CN113793345B - Medical image segmentation method and device based on improved attention module - Google Patents
Medical image segmentation method and device based on improved attention module Download PDFInfo
- Publication number
- CN113793345B CN113793345B CN202111042489.5A CN202111042489A CN113793345B CN 113793345 B CN113793345 B CN 113793345B CN 202111042489 A CN202111042489 A CN 202111042489A CN 113793345 B CN113793345 B CN 113793345B
- Authority
- CN
- China
- Prior art keywords
- feature matrix
- target
- image
- attention
- determining
- 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.)
- Active
Links
- 238000003709 image segmentation Methods 0.000 title claims abstract description 34
- 238000000034 method Methods 0.000 title claims abstract description 34
- 239000011159 matrix material Substances 0.000 claims abstract description 174
- 238000003062 neural network model Methods 0.000 claims abstract description 72
- 230000011218 segmentation Effects 0.000 claims abstract description 36
- 238000003672 processing method Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000005070 sampling Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 35
- 238000005481 NMR spectroscopy Methods 0.000 claims description 19
- 238000007781 pre-processing Methods 0.000 claims description 18
- 238000010586 diagram Methods 0.000 claims description 17
- 238000002591 computed tomography Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000012546 transfer Methods 0.000 claims description 6
- 238000000586 desensitisation Methods 0.000 claims description 3
- 238000004590 computer program Methods 0.000 claims description 2
- 210000004556 brain Anatomy 0.000 description 12
- 230000006870 function Effects 0.000 description 8
- 230000007246 mechanism Effects 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 238000003745 diagnosis Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 238000013421 nuclear magnetic resonance imaging Methods 0.000 description 4
- 208000018737 Parkinson disease Diseases 0.000 description 3
- 230000004913 activation Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 210000003523 substantia nigra Anatomy 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 241001573498 Compacta Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004195 computer-aided diagnosis Methods 0.000 description 1
- 235000013365 dairy product Nutrition 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 208000030533 eye disease Diseases 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000002911 mydriatic effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000002203 pretreatment Methods 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- 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
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
-
- 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
- G06T2207/30016—Brain
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The application discloses an image processing method, an image processing device, computer equipment and a storage medium. The method comprises the following steps: acquiring target tensor data of a target image; inputting the target tensor data into a target neural network model, the target neural network model including an encoder, a decoder, and an attention module; the encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result; and carrying out segmentation feedback on the target image according to the output of the target neural network model. The accuracy of image segmentation can be improved.
Description
Technical Field
The embodiment of the application relates to a machine learning technology, in particular to an image processing method, an image processing device, computer equipment and a storage medium.
Background
Medical image segmentation is an important task in computer-aided diagnosis, but the segmentation work has been challenging due to irregular shape of the target object, blurred boundaries, etc. Thus, for identifying these complex target objects of different dimensions, an accurate and reliable segmentation method is needed.
With the development of deep convolutional neural networks (Convolutional Neural Networks, CNNs), U-Net of U-shaped neural networks and a technical scheme for fusing the U-Net with an attention mechanism are presented. However, the current U-Net combined with the attention mechanism cannot accurately segment targets with different shapes and scales in the medical image, and the accuracy of image segmentation is low.
Disclosure of Invention
The application provides an image processing method, an image processing device, computer equipment and a storage medium, so as to improve the accuracy of image segmentation.
In a first aspect, an embodiment of the present application provides an image processing method, including:
acquiring target tensor data of a target image;
inputting the target tensor data into a target neural network model, the target neural network model including an encoder, a decoder, and an attention module;
the encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result;
and carrying out segmentation feedback on the target image according to the output of the target neural network model.
In a second aspect, an embodiment of the present application further provides an image processing apparatus, including:
the target image tensor acquisition module is used for acquiring target tensor data of the target image;
the segmentation module is used for inputting target tensor data into a target neural network model, and the target neural network model comprises an encoder, a decoder and an attention module;
the encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result;
and the output module is used for carrying out segmentation feedback on the target image according to the output of the target neural network model.
In a third aspect, an embodiment of the present application further provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor executes the program to implement an image processing method as shown in the embodiment of the present application.
In a third aspect, embodiments of the present application also provide a storage medium containing computer-executable instructions which, when executed by a computer processor, are used to perform an image processing method as shown in the embodiments of the present application.
The image processing method provided by the embodiment of the application can acquire the target tensor data of the target image; inputting target tensor data into a target neural network model, and performing downsampling by an encoder of the target neural network model according to the target tensor data to obtain a first feature matrix; the attention module of the target neural network model is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder of the target neural network model is used for up-sampling according to the third feature matrix to obtain an image segmentation result; and carrying out segmentation feedback on the target image according to the output of the target neural network model. Compared with the problem of low accuracy of U-Net image segmentation combined with an attention mechanism at present, the attention module of the target neural network model can process the first feature matrix of the target image based on hole convolution, the range of the receptive field can be improved through hole convolution, further, the region of interest of a user in the target image can be determined more accurately, segmentation is carried out according to the region, and accuracy of image segmentation is improved.
Drawings
Fig. 1 is a flowchart of an image processing method in a first embodiment of the present application;
fig. 2 is a schematic diagram of the structure of an image processing apparatus in a third embodiment of the present application;
fig. 3 is a schematic structural diagram of a computer device in a fourth embodiment of the present application.
