WO2020211293A1 - 一种图像分割方法及装置、电子设备和存储介质 - Google Patents
一种图像分割方法及装置、电子设备和存储介质 Download PDFInfo
- Publication number
- WO2020211293A1 WO2020211293A1 PCT/CN2019/107850 CN2019107850W WO2020211293A1 WO 2020211293 A1 WO2020211293 A1 WO 2020211293A1 CN 2019107850 W CN2019107850 W CN 2019107850W WO 2020211293 A1 WO2020211293 A1 WO 2020211293A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- lung
- data
- lung lobe
- loss function
- segmentation network
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 85
- 238000003709 image segmentation Methods 0.000 title claims abstract description 42
- 238000003860 storage Methods 0.000 title claims abstract description 28
- 210000004072 lung Anatomy 0.000 claims abstract description 440
- 230000011218 segmentation Effects 0.000 claims abstract description 181
- 238000012545 processing Methods 0.000 claims description 126
- 238000005070 sampling Methods 0.000 claims description 116
- 230000006870 function Effects 0.000 claims description 108
- 238000012549 training Methods 0.000 claims description 55
- 238000004590 computer program Methods 0.000 claims description 15
- 230000004992 fission Effects 0.000 claims description 4
- 230000003902 lesion Effects 0.000 abstract description 14
- 208000034693 Laceration Diseases 0.000 abstract 1
- 230000008569 process Effects 0.000 description 18
- 238000010586 diagram Methods 0.000 description 17
- 238000005516 engineering process Methods 0.000 description 11
- 238000004891 communication Methods 0.000 description 10
- 230000009471 action Effects 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 4
- 230000003993 interaction Effects 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 230000005236 sound signal Effects 0.000 description 4
- 208000019693 Lung disease Diseases 0.000 description 3
- 230000001413 cellular effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 229920001690 polydopamine Polymers 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000003759 clinical diagnosis Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 230000002685 pulmonary effect Effects 0.000 description 2
- 230000000241 respiratory effect Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000011282 treatment Methods 0.000 description 2
- 230000002792 vascular Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 210000004204 blood vessel Anatomy 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 210000003437 trachea Anatomy 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
Definitions
- the present disclosure relates to the field of computer vision technology, and in particular to an image segmentation method and device, electronic equipment and storage medium.
- the present disclosure proposes a technical solution for image segmentation.
- an image segmentation method including:
- the location of the target lung lobe in the lung image is determined.
- the lung lobe segmentation network does not rely on manual positioning, but is an adaptive segmentation network trained based on lung lobe data and lung fissure data. Therefore, based on the segmentation network, lung lobe positions can be accurately determined, so as to locate the lesion in time .
- the obtaining the lung lobe segmentation network according to the lung lobe data and the lung fissure data in the lung image includes:
- the determining the location of the target lung lobe in the lung image according to the lung lobe segmentation network includes:
- the location of the target lung lobe in the lung image is determined.
- the manually labeled lung fissure data is added to the input data to perform network training together with the lung lobe data, which can improve the accuracy of segmentation. Since the lung fissure data is used to identify the boundary information of the lung lobes, the lung fissure data is assisted in the training of the lung lobe segmentation network, which strengthens the feature extraction of the lung lobe boundary by the lung lobe segmentation network, so that the lung lobe segmentation after the training is used.
- the network can perform image segmentation more accurately to determine the location of the lung lobes from the lung image, so that the lesion can be located in time according to the location of the lung lobes.
- the use of the lung fissure data in the training of a lung lobe segmentation network containing the lung lobe data to obtain a trained lung lobe segmentation network includes:
- the lung lobe segmentation network is trained through the back propagation of the loss function to obtain a trained lung lobe segmentation network.
- lung fissures are added to the input data for network training, combined with lung fissure data and lung lobe data to obtain a mixed loss function, and the lung lobe segmentation network is trained through the back propagation of the loss function.
- the obtained lung lobe segmentation network can improve the accuracy of segmentation.
- the lung lobe segmentation network after the training can perform image segmentation more accurately to determine the position of the lung lobe from the lung image, so that the lesion can be located in time according to the position of the lung lobe.
- the method further includes: before performing the back propagation of the loss function according to the mixed loss function obtained by combining the lung fissure data and the lung lobe data,
- the hybrid loss function is obtained according to the first loss function, the second loss function, and the third loss function.
- the mixed loss function obtained according to the obtained multiple loss functions is more accurate, and the back propagation of the mixed loss function
- the lung lobe segmentation network is trained, and the lung lobe segmentation network obtained through this training method can improve the segmentation accuracy.
- the lung lobe segmentation network after the training can perform image segmentation more accurately to determine the position of the lung lobe from the lung image, so that the lesion can be located in time according to the position of the lung lobe.
- the method before using the lung fissure data for training of a lung lobe segmentation network containing the lung lobe data, the method further includes:
- the down-sampling processing results and the up-sampling processing results of the same level are subjected to skip connection processing until the processing of all levels is completed, and multi-layer output results corresponding to different resolutions and multi-scale sizes are obtained.
- the multi-layer output result includes: first voxel data used to identify the lung fissure data, and/or second voxel data used to identify the lung lobe data;
- the method further includes: using the first voxel data and/or the second voxel data as training data for training the lung lobe segmentation network.
- multi-level down-sampling processing and corresponding up-sampling processing are performed on the lung image, and the down-sampling processing and up-sampling processing of the same level are skip-connected to obtain corresponding different resolutions and multi-scale sizes.
- the multi-layer output results can improve the segmentation accuracy of the lung lobe segmentation network.
- the lung lobe segmentation network after the training can perform image segmentation more accurately to determine the position of the lung lobe from the lung image, so that the lesion can be located in time according to the position of the lung lobe.
- the skip connection processing of the down-sampling processing result and the up-sampling processing result of the same level includes:
- the down-sampling processing result of the same level and the features of the same scale in the up-sampling processing result are merged to obtain the jump processing result.
- the down-sampling processing result of the same level and the up-sampling processing result of the same scale are merged, which can improve the segmentation accuracy of the lung lobe segmentation network.
- the lung lobe segmentation network after the training can perform image segmentation more accurately to determine the position of the lung lobe from the lung image, so that the lesion can be located in time according to the position of the lung lobe.
- an image segmentation device including:
- the segmentation network obtaining unit is used to obtain the lung lobe segmentation network according to the lung lobe data and the lung fissure data in the lung image;
- the position determining unit is configured to determine the position of the target lung lobe in the lung image according to the lung lobe segmentation network.
- the segmentation network obtaining unit is further configured to:
- the position determining unit is further used for:
- the location of the target lung lobe in the lung image is determined.
- the segmentation network obtaining unit is further configured to:
- the lung lobe segmentation network is trained through the back propagation of the loss function to obtain a trained lung lobe segmentation network.
- the device further includes: a hybrid loss function determining unit, configured to:
- the hybrid loss function is obtained according to the first loss function, the second loss function, and the third loss function.
- the device further includes: a data processing unit, configured to:
- the down-sampling processing results and the up-sampling processing results of the same level are subjected to skip connection processing until the processing of all levels is completed, and multi-layer output results corresponding to different resolutions and multi-scale sizes are obtained.
- the multi-layer output result includes: first voxel data used to identify the lung fissure data, and/or second voxel data used to identify the lung lobe data;
- the device further includes: a data determining unit configured to use the first voxel data and/or the second voxel data as training data for training the lung lobe segmentation network.
- the data processing unit is further configured to:
- the down-sampling processing result of the same level and the features of the same scale in the up-sampling processing result are merged to obtain the jump processing result.
- an electronic device including:
- a memory for storing processor executable instructions
- the processor is configured to execute the above-mentioned image segmentation method.
- a computer-readable storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, the foregoing image segmentation method is implemented.
- the lung lobe segmentation network is obtained according to the lung lobe data and the lung fissure data in the lung image, and the location of the target lung lobe in the lung image is determined according to the lung lobe segmentation network. Since the lung lobe segmentation network does not rely on manual positioning, it is an adaptive segmentation network trained based on lung lobe data and lung fissure data. Therefore, based on the segmentation network, lung lobe positions can be accurately determined, so as to locate the lesion in time.
- Fig. 1 shows a flowchart of an image segmentation method according to an embodiment of the present disclosure.
- Fig. 2 shows a flowchart of an image segmentation method according to an embodiment of the present disclosure.
- Fig. 3 shows a flowchart of an image segmentation method according to an embodiment of the present disclosure.
- Fig. 4 shows a flowchart of a training process according to an embodiment of the present disclosure.
- Fig. 5 shows a block diagram of an image segmentation device according to an embodiment of the present disclosure.
- Fig. 6 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
- FIG. 7 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
- the position of the lung lobe can be determined by human eye recognition. Respiratory doctors often use the infected lobes to assess the severity of the disease and formulate treatments.
- imaging methods can be used to determine the location of the lung lobes. The radiologist will look for nearby slices when encountering lung disease or lesions to determine the diseased lobe. The location of such lesions usually leads to diagnostic errors due to the invisible pulmonary fissures.
- FIG. 1 shows a flowchart of an image segmentation method according to an embodiment of the present disclosure.
- the image segmentation method is applied to an image segmentation device.
- the image segmentation device may be executed by a terminal device or a server or other processing device, where the terminal device may be User Equipment (UE, User Equipment), mobile devices, cellular phones, cordless phones, personal digital assistants (PDAs, Personal Digital Assistant), handheld devices, computing devices, vehicle-mounted devices, wearable devices, etc.
- the image segmentation method can be implemented by a processor calling computer-readable instructions stored in a memory. As shown in Figure 1, the process includes:
- Step S101 Obtain a lung lobe segmentation network according to the lung lobe data and the lung fissure data in the lung image.
- the lung image may be a CT image taken by a hospital
- the lung lobe data and lung fissure data may be manually labeled lung lobe and lung fissure data
- training is performed based on the manually labeled lung lobe and lung fissure data Lung lobe segmentation network to obtain the lung lobe segmentation network after training.
- the lung fissure data is used in the training of the lung lobe segmentation network that contains the lung lobe data, instead of relying only on the lung lobe data itself. Because the lung fissure data is used to identify the boundary information of the lung lobes, the lung fissure data is assisted in the lung lobe segmentation In the network training, the feature extraction of the lung lobe boundary by the lung lobe segmentation network is strengthened, so that the lung lobe segmentation network after the training can perform image segmentation more accurately to determine the position of the lung lobe from the lung image.
- Step S102 Determine the location of the target lung lobe in the lung image according to the lung lobe segmentation network.
- the lung lobe segmentation network is trained based on the manually labeled lung lobe and lung fissure data. After the trained lung lobe segmentation network is obtained, the target lung lobe in the lung image can be determined according to the trained lung lobe segmentation network Location.
