CN114332098A - Carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image - Google Patents
Carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image Download PDFInfo
- Publication number
- CN114332098A CN114332098A CN202111615971.3A CN202111615971A CN114332098A CN 114332098 A CN114332098 A CN 114332098A CN 202111615971 A CN202111615971 A CN 202111615971A CN 114332098 A CN114332098 A CN 114332098A
- Authority
- CN
- China
- Prior art keywords
- image
- magnetic resonance
- neural network
- images
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance images, which comprises the following steps: (1) acquiring a neck multi-sequence magnetic resonance image; (2) registering the multi-sequence magnetic resonance image to obtain a three-dimensional registered image; (3) inputting the registration image into the constructed U-net neural network model to obtain a plaque segmentation image corresponding to the neck; when a U-net neural network model is constructed, modifying channel parameters of the U-net neural network model according to the multiple sequence numbers; and constructing the U-net neural network model by taking the known registration image as an input and taking the corresponding patch mask image as a result. The invention predicts a new neck magnetic resonance image sample by learning a method for automatically predicting the position of an unstable plaque from a multi-sequence magnetic resonance image through a neural network model, judges whether the unstable plaque exists and gives a specific position as a reference for diagnosis of a doctor, thereby improving the diagnosis efficiency of the doctor.
Description
Technical Field
The invention belongs to the technical field of automatic image identification, and particularly relates to a carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance images.
Background
It is well known that shedding of atherosclerotic plaques of the carotid artery can form emboli, block intracranial arteries, cause ischemia of distal brain tissue and lead to cerebral infarction. High resolution magnetic resonance imaging (HR-MRI) has proven to be an effective tool for detecting atherosclerotic vulnerable plaques. The high-resolution magnetic resonance imaging can clearly display the external morphological characteristics, the internal structural components and the position distribution information of the plaque. HR-MRI includes black and bright blood sequences, time of flight sequences (TOF-MRA), in addition to traditional T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and Proton Dense (PDWI).
Magnetic resonance imaging is the only non-invasive imaging technology which can clearly display the whole-body atherosclerotic plaque at present. However, due to the huge number of three-dimensional high-resolution magnetic resonance vascular wall images, 500 images can be obtained for each examiner, and even an experienced professional doctor needs to spend a long time to complete the diagnosis of the examiner, so that the work efficiency is low.
To achieve fast intelligent segmentation of plaque images, researchers began to develop segmentation methods based on automatic recognition:
patent document No. CN109932720A discloses a method for intelligently segmenting intracranial plaque and carotid plaque in a magnetic resonance image, comprising: step S1, acquiring a magnetic resonance image of a blood vessel wall of a user; step S2, preprocessing the magnetic resonance image of the blood vessel wall to obtain a preprocessed image; step S3, segmenting the preset plaque area in the preprocessed image through a pre-trained convolutional neural network model; in step S4, a plaque tissue region segmentation image corresponding to the blood vessel wall magnetic resonance image is output. When the convolutional neural network model is constructed, the whole three-dimensional image magnetic resonance image is used as input, and a large amount of training set data is needed.
In order to overcome the above problem, patent document No. CN111598891A adopts a U-convolution network (U-Net), and has a good segmentation effect even when the amount of image data is small.
However, in the prior art, a single series of magnetic resonance images are adopted for model training, and the accuracy of the obtained neural network model is still not ideal.
Disclosure of Invention
The invention provides a carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance images.
A carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance images comprises the following steps:
(1) acquiring a neck multi-sequence magnetic resonance image;
(2) registering the multi-sequence magnetic resonance image to obtain a registered image;
(3) inputting the registration image into the constructed U-net neural network model to obtain a plaque segmentation image corresponding to the neck;
when a U-net neural network model is constructed, modifying channel parameters of the U-net neural network model according to the multi-sequence number (for example, when the sequence number is N, changing an input _ channel parameter of the U-net into N, and keeping the rest unchanged); and constructing the U-net neural network model by taking the known registration image as an input and taking the corresponding patch mask image as a result.
Preferably, when the U-net neural network model is constructed:
(i) firstly, acquiring a neck multi-sequence magnetic resonance image;
(ii) registering the multi-sequence magnetic resonance images to obtain a three-dimensional registered image;
(iii) labeling unstable carotid plaque areas in the registered images to obtain plaque mask images;
(iv) (iii) forming a training sample set by taking the registration image in the step (ii) as an input and the corresponding mask image as an output, and constructing the U-net neural network model.
