CN111539956A - Cerebral hemorrhage automatic detection method based on brain auxiliary image and electronic medium - Google Patents
Cerebral hemorrhage automatic detection method based on brain auxiliary image and electronic medium Download PDFInfo
- Publication number
- CN111539956A CN111539956A CN202010643129.XA CN202010643129A CN111539956A CN 111539956 A CN111539956 A CN 111539956A CN 202010643129 A CN202010643129 A CN 202010643129A CN 111539956 A CN111539956 A CN 111539956A
- Authority
- CN
- China
- Prior art keywords
- brain
- image
- neural network
- cerebral hemorrhage
- auxiliary image
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
- G06T7/0014—Biomedical image inspection using an image reference approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an automatic cerebral hemorrhage detection method and an electronic medium based on a brain auxiliary image, which are applied to the technical field of image processing, wherein the method comprises the steps of carrying out CT scanning on the head of a patient to obtain a brain CT image; generating a brain auxiliary image according to the brain CT image; inputting the brain CT image and the brain auxiliary image into a pre-trained neural network, automatically detecting a cerebral hemorrhage region in the brain CT image by combining the brain auxiliary image, and outputting a mask comprising the cerebral hemorrhage region and type. The invention combines the brain auxiliary image with the brain CT image scanned by the patient, improves the accuracy of identifying the cerebral hemorrhage area and type of the neural network, reduces the technical requirements of doctors in the diagnosis process and greatly improves the diagnosis efficiency.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a cerebral hemorrhage automatic detection method based on a brain auxiliary image and an electronic medium.
Background
The computer aided detection analysis of the cerebral hemorrhage in the CT image is to identify and analyze the type and the region of the cerebral hemorrhage in the image. Has important significance for the subsequent treatment of the cerebral apoplexy. Image processing software for cerebral hemorrhage in the prior art can detect cerebral hemorrhage images and output types and areas of cerebral hemorrhage, and most of detection is based on an artificial intelligence method, namely, a neural network is constructed firstly, then a data set marked artificially is used, and the data set is used for training to obtain the neural network.
The AI-assisted analysis method for cerebral hemorrhage disclosed currently generally inputs an image of cerebral hemorrhage into a convolutional neural network, and then outputs the image to obtain the cerebral hemorrhage area and the category of cerebral hemorrhage. However, there are many similarities between each type of cerebral hemorrhage, such as a small difference in CT value and many similarities in shape, and training by only inputting images of cerebral hemorrhage does not necessarily lead to a very accurate result, and therefore, a diagnosis is required by a doctor at the time of diagnosis by combining CT values with the brain anatomy.
Meanwhile, if the image of cerebral hemorrhage is input, the spatial structure or the planning information is used as a guide, which is easy to cause errors. The prior art has improved the accuracy of identification by transforming a coordinate system, such as a complex coordinate system along the skull, into the identification system. But after the coordinate system is transformed, the shapes of the bleeding areas are more uniform, and the recognition effect is improved.
Disclosure of Invention
The technical purpose is as follows: aiming at the defect of low detection accuracy of cerebral hemorrhage regions and types in the prior art, the invention discloses an automatic cerebral hemorrhage detection method based on a brain auxiliary image and an electronic medium.
The technical scheme is as follows: in order to achieve the technical purpose, the invention adopts the following technical scheme.
An automatic cerebral hemorrhage detection method based on a brain auxiliary image is characterized by comprising the following steps:
s1, acquiring a brain CT image of the patient: performing CT scanning on the head of a patient, and acquiring a brain CT image through a CT reconstruction algorithm;
s2, acquiring a brain auxiliary image: generating a brain auxiliary image according to the brain CT image, wherein the brain auxiliary image comprises spatial position information of the brain anatomical structure;
s3, training in the automatic cerebral hemorrhage detection neural network, and outputting a processed brain CT image: the brain CT image and the brain auxiliary image are used as input, the two images are input into a pre-trained automatic cerebral hemorrhage detection neural network, the cerebral hemorrhage area in the brain CT image is automatically detected by combining the brain auxiliary image, the output processed brain CT image comprises a mask of the cerebral hemorrhage area and type, the mask is read by a doctor, and the doctor further diagnoses the cerebral hemorrhage by combining the processed brain CT image.
