CN113052785A - Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium - Google Patents
Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium Download PDFInfo
- Publication number
- CN113052785A CN113052785A CN202110320816.2A CN202110320816A CN113052785A CN 113052785 A CN113052785 A CN 113052785A CN 202110320816 A CN202110320816 A CN 202110320816A CN 113052785 A CN113052785 A CN 113052785A
- Authority
- CN
- China
- Prior art keywords
- liver
- deep learning
- dimensional image
- segmentation model
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration by the use of histogram techniques
-
- 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
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The invention provides a method for constructing an automatic liver segmentation model based on deep learning, which comprises the following steps: acquiring a sample liver three-dimensional image, and acquiring a liver segmentation label of the sample liver three-dimensional image; taking the sample liver three-dimensional image and the liver segmentation label as a training set, and carrying out deep learning training on the liver segmentation model iteration to obtain a trained liver segmentation model, wherein a deep learning network adopted by the liver segmentation model is a segmentation network based on the combination of a UNet/VNet and a channel attention mechanism. Related apparatus, computer devices, and readable storage media are also provided. Due to the fact that the UNet/VNet and the channel attention mechanism are combined, more image information of the sample liver three-dimensional image is mined through the UNet/VNet, higher weight is given to important channel information through the channel attention mechanism, a more accurate prediction result is obtained, and an accurate and efficient liver segmentation result can be provided for a doctor.
Description
Technical Field
The invention relates to the technical field of medical image processing, in particular to the technical field of liver segmentation, and specifically relates to a method and a device for constructing an automatic liver segmentation model based on deep learning, computer equipment and a storage medium.
Background
Liver cancer is one of the most common cancer diseases in the world, and at present, liver cancer surpasses gastric cancer and invades the first three of cancer deaths. The liver is a common site of primary or secondary tumor growth, and its heterogeneous and diffuse shape makes it difficult to dissect by segmentation. Therefore, the ability to accurately infer and measure for each segment in the liver is a prerequisite for modern liver surgery.
In clinical diagnosis, it is time consuming to explore spatial information remotely along the Z-axis for liver segmentation, and thus an automated method is needed for efficient solution. Recently, deep Learning has performed well in the field of computer vision, and has been applied in automated Segmentation of Liver and tumor in CT (Sun, C., et al: automated Segmentation of Liver from multiple phase transformed images based on FCNs [ J ]. AI Med.2017,83: 58-66) and automated Segmentation (Tian J, Liu L, Shi Z, et al. automated summary Segmentation from CT volume on Liver Using GLC-UNet [ M ]. Machine Learning in Medical Imaging,10th International kshop, MLMI 2019, Held in connected with MIC CAI 2019, Shenzhen, China, Ococeber 13,2019, progress 892019).
The existing automatic liver segmentation method based on CT is based on deep learning Unnet 2D network (Ronneberger, O., Fischer, P., Brox, T., U-Net: connected for biological image segmentation. in: Navab, N., Hornegger, J., Wells, W.M., Frangi, A.F. (eds.) MICCAI 2015.LNCS, vol.9351, pp.234-241. Springer, Cham (2015)) to segment the liver of CT data through global and local information, and the whole network realizes two tasks of liver segmentation and liver segmentation based on 2D slices. The attention mechanism adopted by the network is a multilayer perceptron, and feature extraction is mainly carried out on multi-level global and local information from coarse to fine. The decoding process of the network finally realizes the result prediction of liver segmentation through multi-level fusion.
However, the current MRI-based automated liver segmentation studies are few, and are still implemented by using conventional algorithm steps (Lebre M A, Vacavant A, Grand-Brochier M, et al. automatic segmentation methods for live and fatty vessels from CT and MRI volumes, applied to the coupled scheme [ J ]. Computers in Biology and Medicine,2019,110(7):42-51), namely liver segmentation, vessel centerline extraction and liver segmentation reconstruction.
Since existing liver segmentation techniques focus on two aspects: deep learning based CT liver segmentation and conventional algorithm based MRI liver segmentation. However, compared with CT, MRI can provide image information of multiple sequence imaging, multiple image modalities, etc., which has great potential superiority for disease diagnosis, and deep learning can automatically mine more potential image information compared with conventional algorithms.