Detailed Description
The application is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not limiting thereof. It should be further noted that, for convenience of description, only some, but not all of the structures related to the present application are shown in the drawings.
Example 1
Fig. 1 is a flowchart of an image processing method according to an embodiment of the present application, where the embodiment is applicable to a case of dividing an area of interest of a user in an image, especially a case of dividing an area of interest of a user such as a lesion in a medical image, where the method may be performed by a computer device that provides an image dividing function for the user, and the computer device may be a personal computer, a notebook computer, a tablet computer or a server. The method specifically comprises the following steps:
step 110, obtaining target tensor data of the target image.
The target image may be an image to be segmented, and the target tensor data is a target image expressed in a vector form. For example, the shape of the target tensor data may be 1×128×128, representing 128 pixels in the horizontal direction and 128 pixels in the vertical direction in a certain color channel of the target image.
On the basis of the above embodiment, acquiring target tensor data of a target image includes:
preprocessing a target image according to a preset preprocessing mode to obtain a preprocessed image, wherein the preprocessing mode comprises an image clipping mode and an image normalization mode; target tensor data is determined from the preprocessed image.
The image cropping mode includes cropping images of different sizes into images of the same size. The image normalization approach involves normalizing images of different formats to a uniform file format. The preprocessing image with uniform specification can be obtained through the preprocessing mode. Since the preprocessed image contains the same number of pixels in the horizontal and vertical directions and the same number of color channels, the target tensor data format obtained from the preprocessed image is also the same. The target tensor data obtained through preprocessing is suitable for the target neural network model, and the segmentation accuracy of the target neural network model can be improved.
On the basis of the above embodiment, the target image is a medical image, and the medical image includes: images are taken using computed tomography images, nuclear magnetic resonance images, high definition cameras.
The high-definition camera can be high-definition camera equipment in the medical field. For example, the high definition camera may be a DRIVE eye disease dataset image acquisition tool. Specifically, a Canon CR5 non-mydriatic 3CCD camera was used to capture images at 45 degrees of field of view. Each image was captured with 8 bits in the color plane of 768×584 pixels. The field of view of each image is circular, with a diameter of about 540 pixels.
On the basis of the above embodiment, acquiring target tensor data of a target image includes:
desensitizing the target image to obtain a desensitized image;
target tensor data for the target image is determined from the desensitized image.
In the embodiment, because the nuclear magnetic resonance scanning image related to the suspected patient with the clinical Parkinson disease is desensitized data with the format of Nii.gz after desensitization treatment, information such as the name of the patient and the like can not be read, and important privacy is not involved, so that the information safety of a user is ensured to the greatest extent.
Step 120, inputting the target tensor data into the target neural network model.
Wherein the target neural network model includes an encoder, a decoder, and an attention module.
The encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result.
The encoder includes a plurality of convolution blocks. The convolutional layer of each convolutional block is followed by an activation function (ReLU) and a Batch Normalization (BN) layer, which are then downsampled using a max-pooling operation, the downsampling step size may be 2. After each downsampling step, the number of characteristic channels will double. The first feature matrix is obtained by a plurality of donation blocks.
In one implementation, the data recall model may be constructed based on data characteristics of clinical brain nuclear magnetic resonance images used for training. Specifically: firstly, an account number and a password with partial user authority are obtained from a PACS system of a hospital, and a computer tomography image of a suspected patient in the PACS system and corresponding tag data are obtained through a communication protocol and stored.
During storage, the brain nuclear magnetic resonance scan images may have different slice numbers according to the thickness of the scan layer, and are stored as a patient-table header. And downsampling the brain nuclear magnetic resonance scanning image through an encoder to obtain a first feature matrix.
After the first feature matrix is obtained, the first feature matrix is input to an attention module (also called an AHA module). And optimizing the first feature matrix in the attention module, and outputting a third feature matrix.
The decoder and the encoder comprise a plurality of symmetrical convolution blocks, and the decoder is used for up-sampling the third feature matrix to restore the original image resolution.
In one implementation, the attention module is configured to perform hole convolution according to the first feature matrix to obtain a second feature matrix, and includes: and carrying out cavity convolution on the first feature matrix according to the plurality of cavity convolutions to obtain a plurality of second feature matrices, wherein each cavity convolution corresponds to the second feature matrix one by one. Accordingly, determining the original attention profile from the second feature matrix comprises: performing mutual matrix multiplication according to the plurality of second feature matrixes to obtain a matrix multiplication result; the original attention map is determined from the matrix multiplication result. Accordingly, determining a hole attention map from the original attention map includes: and carrying out transfer closing operation on the original attention force diagram to obtain a hole attention force diagram. Accordingly, determining a third feature matrix from the first feature matrix and the hole attention map comprises: multiplying the first feature matrix by the hole attention map to obtain a third feature matrix.
In one implementation, the generation process of the hole high order attention matrix is as follows: wherein, at the bottom of the encoder, a feature map (also called a first feature matrix) is obtained from the previous stageWherein H W C represents the height and width of the feature map, respectivelyAnd the channel number, which is sent into 4 shared weight convolution layers (r=1, r=2, r=4, r=8 are respectively set) to generate a multi-scale feature map (also called a second feature matrix) X r＝1,2,4,8 . Scaling the multi-scale feature map (also called second feature matrix) by 1×1 convolution, and performing dot multiplication by the following formula one to obtain a preliminary attention matrix (also called original attention map)/(original attention map)>
Based on the theory of "transitive closure" in graph theory, one can try according to the original attentionObtaining a high-order attention moment arrayThe calculation process is as follows:
first, an original thresholded attention matrix (also called a hole attention map) is obtained according to the following equation II:
where δ is a super-parameter threshold, in the embodiment of the present application, δ may be set to 0.5.
