、装置、计算机设备及非易失性计算机可读存储介质，能够接收待识别的左心室图像之后将其输入到预先建立的左心室测量模型进行测量，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；然后获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径 和内/外腔面积数据，以及所述分类层提取到的心脏相位数据；再根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果。通过以上方式，能够基于图像识别实现更加精确的左心室测量。
首先，为实现上述目的，本申请提供一种基于图像识别的左心室测量方法，所述方法包括：
接收待识别的左心室图像；将所述左心室图像输入到预先建立的左心室测量模型，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
此外，为实现上述目的，本申请还提供一种基于图像识别的左心室测量装置，所述装置包括：
接收模块，用于接收待识别的左心室图像；输入模块，用于将所述左心室图像输入到预先建立的左心室测量模型，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；获取模块，用于获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；查询模块，用于根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
进一步地，本申请还提出一种计算机设备，所述计算机设备包括存储器、处理器，所述存储器上存储有可在所述处理器上运行的计算机可读指令，所述计算机可读指令被所述处理器执行时实现步骤：
接收待识别的左心室图像；将所述左心室图像输入到预先建立的左心室测量模型，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；根据M个方位的室壁、 N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
进一步地，为实现上述目的，本申请还提供一种非易失性计算机可读存储介质，所述非易失性计算机可读存储介质存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行步骤：
接收待识别的左心室图像；将所述左心室图像输入到预先建立的左心室测量模型，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
本申请所提出的基于图像识别的左心室测量方法、装置、计算机设备及非易失性计算机可读存储介质，能够接收待识别的左心室图像之后将其输入到预先建立的左心室测量模型进行测量，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；然后获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据；再根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果。通过以上方式，能够基于图像识别实现更加精确的左心室测量。
附图说明
图1是本申请计算机设备一可选的硬件架构的示意图；
图2是本申请基于图像识别的左心室测量装置一实施例的程序模块示意图；
图3是本申请基于图像识别的左心室测量方法一实施例的流程示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不 用于限定本申请。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
需要说明的是，在本申请中涉及“第一”、“第二”等的描述仅用于描述目的，而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外，各个实施例之间的技术方案可以相互结合，但是必须是以本领域普通技术人员能够实现为基础，当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在，也不在本申请要求的保护范围之内。
参阅图1所示，是本申请计算机设备1一可选的硬件架构的示意图。
本实施例中，所述计算机设备1可包括，但不仅限于，可通过系统总线相互通信连接存储器11、处理器12、网络接口13。
所述计算机设备1通过网络接口13连接网络(图1未标出)，通过网络连接到其他终端设备如移动终端(Mobile Terminal)、用户设备(User Equipment，UE)及便携设备(portable equipment)，PC端等。所述网络可以是企业内部网(Intranet)、互联网(Internet)、全球移动通讯系统(Global System of Mobile communication，GSM)、宽带码分多址(Wideband Code Division Multiple Access，WCDMA)、4G网络、5G网络、蓝牙(Bluetooth)、Wi-Fi、通话网络等无线或有线网络。