Preferably, the brain auxiliary image in S2 includes spatial location information of the brain anatomical structure, specifically including spatial location information of the skull, the dura mater, the arachnoid, the ventricle, and the lobe.
Preferably, in S2, the generating of the brain auxiliary image from the brain CT image is a brain auxiliary image based on a standard model, which includes:
acquiring a standard model including a brain anatomical structure, and taking the standard model as a template;
performing non-rigid registration transformation on the standard model according to the brain CT image to obtain a non-rigid registration coordinate mapping relation;
and outputting the brain auxiliary image based on the standard model according to the non-rigid registered coordinate mapping relation.
Preferably, in S2, the generating of the brain auxiliary image from the brain CT image is a brain auxiliary image based on the skull contour distance by the following specific process:
extracting a skull region from a brain CT image;
calculating the shortest distance from the pixel point to the extracted skull region for all pixel points of the brain image;
and outputting the brain auxiliary image based on the skull contour distance.
Preferably, in S2, the generating the brain auxiliary image according to the brain CT image is an image-based brain auxiliary image of a neural network, and the specific process is as follows:
constructing an image generation neural network: the image generation neural network is a convolution neural network, and the convolution neural network is a backbone network of U-Net or VGG;
pre-training the convolutional neural network: generating a brain auxiliary image based on a standard model or a brain auxiliary image based on a skull contour distance, using a brain CT image as the input of a convolutional neural network, pre-training the convolutional neural network, calculating a loss function of the image which is output by the convolutional neural network and contains a cerebral hemorrhage region and the generated brain auxiliary image based on the standard model or the brain auxiliary image based on the skull contour distance until the loss function is smaller than a threshold value, and completing the pre-training of the convolutional neural network;
outputting a brain auxiliary image: the image containing the cerebral hemorrhage area output by the convolutional neural network is the brain auxiliary image of the neural network generated based on the image.
Preferably, the pre-trained cerebral hemorrhage automatic detection neural network in S3 has a specific structure: a backbone network of U-Net or VGG is used as a backbone network of the cerebral hemorrhage automatic detection neural network, and a convolution network is connected behind the backbone network, wherein a brain CT image is input into the backbone network, and a brain auxiliary image is input into the convolution network.
Preferably, the two images are input into a pre-trained automatic cerebral hemorrhage detection neural network in S3, where the pre-training process of the neural network is as follows:
constructing a sample set: acquiring a plurality of brain CT images in a database, marking bleeding areas and bleeding types in the brain CT images, forming image pairs by the marked brain CT images and corresponding brain auxiliary images, and constructing a sample set with a plurality of image pairs;
inputting a plurality of picture pairs in the sample set into the constructed automatic cerebral hemorrhage detection neural network: a backbone network of U-Net or VGG is used as a backbone network of the cerebral hemorrhage automatic detection neural network, a convolution network is connected behind the backbone network, the brain CT image in the sample set image pair is input into the backbone network, and the brain auxiliary image in the sample set image pair is input into the convolution network;
pre-training the constructed automatic cerebral hemorrhage detection neural network by a gradient descent method: the method comprises the steps that features of a brain CT image are extracted by a backbone network and output to a convolution network, the convolution network carries out convolution processing by combining the features and the brain auxiliary image, and a neural network is trained through a gradient descent method until a loss function of the neural network for automatically detecting cerebral hemorrhage is smaller than a threshold value.
Preferably, the acquiring of the brain auxiliary image in S2 and the training in the automatic brain hemorrhage detecting neural network in S3 and the outputting of the processed brain CT image are implemented by using a total network, where the total network includes an image generating neural network and an automatic brain hemorrhage detecting neural network, the image generating neural network is used to generate and output the brain auxiliary image according to the input brain CT image, the automatic brain hemorrhage detecting neural network is used to automatically detect a brain hemorrhage region in the brain CT image according to the input brain CT image and the brain auxiliary image output by the image generating neural network, a mask including the region and type of the brain hemorrhage in the processed brain CT image is output and is provided for a doctor to read, and the doctor further diagnoses the brain hemorrhage by combining the processed brain CT image.
An electronic medium comprising a memory and a processor, the memory being connected to the processor, the memory storing at least one instruction executable by the processor, the at least one instruction, when executed by the processor, implementing a method for automatic detection of cerebral hemorrhage based on a brain supplementary image as set forth in any one of the above.
