Automatic segmentation method of medical images based on cognitive deep learning

Abstract

The present invention discloses a method for automatic segmentation of medical images based on cognitive deep learning, and relates to the technical field of medical image segmentation. The method includes obtaining different types of medical images, preprocessing them; inputting them into a trained multi-branch convolutional neural network to obtain output features of medical images of the corresponding type of each branch; recognizing the output features through channel attention to obtain output feature fusion weights; merging the output features according to the output features and the output feature fusion weights to obtain a fused and segmented medical image; post-processing the fused and segmented medical image to complete automatic segmentation of the medical image. By using a multi-branch convolutional neural network to process different types of medical images, the processing path and parameters are optimized for each image type, and the accuracy and adaptability of the segmentation are improved; through the channel attention mechanism, the features output by each branch network can be effectively weighted, dynamic fusion of features can be achieved, and information loss and interference can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image segmentation, and specifically to an automatic medical image segmentation method based on cognitive deep learning. Background Art

[0002] In modern medical diagnosis, medical imaging technology plays a vital role and is often used in diagnosis, disease monitoring and treatment plan formulation. Automatic segmentation of medical images is an important research direction in the field of medical image processing. Its goal is to automatically identify and segment lesions or other anatomical structures of interest from complex medical images. Currently, deep learning technology, especially convolutional neural networks (CNNs), has been widely used in the task of automatic segmentation of medical images, improving the accuracy and efficiency of segmentation.

[0003] Although deep learning has made significant progress in medical image segmentation, existing technologies still have some limitations: most existing models are trained and optimized for specific types of medical images, such as CT, MRI or X-ray images. This single image type processing strategy limits the versatility and adaptability of the model, making it difficult to process multimodal medical image data; existing methods often lack an effective feature fusion mechanism when processing multimodal data. Direct fusion of features between different modalities may lead to information loss or interference, affecting the final segmentation accuracy. Summary of the invention

[0004] Based on the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a method for automatic segmentation of medical images based on cognitive deep learning to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for automatic segmentation of medical images based on cognitive deep learning, comprising:

[0006] Acquire different types of medical images, and pre-process the different types of medical images;

[0007] Input the preprocessed different types of medical images into the trained multi-branch convolutional neural network to obtain the output features of the corresponding type of medical images of each branch, wherein each branch processes one type of medical image;

[0008] The output features are recognized through channel attention to obtain output feature fusion weights;

[0009] Merging the output features according to the output features and the output feature fusion weights to obtain a fused segmented medical image;

[0010] The fused and segmented medical image is post-processed to complete automatic segmentation of the medical image.

[0011] The present invention is further configured that the preprocessing of the different types of medical images comprises:

[0012] Normalizing the different types of medical images, and performing contrast enhancement on the normalized medical images;

[0013] Align the pre-processed medical images through elastic transformation so that the key features of different types of medical images are in the same position;

[0014] The different types of medical images after elastic transformation are resampled so that the spatial resolution and size of the different types of medical images are consistent.

[0015] The present invention is further configured such that the training logic of the trained multi-branch convolutional neural network comprises:

[0016] Acquire a historical data set, wherein the historical data set includes a historical medical image data set and a historical segmented medical image data set;

[0017] Branch training is performed according to the historical data set to obtain output features of the medical image, the loss is calculated according to the loss function, and the training is repeated until the loss is less than a threshold value to complete the training.

[0018] The present invention is further configured such that the loss function is: Among them, SL is the loss function, n is the total number of pixels, q j is the true value of pixel j, is the predicted value of pixel j, w1 is the weight coefficient, q t The value logic is:

[0019] The present invention is further configured to recognize the output features through channel attention to obtain output feature fusion weights, including:

[0020] The output features are recorded as the output feature set F, where F∈R C*H*W , where R is the symbol of the real number set, indicating that the elements of the feature map are real numbers, C is the number of channels, H is the height of the output feature, and W is the width of the output feature;

[0021] Perform global average pooling on the output features in the output feature set F of each channel to obtain the channel descriptor;

[0022] The output feature fusion weight is obtained according to the channel descriptor.