需要指出的是，图1仅示出了具有组件11-13的计算机设备1，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。
其中，所述存储器11至少包括一种类型的非易失性计算机可读存储介质，所述非易失性计算机可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，所述存储器11可以是所述计算机设备1的内部存储单元，例如该计算机设备1的硬盘或内存。在另一些实施例中，所述存储器11也可以是所述计算机设备1的外部存储设备，例如该计算机设备1配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。当然，所述存储器11 还可以既包括所述计算机设备1的内部存储单元也包括其外部存储设备。本实施例中，所述存储器11通常用于存储安装于所述计算机设备1的操作系统和各类应用软件，例如基于图像识别的左心室测量装置200的程序代码等。此外，所述存储器11还可以用于暂时地存储已经输出或者将要输出的各类数据。
所述存储器11存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行步骤：
接收待识别的左心室图像；将所述左心室图像输入到预先建立的左心室测量模型，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
所述处理器12在一些实施例中可以是中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器12通常用于控制所述计算机设备1的总体操作，例如执行数据交互或者通信相关的控制和处理等。本实施例中，所述处理器12用于运行所述存储器11中存储的程序代码或者处理数据，例如运行所述的基于图像识别的左心室测量装置200等。
所述网络接口13可包括无线网络接口或有线网络接口，该网络接口13通常用于在所述计算机设备1与其他终端设备如移动终端、用户设备及便携设备，PC端等之间建立通信连接。
本实施例中，所述计算机设备1内安装并运行有基于图像识别的左心室测量装置200时，当所述基于图像识别的左心室测量装置200运行时，能够接收待识别的左心室图像之后将其输入到预先建立的左心室测量模型进行测量，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；然后获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据；再根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果。通过以上方式，能够基于图像识别实现更加精确的左 心室测量。
至此，己经详细介绍了本申请各个实施例的应用环境和相关设备的硬件结构和功能。下面，将基于上述应用环境和相关设备，提出本申请的各个实施例。
首先，本申请提出一种基于图像识别的左心室测量装置200。
参阅图2所示，是本申请基于图像识别的左心室测量装置200一实施例的程序模块图。
本实施例中，所述基于图像识别的左心室测量装置200包括一系列的存储于存储器11上的计算机可读指令，当该计算机可读指令被处理器12执行时，可以实现本申请各实施例的基于图像识别的左心室测量功能。在一些实施例中，基于该计算机可读指令各部分所实现的特定的操作，基于图像识别的左心室测量装置200可以被划分为一个或多个模块。例如，在图2中，所述基于图像识别的左心室测量装置200可以被分割成接收模块201、输入模块202、获取模块203和查询模块204。其中：
所述接收模块201，用于接收待识别的左心室图像。
在本实施例中，所述计算机设备1与用户终端，比如移动终端，PC端等设备连接，然后所述接收模块201则可以通过用户终端接收用户的待识别的左心室图像。当然，在其他实施例中，所述计算机设备1也可以直接提供数据接口接收用户的待识别的左心室图像，或者连接到数据库系统，然后接收从数据库系统发送过来的待识别的左心室图像。在本实施例中，所述待识别的左心室图像可以是MRI，也可以是其他类型的图像比如CT(Computed Tomography，计算机断层扫描)图像等。
所述输入模块202，用于将所述左心室图像输入到预先建立的左心室测量模型进行测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层。
具体地，所述接收模块201接收到待识别的左心室图像之后，所述输入模块202进一步将所述左心室图像输入到预设的左心室测量模型以进行左心室测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层。在本实施例中，通过所述左心室测量模型中的所述图像分割层的对所述左心室图像进行多次下采样和上采样从而实现图像标准化，其中，所述图像分割层包括依次连接的下采样层和上采样层，所述下采样层包括至少第一下采样分层和第二下采样分层的多个下采样分层，所述上采样层包括至少第一上采样分层和第二上采样分层的多个上采样分层。具体地，所述图像分割层执行 所述多次下采样和上采样的过程包括：分别将所述左心室图像输入到第一下采样层和第二下采样层；将所述第一下采样分层的输出数据与所述第一上采样分层的输出数据进行数据合并作为所述第二上采样分层的输入数据；再将所述第二下采样分层的输出数据以及所述第二下采样分层的输出数据进行数据合并作为输入数据输入到所述卷积层。