Therefore, it is desirable to provide a method for constructing an automated liver segmentation model based on deep learning, which can obtain accurate and efficient automated liver segmentation results.
Disclosure of Invention
The present invention is directed to overcoming the above-mentioned shortcomings in the prior art, and providing a method, an apparatus, a computer device and a storage medium for constructing an automated liver segmentation model based on deep learning, which can obtain accurate and efficient automated liver segmentation results.
In order to achieve the above object, in a first aspect of the present invention, there is provided a method for constructing an automated liver segmentation model based on deep learning, comprising the following steps:
(1) acquiring a sample liver three-dimensional image, and acquiring a liver segmentation label of the sample liver three-dimensional image;
(2) and taking the sample liver three-dimensional image and the liver segmentation label as a training set, and performing deep learning training on the liver segmentation model in an iteration mode to obtain a trained liver segmentation model, wherein a deep learning network adopted by the liver segmentation model is a segmentation network based on the combination of UNet/VNet and a channel attention mechanism.
Preferably, in the step (1), the sample liver three-dimensional image is a liver MRI three-dimensional image, a liver B-mode ultrasound three-dimensional image, a liver CT three-dimensional image or a liver MRS three-dimensional image.
Preferably, in the step (1), the step of obtaining a three-dimensional image of a sample liver specifically includes:
preprocessing an original sample liver three-dimensional image to obtain the sample liver three-dimensional image, wherein the preprocessing comprises one or more of histogram equalization, standardization and normalization processing.
Preferably, in step (2), the UNet is UNet2.5D, UNet2D, UNet3D, UNet + +, Res-UNet, Dense U-Net, MultiResUNet, R2U-Net, or Attention UNet.
Preferably, in the step (2), the deep learning network includes a convolutional layer, a pooling layer, an inverse convolutional layer, a cascade layer, and a batch normalization layer, the convolutional layer is connected to the inverse convolutional layer through the pooling layer, the cascade layer is respectively connected to the convolutional layer, the pooling layer, the inverse convolutional layer, and the batch normalization layer through signals, the convolutional layer extracts a feature map of the three-dimensional image of the sample liver, the pooling layer performs a downsampling operation on the feature map, the inverse convolutional layer performs a convolution operation after padding the feature map to enlarge a size of the feature map, the cascade layer combines the feature maps output by different levels, and the batch normalization layer normalizes values of the feature map.
Preferably, in the step (2), the deep learning training includes an encoding process and a decoding process, both of which use the UNet/VNet and the channel attention mechanism, and the decoding process further uses one or more of a multi-level fusion operation and a full supervision operation.
More preferably, in the step (2), the loss function adopted by the fully-supervised operation is a multi-classification cross-entropy loss function.
In a second aspect of the present invention, there is provided an apparatus for constructing an automated liver segmentation model based on deep learning, including:
the liver segmentation model training module is used for carrying out deep learning training on the liver segmentation model iteration by taking the sample liver three-dimensional image and a liver segmentation label of the sample liver three-dimensional image as a training set by adopting a segmentation network based on combination of a UNet/VNet and a channel attention mechanism to obtain a trained liver segmentation model.
In a third aspect of the present invention, there is provided a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method for constructing the automated liver segmentation model based on deep learning when executing the computer program.
In a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program is configured to implement the above-mentioned method for constructing an automated liver segmentation model based on deep learning when being executed by a processor.
By adopting the method, the device, the computer equipment and the storage medium for constructing the automatic liver segmentation model based on the deep learning, because the UNet/VNet and the channel attention mechanism are combined, more image information of the sample liver three-dimensional image is mined through the UNet/VNet, and higher weight is given to important channel information through the channel attention mechanism, so that the characteristic which contributes more to the final prediction result can be extracted.
Drawings
Fig. 1 is a flowchart illustrating a method for constructing an automated liver segmentation model based on deep learning according to an embodiment of the present invention.