[0023] The present invention is further configured such that global average pooling is performed on the output features in the output feature set F of each channel to obtain a channel descriptor, and the calculation logic of the channel descriptor is: Among them, S c is the channel descriptor of the c-th channel, and F(c,l,m) is the element value of the output feature at height l and width m on the c-th channel.

[0024] The present invention is further configured that the output feature fusion weight is obtained according to the channel descriptor, and the calculation logic is: Z c =σ(w2*δ(w3*S c +b1)+b2), where Z c is the feature fusion weight, σ is the sigmoid activation function, δ is the ReLU activation function, w2 and w3 are weight parameters, and b1 and b2 are bias parameters.

[0025] The present invention is further configured that the output feature merging is performed according to the output feature and the output feature fusion weight to obtain a fused segmented medical image, comprising:

[0026] Get the output features of each channel and the corresponding feature fusion weights;

[0027] Calculating a weighted output feature according to the output feature and the feature fusion weight;

[0028] The weighted output features are spliced ​​to obtain a fused medical image, which is then segmented to obtain a fused and segmented medical image.

[0029] The present invention is further configured that the post-processing of the fused segmented medical image comprises:

[0030] Smoothing the segmentation boundaries and filling small holes of the fused segmented medical image through morphological transformation;

[0031] The fused segmented medical images after morphological transformation are blurred by Gaussian filter to reduce image noise.

[0032] The present invention provides a method for automatic segmentation of medical images based on cognitive deep learning. The method obtains different types of medical images and pre-processes the different types of medical images; inputs the pre-processed different types of medical images into a trained multi-branch convolutional neural network to obtain output features of medical images of corresponding types of each branch, wherein each branch processes one type of medical image; recognizes the output features through channel attention to obtain output feature fusion weights; merges the output features according to the output features and the output feature fusion weights to obtain a fused and segmented medical image; and post-processes the fused and segmented medical image to complete automatic segmentation of the medical image. The beneficial effects produced include:

[0033] 1. Improve segmentation accuracy and adaptability: By using a multi-branch convolutional neural network to process different types of medical images, this method can optimize the processing path and parameters for each image type. The customized processing method can more effectively extract the key features of various types of images, thereby improving the accuracy and adaptability of segmentation.

[0034] 2. Dynamic feature fusion: Through the channel attention mechanism, the method can effectively assign weights to the features output by each branch network and realize dynamic fusion of features. The dynamic fusion strategy not only takes into account the characteristics of each type of image, but also dynamically adjusts the fusion strategy according to the specific content, so that the final fused image can retain important information while reducing information loss and interference.

[0035] 3. Optimize the use of computing resources: Through efficient network architecture and post-processing techniques, including global average pooling and feature fusion techniques, this method can optimize the use of computing resources without sacrificing performance.

[0036] 4. Improve the consistency and reliability of segmentation results: Morphological transformation and Gaussian filtering in the post-processing step help smooth segmentation boundaries and fill small holes, which not only improves the visual effect of segmentation, but also improves the consistency and reliability of segmentation results.

[0037] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0039] Figure 1 The present invention is a flowchart of a method for automatic medical image segmentation based on cognitive deep learning, which is an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0041] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0042] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0043] Automatic segmentation methods of medical images based on cognitive deep learning, such as Figure 1 As shown, including:

[0044] Acquire different types of medical images, and pre-process the different types of medical images;

[0045] Input the preprocessed different types of medical images into the trained multi-branch convolutional neural network to obtain the output features of the corresponding type of medical images of each branch, wherein each branch processes one type of medical image;

[0046] The output features are recognized through channel attention to obtain output feature fusion weights;

[0047] Merging the output features according to the output features and the output feature fusion weights to obtain a fused segmented medical image;

[0048] The fused and segmented medical image is post-processed to complete automatic segmentation of the medical image.