例如，所述图像分割层采用CNN神经网络架构，包括下采样层和上采样层，所述下采样层由四个下采样分层组成，所述上采样层由四个上采样分层组成；上采样层的通过一个Dense块连接后续的所述卷积层。具体地，所述下采样层中的每个下采样分层都包括一个Dense块、一个扩展块和一个下采样块，所述Dense块是一个densenet block的基本模块，所述扩展块是将Dense块中扩张率，例如(1，3，5)的卷积块的输出合并起来作为下采样层的输入，所述扩张率是扩张卷积的参数，用来表示卷积核扩张的大小，所述下采样层采用最大池化，例如选用2×2的卷积核；所述上采样分层的每个上采样分层都包括一个Dense块和一个上采样块，其中第一上采样分层的输出与第四下采样组中Dense块的输出合并作为第二上采样分层的输入，第二上采样分层与第三下采样分层中Dense块的输出合并作为第三上采样分层的输入，第三上采样分层与第二下采样分层中Dense块的输出合并作为第四上采样分层的输入，第四上采样分层与第一下采样分层中Dense块的输出合并作为最后一个Dense块的输入，将最后一个Dense块的输出作为下游节点，即所述卷积层的输入。选择这样合并的原因是保持输出大小的一致，这样在做合并时就不会有特征图像大小不一致的问题。下采样层中每个Dense块后做卷积、池化操作，这种结构的可以省参数、省计算，对于图像的特征值重用起到抗过拟合的作用。
一般地，由于图片输入到神经网络中，神经网络将会对图像做卷积再池化的操作，由于在先减少再增大的过程中肯定会有一些信息损失，为了减少由池化带来信息损失，在本实施例中，所述图像分割层的所述下采样层包括扩展块，所述扩展块包括一个卷积核，所述图像分割层通过所述下采样层的扩展块的卷积核对输入的左心室图像进行卷积处理，从而实现下采样过程，其中，所述卷积核采样空洞卷积方式进行卷积。所述空洞卷积可以不通过池化来增大感受野，所述扩展块将膨胀率(1，3，5)的卷积块输出合并连接起来作为下采样块中卷积的输入，上采样路径融合下采样路径部分的输出，这样做是将所述左心室图像中的不同维度的特征融合在了一起，反复并充分利用所述左心室图像的多维度的特征。
接着，所述卷积层进一步提取任务特征，将提取到的特征图作为第三层神经网络的输入。其中，所述卷积层由输入层、多个隐含层和输出层组成，在本实施例中，共设5个隐含层，所述隐含层包括卷积层、非线性层、池化层等。通过所述卷积层对所述图像分割层输出的特征值进一步特征提取，可以提高特征值的收敛度，保证精确度。
然后，所述循环层对所述卷积层提取出来的多维度特征进一步进行特征的识别和判断。在本实施例中，所述循环层包括第一子网络和第二子网络，所述循环层通过将所述第一子网络全连接到每个神经元从而对所述左心室图像的特征值做回归；通过将所述第二子网络全连接到每个神经元从而对所述左心室图像的特征值做分类，所述第一子网络和所述第二子网络都是采用LSTM(Long short-term memory，长短时记忆)神经网络模型结构。
具体地，所述循环层包括具体地是两个标准的LSTM神经网络，所述标准LSTM神经网络包括：一个输入层、一个LSTM细胞层、一个输出层，标准LSTM神经网络可以通过内部反馈存储和利用系统过去时刻的输入输出信息。因此，所述第一子网络的输出与N个神经元全连接输出N个特征图做回归，经回归后得到各项任务的结果，所述各项任务包括M个方位的室壁，N个方位的直径，内腔面积/外腔面积等；所述第二自网络的输出与神经元连接做分类任务，分类任务包括心脏相位等。由于LSTM神经元可以保持记忆，利用之前输入的图像信息优化神经网络的计算过程，因此，可以通过所述循环层以解决心脏相位判断过程中的顺序和时间问题。另外，所述卷积层与所述循环层通过共用隐层的方式交叉使用特征数据，因此，对于所述左心室图像的多维度特征提取，提升了精确度。
最后，所述分类层与所述循环层连接，将所述第二子网络输出的多个特征进行汇总，然后通过所述分类层的分类函数输出针对所述多个特征的进行分类概率的计算，获得特征图中左心室的相位特征。
所述获取模块203，用于获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2。