Before modeling, slicing the obtained multi-sequence three-dimensional registration image, taking the obtained two-dimensional slice image as the input in the step (iv), and taking the corresponding patch mask image as a real prediction result to construct the U-net neural network model.
Because the labeling area in the data is small, only a small part of the two-dimensional slice images contain the labeling area. Preferably, the model is trained using the slice image including the marked unstable plaque region, and after the model meets a set requirement (for example, accuracy), the model is trained by inputting the slice image not including the marked unstable plaque region.
Meanwhile, before modeling, the input two-dimensional slice image needs to be standardized, so that the mean value of the distribution range is 0 and the variance is 1.
Preferably, in the model training process, data enhancement is carried out by one or more operations of randomly turning left and right, randomly amplifying or reducing, adding random Gaussian noise and randomly cutting a region into the same size for each two-dimensional slice image and the mask image thereof.
The plaque mask image can be manually marked by using the existing software (such as Slicer software), and the final plaque mask image format is a binary image. The labeled regions are distinguished from the unlabeled regions by defining each pixel value as 0/255. For example, the dot pixel value corresponding to the labeled region may be set to 0, and the dot pixel values in the remaining non-labeled regions may be set to 255.
When the U-net neural network model is trained, the training process is monitored by using the Focal loss function. The U-net model uses a 'coding-decoding' structure, wherein a coding part extracts a feature map of an image through a convolutional neural network, and a decoding part obtains a segmentation result from the feature map through a deconvolution operation.
Preferably, a test sample set is simultaneously constructed according to the steps (i) to (iii), a training sample set is used for optimizing the model, and the test sample set is used for model verification; when the verification result does not meet the requirement, continuing to train the model; and when the requirements are met, outputting the constructed U-net neural network model.
Preferably, in the model training process, the optimizer uses an Adam algorithm, and the learning rate adjustment strategy uses a cosine annealing method. And taking the value of the loss function of the test set as a judging standard for judging whether the model is good or bad. The Adam algorithm is a self-adaptive motion estimation algorithm, is an extension of a random gradient descent algorithm, and has a self-adaptive mechanism to enable the model training speed to be faster and the robustness to be stronger, reduce the dependence on the hyper-parameter learning rate and reduce the difficulty and time for training a neural network. Chord annealing (Cosine annealing) can reduce the learning rate by a Cosine function. The cosine value of the cosine function firstly slowly decreases with the increase of x, then rapidly decreases, and slowly decreases again. This fall pattern can be coordinated with the learning rate to produce good results in a very efficient computational manner.
Preferably, the multi-sequence magnetic resonance image employed by the present invention is three-dimensional magnetic resonance image data of a T1W sequence and a TOF sequence.
Preferably, a multi-view fusion method can be introduced to optimize the model and the detection method, and specifically comprises the following steps:
a model building stage: slicing the three-dimensional registration image and the corresponding patch mask image by adopting multiple visual angles respectively to obtain multiple groups of two-dimensional slice images and mask slice images, and constructing and obtaining U-net neural network models corresponding to the multiple visual angles respectively;
and (3) carrying out image segmentation: slicing the three-dimensional registration image to be segmented by adopting multiple visual angles, and respectively inputting the two-dimensional slice image of each visual angle into a corresponding U-net neural network model to respectively obtain a plurality of sub-segmented images; after all the sub-segmentation images are obtained, all the sub-segmentation images are fused to obtain a final patch segmentation image; optionally, binarization processing is performed on the sub-segmented image or the fused segmented image before or after fusion to obtain a final patch segmented image.
As a further preference, the invention employs three views for the slicing, respectively coronal, axial and sagittal planes of the three-dimensional magnetic resonance registration image.
Preferably, in the training stage or the actual application stage of the model, the three-dimensional magnetic resonance registration images are respectively sliced from multiple viewing angles and respectively trained to obtain three models (training stage) or respectively input the three models into a corresponding pair of models to be constructed, so as to obtain segmentation results (application stage) of the corresponding viewing angles; in the training stage, the slice data of a plurality of visual angles are used as input, and the mask images of the corresponding visual angles are used as results to realize the construction of the model; in the application stage, the slice data of a plurality of visual angles are respectively input into corresponding models, segmentation results (segmentation images) of the plurality of visual angle models are respectively obtained, and then the segmentation results are fused to obtain a final segmentation result. The advantage of multi-view fusion is that image spatial information can be better utilized, with better performance compared to single-view models.