FIG. 2 is a schematic diagram of a deep learning training method for a liver segmentation model in the embodiment shown in FIG. 1.
FIG. 3 is a diagram of a human labeled liver segmentation label for validating a trained liver segmentation model in the embodiment shown in FIG. 1.
FIG. 4 is a diagram illustrating model predictions for validating a trained liver segmentation model in the embodiment shown in FIG. 1.
Fig. 5 is a schematic frame diagram of an embodiment of the apparatus for constructing an automated liver segmentation model based on deep learning according to the present invention.
Detailed Description
In order to clearly understand the technical contents of the present invention, the following examples are given in detail. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Referring to fig. 1 to 2, in an embodiment of the present invention, a method for constructing an automatic liver segmentation model based on deep learning according to the present invention includes the following steps:
(1) acquiring a sample liver three-dimensional image, and acquiring a liver segmentation label of the sample liver three-dimensional image;
(2) and (2) taking the sample liver three-dimensional image and the liver segmentation label as a training set, and carrying out deep learning training on the liver segmentation model in an iteration way to obtain a trained liver segmentation model, wherein a deep learning network adopted by the liver segmentation model is a segmentation network based on combination of UNet/VNet and a channel attention mechanism (Jie Hu, Li Shen, et al. Squeeze-and-Excitation Networks. [ J ]. IEEE transactions on pattern analysis and machine interaction [ J ].2020,42(8): 2011-.
The liver segmentation model is a machine learning model for performing liver segmentation processing. The trained liver segmentation model can be obtained through deep learning training.
Namely, inputting the sample liver three-dimensional image in the training set into a deep learning network to obtain liver segmented prediction data, then performing difference comparison on the liver segmented prediction data and liver segmented labels in the training set, and iteratively updating the deep learning network according to the difference until an iteration termination condition is met to obtain a liver segmented model.
In the step (1), the three-dimensional image of the liver is a three-dimensional image taken of the liver.
In the step (1), the sample liver three-dimensional image may be any suitable type of liver three-dimensional image, such as a liver MRI (Magnetic Resonance Imaging) three-dimensional image, a liver B-ultrasound (type-B ultrasound) three-dimensional image, a liver CT (Computed Tomography) three-dimensional image, or a liver MRS (Magnetic Resonance Spectroscopy) three-dimensional image.
In the step (1), the step of obtaining a three-dimensional image of a sample liver may specifically adopt any suitable method, and in an embodiment of the present invention, in the step (1), the step of obtaining a three-dimensional image of a sample liver specifically includes:
preprocessing an original sample liver three-dimensional image to obtain the sample liver three-dimensional image, wherein the preprocessing comprises one or more of histogram equalization, standardization and normalization processing.
The original sample liver three-dimensional image is a sample liver three-dimensional image which is not subjected to pretreatment. Histogram equalization processing is carried out, so that the dynamic range of the image gray scale is enhanced, and the contrast of the image is improved; the normalization is to transform the data into a distribution with a mean value of 0 and a standard deviation of 1, and the normalization is to change the data into a certain fixed interval, wherein the interval is [0, 1], so that the solving process is accelerated in the descending process of the model training gradient, and adverse factors such as gradient explosion and the like are also avoided.
In step (2), the UNet may be any suitable UNet, such as UNet2.5D, UNet2D, UNet3D, UNet + +, Res-UNet, Dense U-Net, MultiResUNet, R2U-Net, or Attention UNet, and in one embodiment of the invention, UNet2.5D, as shown in FIG. 2.
In the step (2), the deep learning network may include any suitable configuration, in one embodiment of the present invention, in the step (2), the deep learning network includes a convolutional layer, a pooling layer, an anti-convolutional layer, a cascading layer and a batch normalization layer, the convolution layer is in signal connection with the deconvolution layer through the pooling layer, the cascade layer is in signal connection with the convolution layer, the pooling layer, the deconvolution layer and the batch normalization layer respectively, the convolutional layer extracts a feature map of the sample liver three-dimensional image, the pooling layer performs down-sampling operation on the feature map, the deconvolution layer performs convolution operation after padding the feature map to enlarge the size of the feature map, the cascade layer combines the feature maps output by different levels, and the batch normalization layer normalizes the numerical values of the feature maps.