具体地，所述输入模块202将所述左心室图像输入到所述左心室测量模型之后，所述左心室测量模型则会对所述左心室图像进行一系列的识别操作，然后通过所述循环层中的第一子网络输出提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据； 以及通过所述分类层提取到的心脏相位数据，因此，所述获取模块203可以获取到左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及心脏相位数据。当然，在本实施例中，所述左心室测量模型需要事先被训练，例如，通过大量的相关数据作输入，不断更新权值以及偏置，权值就是所述左心室测量模型的神经元之间的连接权值，在所述左心室测量模型中的图像分割层、卷积层和循环层中表示卷积核中不同位置上的数值，所述偏置是对神经元连接关系的修正，使得神经元的输出更接近真实值。在本实施例中，例如，所述左心室测量模型采用的训练集为DIG-Cardiac公开的145个MRI图像，覆盖年龄广，MRI采样率范围从0.6836到2.0833mm/pixel，心脏病病种全面，训练集具有普遍性，可准确的训练神经网络的权重值，在其他实施例中，可以选用N个心脏病患者的CT(Computed Tomography，计算机断层扫描)图像集作训练集，或者其他类型的图像作为训练集，这里不做限制。
所述查询模块204，用于根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
具体地，当所述获取模块203获取到由所述左心室测量模型输出的对应于所述左心室图像的M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据之后，则会根据这些数据在一个预设的病例数据表中查询出所述左心室图像的测量结果，其中，所述病例数据表中包括以上所述L项数据与测量结果的对应关系。例如，所述计算机设备1接收用户预先建立的病例数据表，所述病例数据表中包括M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据在不同的值的情况下，对应了不同的心脏病例。因此，所述查询模块204可以根据所述获取模块203获取到的所述L项数据在所述病例数据表中查询出对应所述L项数据的测量结果。
从上文可知，所述计算机设备1能够接收待识别的左心室图像之后将其输入到预先建立的左心室测量模型进行测量，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；然后获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据；再根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室 图像的测量结果。通过以上方式，能够基于图像识别实现更加精确的左心室测量。
此外，本申请还提出一种基于图像识别的左心室测量方法，所述方法应用于计算机设备。
参阅图3所示，是本申请基于图像识别的左心室测量方法一实施例的流程示意图。在本实施例中，根据不同的需求，图3所示的流程图中的步骤的执行顺序可以改变，某些步骤可以省略。
步骤S500，接收待识别的左心室图像。
在本实施例中，所述计算机设备与用户终端，比如移动终端，PC端等设备连接，然后通过用户终端接收用户的待识别的左心室图像。当然，在其他实施例中，所述计算机设备也可以直接提供数据接口接收用户的待识别的左心室图像，或者连接到数据库系统，然后接收从数据库系统发送过来的待识别的左心室图像。在本实施例中，所述待识别的左心室图像可以是MRI，也可以是其他类型的图像比如CT(Computed Tomography，计算机断层扫描)图像等。
步骤S502，将所述左心室图像输入到预先建立的左心室测量模型，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层。
具体地，所述计算机设备接收到待识别的左心室图像之后，则进一步将所述左心室图像输入到预设的左心室测量模型以进行左心室测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层。在本实施例中，通过所述左心室测量模型中的所述图像分割层的对所述左心室图像进行多次下采样和上采样从而实现图像标准化，其中，所述图像分割层包括依次连接的下采样层和上采样层，所述下采样层包括至少第一下采样分层和第二下采样分层的多个下采样分层，所述上采样层包括至少第一上采样分层和第二上采样分层的多个上采样分层。具体地，所述图像分割层执行所述多次下采样和上采样的过程包括：分别将所述左心室图像输入到第一下采样层和第二下采样层；将所述第一下采样分层的输出数据与所述第一上采样分层的输出数据进行数据合并作为所述第二上采样分层的输入数据；再将所述第二下采样分层的输出数据以及所述第二下采样分层的输出数据进行数据合并作为输入数据输入到所述卷积层。
例如，所述图像分割层采用CNN神经网络架构，包括下采样层和上采样层，所述下采 样层由四个下采样分层组成，所述上采样层由四个上采样分层组成；上采样层的通过一个Dense块连接后续的所述卷积层。具体地，所述下采样层中的每个下采样分层都包括一个Dense块、一个扩展块和一个下采样块，所述Dense块是一个densenet block的基本模块，所述扩展块是将Dense块中扩张率，例如(1，3，5)的卷积块的输出合并起来作为下采样层的输入，所述扩张率是扩张卷积的参数，用来表示卷积核扩张的大小，所述下采样层采用最大池化，例如选用2×2的卷积核；所述上采样分层的每个上采样分层都包括一个Dense块和一个上采样块，其中第一上采样分层的输出与第四下采样组中Dense块的输出合并作为第二上采样分层的输入，第二上采样分层与第三下采样分层中Dense块的输出合并作为第三上采样分层的输入，第三上采样分层与第二下采样分层中Dense块的输出合并作为第四上采样分层的输入，第四上采样分层与第一下采样分层中Dense块的输出合并作为最后一个Dense块的输入，将最后一个Dense块的输出作为下游节点，即所述卷积层的输入。选择这样合并的原因是保持输出大小的一致，这样在做合并时就不会有特征图像大小不一致的问题。下采样层中每个Dense块后做卷积、池化操作，这种结构的可以省参数、省计算，对于图像的特征值重用起到抗过拟合的作用。