Preferably, the fusion is performed by a weight-sum method.
After the final segmentation result is obtained, in order to ensure the clarity of the output result, morphological operations may be performed on the segmentation result, including: a) and for the three-dimensional binary segmentation result, performing image corrosion operation by using a cross convolution kernel. b) And (4) using a square convolution kernel to carry out image expansion operation on the corrosion operation result.
Compared with the prior art, the invention has the beneficial effects that:
the invention predicts a new neck magnetic resonance image sample by learning a method for automatically predicting the position of an unstable plaque from a multi-sequence magnetic resonance image through a neural network model, judges whether the unstable plaque exists and gives a specific position as a reference for diagnosis of a doctor, thereby improving the diagnosis efficiency of the doctor.
Drawings
FIG. 1 is a flow chart employed in an embodiment of the present invention;
FIG. 2 is a T1W series of two-dimensional slice images;
FIG. 3 is a patch mask image labeled for unstable patch areas;
FIG. 4 is a probability chart of the output of the U-net neural network without binarization processing;
fig. 5 is a probability map after the binarization processing.
Detailed Description
As shown in fig. 1, a carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image includes:
(1) acquiring a neck multi-sequence magnetic resonance image;
(2) registering images of different sequences to obtain three-dimensional image registration;
(3) registering the three-dimensional images, and respectively slicing the three-dimensional images from a coronal plane, an axial plane and a sagittal plane to respectively obtain slice images of three visual angles;
(4) respectively inputting the slice images of the three visual angles into the trained U-net neural network model to respectively obtain segmentation results corresponding to the three visual angles;
(5) and fusing the three segmentation results to obtain a final segmentation result.
In practice, the U-Net model is used as the neural network model of the method. U-Net was published in 2015 by Olaf Ronneberger, Philipp Fischer and Thomas Brox, and is a variant of FCN. The purpose of U-Net is to solve the problems of biomedical images, and since the effect is really good, the U-Net is widely applied to various directions of semantic segmentation, such as satellite image segmentation, industrial flaw detection and the like.
The U-Net network structure is symmetrical, similar to English letter U, so it is called U-Net. The whole picture consists of blue/white frames and arrows of various colors, wherein the blue/white frames represent feature maps; blue arrows represent a 3 × 3 convolution for feature extraction; gray arrows indicate skip-connection for feature fusion; red arrows indicate pooling for dimensionality reduction; the green arrow represents the upsample, used to recover the dimensionality; the cyan arrow represents a 1 × 1 convolution for output of the result.
The U-net model uses a 'coding-decoding' structure, wherein a coding part extracts a feature map of an image through a convolutional neural network, and a decoding part obtains a segmentation result from the feature map through a deconvolution operation. And obtaining the confidence degree of each pixel classified as an unstable plaque area by using sigmoid operation on each pixel of the segmentation result.
The U-Net encoder part uses convolutional layers, batch normalization layers and maximum pooling layers. The function of the batch normalization layer is to increase the robustness of the model.
The U-Net decoder uses convolutional layers, interpolation layers.
Convolution layers in U-Net all use convolution kernels of 3x3 size.
The number of layers and the number of channels of the U-net can be debugged according to specific data, the number of layers and the number of channels are reduced if overfitting is carried out, and the number of layers and the number of channels are increased if the characteristics are not obvious enough.
Focal loss was used as a loss function for the present method. The setting parameter is alpha-0.9 and gamma-0.4.
The actual modeling process is as follows:
(i) acquiring a neck multi-sequence magnetic resonance image;
(ii) registering the images of different sequences to obtain a three-dimensional registered image;
the applicable multi-sequence three-dimensional magnetic resonance image comprises time-of-flight MR blood vessel imaging (3D-TOF), T1 weighted image (T1WI), T1 weighted image enhanced scan (CE-T1WI), T2 weighted image (T2WI), proton density weighted image (PWI) and the like. In the embodiment, a multi-sequence magnetic resonance image is a T1W sequence and a TOF sequence magnetic resonance image, and three-dimensional magnetic resonance images of the T1W sequence and the TOF sequence are registered;
(iii) labeling an unstable carotid plaque area by using a mask to obtain a binary three-dimensional mask image;
in this embodiment, the Slicer software is used to manually frame the region of the unstable plaque in the three-dimensional magnetic resonance image by using a cube. And then, carrying out binarization operation on the marked image, wherein the pixel value of the marked area is set to be 0, and the pixel values of other areas are set to be 255.