The convolutional layer extracts a feature map of the sample liver three-dimensional image by setting the size and the step length of a convolutional kernel, and the batch normalization layer normalizes the numerical values of the feature map to improve the convergence speed of the gradient, accelerate training and alleviate the problem of gradient disappearance.
In the step (2), the deep learning training may include any suitable training process, please refer to fig. 2, and in a specific embodiment of the present invention, in the step (2), the deep learning training includes an encoding process and a decoding process, both of the encoding process and the decoding process employ the UNet/VNet and the channel attention mechanism, and the decoding process further employs one or more of a multi-level fusion operation and a full supervision operation.
In step (2), the loss function adopted by the fully-supervised operation may be any suitable loss function, and in a specific embodiment of the present invention, in step (2), the loss function adopted by the fully-supervised operation is a multi-class cross-entropy loss function.
Fig. 2 is a schematic diagram of a process of obtaining a trained liver segmentation model through deep learning training. As can be seen from fig. 2, the overall training process is such that: inputting an original medical image (namely a sample liver three-dimensional image) into a deep learning network to obtain a prediction result of the deep learning network, comparing the prediction result (namely liver segmented prediction data) with an artificial label (namely a liver segmented label), feeding back the result to the deep learning network, continuously updating the deep learning network by taking the artificial label as a target according to the fed-back comparison information until the prediction result is close to the artificial label, and thus obtaining the trained liver segmented model in the embodiment.
The method for constructing the automatic liver segmentation model based on deep learning of the present invention is specifically described below by taking 500 MRI three-dimensional images of an original sample liver as an example.
1. MRI data pre-processing
DICOM data of 500 original sample liver MRI three-dimensional images are read, and preprocessing is performed in a histogram equalization, standardization and normalization mode. The histogram equalization process enhances the dynamic range of the image gray scale, thereby improving the contrast of the image. Normalization transforms the data into a distribution with a mean of 0 and a standard deviation of 1. Normalization varied the data to a fixed interval [0, 1 ]. The standardization and normalization processing can accelerate the solving process in the descending process of the model training gradient and avoid adverse factors such as gradient explosion and the like. 500 sample liver MRI three-dimensional images are obtained, 350 parts (70%) of the sample liver MRI three-dimensional images and corresponding liver segmentation labels are used as a training set, and 150 parts (30%) of the sample liver MRI three-dimensional images and corresponding liver segmentation labels are used as a verification set.
2. Liver segmentation model construction
Inputting a training set into a segmented network based on combination of UNet2.5D and a channel attention mechanism shown in FIG. 2, specifically, an encoding process comprises 3 levels of UNet2.5D and the channel attention mechanism, a decoding process also comprises 3 levels of UNet2.5D and the channel attention mechanism, a batch size, a learning rate and a kernel initialization parameter are set in a training process, an Adam optimizer is adopted for training, and a multi-classification cross entropy loss function is adopted as a loss function.
(1) Setting model parameters of an encoding process and a decoding process: the convolution layer has convolution kernel size of 3 x 3 and kernel initialization with LeCun homogeneous initializer; the activation function uses a Linear rectification function (ReLU), also called a modified Linear Unit; the convolution kernel size of the deconvolution layer was 3 x 3.
(2) Setting network model output parameters: a convolution layer having convolution kernels with a size of 1 x 1; the activation function uses a Sigmoid function to map the output value between 0 and 1.
(3) Setting parameters of a model training process: batch is set to 8; adopting an Adam optimizer which uses momentum and an adaptive learning rate to accelerate convergence speed, wherein the initial learning rate is set to be 0.0001; the loss function adopts a multi-classification cross entropy loss function; the epoch is set to 300.
(4) And starting training until the model converges, and keeping the optimal training model. Namely, a trained liver segmentation model is obtained.