一般地，由于图片输入到神经网络中，神经网络将会对图像做卷积再池化的操作，由于在先减少再增大的过程中肯定会有一些信息损失，为了减少由池化带来信息损失，在本实施例中，所述图像分割层的所述下采样层包括扩展块，所述扩展块包括一个卷积核，所述图像分割层通过所述下采样层的扩展块的卷积核通过对输入的左心室图像进行卷积处理，从而实现下采样过程，其中，所述卷积核采样空洞卷积方式进行卷积。所述空洞卷积可以不通过池化来增大感受野，所述扩展块将膨胀率(1，3，5)的卷积块输出合并连接起来作为下采样块中卷积的输入，上采样路径融合下采样路径部分的输出，这样做是将所述左心室图像中的不同维度的特征融合在了一起，反复并充分利用所述左心室图像的多维度的特征。
接着，所述卷积层进一步提取任务特征，将提取到的特征图作为第三层神经网络的输入。其中，所述卷积层由输入层、多个隐含层和输出层组成，在本实施例中，共设5个隐含层，所述隐含层包括卷积层、非线性层、池化层等。通过所述卷积层对所述图像分割层输出的特征值进一步特征提取，可以提高特征值的收敛度，保证精确度。
然后，所述循环层对所述卷积层提取出来的多维度特征进一步进行特征的识别和判断。在本实施例中，所述循环层包括第一子网络和第二子网络，所述循环层通过将所述第一子网络全连接到每个神经元从而对所述左心室图像的特征值做回归；通过将所述第二子网络全连接到每个神经元从而对所述左心室图特征值做分类，所述第一子网络和所述第二子网络都是采用LSTM(Long short-term memory，长短时记忆)神经网络模型结构。
具体地，所述循环层包括具体地是两个标准的LSTM神经网络，所述标准LSTM神经网络包括：一个输入层、一个LSTM细胞层、一个输出层，标准LSTM神经网络可以通过内部反馈存储和利用系统过去时刻的输入输出信息。因此，所述第一子网络的输出与N个神经元全连接输出N个特征图做回归，经回归后得到各项任务的结果，所述各项任务包括M个方位的室壁，N个方位的直径，内腔面积/外腔面积等；所述第二自网络的输出与神经元连接做分类任务，分类任务包括心脏相位等。由于LSTM神经元可以保持记忆，利用之前输入的图像信息优化神经网络的计算过程，因此，可以通过所述循环层以解决心脏相位判断过程中的顺序和时间问题。另外，所述卷积层与所述循环层通过共用隐层的方式交叉使用特征数据，因此，对于所述左心室图像的多维度特征提取，提升了精确度。
最后，所述分类层与所述循环层连接，将所述第二子网络输出的多个特征进行汇总，然后通过所述分类层的分类函数输出针对所述多个特征的进行分类概率的计算，获得特征图中左心室的相位特征。
步骤S504，获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2。
具体地，所述计算机设备将所述左心室图像输入到所述左心室测量模型之后，所述左心室测量模型则会对所述左心室图像进行一系列的识别操作，然后通过所述循环层中的第一子网络输出提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据；以及通过所述分类层提取到的心脏相位数据，因此，所述计算机设备可以获取到左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及心脏相位数据。当然，在本实施例中，所述左心室测量模型需要事先被训练，例如，通过大量的相关数据作输入，不断更新权值以及偏置，权值就是所述左心室测量模型的神经元之间的连接权值，在所述左心室测量模型中的图像分割层、卷积层和循环层中表示卷积核中不同位置上的数值，所述偏置 是对神经元连接关系的修正，使得神经元的输出更接近真实值。在本实施例中，例如，所述左心室测量模型采用的训练集为DIG-Cardiac公开的145个MRI图像，覆盖年龄广，MRI采样率范围从0.6836到2.0833mm/pixel，心脏病病种全面，训练集具有普遍性，可准确的训练神经网络的权重值，在其他实施例中，可以选用N个心脏病患者的CT(Computed Tomography，计算机断层扫描)图像集作训练集，或者其他类型的图像作为训练集，这里不做限制。
步骤S506，根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果。其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