Meanwhile, slicing the binary three-dimensional mask image according to the corresponding visual angle according to different visual angles (or coordinate axes) adopted in the step (4); obtaining a corresponding two-dimensional slice mask image; one of the resulting binarized two-dimensional masked slice images is shown in fig. 3.
(iv) Simultaneously, slicing the three-dimensional registration image from a coronal plane, an axial plane and a sagittal plane to respectively obtain two-dimensional slice images of three visual angles; fig. 2 shows a two-dimensional slice image corresponding to the T1W sequence.
The data normalization process is performed for each two-dimensional slice image, in which the mean value of the numerical distribution range is 0 and the variance is 1 by the following formula.
x'＝(x-mean(x))/(std(x))
Wherein x' is the pixel value of a certain point after standardization; x is the pixel value before a certain point is normalized; mean (x) is the average pixel value of the pixel points in the two-dimensional slice image; std (x) is the standard deviation of pixel values of pixel points in the two-dimensional slice image.
(v) Respectively taking the slice images of the three standardized visual angles and the corresponding two-dimensional slice mask images as input and real prediction results to construct a U-net model;
in the process of training the model, because the marked areas in the data are small, only a few slice images contain the marked areas. In this embodiment, in the initial training stage, only the slice image including the annotation data is used for training, and after the model has a certain accuracy, the slice image without the annotation data is added in proportion.
The original image has different resolutions, imaging regions and voxel spacings, and therefore requires normalization. Before the registration operation, the following process can be performed:
first, the N4 bias field correction algorithm is used to correct for intensity inhomogeneities in all sequence magnetic resonance image volumes.
All sequence images are then resampled using B-spline interpolation to obtain a 3D image with approximately isotropic voxel size.
Thereafter, an automatic registration method is applied to match the volumes based on the adaptive mask with 3D rigidity and affine transformation.
The volume of TOF was chosen as the reference image and the T1W, T1, T2 and other applicable sequence images were chosen as the registration images. All sequence volumes are unified to the same spatial volume/pixel size and correspond to the same artery location.
Meanwhile, in order to enhance the generalization ability of the model, data enhancement needs to be performed on the data in the model training process, that is, new training data is obtained by performing some random operations on the original data, which are respectively as follows:
(a) and randomly turning left and right of each image and the label thereof, namely taking a random number between 0 and 1, if the random number is more than 0.5, turning the image left and right, and otherwise, keeping unchanged.
(b) And randomly scaling each image and the label thereof, taking a random number between 0.5 and 1.5 as a scaling ratio, and simultaneously scaling the length and the width of the image and the mask to the random ratio. Random scaling enables the neural network to learn different sized targets, rather than a single size, effectively increasing model accuracy.
(c) Random Gaussian noise is added to each image, namely a random number which is in accordance with normal distribution with the average value of 0 and the standard deviation of 0.05 is added to the value of each pixel of the image.
(d) And (4) random area cutting is carried out on each image and the label thereof, so that the images of all input neural networks have the same size, the batch input is facilitated, and the calculation speed is accelerated. Specifically, a rectangular area of a predetermined size is randomly cut out for each image and its label, and if the size of the image is smaller than the predetermined size, the image is extended (for example, two in a row) and then cut out.
In order to further increase the generalization ability of the model, a data set can be divided into a training set and a test set, the model is optimized by using the training set, and then the generalization ability of the model is judged by using the test set. During model training, the optimizer uses the Adam algorithm. The initial learning rate was set to 0.0001. The learning rate adjustment strategy uses a cosine annealing method. And taking the value of the loss function of the test set as a judging standard for judging whether the model is good or bad.
In this embodiment, the loss function adopted in the training process of the training set is a Focal loss function, which is an improvement of the cross entropy loss function, and improves the effect of small target segmentation.
The Focal loss function formula is as follows, where p is the predicted value, between 0 and 1.
Focal(pt)＝-α(1-pt)γlog pt
y is a mask value of a certain pixel;
when the test set is tested, the value of the loss function of the test set is used as the evaluation standard for the good and bad training of the model, the loss function is not particularly limited, and common loss functions can be adopted.