Has the advantages that:
1. according to the invention, the brain auxiliary image and the brain CT image scanned by the patient are combined for use, the brain auxiliary image comprises the spatial position information of the bleeding area, the brain CT image comprises the pixel value information of the bleeding area, and the combination of the two improves the recognition accuracy of the cerebral bleeding area and type in the automatic cerebral bleeding detection process, reduces the technical requirements of a doctor in the diagnosis process and greatly improves the diagnosis efficiency;
2. according to the method, the brain auxiliary image is obtained by three methods, so that the flexibility and the accuracy of the scheme are greatly improved;
3. according to the scheme, the total network can be adopted to acquire the brain auxiliary image and detect and output information such as the cerebral hemorrhage area, and convenience of the scheme is improved.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a schematic flow diagram of the process of the present invention;
FIG. 3 is a schematic flow diagram of a method employing a regional tissue generator;
fig. 4 is a schematic diagram of cerebral hemorrhage types in a brain auxiliary image;
FIG. 5 is a CT image of the brain for four types of cerebral hemorrhage;
FIG. 6 is a basic flow diagram of non-rigid registration;
FIG. 7 is a schematic diagram of calculating the shortest distance from a pixel point to a skull region;
FIG. 8 is an input brain CT image and an output brain auxiliary image in an image-generating neural network;
FIG. 9 is a schematic flow chart of an automatic neural network for cerebral hemorrhage detection;
fig. 10 is a flow chart of the overall network.
Detailed Description
The automatic detection method for cerebral hemorrhage based on the brain auxiliary image and the electronic medium of the present invention will be further explained and explained with reference to the drawings.
As shown in fig. 5, the existing cerebral hemorrhage includes several types of hemorrhage, such as epidural hemorrhage, subdural hemorrhage, cerebral parenchyma hemorrhage, brainstem hemorrhage, subarachnoid hemorrhage, and the like. In the CT image of the brain of a patient, the pixel value of the normal region is 20-40HU, the pixel value of the region of epidural hemorrhage, subdural hemorrhage, cerebral parenchyma hemorrhage, and brainstem hemorrhage is 45-75HU, and the pixel value of the region of subarachnoid hemorrhage is 40-75HU, so it is difficult to judge the hemorrhage type based on the pixel values alone. Generally, the manual diagnosis is based on the pixel value, and combines the position of the bleeding area and the shape of the bleeding to judge the bleeding area and type.
As shown in fig. 1 and fig. 2, an automatic cerebral hemorrhage detection method based on a brain auxiliary image includes the following steps:
step one, acquiring a brain CT image of a patient: performing CT scanning on the head of a patient to obtain a brain CT image; CT scanning of the patient's head produces an image typically 2.5mm-5mm thick.
Step two, acquiring a brain auxiliary image: generating a brain auxiliary image according to the brain CT image; the brain auxiliary image generated here mainly has the function of giving more information to subsequent processing, and the brain auxiliary image includes spatial position information of the brain anatomical structure, specifically including spatial position information of the skull, the dura mater, the arachnoid, the ventricle and the lobe, as shown in fig. 4. The type of cerebral hemorrhage is usually determined by the shape of the hemorrhage area and the relative spatial position of the hemorrhage area in the skull. The general neural network for image segmentation is mainly based on a convolutional neural network, and the network has invariance of space translation, is beneficial to extracting morphological characteristics, but is insensitive to space positions, so that space position information of a bleeding area is supplemented by combining a brain auxiliary image, and the accuracy of automatic detection of the cerebral bleeding area is improved.
According to the method, the brain auxiliary image is obtained by three methods, so that the flexibility and the accuracy of the scheme are greatly improved; three brain auxiliary image acquisition modes are as follows:
and generating the brain auxiliary image from the brain CT image into the brain auxiliary image based on the standard model.
The standard model comprises a brain anatomical structure, and the standard model is used as a template and is subjected to certain non-rigid registration transformation, so that the anatomical structure which is in line with the brain of the current patient can be obtained, namely a corresponding brain auxiliary image is obtained.