- the human lung is divided into five lung lobes, of which the right lung has three lobes, namely the right upper lobe (RUL, right upper lobe), the right middle lobe (RML, right middle lobe) and the right lower lobe (RLL, right lower lobe), separated by small lung fissure and large lung fissure.
- the left lung has two lobes, the left upper lobe (LUL, left upper lobe) and the left lower lobe (LLL, left lower lobe), which are separated by the large lung fissure. These five lung lobes are functionally independent and have their own bronchial and vascular systems. According to the trained lung lobe segmentation network, the positions of the five lung lobes in the lung image can be determined.
- FIG. 2 shows a flowchart of an image segmentation method according to an embodiment of the present disclosure.
- the image segmentation method is applied to an image segmentation device.
- the image segmentation device may be executed by a terminal device or a server or other processing device, where the terminal device may be User Equipment (UE, User Equipment), mobile devices, cellular phones, cordless phones, personal digital assistants (PDAs, Personal Digital Assistant), handheld devices, computing devices, vehicle-mounted devices, wearable devices, etc.
- the image segmentation method can be implemented by a processor calling computer-readable instructions stored in a memory. As shown in Figure 2, the process includes:
- Step S201 Perform back propagation of the loss function according to the mixed loss function obtained by combining the lung fissure data and the lung lobe data.
- Step S202 Train the lung lobe segmentation network through the back propagation of the loss function to obtain the trained lung lobe segmentation network.
- a mixed loss function combining lung lobe Dice plus lung lobe and lung fissure Cross Entropy can be used as the loss function, and the back propagation of the loss function is used to train each parameter in the lung lobe segmentation network, namely The lung lobe segmentation network performs parameter tuning.
- the first loss function (such as D lobe ) and the second loss function (such as D lobe ) and the second loss function (such as D lobe ) can be obtained from the lung lobe data before performing the back propagation of the loss function according to the mixed loss function obtained by combining the lung fissure data and the lung lobe data.
- Loss function (such as H(p,q) lobe ).
- the third loss function (such as H(p,q) fissure ) is obtained.
- the hybrid loss function is obtained according to the first loss function, the second loss function, and the third loss function.
- a trained lung lobe segmentation network can be obtained according to the manually labeled lung lobe data and lung fissure data. Since the lung fissure data is used to identify the boundary information of the lung lobes, the lung fissure data is assisted in the training of the lung lobe segmentation network, which strengthens the feature extraction of the lung lobe boundary by the lung lobe segmentation network, so that the trained lung lobe segmentation network is used Image segmentation can be performed more accurately to determine the location of lung lobes from lung images.
- lung fissures are used as input for network training, which strengthens the sensitivity of the network model to the location of the lung fissures and raises the degree of attention at the boundaries of the lung lobes. It can improve the segmentation effect at the boundary of different lung lobes and reduce the blur of the boundary.
- Step S203 Determine the location of the target lung lobe in the lung image according to the trained lung lobe segmentation network.
- the lung lobe segmentation network is trained based on the manually labeled lung lobe and lung fissure data. After the trained lung lobe segmentation network is obtained, the target lung lobe in the lung image can be determined according to the trained lung lobe segmentation network Location.
- the human lung is divided into five lung lobes, of which the right lung has three lobes, namely the right upper lobe (RUL, right upper lobe), the right middle lobe (RML, right middle lobe) and the right lower lobe (RLL, right lower lobe), separated by small lung fissure and large lung fissure.
- the left lung has two lobes, the left upper lobe (LUL, left upper lobe) and the left lower lobe (LLL, left lower lobe), which are separated by the large lung fissure. These five lung lobes are functionally independent, and have their own bronchial and vascular systems. According to the trained lung lobe segmentation network, the positions of the five lung lobes in the lung image can be determined.
- FIG. 3 shows a flowchart of an image segmentation method according to an embodiment of the present disclosure.
- the image segmentation method is applied to an image segmentation device.
- the image segmentation device may be executed by a terminal device or a server or other processing device, where the terminal device may be User Equipment (UE, User Equipment), mobile devices, cellular phones, cordless phones, personal digital assistants (PDAs, Personal Digital Assistant), handheld devices, computing devices, vehicle-mounted devices, wearable devices, etc.
- the image segmentation method can be implemented by a processor calling computer-readable instructions stored in a memory. As shown in Figure 3, the process includes:
- Step S301 Input the lung image into the lung lobe segmentation network, and perform multi-level down-sampling processing and corresponding up-sampling processing on the lung image to obtain down-sampling processing results and up-sampling processing results corresponding to different levels.
- Sampling Collecting samples of analog signals. Sampling is to convert signals that are continuous in time and amplitude into discrete signals in time and amplitude under the action of sampling pulses. Sampling is also called the discretization process of the waveform.
- down-sampling for a sample sequence, sampling once at intervals of several samples, and the new sequence obtained is the down-sampling of the original sequence.
- the main purpose of reducing the image is twofold: 1. Make the image fit the size of the display area; 2. Generate a thumbnail of the corresponding image.
- Upsampling It is the inverse process of downsampling. The essence of upsampling is interpolation or difference.
- Image magnification almost always uses interpolation methods, that is, on the basis of the original image pixels, appropriate interpolation algorithms are used to insert new elements between pixels.
- the main purpose of magnifying an image is to magnify the original image so that it can be displayed on a higher resolution display device. It should be pointed out that both up-sampling and down-sampling are re-acquisition of digital signals, and the re-acquisition sampling rate is compared with the original sampling rate of the digital signal (for example, sampled from an analog signal), which is greater than the original signal. It is up-sampling; those that are smaller than the original signal are called down-sampling.
- multiple down-sampling and corresponding up-sampling and skip connections may be: input a CT of a lung image, perform the first-level down-sampling on the CT, and obtain the first down-sampling result, The first down-sampling result is subjected to the second-level down-sampling, and the second down-sampling result is obtained.
- multi-level down-sampling is performed (this disclosure is not limited to four-level down-sampling).
- Taking four-level downsampling as an example perform third down-sampling on the second down-sampling result to obtain the third down-sampling result, and perform fourth-level down-sampling on the third down-sampling result to obtain the fourth down-sampling result, down-sampling
- Upsampling for the second time to get the second upsampling result upsampling the second upsampling result for the third time to get the third upsampling result, and upsampling the third upsampling result for the fourth time to get the fourth upsampling result Sampling results.
- Jump connection is for the same level.
- the down-sampling "first down-sampling result” corresponds to the up-sampling "third up-sampling result”. Therefore, the first down-sampling result and the third up-sampling result are Jump connection.
- Step S302 Perform skip connection processing on the down-sampling processing result and the up-sampling processing result at the same level, until the processing of all levels is completed, to obtain multi-layer output results corresponding to different resolutions and multi-scale sizes.
- the down-sampling processing result of the same level and the features of the same scale in the up-sampling processing result are merged to obtain the jump processing result.
- the segmentation accuracy is improved.
- the multi-layer output result includes: first voxel data for identifying the lung fissure data, and/or second voxel data for identifying the lung lobe data.
- Step S303 Use the first voxel data for identifying lung fissure data and the second voxel data for identifying lung lobe data as the mixed loss function obtained from the training data, and perform the back propagation of the loss function, and pass the loss function Backpropagation trains the lung lobe segmentation network to obtain the trained lung lobe segmentation network.
- down-sampling can also be performed before the training data is input into the lung lobe segmentation network to reduce the amount of data, and use limited computing resources to segment the entire lung. Under the premise of ensuring complete data input, use A more complete network model.
- the data processing speed can be accelerated, and the segmentation speed can be controlled within 2 seconds.
- Step S304 Determine the location of the target lung lobe in the lung image according to the trained lung lobe segmentation network.
- the input lung image CT is the same, the networks used are the same network, and different procedures obtain different data, the above-mentioned step S301-step S302 can be used.
- the processing modes of the lung fissure and lung lobes segmentation processing procedures are the same. The difference in the processing is: for lung fissure data, it can be the first voxel data used to identify lung fissure data; for lung lobe data, it can be It is the second voxel data used to identify lung lobe data.
- the final processing result can be obtained according to the multi-layer output result.
- the process of lung fissure and lung lobes segmentation can be performed simultaneously.
- voxels for an image, if the image is a 2D image, the image can be described as composed of multiple pixels, and the pixels are two-dimensional; if the image is a 3D image, the image can be described as It is composed of multiple voxels.
- the voxels are three-dimensional. In a 3D image, the volume is divided into evenly spaced rows and columns, covering all three different directions (up and down, left and right, inside and outside). This divides the 3D space into cubes, also called voxels (volume elements or volume pixels). Each voxel is defined by a three-dimensional coordinate and the color at that coordinate.
- the lung lobe segmentation network is an end-to-end 3D segmentation network (or called a VNet-based 3D convolutional neural network).
- the present disclosure adopts an end-to-end 3D segmentation network structure to segment the lungs as a whole, which improves spatial perception and can extract more spatial information, thereby improving the segmentation results of each lung lobe.
- the right middle lung lobe has a variable shape and position, and the prediction accuracy is not high. With the present disclosure, the right middle lung lobe region can be accurately segmented. Since the network structure is trained through the back propagation of the mixed loss function, it is a deep learning model.
- the lung image is first input into the lung lobe segmentation network, the lung image is subjected to multi-level down-sampling processing and corresponding up-sampling processing, and the down-sampling processing result and the up-sampling processing result of the same level are subjected to skip connection processing , Until the end of the processing of all levels, the multi-layer output results corresponding to different resolutions and multi-scale sizes are obtained.
- the down-sampling processing result of the same level and the features of the same scale in the up-sampling processing result are merged to obtain the jump processing result, and the segmentation accuracy is improved by fusing the convolution features of the same scale.
- the lung fissure data and the lung lobe data are used for network training, and the loss function is back propagated according to the mixed loss function obtained by combining the lung fissure data and the lung lobe data to realize the training of the lung lobe segmentation network.
- the lung lobe segmentation network obtained through this training method can improve the segmentation accuracy.
- the lung lobe segmentation network after the training can perform image segmentation more accurately to determine the position of the lung lobe from the lung image, so that the lesion can be located in time according to the position of the lung lobe.
- the training process of the lung lobe segmentation network includes the following two processes, the segmentation of the entire lung lobe and the segmentation process of the lung fissure.
- the input lung image CT is the same, and the network used is the same network.
- Fig. 4 is a schematic diagram of the training process according to the present disclosure.
- the lung lobe segmentation network based on the end-to-end 3D structure performs lung lobe segmentation and lung fissure segmentation respectively.