Then, according to the transfer closure theory, carrying out transfer closure calculation on the hollow attention force diagram, and obtaining a high-order attention moment array according to the following formula III
Where M represents the high rank attention of the mth order. And weighting the first feature map through the high-order attention matrix to obtain a third feature matrix, so as to filter noise influence and strengthen useful information.
The attention module described above may be embedded into any "U" -shaped encoder-decoder neural network model for use in medical image segmentation. In order to expand the limited receptive field of local convolution of the convolution layer in the downsampling stage and the deconvolution layer in the upsampling stage, more global information can be effectively captured in a high dimension by adopting multi-scale hole convolution. In addition, the high-order attention mechanism constructs a high-order attention feature map through the transfer closure of the map model so as to acquire stronger relevant context features.
Optionally, the segmentation algorithm used in the embodiment of the application may be a threshold segmentation algorithm, so that the influence of peripheral noise can be removed, the quality of the normalized image is improved, and the robustness of the U-shaped neural network model based on the improved attention module is improved.
And 130, performing segmentation feedback of the target image according to the output of the target neural network model.
On the basis of the above embodiment, the method for performing segmentation feedback of the target image according to the output of the target neural network model includes: and outputting the target image, and determining the highlight region of the target image according to the image segmentation result.
The user can import the target image through the computer equipment, and the computer equipment displays the target image and the image segmentation result after segmenting the target image according to the method. The highlighting may be used to display the segmented user region of interest.
The image processing method provided by the embodiment of the application can acquire the target tensor data of the target image; inputting target tensor data into a target neural network model, and performing downsampling by an encoder of the target neural network model according to the target tensor data to obtain a first feature matrix; the attention module of the target neural network model is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder of the target neural network model is used for up-sampling according to the third feature matrix to obtain an image segmentation result; and carrying out segmentation feedback on the target image according to the output of the target neural network model. Compared with the problem of low accuracy of U-Net image segmentation combined with an attention mechanism at present, the attention module of the target neural network model can process the first feature matrix of the target image based on hole convolution, the range of the receptive field can be improved through hole convolution, further, the region of interest of a user in the target image can be determined more accurately, segmentation is carried out according to the region, and accuracy of image segmentation is improved.
On the basis of the above embodiment, before inputting the tensor data into the target neural network model, the method further includes:
acquiring training tensor data of training images of a training set; and training the target neural network model according to the training tensor data so as to optimize network parameters of the target neural network model.
The training images can be obtained by interfacing with a hospital image archiving and communication system (Picture Archiving and Communication Systems, PACS) and can be expanded based on the images obtained by the interfacing. The expansion mode comprises random overturn, random rotation, random translation, random clipping and Gaussian noise addition. The training tensor data obtained through expansion is used for training the target neural network model, so that the sensitivity of the target neural network model to unstable factors such as rotation, translation, gaussian noise and the like can be improved, and the robustness of the U-shaped neural network model based on the improved attention module is further improved.
Example two
An embodiment II of the present application provides an image processing method, as a further explanation of the above embodiment, including:
and E1, storing medical image information related to the suspected Parkinson patient. The medical image information includes brain magnetic resonance imaging.
The clinical head nuclear magnetic resonance imaging is standard image format data with a format of nii.gz after desensitization treatment.
And E2, preprocessing the target image according to a preset preprocessing mode to obtain a preprocessed image, and determining target tensor data according to the preprocessed image.
The pretreatment method comprises the following steps:
and step T1, obtaining a detection machine model corresponding to brain nuclear magnetic resonance imaging, regulating the nuclear magnetic resonance imaging to obtain a standard image to be detected, and normalizing to obtain the normalized image to be detected.
In this example, all magnetic resonance examinations were performed on a clinical radiology department 3.0-T nuclear magnetic resonance scanner (GE medical company discoery (TM) MR 750) equipped with an 8-channel head matrix coil.
In step T2, since the pixel occupation of the region to be segmented is relatively small, particularly, in this embodiment, the original image is cut in the center, the middle region is taken for further processing, and the nuclear magnetic resonance image of the brain to be detected is converted into Tensor data as the preprocessing data.
Wherein the shape of the Tensor data is 1×128×128.
And E3, training a neural network model to serve as a medical image segmentation model.
Training a "U" shaped neural network model based on an improved attention module includes the steps of:
step S1, constructing a data calling model according to the data characteristics of the clinical brain nuclear magnetic resonance images for training. Specifically:
firstly, an account number and a password with partial user authority are obtained from a PACS system of a hospital, and a computer tomography image of a suspected patient in the PACS system and corresponding tag data are obtained through a communication protocol and stored.
In addition, during storage, the brain nuclear magnetic resonance scan images may have different slice numbers according to the thickness of the scan layer, and are stored as a patient-table.
S2, inputting the image into a downsampling feature extraction network to obtain a first feature matrix;
and S3, correcting the first feature matrix by using a hole high-rank attention mechanism to obtain a second feature matrix.