Non-rigid registration (non-rigid registration) is a commonly used registration method in image processing, and is used for matching two similar images and distinguishing the deformation of the images. This variant is to find a mapping of pixels between two images. Taking a three-dimensional image as an example, the basic flow of non-rigid registration is shown in fig. 6, which shows a current brain image,
is a template image of a standard CT,
for mapping the three-dimensional coordinates obtained by the registration process, the mapping can map the brain tissue existing on the template imageStructure mapping to a coordinate system in the current brain image; the specific process is formulated as follows:
wherein the content of the first and second substances,
the registered image, i.e. the template image is deformed to a shape similar to the current brain image according to the registration procedure of fig. 6.
Taking a three-dimensional image as an example, the method processes the brain CT image with reference to the attached figure 6:
wherein the content of the first and second substances,
for brain-aided images based on a standard model,
is the brain anatomy of a standard model,
to map the three-dimensional coordinates of the brain anatomy of the standard model non-rigidly registered from the brain auxiliary image,
the three-dimensional coordinates of the brain auxiliary image and the type of the brain auxiliary image are respectively represented, and the value of n is 2 for the brain auxiliary image based on the standard model.
And generating a brain auxiliary image according to the brain CT image, wherein the brain auxiliary image is based on the skull contour distance.
The brain auxiliary image may also be a brain auxiliary image based on a skull contour distance generated according to a brain CT image, and the spatial position information of the anatomical structure of the brain, such as spatial position information of the skull, the dura mater, the arachnoid, the ventricles of the brain, the lobes of the brain, and the like, is acquired, for example, a skull region is extracted based on the current brain CT image of the patient, then, the shortest distance from the pixel to the skull region is calculated for all pixels of the brain CT image, and further, the spatial position information corresponding to the anatomical structure of the brain is acquired.
As shown in fig. 7, taking a two-dimensional image as an example, the calculation formula of the shortest distance from a pixel point to a skull region here is:
wherein the content of the first and second substances,
is a two-dimensional coordinate of a pixel point,
is a brain auxiliary image based on the skull contour distance. As shown in fig. 3, here the acquisition of the brain auxiliary image based on the skull contour distance by equation (3) is implemented in the tissue region generator.
Generating a brain auxiliary image from the brain CT image is an image-based brain auxiliary image of a neural network.
As shown in fig. 9, the brain auxiliary image can also be obtained by constructing an image-generating neural network and outputting the image-generating neural network, and the image-generating neural network is first constructed: the image generation neural network is a convolution neural network, and the convolution neural network is a backbone network of U-Net or VGG;
pre-training the convolutional neural network: taking a two-dimensional image as an example, a brain auxiliary image based on a standard model or a brain auxiliary image based on a skull contour distance is generated, i.e.
Taking the brain CT image as the input of the convolutional neural network, pre-training the convolutional neural network, and outputting the packet of the convolutional neural networkImage of area containing cerebral hemorrhage and
and calculating a loss function until the loss function is smaller than a threshold value, completing pre-training of the convolutional neural network, wherein an image including a cerebral hemorrhage region output by the convolutional neural network is a brain auxiliary image based on the image generation neural network, and the image is a brain CT image input in the image generation neural network and an output brain auxiliary image in the attached figure 8. Wherein, the calculation formula of the loss function is as follows:
wherein
Is an image of the area containing cerebral hemorrhage output by the convolutional neural network. The images are all two-dimensional images, where the dimensions of the images
May be (512, n), where the first two dimensions represent two-dimensional coordinates of the image and the last dimension n represents a category of the brain auxiliary image, the value of this dimension n being 1 based on the brain auxiliary image of the skull contour distance; for a brain-aided image based on a standard model, the value of n is 2.
Step three, training in the cerebral hemorrhage automatic detection neural network, and outputting information such as cerebral hemorrhage areas: the brain CT image and the brain auxiliary image are used as input, the two images are input into a pre-trained automatic cerebral hemorrhage detection neural network, the cerebral hemorrhage area in the brain CT image is automatically detected by combining the brain auxiliary image, the output processed brain CT image comprises a mask of the cerebral hemorrhage area and type, the mask is read by a doctor, and the doctor further diagnoses the cerebral hemorrhage by combining the processed brain CT image. The invention combines the brain auxiliary image with the brain CT image scanned by the patient, improves the accuracy of identifying the cerebral hemorrhage area and type of the cerebral hemorrhage automatic detection neural network, and reduces the technical requirements of doctors in the diagnosis process.