- the lung image 111 is the input data of the lung lobe segmentation network, and the lung image 111 can be 3D CT data.
- the output data of the lung lobe segmentation network can be obtained.
- the output data includes two types: lung lobe data 112 and lung fissure data 113.
- the lung lobe segmentation network is trained according to the lung fissure data 113 and the lung lobe data 112. Lung lobe segmentation and lung fissure segmentation can be divided into the following two parts.
- the input of the lung lobe segmentation network is the lung 3D CT data, and multiple down-sampling and corresponding up-sampling processes and skip connections are performed in the lung lobe segmentation network.
- the input original data is a single-channel grayscale image of z ⁇ x ⁇ y, which enters the 3D segmentation network after data preprocessing, and the output is a 6-channel tensor of z ⁇ x ⁇ y size, representing each voxel Which lung lobe or background the location belongs to.
- Each hop connection merges the down-sampling in the network with its corresponding up-sampling, and this level union forms the 3D probability distribution map of the data.
- the lung fissure is used as the target result for training.
- Use end-to-end 3D structure for multi-scale convolution feature fusion In order to make full use of the local information in the lung lobe segmentation network, the multi-gate convolution network structure is used in the lung lobe segmentation network to replace the convolution block in the related technology, and the four-scale network in the lung lobe segmentation network replaces the single-scale network of the related technology. Enhance the feature fusion effect.
- the multi-gate convolution network structure cascades feature maps of different scales before each convolution layer, reducing the feature loss caused by downsampling in traditional FCN, U-Net and other network structures.
- the multi-gate network is used for 3D segmentation, and the segmentation accuracy is improved by fusing the convolution features of different scales, and it has a more accurate prediction effect when extracting the location information of the lung fissure.
- the input of the network is the same as the segmentation of lung lobes, which is a single-channel grayscale image of z ⁇ x ⁇ y, and the output is 4 channel 3D data (where 3 channels are the position information of 3 lung fissures in the human body, and 1 channel is background information).
- the present disclosure uses the mixed loss of the combination of the lobe Dice, the lobe and the cross Entropy as the loss function to optimize the parameters of the network.
- the expression of the Dice loss function is shown in formula (1).
- V represents all voxel points in the 3D image
- p i is the probability that i voxel point is predicted to be the target class, that is, the probability that the voxel point is predicted to be the target lung lobe
- l i is the actual label of the voxel point.
- Multi-Dice weighting is used to modify the boundary, and the Dice loss function during lung lobe training is shown in formula (2).
- D lobe ⁇ 1 D RUL + ⁇ 2 D RML + ⁇ 3 D RLL + ⁇ 4 D LUL + ⁇ 5 D LLL (2)
- D RUL , D RML , D RLL , D LUL , D L represent the respective Dice of the five lung lobes
- ⁇ is an adjustable coefficient used to calibrate the influence of the weight of each lung lobe in the segmentation task on the overall segmentation.
- p(x) is the probability of correct prediction
- q(x) is the probability of wrong prediction
- the loss function of the final network is calculated by formula (4):
- D lobe is the Dice loss function during lung lobe training
- H(p,q) lobe is the Cross Entropy loss function during lung lobe training
- H(p,q) fissure is the Cross Entropy loss function during lung fissure training
- ⁇ 1 , ⁇ 2 , ⁇ 3 are the weights of each loss function, which determine the influence of each part of the training result on the final segmentation result.
- the writing order of the steps does not mean a strict execution order but constitutes any limitation on the implementation process.
- the specific execution order of each step should be based on its function and possibility.
- the inner logic is determined.
- the present disclosure also provides image segmentation devices, electronic equipment, computer-readable storage media, and programs, all of which can be used to implement any image segmentation method provided by the present disclosure.
- image segmentation devices electronic equipment, computer-readable storage media, and programs, all of which can be used to implement any image segmentation method provided by the present disclosure.
- Fig. 5 shows a block diagram of an image segmentation device according to an embodiment of the present disclosure.
- the image segmentation device of an embodiment of the present disclosure includes: a segmentation network obtaining unit 31, which is used to obtain data according to the lung lobe data in the lung image and The lung fissure data obtains a lung lobe segmentation network; the position determining unit 32 is configured to determine the location of the target lung lobe in the lung image according to the lung lobe segmentation network.
- the segmentation network obtaining unit is further configured to: use the lung fissure data in the training of the lung lobe segmentation network containing the lung lobe data to obtain the trained lung lobe segmentation network; the position The determining unit is further configured to determine the location of the target lung lobe in the lung image according to the trained lung lobe segmentation network.
- the segmentation network obtaining unit is further configured to: perform back propagation of the loss function according to the mixed loss function obtained by combining the lung fissure data and the lung lobe data; The back propagation of the loss function trains the lung lobe segmentation network to obtain a trained lung lobe segmentation network.
- the device further includes: a mixed loss function determining unit, configured to: obtain a first loss function and a second loss function according to the lung lobe data; obtain a third loss function according to the lung fission data ; Obtain the hybrid loss function according to the first loss function, the second loss function and the third loss function.
- a mixed loss function determining unit configured to: obtain a first loss function and a second loss function according to the lung lobe data; obtain a third loss function according to the lung fission data ; Obtain the hybrid loss function according to the first loss function, the second loss function and the third loss function.
- the device further includes: a data processing unit, configured to input the lung image into the lung lobe segmentation network, and perform multi-level down-sampling processing and correspondence on the lung image Up-sampling processing, get the down-sampling processing results and up-sampling processing results corresponding to different levels; the down-sampling processing results and up-sampling processing results of the same level are subjected to skip connection processing until the processing of all levels is completed, and the corresponding different resolutions are obtained And multi-scale output results of multiple layers.
- a data processing unit configured to input the lung image into the lung lobe segmentation network, and perform multi-level down-sampling processing and correspondence on the lung image Up-sampling processing, get the down-sampling processing results and up-sampling processing results corresponding to different levels; the down-sampling processing results and up-sampling processing results of the same level are subjected to skip connection processing until the processing of all levels is completed, and the corresponding different resolutions are obtained And multi-scale output results of multiple
- the multi-layer output result includes: first voxel data used to identify the lung fissure data, and/or second voxel data used to identify the lung lobe data;
- the device further includes: a data determining unit, configured to use the first voxel data and/or the second voxel data as training data for training the lung lobe segmentation network.
- the data processing unit is further configured to merge features of the same scale in the down-sampling processing result of the same level and the up-sampling processing result to obtain the jump processing result.
- the functions or modules contained in the device provided in the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments.
- the functions or modules contained in the device provided in the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments.
- the embodiments of the present disclosure also provide a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above-mentioned method when executed by a processor.
- the computer-readable storage medium may be a non-volatile computer-readable storage medium.
- An embodiment of the present disclosure also provides an electronic device, including: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured as the above method.
- the electronic device can be provided as a terminal, server or other form of device.
- Fig. 6 is a block diagram showing an electronic device 800 according to an exemplary embodiment.
- the electronic device 800 may be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and other terminals.
- the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input/output (I/O) interface 812, and a sensor component 814 , And communication component 816.
- the processing component 802 generally controls the overall operations of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations.
- the processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the foregoing method.
- the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components.
- the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.
- the memory 804 is configured to store various types of data to support operations in the electronic device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc.
- the memory 804 can be implemented by any type of volatile or nonvolatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Magnetic Memory, Flash Memory, Magnetic Disk or Optical Disk.
- SRAM static random access memory
- EEPROM electrically erasable programmable read-only memory
- EPROM erasable Programmable Read Only Memory
- PROM Programmable Read Only Memory
- ROM Read Only Memory
- Magnetic Memory Flash Memory
- Magnetic Disk Magnetic Disk or Optical Disk.
- the power supply component 806 provides power for various components of the electronic device 800.
- the power supply component 806 may include a power management system, one or more power supplies, and other components associated with the generation, management, and distribution of power for the electronic device 800.
- the multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
- the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
- the touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation.
- the multimedia component 808 includes a front camera and/or a rear camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
- the audio component 810 is configured to output and/or input audio signals.
- the audio component 810 includes a microphone (MIC).
- the microphone is configured to receive external audio signals.
- the received audio signal may be further stored in the memory 804 or transmitted via the communication component 816.
- the audio component 810 further includes a speaker for outputting audio signals.
- the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module.
- the peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include but are not limited to: home button, volume button, start button, and lock button.
- the sensor component 814 includes one or more sensors for providing the electronic device 800 with various aspects of state evaluation.
- the sensor component 814 can detect the on/off status of the electronic device 800 and the relative positioning of the components.
- the component is the display and the keypad of the electronic device 800.
- the sensor component 814 can also detect the electronic device 800 or the electronic device 800.
- the position of the component changes, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration/deceleration of the electronic device 800, and the temperature change of the electronic device 800.
- the sensor component 814 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact.
- the sensor component 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications.
- the sensor component 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor or a temperature sensor.
- the communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices.
- the electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof.
- the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel.
- the communication component 816 further includes a near field communication (NFC) module to facilitate short-range communication.
- the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
- RFID radio frequency identification
- IrDA infrared data association
- UWB ultra-wideband
- Bluetooth Bluetooth
- the electronic device 800 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field A programmable gate array (FPGA), controller, microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
- ASIC application specific integrated circuits
- DSP digital signal processors
- DSPD digital signal processing devices
- PLD programmable logic devices
- FPGA field A programmable gate array
- controller microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
- a non-volatile computer-readable storage medium such as a memory 804 including computer program instructions, which can be executed by the processor 820 of the electronic device 800 to complete the foregoing method.
- Fig. 7 is a block diagram showing an electronic device 900 according to an exemplary embodiment.
- the electronic device 900 may be provided as a server.
- the electronic device 900 includes a processing component 922, which further includes one or more processors, and a memory resource represented by a memory 932, for storing instructions that can be executed by the processing component 922, such as application programs.
- the application program stored in the memory 932 may include one or more modules each corresponding to a set of instructions.
- the processing component 922 is configured to execute instructions to perform the aforementioned methods.
- the electronic device 900 may also include a power supply component 926 configured to perform power management of the electronic device 900, a wired or wireless network interface 950 configured to connect the electronic device 900 to a network, and an input output (I/O) interface 958 .
- the electronic device 900 can operate based on an operating system stored in the memory 932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
- a non-volatile computer-readable storage medium such as the memory 932 including computer program instructions, which can be executed by the processing component 922 of the electronic device 900 to complete the foregoing method.
- the present disclosure may be a system, method, and/or computer program product.
- the computer program product may include a computer-readable storage medium loaded with computer-readable program instructions for enabling a processor to implement various aspects of the present disclosure.
- the computer-readable storage medium may be a tangible device that can hold and store instructions used by the instruction execution device.