In the step (5), the multi-modal (visual angle) three-dimensional image data is sliced into two-dimensional images according to three directions and sequentially input into a model, and then a column of two-dimensional segmentation result images are combined into a three-dimensional segmentation result according to the original position. The sub-division results (sub-division images) of the plurality of models are weighted-averaged, and the value of each pixel is binarized using a fixed threshold, thereby obtaining a final patch division image.
In the step (5), after obtaining the binary segmentation result (segmentation image), morphological operation can be performed on the binary segmentation result to improve the result precision, and the method specifically comprises the following steps:
and for the three-dimensional binary segmentation result, performing image corrosion operation by using a cross convolution kernel. The aim is to filter out isolated segmented regions.
And (4) using a square convolution kernel to carry out image expansion operation on the corrosion operation result. The object is to expand the segmentation range. Fig. 4 and 5 are plaque distribution probability maps (i.e., segmented images) before and after binarization, respectively. The value of each pixel in the graph indicates the probability of being judged as an unstable plaque, and the whiter the probability is higher.
Claims (10)
1. A carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance images is characterized by comprising the following steps:
(1) acquiring a neck multi-sequence magnetic resonance image;
(2) registering the multi-sequence magnetic resonance image to obtain a three-dimensional registered image;
(3) inputting the registration image into the constructed U-net neural network model to obtain a plaque segmentation image corresponding to the neck;
when a U-net neural network model is constructed, modifying channel parameters of the U-net neural network model according to the multiple sequence numbers; and constructing the U-net neural network model by taking the known registration image as an input and taking the corresponding patch mask image as a result.
2. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 1, wherein the U-net neural network model is constructed by:
(i) firstly, acquiring a neck multi-sequence magnetic resonance image;
(ii) registering the multi-sequence magnetic resonance images to obtain a three-dimensional registered image;
(iii) labeling unstable carotid plaque areas in the registered images to obtain plaque mask images;
(iv) (iii) forming a training sample set by taking the registration image in the step (ii) as an input and the corresponding mask image as an output, and constructing the U-net neural network model.
3. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 2, characterized in that the three-dimensional registration image is sliced, and the obtained two-dimensional slice image is used as the input in step (iv) to construct the U-net neural network model.
4. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 3, characterized in that the model is trained by using the slice image including the marked unstable plaque area, and after the model meets the set requirement, the slice image not including the marked unstable plaque area is input to continue training the model.
5. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 3, characterized in that in the model training process, data enhancement is performed by one or more operations of random left-right turning, random amplification or reduction, random Gaussian noise addition, and random region clipping to the same size for each two-dimensional slice image and its mask image.
6. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 2, characterized in that the Focal loss function is used when training the U-net neural network model by using the training sample set.
7. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 2, characterized in that a test sample set is constructed simultaneously according to the steps (i) - (iii), a training sample set is used for optimizing the model, and the test sample set is used for model verification; when the verification result does not meet the requirement, continuing to train the model; and when the requirements are met, outputting the constructed U-net neural network model.
8. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 2, characterized in that the multi-sequence magnetic resonance image is two or more of three-dimensional time-of-flight MR angiography, T1 weighted image, T1 weighted image enhancement scan, T2 weighted image (T2WI), proton density weighted image.
9. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image as claimed in claim 2, characterized in that: when registering images:
first, the N4 bias field correction algorithm is used to correct intensity inhomogeneities of all sequence magnetic resonance image volumes;
then, resampling all sequence images using B-spline interpolation to obtain a 3D image with approximately isotropic voxel size;
thereafter, an automatic registration method is applied to match these 3D images based on an adaptive mask with 3D rigidity and affine transformation, enabling registration of the images.