[0049] Specifically, obtaining different types of medical images and performing preprocessing involves collecting different types of medical images, including X-rays, which are one of the most common medical images and are used to examine bones and certain soft tissue structures, such as the lungs; computed tomography (CT scan), which uses X-rays and computer technology to generate cross-sectional images, can provide more detailed information about tissue structure, and is widely used to diagnose various diseases; magnetic resonance imaging (MRI), which uses magnetic fields and harmless radio waves to generate detailed images of the body's internal structure, with better display effects on soft tissues and organs, and is often used to examine the brain, spine, joints, etc.; ultrasound, which uses high-frequency sound waves to generate images, is often used to examine the fetus, liver, heart, and blood vessels of pregnant women; positron emission tomography (PET), which provides information about tissue and organ function by detecting the distribution of radioactively labeled drugs in the body, and is often used to diagnose tumors and evaluate brain function. Single photon emission tomography (SPECT), similar to PET scans but using single-photon radioactive tracers, is used to detect cardiovascular and nervous system problems; Using multi-branch convolutional neural networks to process different types of medical images The multi-branch convolutional neural network has multiple parallel convolution branches, each branch processes a different type of medical image; The channel attention mechanism is a mechanism used to learn the importance of different feature channels in a neural network. Channel attention can help the network focus on feature channels that are more important to the task to improve performance; The output feature fusion weights are obtained through the channel attention mechanism and are used to fuse the output features of each branch; The output feature merging will perform weighted fusion of the features according to the output features and the fusion weights to obtain the final fused segmented medical image; Post-processing is a series of operations applied to the fused segmented medical image to remove noise, fill holes, smooth boundaries, etc. to produce the final medical image segmentation result.

[0050] The present invention is further configured that the preprocessing of the different types of medical images comprises:

[0051] Normalizing the different types of medical images, and performing contrast enhancement on the normalized medical images; specifically, normalization is to scale the pixel values ​​of the images to a fixed range to eliminate the brightness differences between different images, and contrast enhancement is to enhance the details and contrast in the image by adjusting the dynamic range of the pixel values ​​to improve the image quality;

[0052] The pre-processed medical images are aligned through elastic transformation so that the key features of different types of medical images are in the same position. Specifically, elastic transformation is an image transformation technology that can locally deform an image to achieve image alignment. In medical imaging, elastic transformation is usually used to align images of different patients so that key features, including organ boundaries, correspond to each other in space, which helps to improve the generalization ability of neural networks.

[0053] The different types of medical images after elastic transformation are resampled to make the spatial resolution and size of different types of medical images consistent; specifically, resampling is to adjust the spatial resolution and size of the image so that different images have the same spatial scale, ensuring that the neural network can process images of different resolutions and sizes, because the neural network usually requires the input to have the same size.

[0054] The present invention is further configured such that the training logic of the trained multi-branch convolutional neural network comprises:

[0055] Acquire a historical data set, wherein the historical data set includes a historical medical image data set and a historical segmented medical image data set; specifically, the historical data set includes a historical medical image data set and a historical segmented medical image data set, the historical medical image data set includes original medical image data, and the historical segmented medical image data set includes corresponding segmentation label data, that is, accurate boundaries of regions of interest in medical images;

[0056] Branch training is performed according to the historical data set to obtain output features of the medical image, and the loss is calculated according to the loss function. The training is repeated until it is less than a loss threshold to complete the training. Specifically, the historical data set is used for training a multi-branch convolutional neural network, and each branch is responsible for processing a type of medical image. By inputting the medical image into the corresponding branch, the network learns to extract the features of the medical image and generates output features that match the segmentation label. The performance of the branch is evaluated using a loss function, and the loss function usually measures the difference between the network output and the segmentation label. In each training iteration, the network adjusts parameters according to the feedback of the loss function to reduce the difference between the prediction and the label. This process is repeated until the loss is reduced to meet a predefined loss threshold.