- the computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
- Computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) Or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanical encoding device, such as a printer with instructions stored thereon
- RAM random access memory
- ROM read-only memory
- EPROM erasable programmable read-only memory
- flash memory flash memory
- SRAM static random access memory
- CD-ROM compact disk read-only memory
- DVD digital versatile disk
- memory stick floppy disk
- mechanical encoding device such as a printer with instructions stored thereon
- the computer-readable storage medium used here is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (for example, light pulses through fiber optic cables), or through wires Transmission of electrical signals.
- the computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to various computing/processing devices, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and/or a wireless network.
- the network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers.
- the network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network, and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing/processing device .
- the computer program instructions used to perform the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, status setting data, or in one or more programming languages.
- Source code or object code written in any combination, the programming language includes object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as "C" language or similar programming languages.
- Computer-readable program instructions can be executed entirely on the user's computer, partly on the user's computer, executed as a stand-alone software package, partly on the user's computer and partly executed on a remote computer, or entirely on the remote computer or server carried out.
- the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to access the Internet connection).
- LAN local area network
- WAN wide area network
- an electronic circuit such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), can be customized by using the status information of the computer-readable program instructions.
- the computer-readable program instructions are executed to realize various aspects of the present disclosure.
- These computer-readable program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine such that when these instructions are executed by the processor of the computer or other programmable data processing device , A device that implements the functions/actions specified in one or more blocks in the flowchart and/or block diagram is produced. It is also possible to store these computer-readable program instructions in a computer-readable storage medium. These instructions make computers, programmable data processing apparatuses, and/or other devices work in a specific manner, so that the computer-readable medium storing instructions includes An article of manufacture, which includes instructions for implementing various aspects of the functions/actions specified in one or more blocks in the flowchart and/or block diagram.
- each block in the flowchart or block diagram may represent a module, program segment, or part of an instruction, and the module, program segment, or part of an instruction contains one or more functions for implementing the specified logical function.
- Executable instructions may also occur in a different order than the order marked in the drawings. For example, two consecutive blocks can actually be executed in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved.
- each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart can be implemented by a dedicated hardware-based system that performs the specified functions or actions Or it can be realized by a combination of dedicated hardware and computer instructions.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
一种图像分割方法及装置、电子设备和存储介质,其中,该方法包括根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络(S101),根据肺叶分割网络,确定肺部图像中目标肺叶所在的位置(S102)。采用本方法,能准确地确定肺叶位置,从而及时对病灶进行定位。
Description
本申请要求在2019年4月18日提交中国专利局、申请号为201910315130.7、发明名称为“一种图像分割方法及装置、电子设备和存储介质”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。
本公开涉及计算机视觉技术领域,尤其涉及一种图像分割方法及装置、电子设备和存储介质。
在临床诊断中,呼吸科医生经常依据被感染的肺叶来做出疾病严重性评估和制定治疗手段;而放射科医生则会在遇到肺部疾病或病变时寻找临近切片才能确定发病肺叶。此类病灶定位通常会由于肺部裂隙不可见的原因而造成诊断失误。如何确定肺叶位置以及时对病灶进行定位,是要解决的问题。然而,相关技术中未存在有效的解决方案。
发明内容
本公开提出了一种图像分割技术方案。
根据本公开的一方面,提供了一种图像分割方法,所述方法包括:
根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络;
根据所述肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
采用本公开,由于该肺叶分割网络不依赖人工定位,而是根据肺叶数据和肺裂数据训练得到的自适应分割网络,因此,基于该分割网络能准确的确定肺叶位置,从而及时对病灶进行定位。
在可能的实现方式中,所述根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络,包括:
将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练中,得到训练后的肺叶分割网络;
所述根据所述肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置,包括:
根据所述训练后的肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
采用本公开,在输入数据中加入手动标注的肺裂数据与肺叶数据一起进行网络训练,可以提高了分割准确度。由于该肺裂数据,是用于标识肺叶的边界信息,因此,将该肺裂数据辅助于肺叶分割网络训练中,强化了肺叶分割网络对肺叶边界的特征提取,使采 用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置,从而根据肺叶的位置,可以及时对病灶进行定位。
在可能的实现方式中,所述将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练中,得到训练后的肺叶分割网络,包括:
根据所述肺裂数据和所述肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播;
通过所述损失函数的反向传播对所述肺叶分割网络进行训练,得到训练后的肺叶分割网络。
采用本公开,在输入数据中加入手动标注的肺裂进行网络训练,结合肺裂数据和肺叶数据得到混合损失函数,通过损失函数的反向传播对肺叶分割网络进行训练,通过这种训练方式所得到的肺叶分割网络,可以提高分割准确度。使采用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置,从而根据肺叶的位置,可以及时对病灶进行定位。
在可能的实现方式中,所述方法还包括:根据所述肺裂数据和所述肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播之前,
根据所述肺叶数据得到第一损失函数和第二损失函数;
根据所述肺裂数据得到第三损失函数;
根据所述第一损失函数、所述第二损失函数和所述第三损失函数得到所述混合损失函数。
采用本公开,分别根据所述肺叶数据和所述肺裂数据得到各自的损失函数后,根据所得到的多个损失函数得到的混合损失函数更为准确,通过该混合损失函数的反向传播对肺叶分割网络进行训练,通过这种训练方式所得到的肺叶分割网络,可以提高分割准确度。使采用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置,从而根据肺叶的位置,可以及时对病灶进行定位。
在可能的实现方式中,所述将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练之前,还包括:
将所述肺部图像输入所述肺叶分割网络中,对所述肺部图像进行多层级的下采样处理和对应的上采样处理,得到对应不同层级的下采样处理结果和上采样处理结果;
将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,直至对所有层级处理结束,得到对应不同分辨率和多尺度大小的多层输出结果。
在可能的实现方式中,所述多层输出结果包括:用于标识所述肺裂数据的第一体素数据,和/或用于标识所述肺叶数据的第二体素数据;
所述方法还包括:将所述第一体素数据和/或所述第二体素数据作为用于训练所述肺叶分割网络的训练数据。
采用本公开,对所述肺部图像进行多层级的下采样处理和对应的上采样处理,以及针对同一层级的下采样处理和上采样处理进行跳跃连接,可以得到对应不同分辨率和多尺度大小的多层输出结果,可以提高肺叶分割网络的分割精度。使采用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置,从而根据肺叶的位置,可以及时对病灶进行定位。