Step S3 comprises the following sub-steps: step S3-1, carrying out cavity convolution on the first feature matrix according to the plurality of cavity convolutions to obtain a plurality of second feature matrices, wherein each cavity convolution corresponds to the second feature matrix one by one, and step S3-2, carrying out mutual matrix multiplication according to the plurality of second feature matrices to obtain a matrix multiplication result; determining an original attention map according to the matrix multiplication result; s2-3, transmitting the original attention force diagram to carry out a closing operation to obtain a hole attention force diagram; and S2-4, multiplying the first feature matrix by the hole attention map to obtain a third feature matrix.
And S4, outputting the third feature matrix obtained in the previous step to an up-sampling segmentation network to obtain a segmentation result.
And S4, training the U-shaped neural network model based on the improved attention module to obtain the trained U-shaped neural network model based on the improved attention module as a hole attention segmentation neural network model.
In step S4-1, the original brain nuclear magnetic resonance image is preprocessed and then converted into Tensor data serving as training data. Specifically:
and extracting pixel values of the brain nuclear magnetic resonance scanning image, normalizing the nuclear magnetic resonance image pixels and cutting off the pixel values, so that the influence of peripheral noise is removed, and the model robustness is improved. The center cropping method was used to preserve intermediate images of the brain with length and width of 128. The pixel matrix is converted into Tensor data by using a Tensor API in the PyTorch framework, so that GPU parallel computing is convenient to use.
And S4-2, expanding the normalized image by using a preset data expansion method to obtain an expanded image.
The data expansion method comprises random overturn, random rotation, random translation, random clipping and Gaussian noise addition.
And S4-3, inputting the expanded image and the label data as processing results into a U-shaped neural network model based on the improved attention module.
In this embodiment, the setting conditions of the relevant parameters when training the hole attention-segmentation neural network model are: the batch size is set to 48, the epoch is 100, the gradient descent method is random gradient descent optimization (SGD), the learning rate adopts an Adam optimizer strategy, the initial value is 0.001, the momentum is 0.9, and the attenuation rate is 3E-5.
And E4, dividing the user attention area according to the target image (medical image) input by the user.
In this embodiment, the activation function is a general activation function (ReLU), and the user attention area is a key division area requiring model learning division, in this embodiment, a black compact (Substantia Nigra Pars Compacta, SNpc).
And E5, displaying the nuclear magnetic resonance scanning image, and displaying a model segmentation result by using a red highlight region to assist a doctor in diagnosis.
All data obtained in this example were scanned by a 3.0T nuclear magnetic resonance scanner (model: discovery TM MR750, GE Healthcare). A total of 188 brain nuclear magnetic resonance imaging screening samples, including in particular 140 Parkinson disease patients and 48 healthy people. The important focusing area of the Parkinson disease is a substantia nigra compact part, so that the classification influence of the Parkinson patient and the healthy person is not needed to be considered in the segmentation task, and all segmentation labels are independently marked by a professional doctor of clinical experience of an eight-year imaging department.
The data are divided into two independent sets, namely training in the development process of algorithm models and other verification data for algorithm verification. Wherein the training set comprises 152 testers 'head nuclear magnetic resonance scan images and the verification set comprises 36 testers' head nuclear magnetic resonance scan images.
Verification by the method or the device is carried out to obtain a 'U' -shaped neural network model based on an improved attention module, wherein the Dairy coefficient (DSC) of the 'U' -shaped neural network model on verification data is 0.8769, the Accuracy (ACC) is 0.9992, the Sensitivity (SE) is 0.8935, the Specificity (SP) is 0.9995, and the area under a subject working characteristic curve (receiver operating characteristic curve, ROC) is 0.9465.
Meanwhile, experiments show that the test process of a batch of 32 images takes about 3 seconds, and all the tests of one patient take about 30 seconds, so that the device has higher time efficiency.
Compared with manual diagnosis, the medical image segmentation device 1 and the medical image segmentation method based on the improved attention module have high accuracy and high efficiency on the related segmentation task of the medical image, and are close to the diagnosis level of expert doctors, so that the medical image segmentation device and the medical image segmentation method based on the improved attention module are particularly suitable for the conditions of lack of basic medical resources and insufficient quantity of expert doctors. In addition, the device can also effectively assist doctors to carry out rapid diagnosis, and is beneficial to improving the diagnosis and treatment efficiency of the image department links of hospitals.
Example III
Fig. 2 is a schematic structural diagram of an image processing apparatus according to a third embodiment of the present application, where the present embodiment is applicable to a case of dividing an area of interest of a user in an image, especially a case of dividing an area of interest of a user such as a lesion in a medical image, and the method may be performed by a computer device that provides an image dividing function for the user, and the computer device may be a personal computer, a notebook computer, a tablet computer, or a server. The device comprises: a target image tensor acquisition module 210, a segmentation module 220, and an output module 230.
A target image tensor acquisition module 210, configured to acquire target tensor data of a target image;
a segmentation module 220 for inputting target tensor data into a target neural network model, the target neural network model comprising an encoder, a decoder, and an attention module;
the encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result;
and the output module 230 is used for carrying out segmentation feedback on the target image according to the output of the target neural network model.
On the basis of the above embodiment, the segmentation module 220 is configured to:
carrying out cavity convolution on the first feature matrix according to the cavity convolutions to obtain a plurality of second feature matrices, wherein each cavity convolution corresponds to the second feature matrix one by one;
performing mutual matrix multiplication according to the plurality of second feature matrixes to obtain a matrix multiplication result; determining an original attention map according to the matrix multiplication result;
transmitting the original attention force diagram to carry out a closing operation to obtain a cavity attention force diagram;
multiplying the first feature matrix by the hole attention map to obtain a third feature matrix.