3. Liver segmentation model validation
Inputting the verification set into the obtained trained liver segmentation model, and quantitatively and qualitatively comparing the model prediction result with the liver segmentation label. Wherein, the average effect of the verification set is measured by using the Dice coefficient during quantification (the Dice coefficient of the verification set is 0.81). The qualitative predictive effect of two-dimensional slices on the validation set is shown in fig. 3 and 4.
Therefore, the invention applies deep learning to liver segmentation of MRI for the first time, explores multi-mode image information, automatically excavates more potential information in the image and provides an end-to-end automatic liver segmentation method.
In a deep learning algorithm applied in the prior art, a 2D-based convolutional neural network ignores spatial information between an MRI upper hierarchy and a MRI lower hierarchy, so that accuracy is reduced, and a 3D-based convolutional neural network causes problems of memory consumption and the like. Therefore, the invention adopts unet2.5d as the basic network to realize efficient automatic liver segmentation for MRI, and has the main advantages that: compared with UNet2D and UNet3D basic networks, UNet2.5D adopted by the method can not only mine spatial information of upper and lower layers of MRI images, but also avoid adverse factors such as memory consumption and the like. In addition, in the prior art, an attention mechanism of a multilayer perceptron is adopted, the importance of channel characteristics and the contribution to a final prediction result are not concerned, the channel characteristic information is fully mined by utilizing the channel attention mechanism in the network model design, and the original characteristics are recalibrated in the channel dimension, namely the importance of the channel characteristics is weighted. And finally, in order to fully utilize the characteristic information of different levels, a full supervision mode is adopted in the decoding process to carry out multi-level optimization, and finally multi-level fusion is carried out to realize an accurate and efficient end-to-end liver segmentation task.
Therefore, the method for constructing the automatic liver segmentation model based on the deep learning mainly comprises data preprocessing and liver segmentation model construction. The liver segmentation model construction comprises an encoding process and a decoding process, wherein the encoding process and the decoding process both adopt the combination of a UNet/VNet and a channel attention mechanism, and the decoding process carries out full supervision operation on each level except for multi-level fusion, namely the output of each level and the liver segmentation label calculation loss participate in the training process together. The full supervision enables different levels of output in the network decoding process to be continuously optimized from coarse to fine, and the prediction accuracy of the model is improved. The training process sets the batch size, the learning rate and the kernel initialization parameters, an Adam optimizer is adopted for training, and a multi-classification cross entropy loss function is adopted as a loss function. The deep learning network is of a single-input multi-output structure, a full supervision mode is adopted in the decoding process during training, and only the final prediction result of multi-level fusion is output in the decoding process during testing.
Referring to fig. 5, in an embodiment of the present invention, the present invention further provides an apparatus for constructing an automated liver segmentation model based on deep learning, including:
the liver segmentation model training module is used for carrying out deep learning training on the liver segmentation model iteration by taking the sample liver three-dimensional image and a liver segmentation label of the sample liver three-dimensional image as a training set by adopting a segmentation network based on combination of a UNet/VNet and a channel attention mechanism to obtain a trained liver segmentation model.
The automatic liver segmentation model construction device based on deep learning may further include any other suitable components, please refer to fig. 3, and in an embodiment of the present invention, the automatic liver segmentation model construction device based on deep learning further includes a preprocessing module, the preprocessing module is in signal connection with the liver segmentation model training module and is configured to preprocess an original sample liver three-dimensional image to obtain the sample liver three-dimensional image, and the preprocessing includes one or more of histogram equalization, normalization, and normalization.
The segmentation network based on combination of UNet/VNet and channel attention mechanism may include any suitable structure, and in a specific embodiment of the present invention, the segmentation network based on combination of UNet/VNet and channel attention mechanism includes a convolutional layer, a pooling layer, an anti-convolutional layer, a cascading layer, and a batch normalization layer, the convolutional layer is connected to the anti-convolutional layer through the pooling layer, the cascading layer is respectively connected to the convolutional layer, the pooling layer, the anti-convolutional layer, and the batch normalization layer, the convolutional layer extracts a feature map of the three-dimensional image of the sample liver, the pooling layer performs a down-sampling operation on the feature map, the anti-convolutional layer performs a convolution operation on the feature map after padding to expand the size of the feature map, and the cascading layer combines the feature maps output by different levels, the batch normalization layer normalizes values of the feature map.