在可能的实现方式中,所述将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,包括:
将所述同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,得到跳跃处理结果。
采用本公开,对于多层输出结果中的每一层,将同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,可以提高肺叶分割网络的分割精度。使采用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置,从而根据肺叶的位置,可以及时对病灶进行定位。
根据本公开的一方面,提供了一种图像分割装置,所述装置包括:
分割网络获得单元,用于根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络;
位置确定单元,用于根据所述肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
在可能的实现方式中,所述分割网络获得单元,进一步用于:
将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练中,得到训练后的肺叶分割网络;
所述位置确定单元,进一步用于:
根据所述训练后的肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
在可能的实现方式中,所述分割网络获得单元,进一步用于:
根据所述肺裂数据和所述肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播;
通过所述损失函数的反向传播对所述肺叶分割网络进行训练,得到训练后的肺叶分割网络。
在可能的实现方式中,所述装置还包括:混合损失函数确定单元,用于:
根据所述肺叶数据得到第一损失函数和第二损失函数;
根据所述肺裂数据得到第三损失函数;
根据所述第一损失函数、所述第二损失函数和所述第三损失函数得到所述混合损失函数。
在可能的实现方式中,所述装置还包括:数据处理单元,用于:
将所述肺部图像输入所述肺叶分割网络中,对所述肺部图像进行多层级的下采样处理和对应的上采样处理,得到对应不同层级的下采样处理结果和上采样处理结果;
将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,直至对所有层级处理结束,得到对应不同分辨率和多尺度大小的多层输出结果。
在可能的实现方式中,所述多层输出结果包括:用于标识所述肺裂数据的第一体素数据,和/或用于标识所述肺叶数据的第二体素数据;
所述装置还包括:数据确定单元,用于:将所述第一体素数据和/或所述第二体素数据作为用于训练所述肺叶分割网络的训练数据。
在可能的实现方式中,所述数据处理单元,进一步用于:
将所述同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,得到跳跃处理结果。
根据本公开的一方面,提供了一种电子设备,包括:
处理器;
用于存储处理器可执行指令的存储器;
其中,所述处理器被配置为:执行上述图像分割方法。
根据本公开的一方面,提供了一种计算机可读存储介质,其上存储有计算机程序指令,所述计算机程序指令被处理器执行时实现上述图像分割方法。
在本公开实施例中,根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络,根据该肺叶分割网络,确定肺部图像中目标肺叶所在的位置。由于该肺叶分割网络不依赖人工定位,而是根据肺叶数据和肺裂数据训练得到的自适应分割网络,因此,基于该分割网络能准确的确定肺叶位置,从而及时对病灶进行定位。
应当理解的是,以上的一般描述和后文的细节描述仅是示例性和解释性的,而非限制本公开。
根据下面参考附图对示例性实施例的详细说明,本公开的其它特征及方面将变得清楚。
此处的附图被并入说明书中并构成本说明书的一部分,这些附图示出了符合本公开的实施例,并与说明书一起用于说明本公开的技术方案。
图1示出根据本公开实施例的图像分割方法的流程图。
图2示出根据本公开实施例的图像分割方法的流程图。
图3示出根据本公开实施例的图像分割方法的流程图。
图4示出根据本公开实施例的训练过程的流程图。
图5示出根据本公开实施例的图像分割装置的框图。
图6示出根据本公开实施例的电子设备的框图。
图7示出根据本公开实施例的电子设备的框图。
以下将参考附图详细说明本公开的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面,但是除非特别指出,不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
本文中术语“和/或”,仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。另外,本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合,例如,包括A、B、C中的至少一种,可以表示包括从A、B和C构成的集合中选择的任意一个或多个元素。
另外,为了更好的说明本公开,在下文的具体实施方式中给出了众多的具体细节。本领域技术人员应当理解,没有某些具体细节,本公开同样可以实施。在一些实例中,对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述,以便于凸显本公开的主旨。
在临床诊断中,一方面,可以通过人眼识别来确定肺叶的位置。呼吸科医生经常依据被感染的肺叶来做出疾病严重性评估和制定治疗手段。另一方面,可以通过影像学方法来确定肺叶的位置。放射科医生会在遇到肺部疾病或病变时寻找临近切片才能确定发病肺叶。此类病灶定位通常会由于肺部裂隙不可见的原因而造成诊断失误。相关技术中,在进行肺叶分割时,需要依赖预先的气管和血管来分割,或是需要用户的交互从而优化分割结果,即需要依赖医生的人工操作来予以识别和优化,因此,不仅得到的肺叶分割结果较差,而且分割速度很慢。
综上所述,无论是采用人眼识别,还是使用影像学方法来确定肺叶的位置,都面临着以下问题:1、大多数肺裂是不完整的,经常无法延伸到肺部边缘,相关研究已经确认 了肺裂的不完整是常见现象;2、肺叶边缘的视觉特征会由于病理学因素影响而产生变化,这些视觉特征包括厚度、位置和形状;3、肺部存在其他裂隙(如副裂和奇裂)可能被误识为大小肺裂。
构建一种可靠且全自动化的肺叶分割网络来确定肺叶的位置,对肺部疾病的诊断、评估和量化有着重要的意义。全自动的肺叶分割方法,还会帮助医生们减少病灶定位的时间并提高定位准确度。
图1示出根据本公开实施例的图像分割方法的流程图,该图像分割方法应用于图像分割装置,例如,图像分割装置可以由终端设备或服务器或其它处理设备执行,其中,终端设备可以为用户设备(UE,User Equipment)、移动设备、蜂窝电话、无绳电话、个人数字处理(PDA,Personal Digital Assistant)、手持设备、计算设备、车载设备、可穿戴设备等。在一些可能的实现方式中,该图像分割方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。如图1所示,该流程包括:
步骤S101、根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络。
本公开一可能的实现方式中,肺部图像可以是医院拍CT的图像,该肺叶数据和肺裂数据可以是手工标注的肺叶和肺裂数据,根据该手工标注的肺叶和肺裂数据来训练肺叶分割网络,得到训练后的肺叶分割网络。
需要指出的是,在肺叶分割的任务中,需要提高分割精度,降低假阳性,区分不同肺叶位置。由于成像和自然生理原因,部分CT影像中可能没有可见的肺裂,视觉上无法区分不同的肺叶,会产生误分割的可能。为此,在输入数据中加入手动标注的肺裂进行网络训练,通过这种方式,可以提高了分割准确度。输入肺叶分割网络的训练数据,不仅包括肺叶数据,还包括肺裂数据。将肺裂数据用于包含肺叶数据的肺叶分割网络训练中,而不是只依赖肺叶数据本身,由于该肺裂数据,是用于标识肺叶的边界信息,因此,将该肺裂数据辅助于肺叶分割网络训练中,强化了肺叶分割网络对肺叶边界的特征提取,使采用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置。
步骤S102、根据肺叶分割网络,确定肺部图像中目标肺叶所在的位置。
本公开一可能的实现方式中,根据该手工标注的肺叶和肺裂数据来训练肺叶分割网络,得到训练后的肺叶分割网络后,可根据训练后的肺叶分割网络,确定肺部图像中目标肺叶所在的位置。就目标肺叶而言,人类的肺被分成五个肺叶,其中右肺有三个肺叶,为右上肺叶(RUL,right upper lobe)、右中肺叶(RML,right middle lobe)和右下肺叶(RLL,right lower lobe),分别被小肺裂和大肺裂所分隔。而左肺有两个肺叶,为左上肺叶(LUL,left upper lobe)和左下肺叶(LLL,left lower lobe),被大肺裂所分隔。这 五个肺叶分别功能性独立,而且有各自的支气管和血管***。根据该训练后的肺叶分割网络,可以确定肺部图像中这五个肺叶在肺部图像中所在的位置。
图2示出根据本公开实施例的图像分割方法的流程图,该图像分割方法应用于图像分割装置,例如,图像分割装置可以由终端设备或服务器或其它处理设备执行,其中,终端设备可以为用户设备(UE,User Equipment)、移动设备、蜂窝电话、无绳电话、个人数字处理(PDA,Personal Digital Assistant)、手持设备、计算设备、车载设备、可穿戴设备等。在一些可能的实现方式中,该图像分割方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。如图2所示,该流程包括:
步骤S201、根据肺裂数据和肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播。
步骤S202、通过损失函数的反向传播对肺叶分割网络进行训练,得到训练后的肺叶分割网络。
本公开一可能实现方式中,可以采用肺叶Dice加肺叶和肺裂Cross Entropy相结合的混合损失函数作为该损失函数,通过该损失函数的反向传播来训练肺叶分割网络中的各个参数,即对该肺叶分割网络进行参数调优。
本公开一可能实现方式中,可以根据肺裂数据和肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播之前,根据肺叶数据得到第一损失函数(如D
lobe)和第二损失函数(如H(p,q)
lobe)。根据肺裂数据得到第三损失函数(如H(p,q)
fissure)。根据所述第一损失函数、所述第二损失函数和所述第三损失函数得到所述混合损失函数。如何计算混合损失函数具体的运算过程在后续的应用示例中具体阐述。
通过步骤S201-步骤S202,可以根据手工标注的肺叶数据和肺裂数据得到训练后的肺叶分割网络。由于该肺裂数据是用于标识肺叶的边界信息,因此,将该肺裂数据辅助于肺叶分割网络训练中,强化了肺叶分割网络对肺叶边界的特征提取,使采用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置。换言之,在肺叶以外,使用手动标注的肺裂作为输入进行网络训练,加强了网络模型对肺裂位置的敏感度,提高在肺叶分界处的关注度。可以提高不同肺叶边界处的分割效果,减少边界模糊。
步骤S203、根据训练后的肺叶分割网络,确定肺部图像中目标肺叶所在的位置。
本公开一可能的实现方式中,根据该手工标注的肺叶和肺裂数据来训练肺叶分割网络,得到训练后的肺叶分割网络后,可根据训练后的肺叶分割网络,确定肺部图像中目标肺叶所在的位置。就目标肺叶而言,人类的肺被分成五个肺叶,其中右肺有三个肺叶,为右上肺叶(RUL,right upper lobe)、右中肺叶(RML,right middle lobe)和右下肺叶 (RLL,right lower lobe),分别被小肺裂和大肺裂所分隔。而左肺有两个肺叶,为左上肺叶(LUL,left upper lobe)和左下肺叶(LLL,left lower lobe),被大肺裂所分隔。这五个肺叶分别功能性独立,而且有各自的支气管和血管***。根据该训练后的肺叶分割网络,可以确定肺部图像中这五个肺叶在肺部图像中所在的位置。
图3示出根据本公开实施例的图像分割方法的流程图,该图像分割方法应用于图像分割装置,例如,图像分割装置可以由终端设备或服务器或其它处理设备执行,其中,终端设备可以为用户设备(UE,User Equipment)、移动设备、蜂窝电话、无绳电话、个人数字处理(PDA,Personal Digital Assistant)、手持设备、计算设备、车载设备、可穿戴设备等。在一些可能的实现方式中,该图像分割方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。如图3所示,该流程包括:
步骤S301、将肺部图像输入肺叶分割网络中,对肺部图像进行多层级的下采样处理和对应的上采样处理,得到对应不同层级的下采样处理结果和上采样处理结果。
采样:是采集模拟信号的样本,采样是将时间上、幅值上都连续的信号,在采样脉冲的作用下,转换成时间、幅值上离散的信号。采样又称为波形的离散化过程。其中,下采样:是对于一个样值序列,间隔几个样值来取样一次,得到的新序列为原序列的下采样。缩小图像(或称为下采样或降采样)的主要目的有两个:1、使得图像符合显示区域的大小;2、生成对应图像的缩略图。上采样:是下采样的逆过程。上采样的实质是内插或差值。图像放大几乎都是采用内插值方法,即在原有图像像素的基础上在像素点之间采用合适的插值算法***新的元素。放大图像(或称为上采样或图像插值)的主要目的是放大原图像,从而可以显示在更高分辨率的显示设备上。需要指出的是,上采样和下采样都是对数字信号的重新采集,重新采集的采样率与原来获得该数字信号(比如从模拟信号采样而来)的采样率比较,大于原信号的,称为上采样;小于原信号的,称为下采样。
本公开一可能实现方式中,多次下采样和对应的上采样及跳跃连接可以为:输入一张肺部图像的CT,对该CT进行第一层下采样,得到第一下采样结果,对第一下采样结果进行第二层下采样,得到第二下采样结果,依次,进行多层级的下采样(本公开不限于四层下采样)。以四层下采样为例,对第二下采样结果进行第三层下采样,得到第三下采样结果,对第三下采样结果进行第四层下采样,得到第四下采样结果,下采样结束后,对下采样最底层的第四下采样结果(本公开中的第四次下采样之后的结果),进行第一次上采样,得到第一上采样结果,对第一上采样结果进行第二次上采样,得到第二上采样结果,对第二上采样结果进行第三次上采样,得到第三上采样结果,对第三上采样结果进行第四次上采样,得到第四上采样结果。