具体地，当所述计算机设备获取到由所述左心室测量模型输出的对应于所述左心室图像的M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据之后，则会根据这些数据在一个预设的病例数据表中查询出所述左心室图像的测量结果，其中，所述病例数据表中包括以上所述L项数据与测量结果的对应关系。例如，所述计算机设备接收用户预先建立的病例数据表，所述病例数据表中包括M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据在不同的值的情况下，对应了不同的心脏病例。因此，所述计算机设备可以根据获取到的所述L项数据在所述病例数据表中查询出对应所述L项数据的测量结果。
本实施例所提出的基于图像识别的左心室测量方法能够接收待识别的左心室图像之后将其输入到预先建立的左心室测量模型进行测量，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；然后获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据；再根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果。通过以上方式，能够基于图像识别实现更加精确的左心室测量。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者 是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种基于图像识别的左心室测量方法，所述方法包括步骤：
    接收待识别的左心室图像；
    将所述左心室图像输入到预先建立的左心室测量模型进行测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；
    获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；
    根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表，查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
  2. 如权利要求1所述的基于图像识别的左心室测量方法，所述方法还包括：通过所述图像分割层的对所述左心室图像进行多次下采样和上采样从而实现图像标准化，其中，所述图像分割层包括依次连接的下采样层和上采样层，所述下采样层包括至少第一下采样分层和第二下采样分层的多个下采样分层，所述上采样层包括至少第一上采样分层和第二上采样分层的多个上采样分层。
  3. 如权利要求2所述的基于图像识别的左心室测量方法，所述多次下采样和上采样的步骤包括：分别将所述左心室图像输入到第一下采样层和第二下采样层；将所述第一下采样分层的输出数据与所述第一上采样分层的输出数据进行数据合并作为所述第二上采样分层的输入数据；再将所述第二下采样分层的输出数据以及所述第二下采样分层的输出数据进行数据合并作为输入数据输入到所述卷积层。
  4. 如权利要求1所述的基于图像识别的左心室测量方法，所述下采样层包括扩展块，所述扩展块包括一个卷积核，所述方法还包括：通过所述下采样层的扩展块的卷积核对输入的左心室图像进行卷积处理从而实现下采样过程，其中，所述卷积核采样空洞卷积方式进行卷积。
  5. 如权利要求1所述的基于图像识别的左心室测量方法，所述循环层包括第一子网络和第二子网络，所述方法还包括：通过将所述第一子网络全连接到每个神经元从而对所述左心室图像的特征值做回归，通过将所述第二子网络全连接到每个神经元从 而对所述左心室图像的特征值做分类。
  6. 如权利要求5所述的基于图像识别的左心室测量方法，所述第一子网络和所述第二子网络都是采用LSTM神经网络模型结构。
  7. 如权利要求1所述的基于图像识别的左心室测量方法，所述卷积层与所述循环层通过共用隐层的方式交叉使用特征数据。
  8. 一种基于图像识别的左心室测量装置，所述装置包括：
    接收模块，用于接收待识别的左心室图像；
    输入模块，用于将所述左心室图像输入到预先建立的左心室测量模型进行测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；
    获取模块，用于获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；
    查询模块，用于根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
  9. 如权利要求8所述的基于图像识别的左心室测量装置，所述输入模块还用于：
    通过所述图像分割层的对所述左心室图像进行多次下采样和上采样从而实现图像标准化，其中，所述图像分割层包括依次连接的下采样层和上采样层，所述下采样层包括至少第一下采样分层和第二下采样分层的多个下采样分层，所述上采样层包括至少第一上采样分层和第二上采样分层的多个上采样分层。
  10. 如权利要求9所述的基于图像识别的左心室测量装置，所述输入模块还用于：