On the basis of the above embodiment, the target image tensor obtaining module 210 is configured to:
preprocessing a target image according to a preset preprocessing mode to obtain a preprocessed image, wherein the preprocessing mode comprises an image clipping mode and an image normalization mode;
target tensor data is determined from the preprocessed image.
In the above embodiment, the target image is a medical image, and the medical image includes: images are taken using computed tomography images, nuclear magnetic resonance images, high definition cameras.
On the basis of the above embodiment, the output module 230 is configured to:
and outputting the target image, and determining the highlight region of the target image according to the image segmentation result.
On the basis of the above embodiment, the target image tensor obtaining module 210 is configured to:
desensitizing the target image to obtain a desensitized image;
target tensor data for the target image is determined from the desensitized image.
On the basis of the embodiment, the training device further comprises a training module. The training module is used for:
acquiring training tensor data of training images of a training set;
and training the target neural network model according to the training tensor data so as to optimize network parameters of the target neural network model.
According to the image processing device provided by the embodiment of the application, the target image tensor acquisition module 210 can acquire target tensor data of a target image; the segmentation module 220 inputs the target tensor data into a target neural network model, and an encoder of the target neural network model performs downsampling according to the target tensor data to obtain a first feature matrix; the attention module of the target neural network model is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder of the target neural network model is used for up-sampling according to the third feature matrix to obtain an image segmentation result; the output module 230 performs segmentation feedback of the target image according to the output of the target neural network model. Compared with the problem of low accuracy of U-Net image segmentation combined with an attention mechanism at present, the attention module of the target neural network model can process the first feature matrix of the target image based on hole convolution, the range of the receptive field can be improved through hole convolution, further, the region of interest of a user in the target image can be determined more accurately, segmentation is carried out according to the region, and accuracy of image segmentation is improved.
The image processing device provided by the embodiment of the application can execute the image processing method provided by any embodiment of the application, and has the corresponding functional modules and beneficial effects of the execution method.
Example IV
Fig. 3 is a schematic structural diagram of a computer device according to a fourth embodiment of the present application, and as shown in fig. 3, the computer device includes a processor 30, a memory 31, an input device 32 and an output device 33; the number of processors 30 in the computer device may be one or more, one processor 30 being taken as an example in fig. 3; the processor 30, the memory 31, the input means 32 and the output means 33 in the computer device may be connected by a bus or by other means, in fig. 3 by way of example.
The memory 31 is a computer-readable storage medium that can be used to store a software program, a computer-executable program, and modules such as program instructions/modules corresponding to the image processing method in the embodiment of the present application (for example, the target image tensor acquisition module 210, the segmentation module 220, and the output module 230 in the image processing apparatus). The processor 30 executes various functional applications of the computer device and data processing, i.e., implements the image processing method described above, by running software programs, instructions, and modules stored in the memory 31.
The memory 31 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, at least one application program required for functions; the storage data area may store data created according to the use of the terminal, etc. In addition, the memory 31 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some examples, memory 31 may further comprise memory remotely located relative to processor 30, which may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input device 32 is operable to receive input numeric or character information and to generate key signal inputs related to user settings and function control of the computer device. The output means 33 may comprise a display device such as a display screen.
Example five
A fifth embodiment of the present application also provides a storage medium containing computer-executable instructions, which when executed by a computer processor, are for performing an image processing method, the method comprising:
acquiring target tensor data of a target image;
inputting the target tensor data into a target neural network model, the target neural network model including an encoder, a decoder, and an attention module;
the encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix according to the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result;
and carrying out segmentation feedback on the target image according to the output of the target neural network model.
On the basis of the foregoing embodiment, the attention module is configured to perform hole convolution according to the first feature matrix to obtain a second feature matrix, and includes:
carrying out cavity convolution on the first feature matrix according to the cavity convolutions to obtain a plurality of second feature matrices, wherein each cavity convolution corresponds to the second feature matrix one by one;
accordingly, determining the original attention profile from the second feature matrix comprises:
performing mutual matrix multiplication according to the plurality of second feature matrixes to obtain a matrix multiplication result; determining an original attention map according to the matrix multiplication result;
accordingly, determining a hole attention map from the original attention map includes:
transmitting the original attention force diagram to carry out a closing operation to obtain a cavity attention force diagram;
accordingly, determining a third feature matrix from the first feature matrix and the hole attention map comprises:
multiplying the first feature matrix by the hole attention map to obtain a third feature matrix.
On the basis of the above embodiment, acquiring target tensor data of a target image includes:
preprocessing a target image according to a preset preprocessing mode to obtain a preprocessed image, wherein the preprocessing mode comprises an image clipping mode and an image normalization mode;
target tensor data is determined from the preprocessed image.
In the above embodiment, the target image is a medical image, and the medical image includes: images are taken using computed tomography images, nuclear magnetic resonance images, high definition cameras.
On the basis of the above embodiment, the method for performing segmentation feedback of the target image according to the output of the target neural network model includes:
and outputting the target image, and determining the highlight region of the target image according to the image segmentation result.
On the basis of the above embodiment, acquiring target tensor data of a target image includes:
desensitizing the target image to obtain a desensitized image;
target tensor data for the target image is determined from the desensitized image.