跳跃连接是针对同一层级,比如,对于第一层,下采样“第一下采样结果”与上采样“第三上采样结果”对应,因此,将第一下采样结果和第三上采样结果进行跳跃连接。
步骤S302、将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,直至对所有层级处理结束,得到对应不同分辨率和多尺度大小的多层输出结果。
本公开一可能实现方式中,将同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,得到跳跃处理结果。通过融合同一尺度的卷积特征,提高分割精度。
本公开一可能实现方式中,该多层输出结果包括:用于标识所述肺裂数据的第一体素数据,和/或用于标识所述肺叶数据的第二体素数据。
步骤S303、将用于标识肺裂数据的第一体素数据和用于标识肺叶数据的第二体素数据作为训练数据所得到的混合损失函数,进行损失函数的反向传播,通过损失函数的反向传播对肺叶分割网络进行训练,得到训练后的肺叶分割网络。
本公开一可能的实现方式中,还可以在该训练数据输入肺叶分割网络前进行下采样,以减小数据量,用有限的计算资源分割整个肺部,在保证完整数据输入的前提下,使用更为完备的网络模型。通过在该训练数据输入肺叶分割网络前进行下采样,可以加快对数据的处理速度,将分割速度控制在2秒以内。
步骤S304、根据训练后的肺叶分割网络,确定肺部图像中目标肺叶所在的位置。
本公开中,对于肺裂和肺叶分割处理流程,输入的肺部图像CT是一样的,采用的网络都是同一个网络,不同流程得到不同的数据,都可以采用上述步骤S301-步骤S302的处理流程。肺裂和肺叶分割处理流程二者的处理模式是相同的,其处理中的不同之处在于:对于肺裂数据,可以是用于标识肺裂数据的第一体素数据;对于肺叶数据,可以是用于标识肺叶数据的第二体素数据。也就是说,根据多层输出结果可以得到最终处理结果,从该最终处理结果中所提取并用于训练的数据是两种,一种是针对肺裂的第一体素数据,另一种是针对肺叶的第二体素数据。肺裂和肺叶分割处理流程可以是同时进行的。
就体素而言,对于一副图像,如果该图像为2D图像,则该图像可以描述为由多个像素构成,像素是二维的;如果该图像为3D图像,则该图像可以描述为由多个体素所构成,体素是三维的,在3D图像中,体积分为均匀间隔的行和列,涵盖所有三个不同的方向(上下、左右,内外)。这将3D空间划分成立方体,也称为体素(体积元素或体积像素)。每个体素由三维坐标和该坐标处的颜色定义。
本公开中,该肺叶分割网络为端到端的3D分割网络(或称为基于VNet的3D卷积神经网络)。为了实现更精确的分割,本公开采用端到端的3D分割网络结构,对肺部进行整体分割,提高了在空间上的感知能力,可以提取更多的空间信息,从而提高各肺叶的分割 结果。右中肺叶因其形状和位置多变,预测准确度不高,采用本公开,可以对该右中肺叶区域进行精准分割。由于通过混合损失函数的反向传播来训练该网络结构,是一种深度学习模型,与相关技术相比不需要医生在分割结果上进行额外的工作,即无需医生的交互和修改才能得到较为准确的肺叶分割结果,而是利用该网络结构进行全自动的肺叶分割,在保证分割精度的同时减少医生的工作量,提高分割的处理效率。
采用本公开,首先将肺部图像输入肺叶分割网络中,对肺部图像进行多层级的下采样处理和对应的上采样处理,将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,直至对所有层级处理结束,得到对应不同分辨率和多尺度大小的多层输出结果。将同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,得到跳跃处理结果,通过融合同一尺度的卷积特征的方式来提高分割精度。然后,将肺裂数据和肺叶数据用于网络训练,根据肺裂数据和肺叶数据相结合所得到的混合损失函数进行损失函数的反向传播,实现对肺叶分割网络的训练。通过这种训练方式所得到的肺叶分割网络,可以提高分割准确度。使采用该训练后的肺叶分割网络能更精确的进行图像分割,以从肺部图像中确定出肺叶的位置,从而根据肺叶的位置,可以及时对病灶进行定位。
应用示例:
该肺叶分割网络的训练过程包含以下两个流程,肺叶整体的分割和肺裂的分割流程。对于肺裂和肺叶分割处理流程,输入的肺部图像CT是一样的,采用的网络都是同一个网络。图4为根据本公开的训练过程示意图。如图4所示,基于端到端3D结构的肺叶分割网络,分别进行肺叶分割和肺裂分割工作,肺部图像111为肺叶分割网络的输入数据,肺部图像111可以为3维CT数据,通过上采样、对应的下采样及同层的跳跃连接,可以得到肺叶分割网络的输出数据,输出数据包括两种:肺叶数据112和肺裂数据113。然后根据肺裂数据113和肺叶数据112训练该肺叶分割网络。肺叶分割和肺裂分割工作具体可分为以下两个部分。
一、肺叶分割
基于端到端3D结构的肺叶分割网络,如图4所示,肺叶分割网络的输入为肺部3维CT数据,在肺叶分割网络中进行多次下采样和对应的上采样过程以及跳跃连接,产生不同分辨率和多尺度大小的多层输出,将这些多尺度输出结合在一起就得到了最终分割结果。其中,输入的原始数据是z×x×y的单通道灰度图像,经过数据预处理之后进入3D分割网络,而输出则为z×x×y尺寸的6通道张量,分别代表每个体素位置属于哪个肺叶或是背景。其中每个跳跃连接都是将网络中下采样与其相对应的上采样相融合,这样的级联合 成了该数据的3D概率分布图。
二、肺裂分割
基于端到端3D结构的肺叶分割网络,如图4所示,为了提取肺部边界的信息,使用肺裂作为目标结果进行训练。利用端到端3D结构进行多尺度的卷积特征融合。为了充分利用肺叶分割网络中的局部信息,在肺叶分割网络中采用多栅卷积的网络结构代替相关技术中的卷积块,肺叶分割网络中用四尺度代替了相关技术的单尺度网络,以增强特征融合效果。多栅卷积的网络结构在每个卷积层之前将不同尺度的特征图进行级联,降低了传统FCN、U-Net等网络结构由于降采样造成的特征损失。将多栅网络用于3D分割,通过融合不同尺度的卷积特征,提高分割精度,在提取肺裂位置信息时有更准确的预测效果。网络的输入与肺叶分割相同,是z×x×y的单通道灰度图像,输出为4通道3D数据(其中3通道为人体内3条肺裂的位置信息,1通道为背景信息)。
训练过程中,本公开采用肺叶Dice加肺叶和肺裂Cross Entropy相结合的混合损失作为损失函数对网络进行参数调优。其中Dice损失函数的表达式如公式(1)所示。
其中,V表示3D图像中的所有体素点,p
i为i体素点被预测为目标类的概率,即预测为目标肺叶的概率;l
i为该体素点的实际标签。采用多Dice的加权以修正边界,肺叶训练时的Dice损失函数如公式(2)所示。
D
lobe=α
1D
RUL+α
2D
RML+α
3D
RLL+α
4D
LUL+α
5D
LLL (2)
其中,D
RUL,D
RML,D
RLL,D
LUL,D
L表示五个肺叶分别的Dice,α是可调节系数,用来标定各肺叶在分割任务中的权重对整体分割的影响。
Cross Entropy损失函数的表达式如公式(3)所示:
H(p,q)=-∑
x∈Xp(x)logq(x) (3)
其中p(x)为预测正确的概率,q(x)为预测错误的概率。
最终网络的损失函数采用公式(4)计算得到:
Loss=β
1D
lobe+β
2H(p,q)
lobe+β
3H(p,q)
fissure (4)
其中,D
lobe为肺叶训练时的Dice损失函数,H(p,q)
lobe为肺叶训练时的Cross Entropy损失函数,H(p,q)
fissure为肺裂训练时的Cross Entropy损失函数,β
1、β
2、β
3是各个损失函数的权重,决定各部分训练结果对最终分割结果对影响。
本领域技术人员可以理解,在具体实施方式的上述方法中,各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定,各步骤的具体执行顺序应当以其功 能和可能的内在逻辑确定。
本公开提及的上述各个方法实施例,在不违背原理逻辑的情况下,均可以彼此相互结合形成结合后的实施例,限于篇幅,本公开不再赘述。
此外,本公开还提供了图像分割装置、电子设备、计算机可读存储介质、程序,上述均可用来实现本公开提供的任一种图像分割方法,相应技术方案和描述和参见方法部分的相应记载,不再赘述。
图5示出根据本公开实施例的图像分割装置的框图,如图5所示,本公开实施例的图像分割装置,包括:分割网络获得单元31,用于根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络;位置确定单元32,用于根据所述肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
本公开可能的实现方式中,所述分割网络获得单元,进一步用于:将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练中,得到训练后的肺叶分割网络;所述位置确定单元,进一步用于:根据所述训练后的肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
本公开可能的实现方式中,所述分割网络获得单元,进一步用于:根据所述肺裂数据和所述肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播;通过所述损失函数的反向传播对所述肺叶分割网络进行训练,得到训练后的肺叶分割网络。
本公开可能的实现方式中,所述装置还包括:混合损失函数确定单元,用于:根据所述肺叶数据得到第一损失函数和第二损失函数;根据所述肺裂数据得到第三损失函数;根据所述第一损失函数、所述第二损失函数和所述第三损失函数得到所述混合损失函数。
本公开可能的实现方式中,所述装置还包括:数据处理单元,用于:将所述肺部图像输入所述肺叶分割网络中,对所述肺部图像进行多层级的下采样处理和对应的上采样处理,得到对应不同层级的下采样处理结果和上采样处理结果;将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,直至对所有层级处理结束,得到对应不同分辨率和多尺度大小的多层输出结果。
本公开可能的实现方式中,所述多层输出结果包括:用于标识所述肺裂数据的第一体素数据,和/或用于标识所述肺叶数据的第二体素数据;所述装置还包括:数据确定单元,用于:将所述第一体素数据和/或所述第二体素数据作为用于训练所述肺叶分割网络的训练数据。
本公开可能的实现方式中,所述数据处理单元,进一步用于:将所述同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,得到跳跃处理结果。
在一些实施例中,本公开实施例提供的装置具有的功能或包含的模块可以用于执行上文方法实施例描述的方法,其具体实现可以参照上文方法实施例的描述,为了简洁,这里不再赘述。
本公开实施例还提出一种计算机可读存储介质,其上存储有计算机程序指令,所述计算机程序指令被处理器执行时实现上述方法。计算机可读存储介质可以是非易失性计算机可读存储介质。
本公开实施例还提出一种电子设备,包括:处理器;用于存储处理器可执行指令的存储器;其中,所述处理器被配置为上述方法。
电子设备可以被提供为终端、服务器或其它形态的设备。
图6是根据一示例性实施例示出的一种电子设备800的框图。例如,电子设备800可以是移动电话,计算机,数字广播终端,消息收发设备,游戏控制台,平板设备,医疗设备,健身设备,个人数字助理等终端。
参照图6,电子设备800可以包括以下一个或多个组件:处理组件802,存储器804,电源组件806,多媒体组件808,音频组件810,输入/输出(I/O)的接口812,传感器组件814,以及通信组件816。
处理组件802通常控制电子设备800的整体操作,诸如与显示,电话呼叫,数据通信,相机操作和记录操作相关联的操作。处理组件802可以包括一个或多个处理器820来执行指令,以完成上述的方法的全部或部分步骤。此外,处理组件802可以包括一个或多个模块,便于处理组件802和其他组件之间的交互。例如,处理组件802可以包括多媒体模块,以方便多媒体组件808和处理组件802之间的交互。
存储器804被配置为存储各种类型的数据以支持在电子设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令,联系人数据,电话簿数据,消息,图片,视频等。存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现,如静态随机存取存储器(SRAM),电可擦除可编程只读存储器(EEPROM),可擦除可编程只读存储器(EPROM),可编程只读存储器(PROM),只读存储器(ROM),磁存储器,快闪存储器,磁盘或光盘。
电源组件806为电子设备800的各种组件提供电力。电源组件806可以包括电源管理***,一个或多个电源,及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中,屏幕可以包括液晶显示器(LCD)和触摸面板(TP)。如果屏幕包括触摸面板,屏幕可以被实现为触摸屏,以接收来自用户的输入信号。触摸面板包括一个或多 个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界,而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中,多媒体组件808包括一个前置摄像头和/或后置摄像头。当电子设备800处于操作模式,如拍摄模式或视频模式时,前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜***或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如,音频组件810包括一个麦克风(MIC),当电子设备800处于操作模式,如呼叫模式、记录模式和语音识别模式时,麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在存储器804或经由通信组件816发送。在一些实施例中,音频组件810还包括一个扬声器,用于输出音频信号。