    分别将所述左心室图像输入到第一下采样层和第二下采样层；将所述第一下采样分层的输出数据与所述第一上采样分层的输出数据进行数据合并作为所述第二上采样分层的输入数据；再将所述第二下采样分层的输出数据以及所述第二下采样分层的输出数据进行数据合并作为输入数据输入到所述卷积层。
  11. 如权利要求8所述的基于图像识别的左心室测量装置，所述下采样层包括扩展块，所述扩展块包括一个卷积核，所述输入模块还用于：
    通过所述下采样层的扩展块的卷积核对输入的左心室图像进行卷积处理从而实现下采样过程，其中，所述卷积核采样空洞卷积方式进行卷积。
  12. 如权利要求8所述的基于图像识别的左心室测量装置，所述循环层包括第一子网络和第二子网络，所述输入模块还用于：
    通过将所述第一子网络全连接到每个神经元从而对所述左心室图像的特征值做回归，通过将所述第二子网络全连接到每个神经元从而对所述左心室图像的特征值做分类。
  13. 如权利要求12所述的基于图像识别的左心室测量装置，所述第一子网络和所述第二子网络都是采用LSTM神经网络模型结构。
  14. 如权利要求8所述的基于图像识别的左心室测量装置，所述卷积层与所述循环层通过共用隐层的方式交叉使用特征数据。
  15. 一种计算机设备，所述计算机设备包括存储器、处理器，所述存储器上存储有可在所述处理器上运行的计算机可读指令，所述计算机可读指令被所述处理器执行时实现步骤：
    接收待识别的左心室图像；
    将所述左心室图像输入到预先建立的左心室测量模型进行测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；
    获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；
    根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表，查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
  16. 如权利要求15所述的计算机设备，所述计算机可读指令被所述处理器执行时还实现步骤：
    通过所述图像分割层的对所述左心室图像进行多次下采样和上采样从而实现图像标准化，其中，所述图像分割层包括依次连接的下采样层和上采样层，所述下采样层包括至少第一下采样分层和第二下采样分层的多个下采样分层，所述上采样层包括至 少第一上采样分层和第二上采样分层的多个上采样分层。
  17. 如权利要求16所述的计算机设备，所述多次下采样和上采样的步骤包括：分别将所述左心室图像输入到第一下采样层和第二下采样层；将所述第一下采样分层的输出数据与所述第一上采样分层的输出数据进行数据合并作为所述第二上采样分层的输入数据；再将所述第二下采样分层的输出数据以及所述第二下采样分层的输出数据进行数据合并作为输入数据输入到所述卷积层。
  18. 一种非易失性计算机可读存储介质，所述非易失性计算机可读存储介质存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行步骤：
    接收待识别的左心室图像；
    将所述左心室图像输入到预先建立的左心室测量模型进行测量，其中，所述左心室测量模型包括依次连接的图像分割层、卷积层、循环层以及分类层；
    获取所述循环层提取到的左心室的M个方位的室壁、N个方位的直径和内/外腔面积数据，以及所述分类层提取到的心脏相位数据共计L项数据，其中L＝M+N+2；
    根据M个方位的室壁、N个方位的直径、内/外腔面积以及心脏相位数据以及预设的病例数据表，查询出所述左心室图像的测量结果，其中，所述病例数据查询表包括所述L项数据与测量结果的对应关系。
  19. 如权利要求18所述的非易失性计算机可读存储介质，所述计算机可读指令被所述处理器执行时还实现步骤：
    通过所述图像分割层的对所述左心室图像进行多次下采样和上采样从而实现图像标准化，其中，所述图像分割层包括依次连接的下采样层和上采样层，所述下采样层包括至少第一下采样分层和第二下采样分层的多个下采样分层，所述上采样层包括至少第一上采样分层和第二上采样分层的多个上采样分层。
  20. 如权利要求19所述的非易失性计算机可读存储介质，所述多次下采样和上采样的步骤包括：分别将所述左心室图像输入到第一下采样层和第二下采样层；将所述第一下采样分层的输出数据与所述第一上采样分层的输出数据进行数据合并作为所述第二上采样分层的输入数据；再将所述第二下采样分层的输出数据以及所述第二下采 样分层的输出数据进行数据合并作为输入数据输入到所述卷积层。
PCT/CN2019/117552 2019-09-16 2019-11-12 基于图像识别的左心室测量方法、装置以及计算机设备 WO2021051555A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910871715.7 2019-09-16
CN201910871715.7A CN110731777B (zh) 2019-09-16 2019-09-16 基于图像识别的左心室测量方法、装置以及计算机设备