The liver segmentation model training module may include any suitable configuration, and preferably includes an encoding module and a decoding module, the encoding module is in signal connection with the decoding module, the encoding module and the decoding module respectively include a plurality of successively signal-connected UNet/VNet + channel attention mechanism modules, the UNet/VNet + channel attention mechanism module at the most downstream of the encoding module is in signal connection with the UNet/VNet + channel attention mechanism module at the most upstream of the decoding module, the decoding module further includes one or more of a multi-level fusion module and a full supervision module, the multi-level fusion module respectively is in signal connection with the UNet/VNet + channel attention mechanism module of the decoding module and is used for fusing feature maps output by the UNet/VNet + channel attention mechanism module of the decoding module, the full supervision module is respectively in signal connection with the UNet/VNet + channel attention mechanism module of the decoding module and used for calculating loss and optimizing the feature map output by the UNet/VNet + channel attention mechanism module of the decoding module and the liver segmentation label. Referring to fig. 2 and 3, in an embodiment of the present invention, the UNet/VNet + channel attention mechanism module is an UNet2.5d + channel attention mechanism module, the encoding module includes 3 levels of the UNet2.5d + channel attention mechanism modules, and the decoding module also includes 3 levels of the UNet2.5d + channel attention mechanism modules.
The loss function employed by the fully supervised module may be any suitable loss function, and in a specific embodiment of the present invention, the loss function employed by the fully supervised module is a multi-class cross entropy loss function.
For other specific limitations of the device for constructing the automated liver segmentation model based on deep learning, reference may be made to the above limitations on the method for constructing the automated liver segmentation model based on deep learning, and details are not repeated here. The modules in the device for constructing the automatic liver segmentation model based on deep learning can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In an embodiment of the present invention, the invention further provides a computer device, which includes a memory and a processor, where the memory stores a computer program, and the processor implements the method for constructing the automated liver segmentation model based on deep learning when executing the computer program.
In an embodiment of the present invention, the present invention further provides a computer readable storage medium, on which a computer program is stored, where the computer program is executed by a processor to implement the method for constructing an automated liver segmentation model based on deep learning.
Therefore, by adopting the method, the device, the computer equipment and the storage medium for constructing the automatic liver segmentation model based on the deep learning, due to the combination of the UNet/VNet and the channel attention mechanism, more image information of the sample liver three-dimensional image is mined through the UNet/VNet, and higher weight is given to important channel information through the channel attention mechanism, so that the characteristic which contributes more to the final prediction result is extracted.
It will thus be seen that the objects of the invention have been fully and effectively accomplished. The functional and structural principles of the present invention have been shown and described in the embodiments, and the embodiments may be modified without departing from the principles. Therefore, this invention includes all modifications encompassed within the spirit and scope of the claims.
Claims (10)
1. A method for constructing an automatic liver segmentation model based on deep learning is characterized by comprising the following steps:
(1) acquiring a sample liver three-dimensional image, and acquiring a liver segmentation label of the sample liver three-dimensional image;
(2) and taking the sample liver three-dimensional image and the liver segmentation label as a training set, and performing deep learning training on the liver segmentation model in an iteration mode to obtain a trained liver segmentation model, wherein a deep learning network adopted by the liver segmentation model is a segmentation network based on the combination of UNet/VNet and a channel attention mechanism.
2. The method for constructing an automated liver segmentation model based on deep learning of claim 1, wherein in the step (1), the sample liver three-dimensional image is a liver MRI three-dimensional image, a liver B-mode ultrasound three-dimensional image, a liver CT three-dimensional image or a liver MRS three-dimensional image.
3. The method for constructing an automated liver segmentation model based on deep learning according to claim 1, wherein in the step (1), the step of obtaining a three-dimensional image of a sample liver specifically comprises:
preprocessing an original sample liver three-dimensional image to obtain the sample liver three-dimensional image, wherein the preprocessing comprises one or more of histogram equalization, standardization and normalization processing.