I/O接口812为处理组件802和***接口模块之间提供接口,上述***接口模块可以是键盘,点击轮,按钮等。这些按钮可包括但不限于:主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件814包括一个或多个传感器,用于为电子设备800提供各个方面的状态评估。例如,传感器组件814可以检测到电子设备800的打开/关闭状态,组件的相对定位,例如所述组件为电子设备800的显示器和小键盘,传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变,用户与电子设备800接触的存在或不存在,电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器,被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器,如CMOS或CCD图像传感器,用于在成像应用中使用。在一些实施例中,该传感器组件814还可以包括加速度传感器,陀螺仪传感器,磁传感器,压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络,如WiFi,2G或3G,或它们的组合。在一个示例性实施例中,通信组件816经由广播信道接收来自外部广播管理***的广播信号或广播相关信息。在一个示例性实施例中,所述通信组件816还包括近场通信(NFC)模块,以促进短程通信。例如,在NFC模块可基于射频识别(RFID)技术,红外数据协会(IrDA)技术,超宽带(UWB)技术,蓝牙(BT)技术和其他技术来实现。
在示例性实施例中,电子设备800可以被一个或多个应用专用集成电路(ASIC)、数字信号处理器(DSP)、数字信号处理设备(DSPD)、可编程逻辑器件(PLD)、现场可编程门阵列(FPGA)、控制器、微控制器、微处理器或其他电子元件实现,用于执行上 述方法。
在示例性实施例中,还提供了一种非易失性计算机可读存储介质,例如包括计算机程序指令的存储器804,上述计算机程序指令可由电子设备800的处理器820执行以完成上述方法。
图7是根据一示例性实施例示出的一种电子设备900的框图。例如,电子设备900可以被提供为一服务器。参照图7,电子设备900包括处理组件922,其进一步包括一个或多个处理器,以及由存储器932所代表的存储器资源,用于存储可由处理组件922的执行的指令,例如应用程序。存储器932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外,处理组件922被配置为执行指令,以执行上述方法。
电子设备900还可以包括一个电源组件926被配置为执行电子设备900的电源管理,一个有线或无线网络接口950被配置为将电子设备900连接到网络,和一个输入输出(I/O)接口958。电子设备900可以操作基于存储在存储器932的操作***,例如Windows ServerTM,Mac OS XTM,UnixTM,LinuxTM,FreeBSDTM或类似。
在示例性实施例中,还提供了一种非易失性计算机可读存储介质,例如包括计算机程序指令的存储器932,上述计算机程序指令可由电子设备900的处理组件922执行以完成上述方法。
本公开可以是***、方法和/或计算机程序产品。计算机程序产品可以包括计算机可读存储介质,其上载有用于使处理器实现本公开的各个方面的计算机可读程序指令。
计算机可读存储介质可以是可以保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是――但不限于――电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括:便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、静态随机存取存储器(SRAM)、便携式压缩盘只读存储器(CD-ROM)、数字多功能盘(DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身,诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(例如,通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里所描述的计算机可读程序指令可以从计算机可读存储介质下载到各个计算/处理设备,或者通过网络、例如因特网、局域网、广域网和/或无线网下载到外部计算机或外部存储设备。网络可以包括铜传输电缆、光纤传输、无线传输、路由器、防火墙、交换机、网关计算机和/或边缘服务器。每个计算/处理设备中的网络适配卡或者网络接口从 网络接收计算机可读程序指令,并转发该计算机可读程序指令,以供存储在各个计算/处理设备中的计算机可读存储介质中。
用于执行本公开操作的计算机程序指令可以是汇编指令、指令集架构(ISA)指令、机器指令、机器相关指令、微代码、固件指令、状态设置数据、或者以一种或多种编程语言的任意组合编写的源代码或目标代码,所述编程语言包括面向对象的编程语言—诸如Smalltalk、C++等,以及常规的过程式编程语言—诸如“C”语言或类似的编程语言。计算机可读程序指令可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中,远程计算机可以通过任意种类的网络—包括局域网(LAN)或广域网(WAN)—连接到用户计算机,或者,可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。在一些实施例中,通过利用计算机可读程序指令的状态信息来个性化定制电子电路,例如可编程逻辑电路、现场可编程门阵列(FPGA)或可编程逻辑阵列(PLA),该电子电路可以执行计算机可读程序指令,从而实现本公开的各个方面。
这里参照根据本公开实施例的方法、装置(***)和计算机程序产品的流程图和/或框图描述了本公开的各个方面。应当理解,流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合,都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器,从而生产出一种机器,使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时,产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中,这些指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作,从而,存储有指令的计算机可读介质则包括一个制造品,其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
也可以把计算机可读程序指令加载到计算机、其它可编程数据处理装置、或其它设备上,使得在计算机、其它可编程数据处理装置或其它设备上执行一系列操作步骤,以产生计算机实现的过程,从而使得在计算机、其它可编程数据处理装置、或其它设备上执行的指令实现流程图和/或框图中的一个或多个方框中规定的功能/动作。
附图中的流程图和框图显示了根据本公开的多个实施例的***、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上,流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分,所述模块、程序段或指令的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。在有些作为替换的实现中,方框中所标注 的功能也可以以不同于附图中所标注的顺序发生。例如,两个连续的方框实际上可以基本并行地执行,它们有时也可以按相反的顺序执行,这依所涉及的功能而定。也要注意的是,框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合,可以用执行规定的功能或动作的专用的基于硬件的***来实现,或者可以用专用硬件与计算机指令的组合来实现。
在不违背逻辑的情况下,本申请不同实施例之间可以相互结合,不同实施例描述有所侧重,为侧重描述的部分可以参见其他实施例的记载。以上已经描述了本公开的各实施例,上述说明是示例性的,并非穷尽性的,并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下,对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择,旨在最好地解释各实施例的原理、实际应用或对市场中技术的技术改进,或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。
Claims (17)
- 一种图像分割方法,其特征在于,所述方法包括:根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络;根据所述肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
- 根据权利要求1所述的方法,其中,所述根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络,包括:将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练中,得到训练后的肺叶分割网络;所述根据所述肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置,包括:根据所述训练后的肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
- 根据权利要求2所述的方法,其特征在于,所述将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练中,得到训练后的肺叶分割网络,包括:根据所述肺裂数据和所述肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播;通过所述损失函数的反向传播对所述肺叶分割网络进行训练,得到训练后的肺叶分割网络。
- 根据权利要求3所述的方法,其特征在于,所述方法还包括:根据所述肺裂数据和所述肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播之前,根据所述肺叶数据得到第一损失函数和第二损失函数;根据所述肺裂数据得到第三损失函数;根据所述第一损失函数、所述第二损失函数和所述第三损失函数得到所述混合损失函数。
- 根据权利要求2至4中任一项所述的方法,其特征在于,所述将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练之前,还包括:将所述肺部图像输入所述肺叶分割网络中,对所述肺部图像进行多层级的下采样处理和对应的上采样处理,得到对应不同层级的下采样处理结果和上采样处理结果;将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,直至对所有层级处理结束,得到对应不同分辨率和多尺度大小的多层输出结果。
- 根据权利要求1-5所述的方法,其特征在于,所述多层输出结果包括:用于标识所述肺裂数据的第一体素数据,和/或用于标识所述肺叶数据的第二体素数据;所述方法还包括:将所述第一体素数据和/或所述第二体素数据作为用于训练所述肺叶分割网络的训练数据。
- 根据权利要求1-5所述的方法,其特征在于,所述将同一层级的下采样处理结果 和上采样处理结果进行跳跃连接处理,包括:将所述同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,得到跳跃处理结果。
- 一种图像分割装置,其特征在于,所述装置包括:分割网络获得单元,用于根据肺部图像中的肺叶数据和肺裂数据得到肺叶分割网络;位置确定单元,用于根据所述肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
- 根据权利要求8所述的装置,其中,所述分割网络获得单元,进一步用于:将所述肺裂数据用于包含所述肺叶数据的肺叶分割网络训练中,得到训练后的肺叶分割网络;所述位置确定单元,进一步用于:根据所述训练后的肺叶分割网络,确定所述肺部图像中目标肺叶所在的位置。
- 根据权利要求9所述的装置,其特征在于,所述分割网络获得单元,进一步用于:根据所述肺裂数据和所述肺叶数据相结合所得到的混合损失函数,进行损失函数的反向传播;通过所述损失函数的反向传播对所述肺叶分割网络进行训练,得到训练后的肺叶分割网络。
- 根据权利要求10所述的装置,其特征在于,所述装置还包括:混合损失函数确定单元,用于:根据所述肺叶数据得到第一损失函数和第二损失函数;根据所述肺裂数据得到第三损失函数;根据所述第一损失函数、所述第二损失函数和所述第三损失函数得到所述混合损失函数。
- 根据权利要求9至11中任一项所述的装置,其特征在于,所述装置还包括:数据处理单元,用于:将所述肺部图像输入所述肺叶分割网络中,对所述肺部图像进行多层级的下采样处理和对应的上采样处理,得到对应不同层级的下采样处理结果和上采样处理结果;将同一层级的下采样处理结果和上采样处理结果进行跳跃连接处理,直至对所有层级处理结束,得到对应不同分辨率和多尺度大小的多层输出结果。
- 根据权利要求8-12所述的装置,其特征在于,所述多层输出结果包括:用于标识所述肺裂数据的第一体素数据,和/或用于标识所述肺叶数据的第二体素数据;所述装置还包括:数据确定单元,用于:将所述第一体素数据和/或所述第二体素数 据作为用于训练所述肺叶分割网络的训练数据。
- 根据权利要求8-12所述的装置,其特征在于,所述数据处理单元,进一步用于:将所述同一层级的下采样处理结果和上采样处理结果中同一尺度的特征进行融合,得到跳跃处理结果。
- 一种电子设备,其特征在于,包括:处理器;用于存储处理器可执行指令的存储器;其中,所述处理器被配置为:执行权利要求1至7中任意一项所述的方法。
- 一种计算机可读存储介质,其上存储有计算机程序指令,其特征在于,所述计算机程序指令被处理器执行时实现权利要求1至7中任意一项所述的方法。
- 一种计算机程序,其特征在于,所述计算机程序包括计算机可读代码,当所述计算机可读代码在电子设备中运行时,所述电子设备中的处理器执行用于实现权利要求1至7中的任意一项所述的方法。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2021534283A JP2022515722A (ja) | 2019-04-18 | 2019-09-25 | 画像セグメンテーション方法及び装置、電子機器並びに記憶媒体 |