Publications (1)

Publication Number Publication Date
WO2021051555A1 true WO2021051555A1 (zh) 2021-03-25

Family

ID=69267939

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/117552 WO2021051555A1 (zh) 2019-09-16 2019-11-12 基于图像识别的左心室测量方法、装置以及计算机设备

Country Status (2)

Country Link
CN (1) CN110731777B (zh)
WO (1) WO2021051555A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113298786B (zh) * 2021-05-26 2022-02-11 北京长木谷医疗科技有限公司 图像识别及模型训练的方法、真臼位置识别的方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1914617A (zh) * 2004-02-03 2007-02-14 美国西门子医疗解决公司 与心脏有关的疾病和病症的自动诊断和决策支持系统和方法
US20080319308A1 (en) * 2007-05-22 2008-12-25 Worcester Polytechnic Institute Patient-specific image-based computational modeling and techniques for human heart surgery optimization
CN109285157A (zh) * 2018-07-24 2019-01-29 深圳先进技术研究院 左心室心肌分割方法、装置及计算机可读存储介质
CN109598727A (zh) * 2018-11-28 2019-04-09 北京工业大学 一种基于深度神经网络的ct图像肺实质三维语义分割方法

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6795732B2 (en) * 2001-10-30 2004-09-21 Medtronic, Inc. Implantable medical device employing sonomicrometer output signals for detection and measurement of cardiac mechanical function
US20110105931A1 (en) * 2007-11-20 2011-05-05 Siemens Medical Solutions Usa, Inc. System for Determining Patient Heart related Parameters for use in Heart Imaging
CN102397070B (zh) * 2011-01-26 2013-06-19 嘉兴医孚软件有限公司 一种全自动分割量化心脏磁共振图像左心室的方法
US9402549B2 (en) * 2013-11-27 2016-08-02 General Electric Company Methods and apparatus to estimate ventricular volumes
US9949643B2 (en) * 2014-10-18 2018-04-24 International Business Machines Corporation Automatic visualization of regional functional parameters of left ventricle from cardiac imaging
JP6993334B2 (ja) * 2015-11-29 2022-01-13 アーテリーズ インコーポレイテッド 自動化された心臓ボリュームセグメンテーション
KR102078525B1 (ko) * 2017-11-22 2020-02-19 서울대학교병원 스마트장치를 이용한 심혈관 질환의 진단정보 제공방법 및 이를 위한 심음 애플리케이션
US10595727B2 (en) * 2018-01-25 2020-03-24 Siemens Healthcare Gmbh Machine learning-based segmentation for cardiac medical imaging
CN109685803B (zh) * 2018-12-14 2020-10-23 深圳先进技术研究院 一种左心室图像分割方法、装置、设备及存储介质
CN109785334A (zh) * 2018-12-17 2019-05-21 深圳先进技术研究院 心脏磁共振图像分割方法、装置、终端设备及存储介质
CN109584254B (zh) * 2019-01-07 2022-12-20 浙江大学 一种基于深层全卷积神经网络的心脏左心室分割方法
CN110197492A (zh) * 2019-05-23 2019-09-03 山东师范大学 一种心脏mri左心室分割方法及系统
CN110163876B (zh) * 2019-05-24 2021-08-17 山东师范大学 基于多特征融合的左心室分割方法、系统、设备及介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1914617A (zh) * 2004-02-03 2007-02-14 美国西门子医疗解决公司 与心脏有关的疾病和病症的自动诊断和决策支持系统和方法
US20080319308A1 (en) * 2007-05-22 2008-12-25 Worcester Polytechnic Institute Patient-specific image-based computational modeling and techniques for human heart surgery optimization
CN109285157A (zh) * 2018-07-24 2019-01-29 深圳先进技术研究院 左心室心肌分割方法、装置及计算机可读存储介质
CN109598727A (zh) * 2018-11-28 2019-04-09 北京工业大学 一种基于深度神经网络的ct图像肺实质三维语义分割方法

Also Published As

Publication number Publication date
CN110731777A (zh) 2020-01-31
CN110731777B (zh) 2023-07-25

Similar Documents

Publication Publication Date Title
JP7058373B2 (ja) 医療画像に対する病変の検出及び位置決め方法、装置、デバイス、及び記憶媒体
US20200104984A1 (en) Methods and devices for reducing dimension of eigenvectors
CN108629784A (zh) 一种基于深度学习的ct图像颅内血管分割方法及系统
CN104899579A (zh) 人脸识别方法和装置
CN110765882B (zh) 一种视频标签确定方法、装置、服务器及存储介质
US20220207744A1 (en) Image processing method and apparatus
Dezaki et al. Deep residual recurrent neural networks for characterisation of cardiac cycle phase from echocardiograms
CN109817339A (zh) 基于大数据的患者分组方法和装置
CN112926537B (zh) 图像处理方法、装置、电子设备以及存储介质
US20220084679A1 (en) Deep neural network pre-training method for classifying electrocardiogram (ecg) data
Xie et al. Optic disc and cup image segmentation utilizing contour-based transformation and sequence labeling networks
CN114783580B (zh) 一种医疗数据质量评估方法及系统
CN114886404A (zh) 一种心律数据分类方法、装置、电子设备及存储介质
WO2021051555A1 (zh) 基于图像识别的左心室测量方法、装置以及计算机设备
WO2021000674A1 (zh) 细胞图片识别方法、系统、计算机设备及可读存储介质
CN113052236A (zh) 一种基于NASNet的肺炎图像分类方法
CN113469091A (zh) 人脸识别方法、训练方法、电子设备及存储介质
CN117036894A (zh) 基于深度学习的多模态数据分类方法、装置及计算机设备
CN112541900A (zh) 基于卷积神经网络的检测方法、装置、计算机设备及存储介质
CN116433970A (zh) 甲状腺结节分类方法、系统、智能终端及存储介质
Xie et al. Automatic neuron tracing in volumetric microscopy images with anisotropic path searching
CN114864093B (zh) 基于眼底图像进行疾病预测的装置、方法及存储介质
Zhang et al. An intelligent method of cancer prediction based on mobile cloud computing
CN114462522A (zh) 肺癌生存期预测模型训练、预测方法、系统、装置及介质
JP2024508852A (ja) 医療画像における病変分析方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19946225

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 20.07.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19946225

Country of ref document: EP

Kind code of ref document: A1