On the basis of the above embodiment, before inputting the tensor data into the target neural network model, the method further includes:
acquiring training tensor data of training images of a training set;
and training the target neural network model according to the training tensor data so as to optimize network parameters of the target neural network model.
Of course, the storage medium containing the computer executable instructions provided in the embodiments of the present application is not limited to the method operations described above, and may also perform the related operations in the image processing method provided in any embodiment of the present application.
From the above description of embodiments, it will be clear to a person skilled in the art that the present application may be implemented by means of software and necessary general purpose hardware, but of course also by means of hardware, although in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as a floppy disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a FLASH Memory (FLASH), a hard disk or an optical disk of a computer, etc., and include several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments of the present application.
It should be noted that, in the above-mentioned embodiments of the search apparatus, each unit and module included are only divided according to the functional logic, but not limited to the above-mentioned division, as long as the corresponding functions can be implemented; in addition, the specific names of the functional units are also only for distinguishing from each other, and are not used to limit the protection scope of the present application.
Note that the above is only a preferred embodiment of the present application and the technical principle applied. It will be understood by those skilled in the art that the present application is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the application. Therefore, while the application has been described in connection with the above embodiments, the application is not limited to the embodiments, but may be embodied in many other equivalent forms without departing from the spirit or scope of the application, which is set forth in the following claims.
Claims (9)
1. An image processing method, comprising:
acquiring target tensor data of a target image;
inputting the target tensor data to a target neural network model, the target neural network model comprising an encoder, a decoder, and an attention module;
the encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix based on the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result;
carrying out segmentation feedback on the target image according to the output of the target neural network model;
correspondingly, the attention module is configured to perform hole convolution according to the first feature matrix to obtain a second feature matrix, and includes: carrying out cavity convolution on the first feature matrix according to the cavity convolutions to obtain a plurality of second feature matrices, wherein each cavity convolution corresponds to the second feature matrix one by one;
correspondingly, the determining the original attention map according to the second feature matrix includes: performing mutual matrix multiplication according to the plurality of second feature matrixes to obtain a matrix multiplication result; determining an original attention map according to the matrix multiplication result;
accordingly, the determining a hole attention map according to the original attention map includes: carrying out transfer closing operation on the original attention force diagram to obtain a hole attention force diagram;
correspondingly, the determining a third feature matrix according to the first feature matrix and the hole attention map comprises: multiplying the first feature matrix with the hole attention map to obtain a third feature matrix.
2. The method of claim 1, wherein obtaining target tensor data for the target image comprises:
preprocessing the target image according to a preset preprocessing mode to obtain a preprocessed image, wherein the preprocessing mode comprises an image cutting mode and an image normalization mode;
and determining target tensor data according to the preprocessed image.
3. The method of claim 1, wherein the target image is a medical image, the medical image comprising: images are taken using computed tomography images, nuclear magnetic resonance images, high definition cameras.
4. The method of claim 1, wherein the performing segmentation feedback of the target image based on the output of the target neural network model comprises:
and outputting the target image, and determining a highlight region of the target image according to the image segmentation result.
5. The method of claim 1, wherein obtaining target tensor data for the target image comprises:
desensitizing the target image to obtain a desensitized image;
and determining target tensor data of the target image according to the desensitization image.
6. The method of claim 1, further comprising, prior to inputting the tensor data into a target neural network model:
acquiring training tensor data of training images of a training set;
and training the target neural network model according to the training tensor data so as to optimize network parameters of the target neural network model.
7. An image processing apparatus, comprising:
the target image tensor acquisition module is used for acquiring target tensor data of the target image;
a segmentation module for inputting the target tensor data into a target neural network model, the target neural network model comprising an encoder, a decoder, and an attention module;
the encoder is used for carrying out downsampling according to the target tensor data to obtain a first feature matrix; the attention module is used for carrying out cavity convolution according to the first feature matrix to obtain a second feature matrix; determining an original attention map from the second feature matrix; determining a hole attention map from the original attention map; determining a third feature matrix based on the first feature matrix and the hole attention map; outputting the third feature matrix to a value decoder; the decoder is used for up-sampling according to the third feature matrix to obtain an image segmentation result;
the output module is used for carrying out segmentation feedback on the target image according to the output of the target neural network model;
correspondingly, the attention module is configured to perform hole convolution according to the first feature matrix to obtain a second feature matrix, and includes: carrying out cavity convolution on the first feature matrix according to the cavity convolutions to obtain a plurality of second feature matrices, wherein each cavity convolution corresponds to the second feature matrix one by one;
correspondingly, the determining the original attention map according to the second feature matrix includes: performing mutual matrix multiplication according to the plurality of second feature matrixes to obtain a matrix multiplication result; determining an original attention map according to the matrix multiplication result;
accordingly, the determining a hole attention map according to the original attention map includes: carrying out transfer closing operation on the original attention force diagram to obtain a hole attention force diagram;
correspondingly, the determining a third feature matrix according to the first feature matrix and the hole attention map comprises: multiplying the first feature matrix with the hole attention map to obtain a third feature matrix.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the image processing method according to any of claims 1-6 when executing the program.