[0057] The present invention is further configured such that the loss function is: Among them, SL is the loss function, n is the total number of pixels, q j is the true value of pixel j, is the predicted value of pixel j, w1 is the weight coefficient, q t The value logic is:

[0058] The present invention is further configured to recognize the output features through channel attention to obtain output feature fusion weights, including:

[0059] The output features are recorded as the output feature set F, where F∈R C*H*W , where R is the symbol of the real number set, indicating that the elements of the feature map are real numbers, C is the number of channels, H is the height of the output feature, and W is the width of the output feature;

[0060] Perform global average pooling on the output features in the output feature set F of each channel to obtain a channel descriptor; specifically, for the output features of each channel, perform a global average pooling operation, which is an average pooling operation performed on the entire feature map, taking the average of all elements of each channel to obtain a feature descriptor of the channel. For C channels, C channel descriptors are obtained;

[0061] The output feature fusion weight is obtained according to the channel descriptor; specifically, the channel descriptor is used as input, and the output feature fusion weight is obtained through a preset calculation logic.

[0062] The present invention is further configured such that global average pooling is performed on the output features in the output feature set F of each channel to obtain a channel descriptor, and the calculation logic of the channel descriptor is: Among them, S c is the channel descriptor of the c-th channel, and F(c,l,m) is the element value of the output feature at height l and width m on the c-th channel.

[0063] The present invention is further configured that the output feature fusion weight is obtained according to the channel descriptor, and the calculation logic is: Z c =σ(w2*δ(w3*S c +b1)+b2), where Z cis the feature fusion weight, σ is the sigmoid activation function, δ is the ReLU activation function, w2 and w3 are weight parameters, b1 and b2 are bias parameters; specifically, the values ​​of weight parameters w2 and w3, and bias parameters b1 and b2 are learned through the back propagation algorithm during the training process, so that the network can better learn the relationship between features according to the input data. The weight parameters w2 and w3, and bias parameters b1 and b2 are initialized randomly, and then iteratively optimized during the training process through the back propagation algorithm. ReLU is the rectified linear unit activation function, which is used to ensure that the output is non-negative. The sigmoid activation function converts w3*S c +b1 is mapped to (0,1); through the above calculation logic, the output feature fusion weight of each channel can be calculated according to the channel descriptor, so as to perform weighted fusion on the output features. The weighted fusion method can make the network pay more attention to the feature channels that are more important to the final task, thereby improving the performance and generalization ability of the network.

[0064] The present invention is further configured that the output feature merging is performed according to the output feature and the output feature fusion weight to obtain a fused segmented medical image, comprising:

[0065] Get the output features of each channel and the corresponding feature fusion weights; specifically, for each channel k, get the corresponding output feature F k and feature fusion weight Z c .

[0066] The weighted output feature is calculated according to the output feature and the feature fusion weight; specifically, for each channel k, the output feature F k Multiply by the feature fusion weight Z c Get weighted output features The weighted output feature considers the importance of each channel and weights the features of the corresponding channels according to the weight;

[0067] The weighted output features are spliced ​​to obtain a fused medical image, which is then segmented to obtain a fused and segmented medical image. Specifically, all weighted output features are spliced ​​to obtain a fused and segmented medical image. Splice along the channel dimension to obtain the fused feature map F sFurthermore, splicing refers to connecting multiple tensors along a certain dimension to form a larger tensor. In an embodiment of the present invention, splicing is performed on the channel dimension, that is, the feature maps of multiple channels are connected in the channel direction. Specifically, in the medical image segmentation task, a multi-branch convolutional neural network is used as a model. During the feature extraction process, the multi-branch convolutional neural network will generate feature maps of multiple channels. Each channel represents a different feature representation. Therefore, when merging different features, the information learned by each branch is retained to make full use of features of different types and scales. It is necessary to splice the feature maps of each channel together according to the channel dimension to form a more comprehensive feature representation so that the model can understand the input image more comprehensively.