4. The method for constructing an automated liver segmentation model based on deep learning according to claim 1, wherein in the step (2), the UNet is UNet2.5d, UNet2D, UNet3D, UNet + +, Res-UNet, Dense U-Net, multiresuunt, R2U-Net or Attention UNet.
5. The method for constructing the automatic liver segmentation model based on deep learning according to claim 1, characterized in that, in the step (2), the deep learning network comprises a convolutional layer, a pooling layer, a deconvolution layer, a cascade layer and a batch normalization layer, the convolution layer is in signal connection with the deconvolution layer through the pooling layer, the cascade layer is in signal connection with the convolution layer, the pooling layer, the deconvolution layer and the batch normalization layer respectively, the convolutional layer extracts a feature map of the sample liver three-dimensional image, the pooling layer performs down-sampling operation on the feature map, the deconvolution layer performs convolution operation after padding the feature map to enlarge the size of the feature map, the cascade layer combines the feature maps output by different levels, and the batch normalization layer normalizes the numerical values of the feature maps.
6. The method for constructing an automated liver segmentation model based on deep learning according to claim 1, wherein in the step (2), the deep learning training comprises an encoding process and a decoding process, the encoding process and the decoding process both use the UNet/VNet and the channel attention mechanism, and the decoding process further uses one or more of a multi-level fusion operation and a full supervision operation.
7. The method for constructing an automated liver segmentation model based on deep learning of claim 6, wherein in the step (2), the loss function adopted by the fully supervised operation is a multi-class cross entropy loss function.
8. An automatic liver segmentation model construction device based on deep learning is characterized by comprising the following steps:
the liver segmentation model training module is used for carrying out deep learning training on the liver segmentation model iteration by taking the sample liver three-dimensional image and a liver segmentation label of the sample liver three-dimensional image as a training set by adopting a segmentation network based on combination of a UNet/VNet and a channel attention mechanism to obtain a trained liver segmentation model.
9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program implements the method of constructing an automated deep learning based liver segmentation model according to any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the method of constructing an automated liver segmentation model based on deep learning according to any one of claims 1 to 7.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110320816.2A CN113052785A (en) | 2021-03-25 | 2021-03-25 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
| CN202110807571.6A CN113436127A (en) | 2021-03-25 | 2021-07-16 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110320816.2A CN113052785A (en) | 2021-03-25 | 2021-03-25 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113052785A true CN113052785A (en) | 2021-06-29 |
Family
ID=76515507
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110320816.2A Withdrawn CN113052785A (en) | 2021-03-25 | 2021-03-25 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
| CN202110807571.6A Pending CN113436127A (en) | 2021-03-25 | 2021-07-16 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
Family Applications After (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110807571.6A Pending CN113436127A (en) | 2021-03-25 | 2021-07-16 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN113052785A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113436185A (en) * | 2021-03-25 | 2021-09-24 | 上海志御软件信息有限公司 | Quantitative analysis method, device, computer equipment and storage medium for three-dimensional structural characteristics of hepatic vein blood vessels |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114494482A (en) * | 2021-12-24 | 2022-05-13 | 中国人民解放军总医院第一医学中心 | Method for generating CT blood vessel imaging based on flat scanning CT |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211052A (en) * | 2019-03-29 | 2019-09-06 | 北京工业大学 | A kind of single image to the fog method based on feature learning |
| CN110176004A (en) * | 2019-04-24 | 2019-08-27 | 艾瑞迈迪科技石家庄有限公司 | A kind of liver subsection method and system |
| CN110210485A (en) * | 2019-05-13 | 2019-09-06 | 常熟理工学院 | The image, semantic dividing method of Fusion Features is instructed based on attention mechanism |