10. The carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance imaging according to claim 3, characterized in that, in the model construction stage or when the constructed U-net neural network model is used for image segmentation:
a model building stage: slicing the three-dimensional registration image and the corresponding patch mask image by adopting multiple visual angles respectively to obtain multiple groups of two-dimensional slice images and mask slice images, and constructing and obtaining U-net neural network models corresponding to the multiple visual angles respectively;
and (3) carrying out image segmentation: slicing the three-dimensional registration image to be segmented by adopting multiple visual angles, and respectively inputting the two-dimensional slice image of each visual angle into a corresponding U-net neural network model to respectively obtain a sub-segmented image; after all the sub-segmentation images are obtained, all the sub-segmentation images are fused to obtain a final patch segmentation image; optionally, binarization processing is performed on the sub-segmented image or the fused segmented image before or after fusion to obtain a final patch segmented image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111615971.3A CN114332098A (en) | 2021-12-27 | 2021-12-27 | Carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111615971.3A CN114332098A (en) | 2021-12-27 | 2021-12-27 | Carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114332098A true CN114332098A (en) | 2022-04-12 |
Family
ID=81012416
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111615971.3A Pending CN114332098A (en) | 2021-12-27 | 2021-12-27 | Carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114332098A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114792315A (en) * | 2022-06-22 | 2022-07-26 | 浙江太美医疗科技股份有限公司 | Medical image visual model training method and device, electronic equipment and storage medium |
| CN116681706A (en) * | 2023-08-04 | 2023-09-01 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Medical image processing method and device, electronic equipment and storage medium |
| CN117115187A (en) * | 2023-10-24 | 2023-11-24 | 北京联影智能影像技术研究院 | Carotid artery wall segmentation method, carotid artery wall segmentation device, carotid artery wall segmentation computer device, and carotid artery wall segmentation storage medium |
-
2021
- 2021-12-27 CN CN202111615971.3A patent/CN114332098A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114792315A (en) * | 2022-06-22 | 2022-07-26 | 浙江太美医疗科技股份有限公司 | Medical image visual model training method and device, electronic equipment and storage medium |
| CN116681706A (en) * | 2023-08-04 | 2023-09-01 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Medical image processing method and device, electronic equipment and storage medium |
| CN116681706B (en) * | 2023-08-04 | 2023-11-10 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Medical image processing method and device, electronic equipment and storage medium |
| CN117115187A (en) * | 2023-10-24 | 2023-11-24 | 北京联影智能影像技术研究院 | Carotid artery wall segmentation method, carotid artery wall segmentation device, carotid artery wall segmentation computer device, and carotid artery wall segmentation storage medium |
| CN117115187B (en) * | 2023-10-24 | 2024-02-09 | 北京联影智能影像技术研究院 | Carotid artery wall segmentation method, carotid artery wall segmentation device, carotid artery wall segmentation computer device, and carotid artery wall segmentation storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114332098A (en) | Carotid artery unstable plaque segmentation method based on multi-sequence magnetic resonance image | |
| CN109978037B (en) | Image processing method, model training method, device and storage medium | |
| CN109035172B (en) | Non-local mean ultrasonic image denoising method based on deep learning | |
| CN110889852A (en) | Liver segmentation method based on residual error-attention deep neural network | |
| CN110705555A (en) | Abdomen multi-organ nuclear magnetic resonance image segmentation method, system and medium based on FCN | |
| CN103810473B (en) | A kind of target identification method of human object based on HMM | |
| WO2019071976A1 (en) | Panoramic image saliency detection method based on regional growth and eye movement model | |
| CN115496771A (en) | Brain tumor segmentation method based on brain three-dimensional MRI image design | |
| CN112446891A (en) | Medical image segmentation method based on U-Net network brain glioma | |
| Venkatesh et al. | Intensity inhomogeneity correction of MRI images using InhomoNet | |
| CN114266939B (en) | Brain extraction method based on ResTLU-Net model | |
| CN113362331A (en) | Image segmentation method and device, electronic equipment and computer storage medium | |
| CN113160265A (en) | Construction method of prediction image for brain corpus callosum segmentation for corpus callosum state evaluation | |
| CN110648331A (en) | Detection method for medical image segmentation, medical image segmentation method and device | |
| CN114170244A (en) | Brain glioma segmentation method based on cascade neural network structure | |
| CN117392375A (en) | Target detection algorithm for tiny objects | |
| CN112184731A (en) | Multi-view stereo depth estimation method based on antagonism training | |
| Kezia et al. | A color-texture based segmentation method to extract object from background | |
| CN110991408A (en) | Method and device for segmenting white matter high signal based on deep learning method | |
| CN115719357A (en) | Multi-structure segmentation method for brain medical image | |
| CN112330660B (en) | Sperm tail detection method and system based on neural network | |
| Carmo et al. | Extended 2d volumetric consensus hippocampus segmentation | |
| Stitt et al. | Fuzzy c-means image segmentation of side-scan sonar images. | |
| CN117612050A (en) | Small target object detection method for construction site | |
| CN112330642B (en) | Pancreas image segmentation method and system based on double-input full convolution network |
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 |