9. A storage medium containing computer executable instructions which, when executed by a computer processor, are for performing the image processing method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111042489.5A CN113793345B (en) | 2021-09-07 | 2021-09-07 | Medical image segmentation method and device based on improved attention module |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111042489.5A CN113793345B (en) | 2021-09-07 | 2021-09-07 | Medical image segmentation method and device based on improved attention module |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113793345A CN113793345A (en) | 2021-12-14 |
| CN113793345B true CN113793345B (en) | 2023-10-31 |
Family
ID=78879628
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111042489.5A Active CN113793345B (en) | 2021-09-07 | 2021-09-07 | Medical image segmentation method and device based on improved attention module |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113793345B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111681273B (en) * | 2020-06-10 | 2023-02-03 | 创新奇智(青岛)科技有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
| CN114387436B (en) * | 2021-12-28 | 2022-10-25 | 北京安德医智科技有限公司 | Wall coronary artery detection method and device, electronic device and storage medium |
| CN114565628B (en) * | 2022-03-23 | 2022-09-13 | 中南大学 | Image segmentation method and system based on boundary perception attention |
| CN115731243B (en) * | 2022-11-29 | 2024-02-09 | 北京长木谷医疗科技股份有限公司 | Spine image segmentation method and device based on artificial intelligence and attention mechanism |
Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018036293A1 (en) * | 2016-08-26 | 2018-03-01 | 杭州海康威视数字技术股份有限公司 | Image segmentation method, apparatus, and fully convolutional network system |
| CN109191472A (en) * | 2018-08-28 | 2019-01-11 | 杭州电子科技大学 | Based on the thymocyte image partition method for improving U-Net network |
| CN109685813A (en) * | 2018-12-27 | 2019-04-26 | 江西理工大学 | A kind of U-shaped Segmentation Method of Retinal Blood Vessels of adaptive scale information |
| CN110059586A (en) * | 2019-03-29 | 2019-07-26 | 电子科技大学 | A kind of Iris Location segmenting system based on empty residual error attention structure |
| WO2019153908A1 (en) * | 2018-02-11 | 2019-08-15 | 北京达佳互联信息技术有限公司 | Image recognition method and system based on attention model |
| CN111612807A (en) * | 2020-05-15 | 2020-09-01 | 北京工业大学 | Small target image segmentation method based on scale and edge information |
| WO2020177651A1 (en) * | 2019-03-01 | 2020-09-10 | 华为技术有限公司 | Image segmentation method and image processing device |
| WO2020215985A1 (en) * | 2019-04-22 | 2020-10-29 | 腾讯科技（深圳）有限公司 | Medical image segmentation method and device, electronic device and storage medium |
| CN111898709A (en) * | 2020-09-30 | 2020-11-06 | 中国人民解放军国防科技大学 | Image classification method and device |
| CN112017198A (en) * | 2020-10-16 | 2020-12-01 | 湖南师范大学 | Right ventricle segmentation method and device based on self-attention mechanism multi-scale features |
| CN112017191A (en) * | 2020-08-12 | 2020-12-01 | 西北大学 | Method for establishing and segmenting liver pathology image segmentation model based on attention mechanism |
| WO2020244108A1 (en) * | 2019-06-05 | 2020-12-10 | Boe Technology Group Co., Ltd. | Methods and apparatuses for semantically segmenting input image, and computer-program product |
| CN112101363A (en) * | 2020-09-02 | 2020-12-18 | 河海大学 | Full convolution semantic segmentation system and method based on cavity residual error and attention mechanism |
| WO2021031066A1 (en) * | 2019-08-19 | 2021-02-25 | 中国科学院深圳先进技术研究院 | Cartilage image segmentation method and apparatus, readable storage medium, and terminal device |
| CN112541918A (en) * | 2020-12-23 | 2021-03-23 | 山东师范大学 | Three-dimensional medical image segmentation method based on self-attention mechanism neural network |
| CN112750142A (en) * | 2021-01-26 | 2021-05-04 | 上海工程技术大学 | Ultrasonic image segmentation system and method based on side window attention mechanism |
| GB202104506D0 (en) * | 2021-03-30 | 2021-05-12 | Ucl Business Plc | Medical Image Analysis Using Neural Networks |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8209120B2 (en) * | 1997-10-22 | 2012-06-26 | American Vehicular Sciences Llc | Vehicular map database management techniques |
| US11195057B2 (en) * | 2014-03-18 | 2021-12-07 | Z Advanced Computing, Inc. | System and method for extremely efficient image and pattern recognition and artificial intelligence platform |
| US10755391B2 (en) * | 2018-05-15 | 2020-08-25 | Adobe Inc. | Digital image completion by learning generation and patch matching jointly |
| US10776671B2 (en) * | 2018-05-25 | 2020-09-15 | Adobe Inc. | Joint blur map estimation and blur desirability classification from an image |
| US11461998B2 (en) * | 2019-09-25 | 2022-10-04 | Samsung Electronics Co., Ltd. | System and method for boundary aware semantic segmentation |
-
2021
- 2021-09-07 CN CN202111042489.5A patent/CN113793345B/en active Active
Patent Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018036293A1 (en) * | 2016-08-26 | 2018-03-01 | 杭州海康威视数字技术股份有限公司 | Image segmentation method, apparatus, and fully convolutional network system |
| WO2019153908A1 (en) * | 2018-02-11 | 2019-08-15 | 北京达佳互联信息技术有限公司 | Image recognition method and system based on attention model |
| CN109191472A (en) * | 2018-08-28 | 2019-01-11 | 杭州电子科技大学 | Based on the thymocyte image partition method for improving U-Net network |