| CN110428427B (en) * | 2019-07-16 | 2023-05-12 | 东南大学 | Semi-supervised renal artery segmentation method based on dense bias network and self-encoder |
| CN110428428B (en) * | 2019-07-26 | 2022-03-25 | 长沙理工大学 | Image semantic segmentation method, electronic equipment and readable storage medium |
| CN110852317A (en) * | 2019-09-16 | 2020-02-28 | 浙江工业大学 | Small-scale target detection method based on weak edge |
| CN111161241B (en) * | 2019-12-27 | 2024-04-23 | 联想(北京)有限公司 | Liver image identification method, electronic equipment and storage medium |
| CN111340812A (en) * | 2020-02-20 | 2020-06-26 | 中南大学 | Interactive liver image segmentation method based on deep neural network |
| CN111627019B (en) * | 2020-06-03 | 2023-03-14 | 西安理工大学 | Liver tumor segmentation method and system based on convolutional neural network |
| CN112241766B (en) * | 2020-10-27 | 2023-04-18 | 西安电子科技大学 | Liver CT image multi-lesion classification method based on sample generation and transfer learning |
-
2021
- 2021-03-25 CN CN202110320816.2A patent/CN113052785A/en not_active Withdrawn
- 2021-07-16 CN CN202110807571.6A patent/CN113436127A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113436185A (en) * | 2021-03-25 | 2021-09-24 | 上海志御软件信息有限公司 | Quantitative analysis method, device, computer equipment and storage medium for three-dimensional structural characteristics of hepatic vein blood vessels |
| CN113436185B (en) * | 2021-03-25 | 2022-11-15 | 上海志御软件信息有限公司 | Liver vein blood vessel three-dimensional structure characteristic quantitative analysis method, device, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113436127A (en) | 2021-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chaitanya et al. | Semi-supervised task-driven data augmentation for medical image segmentation | |
| Lather et al. | Investigating brain tumor segmentation and detection techniques | |
| Niyas et al. | Medical image segmentation with 3D convolutional neural networks: A survey | |
| Zhao et al. | A novel U-Net approach to segment the cardiac chamber in magnetic resonance images with ghost artifacts | |
| EP3852054A1 (en) | Method and system for automatically detecting anatomical structures in a medical image | |
| Ypsilantis et al. | Recurrent convolutional networks for pulmonary nodule detection in CT imaging | |
| CN115170582A (en) | Liver image segmentation method based on multi-scale feature fusion and grid attention mechanism | |
| CN113436127A (en) | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium | |
| EP4030385A1 (en) | Devices and process for synthesizing images from a source nature to a target nature | |
| CN112819831A (en) | Segmentation model generation method and device based on convolution Lstm and multi-model fusion | |
| La Rosa | A deep learning approach to bone segmentation in CT scans | |
| Mukherjee et al. | Domain adapted multitask learning for segmenting amoeboid cells in microscopy | |
| US11776128B2 (en) | Automatic detection of lesions in medical images using 2D and 3D deep learning networks | |
| Ibrahim et al. | Deep learning based Brain Tumour Classification based on Recursive Sigmoid Neural Network based on Multi-Scale Neural Segmentation | |
| CN117409201A (en) | MR medical image colorectal cancer segmentation method and system based on semi-supervised learning | |
| CN116759076A (en) | Unsupervised disease diagnosis method and system based on medical image | |
| Asma-Ull et al. | Data efficient segmentation of various 3d medical images using guided generative adversarial networks | |
| Baumgartner et al. | Fully convolutional networks in medical imaging: Applications to image enhancement and recognition | |
| Dickson et al. | An integrated 3D-sparse deep belief network with enriched seagull optimization algorithm for liver segmentation | |
| Ammari et al. | Deep-active-learning approach towards accurate right ventricular segmentation using a two-level uncertainty estimation | |
| Devi et al. | Hybrid convolutional neural network based segmentation of visceral and subcutaneous adipose tissue from abdominal magnetic resonance images | |
| Zhao et al. | Automated breast lesion segmentation from ultrasound images based on ppu-net | |
| Larsen | Exploring generative adversarial networks to improve prostate segmentation on mri | |
| Johansen et al. | Medical image segmentation: A general u-net architecture and novel capsule network approaches | |
| Ramya et al. | Design of Novel Brain Tumour Segmentation System using Hybrid Heuristic-aided Multiscale Self-Guided Attention Mechanism based Adaptive Unet+++ with 3D Brain MRI Images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210629 |
|
| WW01 | Invention patent application withdrawn after publication |