[0068] The present invention is further configured that the post-processing of the fused segmented medical image comprises:

[0069] The fused segmented medical image is subjected to morphological transformation to smooth the segmentation boundary and fill small holes; specifically, morphological transformation is an operation based on image shape, which is often used to smooth the boundary and fill holes. In an embodiment of the present invention, morphological operations, including dilation, erosion, opening operation, closing operation, etc., are used to eliminate small holes and noise in the image and smooth the segmentation boundary, so that the segmentation result is more continuous and accurate;

[0070] The fused segmented medical image after morphological transformation is blurred by a Gaussian filter to reduce image noise. Specifically, the Gaussian filter is a commonly used image blurring method, which can effectively reduce the noise in the image. In an embodiment of the present invention, a Gaussian filter can be applied to blur the segmentation result after morphological transformation to further reduce the noise in the image and make the segmentation boundary smoother and more natural.

[0071] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0072] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0073] In this application, "at least one" means one or more, and "more" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0074] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0075] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0076] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0077] In the several embodiments provided in the present application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0078] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0079] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0080] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0081] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A medical image automatic segmentation method based on cognitive deep learning, characterized in that: include: Acquire different types of medical images, and pre-process the different types of medical images; Input the preprocessed different types of medical images into the trained multi-branch convolutional neural network to obtain the output features of the corresponding type of medical images of each branch, wherein each branch processes one type of medical image; The output features are recognized by channel attention to obtain output feature fusion weights, including: recording the output features as an output feature set F, where: , where R is the symbol of the real number set, indicating that the elements of the feature map are real numbers, C is the number of channels, H is the height of the output feature, and W is the width of the output feature; perform global average pooling on the output features in the output feature set F of each channel to obtain the channel descriptor. The calculation logic of the channel descriptor is: ,in, is the channel descriptor of the cth channel, is the height of the output feature on the cth channel , the element value at the width m; obtain the output feature fusion weight according to the channel descriptor, and the calculation logic of the feature fusion weight is: ,in, is the feature fusion weight, is the sigmoid activation function, is the ReLU activation function, and is the weight parameter, and is the bias parameter; Merging the output features according to the output features and the output feature fusion weights to obtain a fused segmented medical image; The fused and segmented medical image is post-processed to complete automatic segmentation of the medical image.

2. The method for automatic medical image segmentation based on cognitive deep learning according to claim 1, characterized in that: The preprocessing of the different types of medical images includes: Normalizing the different types of medical images, and performing contrast enhancement on the normalized medical images; Align the pre-processed medical images through elastic transformation so that the key features of different types of medical images are in the same position; The different types of medical images after elastic transformation are resampled so that the spatial resolution and size of the different types of medical images are consistent.

3. The method for automatic segmentation of medical images based on cognitive deep learning according to claim 1, characterized in that: The training logic of the trained multi-branch convolutional neural network includes: Acquire a historical data set, wherein the historical data set includes a historical medical image data set and a historical segmented medical image data set; Branch training is performed according to the historical data set to obtain output features of the medical image, the loss is calculated according to the loss function, and the training is repeated until the loss is less than a threshold value to complete the training.

4. The method for automatic segmentation of medical images based on cognitive deep learning according to claim 3 is characterized in that: The loss function is: , where SL is the loss function, n is the total number of pixels, is the true value of pixel j, is the predicted value of pixel j, is the weight coefficient, The value logic is: .

5. The method for automatic medical image segmentation based on cognitive deep learning according to claim 1, characterized in that: The step of merging the output features according to the output features and the output feature fusion weights to obtain a fused segmented medical image includes: Get the output features of each channel and the corresponding feature fusion weights; Calculating a weighted output feature according to the output feature and the feature fusion weight; The weighted output features are spliced ​​to obtain a fused medical image, which is then segmented to obtain a fused and segmented medical image.

6. The method for automatic segmentation of medical images based on cognitive deep learning according to claim 1, characterized in that: The post-processing of the fused segmented medical image comprises: Smoothing the segmentation boundaries and filling small holes of the fused segmented medical image through morphological transformation; The fused segmented medical images after morphological transformation are blurred by Gaussian filter to reduce image noise.

Images