The pre-trained automatic cerebral hemorrhage detection neural network has the specific structure as follows: a backbone network of U-Net or VGG is used as a backbone network of the cerebral hemorrhage automatic detection neural network, and a convolution network is connected behind the backbone network, wherein a brain CT image is input into the backbone network, and a brain auxiliary image is input into the convolution network.
Taking a two-dimensional image as an example, the training process of the neural network for automatically detecting cerebral hemorrhage is as follows:
constructing a sample set: acquiring a plurality of brain CT images in a database, marking bleeding areas and bleeding types in the brain CT images, forming image pairs by the marked brain CT images and corresponding brain auxiliary images, and constructing a sample set with a plurality of image pairs; marking the bleeding areas and categories of the brain CT images in the database to obtain a bleeding area mask of the brain CT, wherein in the brain CT image of a patient, the pixel value of a normal area is 20-40HU, the pixel value of an epidural bleeding area, a subdural bleeding area, a brain parenchyma bleeding area and a brain stem bleeding area is 45-75HU, and the pixel value of a subarachnoid bleeding area is 40-75 HU.
Inputting a plurality of picture pairs in the sample set into the constructed automatic cerebral hemorrhage detection neural network: a backbone network of U-Net or VGG is used as a backbone network of the cerebral hemorrhage automatic detection neural network, a convolution network is connected behind the backbone network, the brain CT image in the sample set image pair is input into the backbone network, and the brain auxiliary image in the sample set image pair is input into the convolution network;
pre-training the constructed automatic cerebral hemorrhage detection neural network by a gradient descent method: the method comprises the steps that features of a brain CT image are extracted by a backbone network and output to a convolution network, the convolution network carries out convolution processing by combining the features and a brain auxiliary image, and the automatic cerebral hemorrhage detection neural network is trained through a gradient descent method until a loss function of the automatic cerebral hemorrhage detection neural network is smaller than a threshold value. The loss function of the automatic cerebral hemorrhage detection neural network can be a multi-class cross entropy function, and the automatic cerebral hemorrhage detection neural network can be stored for later use after pre-training is finished. The calculation formula of the multi-class cross entropy function is as follows:
wherein the content of the first and second substances,
is a cerebral hemorrhage area output by the automatic cerebral hemorrhage detection neural network,
is an artificially marked cerebral hemorrhage area,
there are 5 types of cerebral hemorrhage, i.e., epidural hemorrhage, subdural hemorrhage, cerebral parenchyma hemorrhage, brainstem hemorrhage, and subarachnoid hemorrhage.
The scheme further provides that a total network is adopted to achieve the purposes of obtaining a brain auxiliary image, training in the automatic brain hemorrhage detection neural network and outputting the processed brain CT image, as shown in figure 10, wherein the total network comprises an image generation neural network and a brain hemorrhage automatic detection neural network, the image generation neural network is used for generating the brain auxiliary image according to the input brain CT image and outputting the brain auxiliary image, the brain hemorrhage automatic detection neural network is used for automatically detecting a brain hemorrhage area in the brain CT image according to the input brain CT image and the brain auxiliary image output by the image generation neural network, the output processed brain CT image comprises a mask of the brain hemorrhage area and type, the mask is read by a doctor, and the doctor further diagnoses the brain hemorrhage by combining the processed brain CT image. The loss function in the overall network pre-training process is:
wherein the content of the first and second substances,
is a cerebral hemorrhage area output by the automatic cerebral hemorrhage detection neural network,
is an artificially marked cerebral hemorrhage area,
in two-dimensional coordinates of the image and the type of cerebral hemorrhage,
is an image of the area containing cerebral hemorrhage output by the convolutional neural network,
to generate a brain auxiliary image based on a standard model or a brain auxiliary image based on a contour distance of the skull,
representing the two-dimensional coordinates of the image and the class of brain-assist images,
as the weight coefficient, in the present scheme
The value is 1. After the pre-training is finished, keeping a total network and a weight coefficient; the total network is adopted to realize acquisition of the brain auxiliary image and detection and output of information such as the cerebral hemorrhage area, and convenience of the scheme is improved.
The invention also discloses an electronic medium, which comprises a memory and a processor, wherein the memory is connected with the processor, the memory stores at least one instruction which can be executed by the processor, and when the at least one instruction is executed by the processor, the automatic cerebral hemorrhage detection method based on the brain auxiliary image is realized.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.
Claims (9)
1. An automatic cerebral hemorrhage detection method based on a brain auxiliary image is characterized by comprising the following steps:
s1, acquiring a brain CT image of the patient: performing CT scanning on the head of a patient, and acquiring a brain CT image through a CT reconstruction algorithm;
s2, acquiring a brain auxiliary image: generating a brain auxiliary image according to the brain CT image, wherein the brain auxiliary image comprises spatial position information of the brain anatomical structure;
s3, training in the automatic cerebral hemorrhage detection neural network, and outputting a processed brain CT image: the brain CT image and the brain auxiliary image are used as input, the two images are input into a pre-trained automatic cerebral hemorrhage detection neural network, the cerebral hemorrhage area in the brain CT image is automatically detected by combining the brain auxiliary image, the output processed brain CT image comprises a mask of the cerebral hemorrhage area and type, the mask is read by a doctor, and the doctor further diagnoses the cerebral hemorrhage by combining the processed brain CT image.
2. The method for automatically detecting cerebral hemorrhage based on the brain auxiliary image according to claim 1, wherein the brain auxiliary image in S2 includes spatial location information of the brain anatomical structure, specifically including spatial location information of the skull, the dura mater, the arachnoid, the ventricle and the lobe.
3. The method for automatically detecting cerebral hemorrhage according to claim 1, wherein the generating of the cerebral auxiliary image from the cerebral CT image in S2 is based on a standard model of the cerebral auxiliary image, and the specific process is as follows:
acquiring a standard model including a brain anatomical structure, and taking the standard model as a template;
performing non-rigid registration transformation on the standard model according to the brain CT image to obtain a non-rigid registration coordinate mapping relation;
and outputting the brain auxiliary image based on the standard model according to the non-rigid registered coordinate mapping relation.
4. The method for automatically detecting cerebral hemorrhage based on the brain auxiliary image according to claim 1, wherein the generating of the brain auxiliary image from the brain CT image in S2 is based on the skull contour distance, and the specific process is as follows:
extracting a skull region from a brain CT image;
calculating the shortest distance from the pixel point to the extracted skull region for all pixel points of the brain image;
and outputting the brain auxiliary image based on the skull contour distance.
5. The method according to claim 1, wherein the generating of the brain auxiliary image from the brain CT image in S2 is an image-based neural network brain auxiliary image, and comprises:
constructing an image generation neural network: the image generation neural network is a convolution neural network, and the convolution neural network is a backbone network of U-Net or VGG;
pre-training the convolutional neural network: generating a brain auxiliary image based on a standard model or a brain auxiliary image based on a skull contour distance, using a brain CT image as the input of a convolutional neural network, pre-training the convolutional neural network, calculating a loss function of the image which is output by the convolutional neural network and contains a cerebral hemorrhage region and the generated brain auxiliary image based on the standard model or the brain auxiliary image based on the skull contour distance until the loss function is smaller than a threshold value, and completing the pre-training of the convolutional neural network;
outputting a brain auxiliary image: the image containing the cerebral hemorrhage area output by the convolutional neural network is the brain auxiliary image of the neural network generated based on the image.
6. The method for automatically detecting cerebral hemorrhage based on the brain auxiliary image as claimed in claim 1, wherein the pre-trained cerebral hemorrhage automatic detection neural network in S3 has a specific structure: a backbone network of U-Net or VGG is used as a backbone network of the cerebral hemorrhage automatic detection neural network, and a convolution network is connected behind the backbone network, wherein a brain CT image is input into the backbone network, and a brain auxiliary image is input into the convolution network.
7. The method for automatically detecting cerebral hemorrhage based on brain auxiliary image according to claim 1, wherein the two images are input into a pre-trained automatic cerebral hemorrhage detecting neural network in S3, wherein the pre-training process of the neural network is as follows:
constructing a sample set: acquiring a plurality of brain CT images in a database, marking bleeding areas and bleeding types in the brain CT images, forming image pairs by the marked brain CT images and corresponding brain auxiliary images, and constructing a sample set with a plurality of image pairs;
inputting a plurality of picture pairs in the sample set into the constructed automatic cerebral hemorrhage detection neural network: a backbone network of U-Net or VGG is used as a backbone network of the cerebral hemorrhage automatic detection neural network, a convolution network is connected behind the backbone network, the brain CT image in the sample set image pair is input into the backbone network, and the brain auxiliary image in the sample set image pair is input into the convolution network;
pre-training the constructed automatic cerebral hemorrhage detection neural network by a gradient descent method: the method comprises the steps that features of a brain CT image are extracted by a backbone network and output to a convolution network, the convolution network carries out convolution processing by combining the features and the brain auxiliary image, and a neural network is trained through a gradient descent method until a loss function of the neural network for automatically detecting cerebral hemorrhage is smaller than a threshold value.
8. The method according to claim 1, wherein the cerebral hemorrhage is detected automatically based on the auxiliary brain image, it is characterized in that the brain auxiliary image acquired in the step S2 and the brain CT image trained and output in the step S3 in the automatic cerebral hemorrhage detection neural network are realized by adopting a total network, wherein the total network comprises an image generation neural network and a cerebral hemorrhage automatic detection neural network, the image generation neural network is used for generating and outputting a brain auxiliary image according to an input brain CT image, the cerebral hemorrhage automatic detection neural network is used for automatically detecting a cerebral hemorrhage area in the brain CT image according to the input brain CT image and the brain auxiliary image output by the image generation neural network, and outputting a mask containing the cerebral hemorrhage area and type in the processed brain CT image, and the doctor can read the image for further diagnosis of cerebral hemorrhage by combining the processed cerebral CT image.
9. An electronic medium, comprising: the device comprises a memory and a processor, wherein the memory is connected with the processor, the memory stores at least one instruction which can be executed by the processor, and when the at least one instruction is executed by the processor, the method for automatically detecting cerebral hemorrhage based on the brain auxiliary image realizes the method for automatically detecting cerebral hemorrhage according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010643129.XA CN111539956B (en) | 2020-07-07 | 2020-07-07 | Cerebral hemorrhage automatic detection method based on brain auxiliary image and electronic medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010643129.XA CN111539956B (en) | 2020-07-07 | 2020-07-07 | Cerebral hemorrhage automatic detection method based on brain auxiliary image and electronic medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111539956A true CN111539956A (en) | 2020-08-14 |
| CN111539956B CN111539956B (en) | 2020-12-29 |
Family
ID=71979790
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010643129.XA Active CN111539956B (en) | 2020-07-07 | 2020-07-07 | Cerebral hemorrhage automatic detection method based on brain auxiliary image and electronic medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111539956B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111798535A (en) * | 2020-09-09 | 2020-10-20 | 南京安科医疗科技有限公司 | CT image enhancement display method and computer readable storage medium |
| CN112116625A (en) * | 2020-08-25 | 2020-12-22 | 澳门科技大学 | Automatic heart CT image segmentation method, device and medium based on contradiction marking method |
| CN113076987A (en) * | 2021-03-29 | 2021-07-06 | 北京长木谷医疗科技有限公司 | Osteophyte identification method, device, electronic equipment and storage medium |
| CN113768528A (en) * | 2021-09-26 | 2021-12-10 | 华中科技大学 | CT image cerebral hemorrhage auxiliary positioning system |
| CN114549532A (en) * | 2022-04-27 | 2022-05-27 | 珠海市人民医院 | Cerebral ischemia auxiliary analysis method and system based on medical image processing |
| CN115187600A (en) * | 2022-09-13 | 2022-10-14 | 杭州涿溪脑与智能研究所 | Brain hemorrhage volume calculation method based on neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2224401A1 (en) * | 2009-02-27 | 2010-09-01 | Medicsight PLC | Computer-aided detection of lesions |
| CN108369642A (en) * | 2015-12-18 | 2018-08-03 | 加利福尼亚大学董事会 | Acute disease feature is explained and quantified according to head computer tomography |
| CN110880366A (en) * | 2019-12-03 | 2020-03-13 | 上海联影智能医疗科技有限公司 | Medical image processing system |
| CN110934606A (en) * | 2019-10-31 | 2020-03-31 | 上海杏脉信息科技有限公司 | Cerebral apoplexy early-stage flat-scan CT image evaluation system and method and readable storage medium |
| WO2020111463A1 (en) * | 2018-11-29 | 2020-06-04 | 주식회사 휴런 | System and method for estimating aspect score |
-
2020
- 2020-07-07 CN CN202010643129.XA patent/CN111539956B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2224401A1 (en) * | 2009-02-27 | 2010-09-01 | Medicsight PLC | Computer-aided detection of lesions |
| CN108369642A (en) * | 2015-12-18 | 2018-08-03 | 加利福尼亚大学董事会 | Acute disease feature is explained and quantified according to head computer tomography |
| WO2020111463A1 (en) * | 2018-11-29 | 2020-06-04 | 주식회사 휴런 | System and method for estimating aspect score |
| CN110934606A (en) * | 2019-10-31 | 2020-03-31 | 上海杏脉信息科技有限公司 | Cerebral apoplexy early-stage flat-scan CT image evaluation system and method and readable storage medium |
| CN110880366A (en) * | 2019-12-03 | 2020-03-13 | 上海联影智能医疗科技有限公司 | Medical image processing system |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112116625A (en) * | 2020-08-25 | 2020-12-22 | 澳门科技大学 | Automatic heart CT image segmentation method, device and medium based on contradiction marking method |
| CN111798535A (en) * | 2020-09-09 | 2020-10-20 | 南京安科医疗科技有限公司 | CT image enhancement display method and computer readable storage medium |
| CN113076987A (en) * | 2021-03-29 | 2021-07-06 | 北京长木谷医疗科技有限公司 | Osteophyte identification method, device, electronic equipment and storage medium |
| CN113768528A (en) * | 2021-09-26 | 2021-12-10 | 华中科技大学 | CT image cerebral hemorrhage auxiliary positioning system |
| CN114549532A (en) * | 2022-04-27 | 2022-05-27 | 珠海市人民医院 | Cerebral ischemia auxiliary analysis method and system based on medical image processing |
| CN115187600A (en) * | 2022-09-13 | 2022-10-14 | 杭州涿溪脑与智能研究所 | Brain hemorrhage volume calculation method based on neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111539956B (en) | 2020-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111539956B (en) | Cerebral hemorrhage automatic detection method based on brain auxiliary image and electronic medium | |
| US11430140B2 (en) | Medical image generation, localizaton, registration system | |
| CN110232383B (en) | Focus image recognition method and focus image recognition system based on deep learning model | |
| EP3046478B1 (en) | Image analysis techniques for diagnosing diseases | |
| US20200367853A1 (en) | Apparatus for ai-based automatic ultrasound diagnosis of liver steatosis and remote medical diagnosis method using the same | |
| CN111047605B (en) | Construction method and segmentation method of vertebra CT segmentation network model | |
| CN111681230A (en) | System and method for scoring high-signal of white matter of brain | |
| CN112365464A (en) | GAN-based medical image lesion area weak supervision positioning method | |
| US20230005140A1 (en) | Automated detection of tumors based on image processing | |
| CN112465771B (en) | Spine nuclear magnetic resonance image analysis method and device and computer equipment | |
| CN110738643A (en) | Method for analyzing cerebral hemorrhage, computer device and storage medium | |
| CN112184720B (en) | Method and system for segmenting internal rectus muscle and optic nerve of CT image | |
| JP2019028887A (en) | Image processing method | |
| CN110334566A (en) | Fingerprint extraction method inside and outside a kind of OCT based on three-dimensional full convolutional neural networks | |
| CN113096137A (en) | Adaptive segmentation method and system for OCT (optical coherence tomography) retinal image field | |
| CN110490841B (en) | Computer-aided image analysis method, computer device and storage medium | |
| CN113822323A (en) | Brain scanning image identification processing method, device, equipment and storage medium | |
| CN113298773A (en) | Heart view identification and left ventricle detection device and system based on deep learning | |
| CN114066798A (en) | Brain tumor nuclear magnetic resonance image data synthesis method based on deep learning | |
| CN109816665B (en) | Rapid segmentation method and device for optical coherence tomography image | |
| CN113989269B (en) | Traditional Chinese medicine tongue image tooth trace automatic detection method based on convolutional neural network multi-scale feature fusion | |
| CN111986205A (en) | Vessel tree generation and lesion identification method, device, equipment and readable storage medium | |
| CN116168097A (en) | Method, device, equipment and medium for constructing CBCT sketching model and sketching CBCT image | |
| CN111598904B (en) | Image segmentation method, device, equipment and storage medium | |
| CN114360695A (en) | Mammary gland ultrasonic scanning analysis auxiliary system, medium and equipment |
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 |