KR1020217018707A KR20210107667A (ko) | 2019-04-18 | 2019-09-25 | 이미지 분할 방법 및 장치, 전자 기기 및 저장 매체 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910315130.7A CN110060262A (zh) | 2019-04-18 | 2019-04-18 | 一种图像分割方法及装置、电子设备和存储介质 |
CN201910315130.7 | 2019-04-18 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2020211293A1 true WO2020211293A1 (zh) | 2020-10-22 |
Family
ID=67319606
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2019/107850 WO2020211293A1 (zh) | 2019-04-18 | 2019-09-25 | 一种图像分割方法及装置、电子设备和存储介质 |
Country Status (5)
Country | Link |
---|---|
JP (1) | JP2022515722A (zh) |
KR (1) | KR20210107667A (zh) |
CN (1) | CN110060262A (zh) |
TW (1) | TWI713054B (zh) |
WO (1) | WO2020211293A1 (zh) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112508974A (zh) * | 2020-12-14 | 2021-03-16 | 北京达佳互联信息技术有限公司 | 图像分割模型的训练方法、装置、电子设备和存储介质 |
CN112561917A (zh) * | 2020-12-22 | 2021-03-26 | 上海联影智能医疗科技有限公司 | 图像的分段方法、***、电子设备及可读存储介质 |
CN112560945A (zh) * | 2020-12-14 | 2021-03-26 | 珠海格力电器股份有限公司 | 一种基于情绪识别的设备控制方法及*** |
CN112651969A (zh) * | 2021-02-08 | 2021-04-13 | 福州大学 | 结合多信息融合网络和区域增长的气管树分级提取方法 |
CN112749801A (zh) * | 2021-01-22 | 2021-05-04 | 上海商汤智能科技有限公司 | 神经网络训练和图像处理方法及装置 |
CN112950553A (zh) * | 2021-02-05 | 2021-06-11 | 慧影医疗科技(北京)有限公司 | 一种多尺度肺叶分割方法、***、存储介质及电子设备 |
CN113762265A (zh) * | 2021-08-27 | 2021-12-07 | 慧影医疗科技(北京)有限公司 | 肺炎的分类分割方法及*** |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110060262A (zh) * | 2019-04-18 | 2019-07-26 | 北京市商汤科技开发有限公司 | 一种图像分割方法及装置、电子设备和存储介质 |
CN110689548B (zh) * | 2019-09-29 | 2023-01-17 | 浪潮电子信息产业股份有限公司 | 一种医学图像分割方法、装置、设备及可读存储介质 |
CN110942453A (zh) * | 2019-11-21 | 2020-03-31 | 山东众阳健康科技集团有限公司 | 一种基于神经网络的ct影像肺叶识别方法 |
CN111563902B (zh) * | 2020-04-23 | 2022-05-24 | 华南理工大学 | 一种基于三维卷积神经网络的肺叶分割方法及*** |
CN113870167A (zh) * | 2020-06-11 | 2021-12-31 | 通用电气精准医疗有限责任公司 | 用于对肺部图像进行分割的方法、***以及存储介质 |
CN111738998B (zh) * | 2020-06-12 | 2023-06-23 | 深圳技术大学 | 病灶位置动态检测方法及装置、电子设备和存储介质 |
CN112348794A (zh) * | 2020-11-05 | 2021-02-09 | 南京天智信科技有限公司 | 基于注意力增强u型网络的超声乳腺肿瘤自动分割方法 |
CN112801964B (zh) * | 2021-01-20 | 2022-02-22 | 中国人民解放军总医院 | 肺部ct图像的多标签智能检测方法、装置、设备和介质 |
CN113496496B (zh) * | 2021-07-07 | 2023-04-07 | 中南大学 | 基于多种损失和多尺度特征的mri影像海马体区域分割方法 |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020114503A1 (en) * | 2001-02-17 | 2002-08-22 | Siemens Aktiengesellschaft | Method and apparatus for processing a computed tomography image of a lung obtained using contrast agent |
CN107909581A (zh) * | 2017-11-03 | 2018-04-13 | 杭州依图医疗技术有限公司 | Ct影像的肺叶段分割方法、装置、***、存储介质及设备 |
CN110060262A (zh) * | 2019-04-18 | 2019-07-26 | 北京市商汤科技开发有限公司 | 一种图像分割方法及装置、电子设备和存储介质 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104700118A (zh) * | 2015-03-18 | 2015-06-10 | 中国科学院自动化研究所 | 基于卷积神经网络的肺结节良恶性预测方法 |
CN107230204B (zh) * | 2017-05-24 | 2019-11-22 | 东北大学 | 一种从胸部ct图像中提取肺叶的方法及装置 |
EP3633612A4 (en) * | 2017-06-30 | 2020-06-03 | Shanghai United Imaging Healthcare Co., Ltd. | IMAGE SEGMENTATION METHOD AND SYSTEM |
CN107784647B (zh) * | 2017-09-29 | 2021-03-09 | 华侨大学 | 基于多任务深度卷积网络的肝脏及其肿瘤分割方法及*** |
CN109636808B (zh) * | 2018-11-27 | 2022-08-12 | 杭州健培科技有限公司 | 一种基于全卷积神经网络的肺叶分割方法 |
-
2019
- 2019-04-18 CN CN201910315130.7A patent/CN110060262A/zh active Pending
- 2019-09-25 KR KR1020217018707A patent/KR20210107667A/ko unknown
- 2019-09-25 JP JP2021534283A patent/JP2022515722A/ja active Pending
- 2019-09-25 WO PCT/CN2019/107850 patent/WO2020211293A1/zh active Application Filing
- 2019-10-21 TW TW108137871A patent/TWI713054B/zh active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020114503A1 (en) * | 2001-02-17 | 2002-08-22 | Siemens Aktiengesellschaft | Method and apparatus for processing a computed tomography image of a lung obtained using contrast agent |
CN107909581A (zh) * | 2017-11-03 | 2018-04-13 | 杭州依图医疗技术有限公司 | Ct影像的肺叶段分割方法、装置、***、存储介质及设备 |
CN110060262A (zh) * | 2019-04-18 | 2019-07-26 | 北京市商汤科技开发有限公司 | 一种图像分割方法及装置、电子设备和存储介质 |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112508974A (zh) * | 2020-12-14 | 2021-03-16 | 北京达佳互联信息技术有限公司 | 图像分割模型的训练方法、装置、电子设备和存储介质 |
CN112560945A (zh) * | 2020-12-14 | 2021-03-26 | 珠海格力电器股份有限公司 | 一种基于情绪识别的设备控制方法及*** |
CN112508974B (zh) * | 2020-12-14 | 2024-06-11 | 北京达佳互联信息技术有限公司 | 图像分割模型的训练方法、装置、电子设备和存储介质 |
CN112561917A (zh) * | 2020-12-22 | 2021-03-26 | 上海联影智能医疗科技有限公司 | 图像的分段方法、***、电子设备及可读存储介质 |
CN112749801A (zh) * | 2021-01-22 | 2021-05-04 | 上海商汤智能科技有限公司 | 神经网络训练和图像处理方法及装置 |
CN112950553A (zh) * | 2021-02-05 | 2021-06-11 | 慧影医疗科技(北京)有限公司 | 一种多尺度肺叶分割方法、***、存储介质及电子设备 |
CN112651969A (zh) * | 2021-02-08 | 2021-04-13 | 福州大学 | 结合多信息融合网络和区域增长的气管树分级提取方法 |
CN113762265A (zh) * | 2021-08-27 | 2021-12-07 | 慧影医疗科技(北京)有限公司 | 肺炎的分类分割方法及*** |
CN113762265B (zh) * | 2021-08-27 | 2024-05-07 | 慧影医疗科技(北京)股份有限公司 | 肺炎的分类分割方法及*** |
Also Published As
Publication number | Publication date |
---|---|
TW202040587A (zh) | 2020-11-01 |
TWI713054B (zh) | 2020-12-11 |
CN110060262A (zh) | 2019-07-26 |
JP2022515722A (ja) | 2022-02-22 |
KR20210107667A (ko) | 2021-09-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020211293A1 (zh) | 一种图像分割方法及装置、电子设备和存储介质 | |
TWI779238B (zh) | 影像處理方法及裝置、電子設備和電腦可讀取的記錄媒體 | |
TWI755853B (zh) | 圖像處理方法、電子設備和電腦可讀儲存介質 | |
TWI754375B (zh) | 圖像處理方法、電子設備、電腦可讀儲存介質 | |
WO2022151755A1 (zh) | 目标检测方法及装置、电子设备、存储介质、计算机程序产品和计算机程序 | |
WO2022127071A1 (zh) | 网络训练方法、图像分割方法、装置、设备、介质及产品 | |
CN112767329B (zh) | 图像处理方法及装置、电子设备 | |
TW202131281A (zh) | 圖像處理方法及圖像處理裝置、電子設備和電腦可讀儲存媒介 | |
JP2022535219A (ja) | 画像分割方法及び装置、電子機器並びに記憶媒体 | |
WO2021259391A2 (zh) | 图像处理方法及装置、电子设备和存储介质 | |
CN110210487A (zh) | 一种图像分割方法及装置、电子设备和存储介质 | |
CN114820584B (zh) | 肺部病灶定位装置 | |
CN112115894B (zh) | 手部关键点检测模型的训练方法、装置及电子设备 | |
WO2021259390A2 (zh) | 一种冠脉钙化斑块检测方法及装置 | |
KR102456264B1 (ko) | 비디오 클립 추출 방법, 비디오 클립 추출 장치 및 저장매체 | |
WO2022032998A1 (zh) | 图像处理方法及装置、电子设备、存储介质和程序产品 | |
WO2022242046A1 (zh) | 医学图像的展示方法及装置、电子设备、存储介质和计算机程序 | |
CN115423819A (zh) | 肺血管的分割方法及装置、电子设备和存储介质 | |
CN114387436B (zh) | 壁冠状动脉检测方法及装置、电子设备和存储介质 | |
CN111724361A (zh) | 实时显示病灶的方法及装置、电子设备和存储介质 | |
WO2022193456A1 (zh) | 目标跟踪方法及装置、电子设备和存储介质 | |
KR20220012407A (ko) | 이미지 분할 방법 및 장치, 전자 기기 및 저장 매체 | |
CN113553460B (zh) | 影像检索方法及装置、电子设备和存储介质 | |
WO2023230927A1 (zh) | 图像处理方法、装置及可读存储介质 | |
CN115471474A (zh) | 右心室分割、心肺分析方法及装置、电子设备和存储介质 |
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: 19925242 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2021534283 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 19925242 Country of ref document: EP Kind code of ref document: A1 |