| CN109685813A (en) * | 2018-12-27 | 2019-04-26 | 江西理工大学 | A kind of U-shaped Segmentation Method of Retinal Blood Vessels of adaptive scale information |
| WO2020177651A1 (en) * | 2019-03-01 | 2020-09-10 | 华为技术有限公司 | Image segmentation method and image processing device |
| CN110059586A (en) * | 2019-03-29 | 2019-07-26 | 电子科技大学 | A kind of Iris Location segmenting system based on empty residual error attention structure |
| WO2020215985A1 (en) * | 2019-04-22 | 2020-10-29 | 腾讯科技（深圳）有限公司 | Medical image segmentation method and device, electronic device and storage medium |
| WO2020244108A1 (en) * | 2019-06-05 | 2020-12-10 | Boe Technology Group Co., Ltd. | Methods and apparatuses for semantically segmenting input image, and computer-program product |
| WO2021031066A1 (en) * | 2019-08-19 | 2021-02-25 | 中国科学院深圳先进技术研究院 | Cartilage image segmentation method and apparatus, readable storage medium, and terminal device |
| CN111612807A (en) * | 2020-05-15 | 2020-09-01 | 北京工业大学 | Small target image segmentation method based on scale and edge information |
| CN112017191A (en) * | 2020-08-12 | 2020-12-01 | 西北大学 | Method for establishing and segmenting liver pathology image segmentation model based on attention mechanism |
| CN112101363A (en) * | 2020-09-02 | 2020-12-18 | 河海大学 | Full convolution semantic segmentation system and method based on cavity residual error and attention mechanism |
| CN111898709A (en) * | 2020-09-30 | 2020-11-06 | 中国人民解放军国防科技大学 | Image classification method and device |
| CN112017198A (en) * | 2020-10-16 | 2020-12-01 | 湖南师范大学 | Right ventricle segmentation method and device based on self-attention mechanism multi-scale features |
| CN112541918A (en) * | 2020-12-23 | 2021-03-23 | 山东师范大学 | Three-dimensional medical image segmentation method based on self-attention mechanism neural network |
| CN112750142A (en) * | 2021-01-26 | 2021-05-04 | 上海工程技术大学 | Ultrasonic image segmentation system and method based on side window attention mechanism |
| GB202104506D0 (en) * | 2021-03-30 | 2021-05-12 | Ucl Business Plc | Medical Image Analysis Using Neural Networks |
Non-Patent Citations (3)
| Title |
|---|
| A Multi-Scaled Receptive Field Learning Approach for Medical Image Segmentation;P. Guo, X. Su, H. Zhang, M. Wang and F. Bao;《ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)》;全文 * |
| Deep Learning with Skip Connection Attention for Choroid Layer Segmentation in OCT Images;X. Mao et al.;《2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)》;全文 * |
| 基于多尺度密集网络与两级残差注意力的乳腺肿瘤分割方法研究;赵引;《CNKI优秀硕士学位论文全文库》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113793345A (en) | 2021-12-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113793345B (en) | Medical image segmentation method and device based on improved attention module | |
| Mahapatra et al. | Image super-resolution using progressive generative adversarial networks for medical image analysis | |
| US11908580B2 (en) | Image classification method, computer-readable storage medium, and computer device | |
| US11645748B2 (en) | Three-dimensional automatic location system for epileptogenic focus based on deep learning | |
| US20190191988A1 (en) | Screening method for automated detection of vision-degenerative diseases from color fundus images | |
| WO2018120942A1 (en) | System and method for automatically detecting lesions in medical image by means of multi-model fusion | |
| US11449717B2 (en) | System and method for identification and localization of images using triplet loss and predicted regions | |
| KR20190136577A (en) | Method for classifying images using deep neural network and apparatus using the same | |
| CN113240655B (en) | Method, storage medium and device for automatically detecting type of fundus image | |
| WO2023207743A1 (en) | Image detection method and apparatus, and computer device, storage medium and program product | |
| CN113744271A (en) | Neural network-based automatic optic nerve segmentation and compression degree measurement and calculation method | |
| CN113782184A (en) | Cerebral apoplexy auxiliary evaluation system based on facial key point and feature pre-learning | |
| WO2023047118A1 (en) | A computer-implemented method of enhancing object detection in a digital image of known underlying structure, and corresponding module, data processing apparatus and computer program | |
| CN112614573A (en) | Deep learning model training method and device based on pathological image labeling tool | |
| CN116091522A (en) | Medical image segmentation method, device, equipment and readable storage medium | |
| JP2020010804A (en) | Medical image diagnosis support apparatus, program, and medical image diagnosis support method | |
| CN112862785B (en) | CTA image data identification method, device and storage medium | |
| CN112862786B (en) | CTA image data processing method, device and storage medium | |
| CN111369564B (en) | Image processing method, model training method and model training device | |
| CN112967246A (en) | X-ray image auxiliary device and method for clinical decision support system | |
| CN116258717B (en) | Lesion recognition method, device, apparatus and storage medium | |
| KR102442591B1 (en) | Method, program, and apparatus for generating label | |
| CN117392468B (en) | Cancer pathology image classification system, medium and equipment based on multi-example learning | |
| CN116580446B (en) | Iris characteristic recognition method and system for vascular diseases | |
| CN112862787B (en) | CTA image data processing method, device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |