CN113487536A

CN113487536A - Image segmentation method, computer device and storage medium

Info

Publication number: CN113487536A
Application number: CN202110607453.0A
Authority: CN
Inventors: 姜娈; 霍璐
Original assignee: Shanghai United Imaging Intelligent Healthcare Co Ltd
Current assignee: Shanghai United Imaging Intelligent Healthcare Co Ltd
Priority date: 2021-06-01
Filing date: 2021-06-01
Publication date: 2021-10-08

Abstract

The present application relates to an image segmentation method, a computer device, and a storage medium. The method comprises the following steps: partitioning the fat suppression magnetic resonance image, and performing contrast-limiting adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat suppression magnetic resonance image; the fat suppression magnetic resonance image includes a region of interest; inputting the processed fat suppression magnetic resonance image into a preset segmentation network, and obtaining a mask image of the region of interest through the segmentation network; and obtaining a segmentation result of the region of interest in the fat suppression magnetic resonance image according to the fat suppression magnetic resonance image and the mask image. By adopting the method, the segmentation result of the region of interest in the fat suppression magnetic resonance image can be accurately obtained.

Description

Image segmentation method, computer device and storage medium

Technical Field

The present application relates to the field of medical image technology, and in particular, to an image segmentation method, a computer device, and a storage medium.

Background

The segmentation of the lesion region of the medical image is of great significance in a computer-aided diagnosis system, the lesion segmented in the medical image can be accurately detected through the computer-aided diagnosis system, for example, early detection and diagnosis of breast cancer can effectively improve the cure rate of breast cancer, and the segmentation of breast tissues and glandular tissues is of great significance in the computer-aided diagnosis system based on three-dimensional magnetic resonance images.

At present, most segmentation methods realize the segmentation of tissues and focuses based on non-fat-suppression magnetic resonance images, and the tissues and focuses in the fat-suppression magnetic resonance images are difficult to be accurately segmented.

Disclosure of Invention

In view of the above, it is necessary to provide an image segmentation method, a computer device, and a storage medium capable of accurately segmenting tissues and lesions in a fat-suppressed magnetic resonance image in view of the above technical problems.

A method of image segmentation, the method comprising:

partitioning the fat suppression magnetic resonance image, and performing contrast-limited adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat suppression magnetic resonance image; the fat-suppressed magnetic resonance image comprises a region of interest;

inputting the processed fat suppression magnetic resonance image into a preset segmentation network, and obtaining a mask image of the region of interest through the segmentation network;

and obtaining a segmentation result of the region of interest in the fat suppression magnetic resonance image according to the fat suppression magnetic resonance image and the mask image.

In one embodiment, the split network comprises: an encoding unit and a decoding unit; the step of inputting the processed fat suppression magnetic resonance image into a preset segmentation network, and obtaining a mask image of the region of interest through the segmentation network includes:

inputting the processed fat-suppression magnetic resonance image into the encoding unit, performing boundary filling on the processed fat-suppression magnetic resonance image by adopting filling convolution, and performing down-sampling on the fat-suppression magnetic resonance image after the boundary filling to obtain a down-sampling feature map corresponding to the processed fat-suppression magnetic resonance image;

and inputting the downsampled feature map into the decoding unit, filling the boundary of the downsampled feature map by adopting the filling convolution, and upsampling the downsampled feature map after the boundary is filled to obtain a mask image of the region of interest.

In one embodiment, the encoding unit includes a plurality of convolution blocks, each convolution block being composed of a convolution layer and a pooling layer, and each convolution layer connecting a normalization layer and an activation function layer of the segmentation network; the convolutional layer comprises the filler convolution; the filling convolution is used for filling the boundary of the fat suppression magnetic resonance image and increasing the size of an image matrix corresponding to the fat suppression magnetic resonance image; the decoding unit comprises a plurality of deconvolution blocks, each deconvolution block consists of a convolution layer and a deconvolution layer, and each convolution layer is connected with the normalization layer and the activation function layer; wherein the number of the volume blocks is the same as the number of the anti-volume blocks; the function adopted by the normalization layer is an example normalization function; the function adopted by the activation function layer is a linear unit function with leakage correction.

In one embodiment, the split network further comprises a connection unit; the coding unit is connected with the decoding unit through the connecting unit; the inputting the downsampling feature map into the decoding unit, and upsampling the downsampling feature map to obtain a mask image of the region of interest includes:

connecting the down-sampling feature map output by the first convolution block with the up-sampling feature map output by the first anti-convolution block through the connecting unit to obtain a connected feature map; the first deconvolution block is a deconvolution block corresponding to the first convolution block;

inputting the connected feature map into a second deconvolution block for up-sampling to obtain a mask image of the region of interest; the second deconvolution block is the next deconvolution block adjacent to the first deconvolution block.

In one embodiment, the number of the convolution blocks and the number of the deconvolution blocks are determined according to the size of the fat suppression magnetic resonance image, the size of the up-sampling feature map, and the size of the down-sampling feature map.

In one embodiment, before the block processing is performed on the fat-suppressed magnetic resonance image and the contrast-limited adaptive histogram equalization processing is performed on each image block as a unit to obtain the processed fat-suppressed magnetic resonance image, the method further includes:

and resampling the fat-suppressed magnetic resonance image according to the size of the fat-suppressed magnetic resonance image and the space between each voxel of the fat-suppressed magnetic resonance image to obtain a resampled magnetic resonance image.

In one embodiment, the training process of the segmentation network includes:

acquiring a sample fat inhibition magnetic resonance image and a gold standard image corresponding to the sample fat inhibition magnetic resonance image; the sample fat suppression magnetic resonance image comprises a sample region of interest;

partitioning the sample fat suppression magnetic resonance image, and performing contrast-limiting adaptive histogram equalization processing by taking each sample image block as a unit to obtain a processed sample fat suppression magnetic resonance image;

inputting the processed sample fat suppression magnetic resonance image into an initial segmentation network, and obtaining a sample mask image of the sample region of interest through the initial segmentation network;

obtaining a loss function value of the initial segmentation network according to the sample mask image and the gold standard image; the loss function of the initial segmentation network comprises a Dice loss function and a cross entropy loss function; the value of the loss function of the initial segmentation network is the sum of the value of the Dice loss function and the value of the cross entropy loss function;

and training the initial segmentation network according to the value of the loss function of the initial segmentation network to obtain the segmentation network.

A method of image segmentation, the method comprising:

carrying out blocking processing on the fat-suppressed mammary gland image containing the breast, and carrying out limited contrast self-adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat-suppressed mammary gland image;

inputting the processed fat-suppressed mammary gland image into a preset breast segmentation network, and acquiring a breast mask image through the breast segmentation network;

acquiring a breast segmentation image according to the breast mask image and the fat-suppressed mammary gland image;

inputting the breast segmentation image into a preset gland segmentation network, and obtaining a gland mask image through the gland segmentation network;

acquiring a gland segmentation image according to the gland mask image and the fat-suppressed mammary gland image or according to the gland mask image and the breast segmentation image;

and acquiring the breast density corresponding to the fat-suppressed breast image according to the breast segmentation image and the gland segmentation image.

An image segmentation apparatus, the apparatus comprising:

the first acquisition module is used for carrying out blocking processing on the fat-suppressed magnetic resonance image and carrying out contrast-limiting adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat-suppressed magnetic resonance image; the fat-suppressed magnetic resonance image comprises a region of interest;

the first segmentation module is used for inputting the processed fat suppression magnetic resonance image into a preset segmentation network and obtaining a mask image of the region of interest through the segmentation network;

and the second acquisition module is used for obtaining a segmentation result of the region of interest in the fat suppression magnetic resonance image according to the fat suppression magnetic resonance image and the mask image.

An image segmentation apparatus, the apparatus comprising:

the first acquisition module is used for carrying out blocking processing on the fat-suppressed mammary gland image containing the breast, and carrying out contrast-limited adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat-suppressed mammary gland image;

the first segmentation module is used for inputting the processed fat-suppressed breast image into a preset breast segmentation network and acquiring a breast mask image through the breast segmentation network;

a second acquisition module, configured to acquire a breast segmentation image according to the breast mask image and the fat-suppressed breast image;

the second segmentation module is used for inputting the breast segmentation image into a preset gland segmentation network and obtaining a gland mask image through the gland segmentation network;

a third obtaining module, configured to obtain a gland segmentation image according to the gland mask image and the fat-suppressed breast image, or according to the gland mask image and the breast segmentation image;

and the fourth acquisition module is used for acquiring the breast density corresponding to the fat-suppressed breast image according to the breast segmentation image and the gland segmentation image.

A computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:

A computer-readable storage medium, having stored thereon a computer program which, when executed by a processor, performs the steps of:

The image segmentation method, the device, the computer equipment and the storage medium can perform the contrast-limiting adaptive histogram equalization processing on the fat-suppression magnetic resonance image by taking each image block as a unit through the block processing of the fat-suppression magnetic resonance image to obtain the processed fat-suppression magnetic resonance image, can enhance the contrast of the fat-suppression magnetic resonance image and reduce the noise points of the fat-suppression magnetic resonance image due to the contrast-limiting adaptive histogram equalization processing, thus the processed fat-suppression magnetic resonance image is input into the preset segmentation network, the processed fat-suppression magnetic resonance image can be accurately segmented through the segmentation network, the mask image of the region of interest in the fat-suppression magnetic resonance image can be accurately obtained, the accuracy of obtaining the mask image of the region of interest is improved, and the fat-suppression magnetic resonance image and the obtained mask image of the region of interest can be obtained according to the fat-suppression magnetic resonance image, the segmentation result of the region of interest in the fat-suppressed magnetic resonance image is accurately obtained.

Drawings

FIG. 1 is a diagram of an exemplary embodiment of an image segmentation method;

FIG. 2 is a flow diagram illustrating a method for image segmentation in one embodiment;

FIG. 2a is a diagram illustrating image contrast with CLAHE pre-processing in one embodiment;

FIG. 3 is a flow diagram illustrating a method for image segmentation in one embodiment;

FIG. 3a is a schematic diagram of a split network in one embodiment;

FIG. 4 is a flow diagram illustrating a method for image segmentation in one embodiment;

FIG. 5 is a flow diagram illustrating a method for image segmentation in one embodiment;

FIG. 6 is a flow diagram illustrating a method for image segmentation in one embodiment;

FIG. 6a is a schematic diagram of a segmentation method for a breast image with suppressed breast fat in an embodiment;

FIG. 7 is a graph illustrating the results of breast and gland segmentation in one embodiment;

FIG. 8 is a graphical illustration of the correlation of the segmentation result volume to the gold standard volume in one embodiment;

FIG. 9 is a block diagram showing the structure of an image segmentation apparatus according to an embodiment;

FIG. 10 is a block diagram showing an example of the structure of an image segmentation apparatus.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The image segmentation method provided by the embodiment of the application can be applied to computer equipment shown in fig. 1. The computer device comprises a processor, a memory connected by a system bus, the memory having stored therein a computer program which, when executed by the processor, is operable to perform the steps of the method embodiments described below. Optionally, the computer device may further comprise a network interface, a display screen and an input means. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a nonvolatile storage medium storing an operating system and a computer program, and an internal memory. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. Optionally, the computer device may be a server, a personal computer, a personal digital assistant, other terminal devices such as a tablet computer, a mobile phone, and the like, or a cloud or a remote server, and the specific form of the computer device is not limited in the embodiment of the present application.

At present, for the segmentation of breast and gland tissues in a three-dimensional magnetic resonance image, manual segmentation and semi-automatic segmentation methods which need user assistance are complex, have low segmentation efficiency, and have great difference between observers and between observers themselves. In the segmentation method based on deep learning, segmentation of tissues and lesion areas depends on manual labeling of doctors, which consumes time and labor, and most of the existing methods realize the segmentation of the tissues and the lesions based on non-fat-suppression magnetic resonance images, while in the clinical breast dynamic enhanced magnetic resonance imaging (DCE-MRI) scanning process, images before and after the injection of a contrast agent are generally obtained by adopting a fat suppression imaging technology, the obtained fat suppression magnetic resonance images have high noise level, low contrast and uneven fat suppression, and the traditional segmentation method cannot accurately segment the tissues and the lesions in the fat suppression magnetic resonance images. Therefore, it is necessary to provide an image segmentation method for accurately segmenting tissues and lesions in fat-suppressed magnetic resonance images.

In one embodiment, as shown in fig. 2, an image segmentation method is provided, which is described by taking the example that the method is applied to the computer device in fig. 1, and comprises the following steps:

s201, performing blocking processing on the fat-suppressed magnetic resonance image, and performing contrast-limited adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat-suppressed magnetic resonance image; the fat-suppressed magnetic resonance image includes a region of interest.

The fat-suppressed magnetic resonance image is an image acquired by injecting a contrast medium into a scan subject to suppress fat at a scan site of the scan subject. The fat-suppressed magnetic resonance image in the present application is a fat-suppressed dynamic-enhancement magnetic resonance image.

Specifically, the computer device performs block processing on the acquired fat-suppressed magnetic resonance image including the region of interest, and performs Contrast-Limited Adaptive Histogram Equalization (CLAHE) processing on each image block after the block processing as a unit to obtain a processed fat-suppressed magnetic resonance image. It should be noted here that the contrast-limited adaptive histogram equalization processing is performed on the acquired fat-suppressed magnetic resonance image in units of blocks, and is a preprocessing operation added to the features of the fat-suppressed magnetic resonance image, such as relatively high noise level, low intensity contrast, and uneven fat suppression. Optionally, the computer device may perform blocking processing on the fat-suppressed magnetic resonance image, perform histogram equalization with the image block as a unit, determine inter-block voxel values by using linear difference values, and simultaneously achieve contrast enhancement and noise suppression, thereby improving the accuracy of segmentation. Illustratively, the image contrast map processed by CLAHE can be shown in fig. 2a, and as can be seen from fig. 2a, the image contrast processed by CLAHE is enhanced and the noise points are reduced. Alternatively, the region of interest may be the breast, or a region of the scanning object including a tumor, for example, the abdomen, lymph, or the like. Alternatively, the computer device may acquire the fat-suppressed magnetic resonance image from a magnetic resonance device, or may acquire the fat-suppressed magnetic resonance image from a PACS (Picture Archiving and Communication Systems) server.

S202, inputting the processed fat suppression magnetic resonance image into a preset segmentation network, and obtaining a mask image of the region of interest through the segmentation network.

Specifically, the computer device inputs the acquired fat-suppressed magnetic resonance image into a preset segmentation network, and obtains a mask image of the region of interest through the segmentation network. It is understood that the obtained mask image of the region of interest is a binary image. In general, in the convolution operation of the neural network, since the voxel point at the edge of the input image is not located at the center of the convolution kernel, and the convolution kernel cannot extend beyond the edge of the input image, information at the edge of the input image may be omitted, which may cause the size of the input image and the size of the output image to be inconsistent. Optionally, the filling convolution is usually filled with "0", and when the convolution kernel is operated on the fat-suppressed magnetic resonance image, the filling convolution can extend to the pseudo voxels beyond the edges of the fat-suppressed magnetic resonance image, so that the output image of the segmentation network and the input fat-suppressed magnetic resonance image have the same size.

And S203, obtaining a segmentation result of the region of interest in the fat-suppressed magnetic resonance image according to the fat-suppressed magnetic resonance image and the mask image.

Specifically, the computer device obtains a segmentation result of the region of interest in the fat-suppressed magnetic resonance image according to the fat-suppressed magnetic resonance image and the mask image of the region of interest. Optionally, the computer device may perform dot product on each pixel point of the fat-suppressed magnetic resonance image and the mask image of the region of interest to obtain a segmentation result of the region of interest in the fat-suppressed magnetic resonance image; or the fat-suppressed magnetic resonance image and the mask image of the region of interest may be superimposed to obtain a segmentation result of the region of interest in the fat-suppressed magnetic resonance image.

In the image segmentation method, the fat-suppressed magnetic resonance image is segmented, the contrast-limited adaptive histogram equalization treatment can be carried out by taking each image block as a unit to obtain the processed fat-suppressed magnetic resonance image, the contrast of the fat-suppressed magnetic resonance image can be enhanced and the noise points of the fat-suppressed magnetic resonance image can be reduced due to the contrast-limited adaptive histogram equalization treatment, so that the processed fat-suppressed magnetic resonance image is input into a preset segmentation network, the processed fat-suppressed magnetic resonance image can be accurately segmented through the segmentation network, the mask image of the region of interest in the fat-suppressed magnetic resonance image is accurately obtained, the accuracy of obtaining the mask image of the region of interest is improved, and the mask image of the region of interest can be obtained according to the fat-suppressed magnetic resonance image and the obtained mask image of the region of interest, the segmentation result of the region of interest in the fat-suppressed magnetic resonance image is accurately obtained.

In the above-mentioned scene where the processed fat-suppressed magnetic resonance image is input into a preset segmentation network, and a mask image of the region of interest is obtained through the segmentation network, in an embodiment, the segmentation network includes: an encoding unit and a decoding unit; as shown in fig. 3, S202 includes:

and S301, inputting the processed fat-suppression magnetic resonance image into an encoding unit, filling the boundary of the processed fat-suppression magnetic resonance image by adopting filling convolution, and performing down-sampling on the fat-suppression magnetic resonance image after the boundary is filled to obtain a down-sampling feature map corresponding to the processed fat-suppression magnetic resonance image.

Specifically, the computer device inputs the processed fat-suppressed magnetic resonance image into an encoding unit of a segmentation network, performs boundary filling on the processed fat-suppressed magnetic resonance image by using filling convolution, and performs downsampling on the fat-suppressed magnetic resonance image after the boundary filling to obtain a downsampling feature map corresponding to the processed fat-suppressed magnetic resonance image. Optionally, as shown in fig. 3a, the encoding unit includes a plurality of convolution blocks, each convolution block is composed of a convolution layer and a pooling layer, and each convolution layer connects the normalization layer and the activation function layer of the partition network, and the convolution layer includes a padding convolution; the filling convolution is used for filling the boundary of the processed fat-suppression magnetic resonance image, the size of an image matrix corresponding to the processed fat-suppression magnetic resonance image is increased, the function adopted by the normalization layer is an example normalization function, and the function adopted by the activation function layer is a linear unit function with leakage correction. Illustratively, as an implementation manner, each of the above mentioned convolution blocks may be composed of two 3 × 3 × 3 convolution layers and one 2 × 2 × 2 max pooling layer, and each convolution layer is followed by a Normalization layer and a modified Linear Unit (ReLU). It is understood that the boundary filling operation is performed on the processed fat-suppressed magnetic resonance image by using filling convolution to fill the convolution layer included in each convolution block. Alternatively, the number of downsampling performed on the boundary-filled fat-suppressed magnetic resonance image by the encoding unit may be determined according to the size of the fat-suppressed magnetic resonance image, the size of the obtained downsampling feature map, and the size of the upsampling feature map, and in this embodiment, the number of downsampling may be set to 5.

S302, inputting the downsampled feature map into a decoding unit, filling the boundary of the downsampled feature map by adopting filling convolution, and upsampling the downsampled feature map after the boundary is filled to obtain a mask image of the region of interest.

Specifically, the computer device inputs the obtained downsampling feature map corresponding to the processed fat-suppressed magnetic resonance image into a decoding unit of a segmentation network, performs boundary filling on the obtained downsampling feature map by using filling convolution, and performs upsampling on the downsampling feature map after the boundary filling to obtain a mask image of the region of interest. Optionally, with continued reference to fig. 3a, the decoding unit includes a plurality of deconvolution blocks, each deconvolution block is composed of a convolution layer and a deconvolution layer, each convolution layer connects the normalization layer and the activation function layer, a function used by the normalization layer is an example normalization function, and a function used by the activation function layer is a linear unit function with leakage correction. Optionally, the number of the rolling blocks included in the encoding unit is the same as the number of the reverse rolling blocks included in the decoding unit. That is, the number of downsampling and the number of upsampling by the segmentation network correspond. Alternatively, the number of the above-mentioned convolution blocks and the above-mentioned inverse convolution blocks may be determined in accordance with the size of the fat-suppressed magnetic resonance image, the size of the above-mentioned up-sampling feature map, and the size of the above-mentioned down-sampling feature map. It will be appreciated that the decoding unit described above is an upsampling process, the decoding unit comprising a plurality of deconvolution blocks, and the upsampling may be implemented by a 2 x 2 deconvolution. Optionally, the last layer of the decoding unit may map the upsampled feature map to an output layer of the entire segmentation network through convolution of 1 × 1 × 1, and the output image is a classification result of each pixel point of the input image, that is, a classification result of a background voxel or a foreground voxel.

It should be noted that the function used by the Normalization layer IN this embodiment is an example Normalization function because the GPU memory of the computer device limits the size of the Batch (Batch) IN the training process, and therefore, the example Normalization (IN) is used instead of the Batch Normalization (Batch Normalization). When the region of interest of the fat-suppressed magnetic resonance image is segmented, background voxels (voxel values are approximately 0) in the fat-suppressed magnetic resonance image are more and can become negative values after preprocessing, and in the learning process based on the gradient, if a ReLU function is used, the gradient parameters of neurons are always 0 and will not be activated in the training process at the back, so that the training is slow, so that a leakage-corrected linear unit (Leaky ReLU) function is used to replace the ReLU function, and when the input voxel value is a negative value, the gradient parameters of the neurons are not 0 (but a small number), so that the neurons can be activated all the time, and the learning speed of the network is accelerated.

In this embodiment, the computer device inputs the processed fat-suppressed magnetic resonance image into the encoding unit of the segmentation network, the boundary filling can be performed on the processed fat-suppressed magnetic resonance image by using the filling convolution through the encoding unit, the size of the matrix corresponding to the processed fat-suppressed magnetic resonance image can be increased, and it is ensured that information at the edge of the processed fat-suppressed magnetic resonance image is not missed, so that the processed fat-suppressed magnetic resonance image can completely retain the information of the processed fat-suppressed magnetic resonance image, the boundary-filled fat-suppressed magnetic resonance image can be accurately downsampled, the accuracy of obtaining the downsampled feature map corresponding to the processed fat-suppressed magnetic resonance image is improved, and the obtained downsampled feature map can be input into the decoding unit of the segmentation network, the obtained down-sampling feature map can be subjected to boundary filling by adopting filling convolution through the decoding unit, so that the down-sampling feature map after the boundary filling can completely retain the information of the down-sampling feature map, the down-sampling feature map after the boundary filling can be accurately up-sampled, and the accuracy of the mask image of the region of interest of the processed fat suppression magnetic resonance image is improved.

In a scene where the obtained downsampled feature map is input to a decoding unit of a segmentation network, and the downsampled feature map is upsampled to obtain a mask image of a region of interest of a fat-suppressed image, the segmentation network further includes a connection unit, and the encoding unit is connected to the encoding unit through the connection unit, in one embodiment, as shown in fig. 4, the S302 includes:

s401, connecting the down-sampling feature map output by the first convolution block with the up-sampling feature map output by the first anti-convolution block through a connecting unit to obtain a connected feature map; the first reverse volume block is a reverse volume block corresponding to the first volume block.

Specifically, the computer device connects the downsampled feature map output by the first convolution block of the encoding unit and the upsampled feature map output by the first deconvolution block of the decoding unit through a connection unit of a split network to obtain a connected feature map. Wherein the first deconvolution is a deconvolution block corresponding to the first convolution block. For example, with continued reference to fig. 3a, the connection unit of the partition network may be the dotted line shown in fig. 3a, the first convolution block may be the last convolution block from top to bottom in the down-sampling process (left side) in fig. 3a, and the first de-convolution block may be the first de-convolution block from bottom to top in the up-sampling process (right side) in fig. 3 a. Optionally, the computer device may splice the downsampling feature map output by the first convolution block and the upsampling feature map output by the first anti-convolution block through the connection unit, so as to obtain a connected feature map.

S402, inputting the connected feature map into a second deconvolution block for upsampling to obtain a mask image of the region of interest; the second deconvolution block is the next deconvolution block adjacent to the first deconvolution block.

Specifically, the computer device inputs the connected feature map into a second deconvolution block for sampling, so as to obtain a mask image of the region of interest of the fat-suppressed magnetic resonance image. Wherein the second deconvolution block is the next deconvolution block adjacent to the first deconvolution block. The second deconvolution block may illustratively be the second deconvolution block from bottom to top in the upsampling process (right side) of fig. 3 a.

In this embodiment, the computer device connects the downsampling feature map output by the first convolution block of the encoding unit and the upsampling feature map output by the first deconvolution block of the decoding unit through the connection unit of the segmentation network, and the obtained connected feature map can avoid loss of detail information in the downsampling and upsampling processes, and can obtain a connected feature map with higher accuracy, so that the connected feature map with higher accuracy can be input into the second deconvolution block of the decoding unit to be upsampled, a mask image of a region of interest with higher accuracy is obtained, and the segmentation accuracy of the fat-suppressed image is improved.

In some scenarios, before the blocking the fat-suppressed magnetic resonance image and performing the contrast-limited adaptive histogram equalization on each image block unit to obtain the processed fat-suppressed magnetic resonance image, the computer device may further perform resampling on the fat-suppressed magnetic resonance image and perform blocking on the resampled magnetic resonance image, where in an embodiment, before S201, the method further includes: and resampling the fat-suppressed magnetic resonance image according to the size of the fat-suppressed magnetic resonance image and the distance between voxels of the fat-suppressed magnetic resonance image to obtain a resampled magnetic resonance image.

Specifically, the computer device performs resampling on the fat-suppressed magnetic resonance image according to the size of the fat-suppressed magnetic resonance image and the distance between voxels of the fat-suppressed magnetic resonance image, and obtains a resampled magnetic resonance image. It should be noted that, for the anisotropic fat-suppressed magnetic resonance image, since the voxel pitches in the respective directions are different, the direction with the higher resolution of the image may be down-sampled to make the resolution thereof coincide with the other directions, and then the anisotropic fat-suppressed magnetic resonance image in all directions may be re-sampled according to the processing method of the isotropic image. Optionally, the computer device may resample the fat-suppressed magnetic resonance image by using a third-order spline interpolation, or resample the fat-suppressed magnetic resonance image by using a nearest neighbor interpolation.

In this embodiment, the computer device can resample the fat-suppressed magnetic resonance image according to the size of the fat-suppressed magnetic resonance image and the distance between voxels of the fat-suppressed magnetic resonance image, obtain a resampled magnetic resonance image, and ensure the accuracy of the obtained resampled magnetic resonance image.

In the above scenario of inputting the fat-suppressed mr image into the segmentation network, the segmentation network needs to be trained in advance, and in one embodiment, as shown in fig. 5, the training process of the segmentation network includes:

s501, acquiring a sample fat suppression magnetic resonance image and a gold standard image corresponding to the sample fat suppression magnetic resonance image; the sample fat suppression magnetic resonance image includes a sample region of interest.

Specifically, the computer device acquires a sample fat suppression magnetic resonance image and a gold standard image corresponding to the sample fat suppression magnetic resonance image. Wherein the sample fat suppression magnetic resonance image comprises a sample sensitive region of interest. Alternatively, the sample region of interest may be the breast, or may be a region of the scan subject that includes a tumor, such as the abdomen, lymph, or the like. Alternatively, the computer device may acquire the sample fat suppression magnetic resonance image from a magnetic resonance device, or may acquire the sample fat suppression magnetic resonance image from a PACS (Picture Archiving and Communication Systems) server. Optionally, the computer device may employ data amplification operations such as mirror image transformation, scale-scaling transformation, rotational translation, and the like, to expand the volume of the sample fat suppression magnetic resonance image and prevent overfitting of the model.

And S502, performing blocking processing on the sample fat suppression magnetic resonance image, and performing contrast-limiting adaptive histogram equalization processing by taking each sample image block as a unit to obtain a processed sample fat suppression magnetic resonance image.

Specifically, the computer device performs block processing on the obtained sample fat suppression magnetic resonance image, and performs Contrast Limited Adaptive Histogram Equalization (CLAHE) processing on each block of the sample image after the block processing, so as to obtain a processed sample fat suppression magnetic resonance image. It should be noted here that the contrast-limited adaptive histogram equalization processing is performed on the acquired sample fat-suppressed magnetic resonance image in units of blocks, which is a preprocessing operation added to the characteristics of the sample fat-suppressed magnetic resonance image, such as relatively high noise level, low intensity contrast, and uneven fat suppression. Optionally, the computer device may perform blocking processing on the sample fat suppression magnetic resonance image, perform histogram equalization with the image block as a unit, determine a pixel value between blocks by using a linear difference value, and simultaneously realize contrast enhancement and noise suppression, thereby improving the accuracy of segmentation. Optionally, the computer device may perform resampling processing on the sample fat suppression magnetic resonance image before performing blocking processing on the sample fat suppression magnetic resonance image, and optionally, the computer device may perform resampling processing on the sample fat suppression magnetic resonance image by using a third-order spline interpolation. Optionally, the computer device may also perform resampling processing on the gold standard image corresponding to the sample fat suppression magnetic resonance image by using nearest neighbor interpolation.

And S503, inputting the processed sample fat suppression magnetic resonance image into an initial segmentation network, and obtaining a sample mask image of the sample region of interest through the initial segmentation network.

Specifically, the computer device inputs the processed sample fat suppression magnetic resonance image obtained as described above into an initial segmentation network, and obtains a sample mask image of a sample region of interest through the initial segmentation network. Optionally, the network structure of the initial segmentation network may refer to the description of the above embodiments, and this embodiment is not described herein again.

S504, obtaining a loss function value of the initial segmentation network according to the sample mask image and the gold standard image; the loss function of the initial segmentation network comprises a Dice loss function and a cross entropy loss function; the value of the loss function of the initial segmentation network is the sum of the value of the Dice loss function and the value of the cross entropy loss function.

Specifically, the computer device obtains the value of the loss function of the initial segmentation network according to the obtained sample mask image and the gold standard image corresponding to the sample fat suppression magnetic resonance image. The loss function of the initial segmentation network comprises a Dice loss function and a cross entropy loss function; the value of the loss function of the initial segmentation network is the sum of the value of the Dice loss function and the value of the cross entropy loss function. It can be understood that the Dice loss function can directly optimize the segmentation similarity and can solve the problem of unbalanced training sample classes, but the gradient form is complex, and the problem that the network is difficult to converge due to severe gradient change is easy to occur in the back propagation process. The cross entropy loss function can measure the difference between the gold standard image and the segmentation result in the segmentation task, the smaller the cross entropy value is, the better the segmentation effect is, but the problem of category imbalance cannot be solved, so the stability of the training process and the accuracy of segmentation can be effectively improved by combining the Dice loss function and the cross entropy loss function. Alternatively, the expressions of the Dice loss function and the cross entropy loss function may be as shown in the following equations (1) and (2):

where K denotes the number of classes (in this patent, K is 2, representing foreground and background), I denotes the set of voxels in each Batch, u denotes the Softmax output probability value, and v denotes the one-hot code value of the gold standard.

And S505, training the initial segmentation network according to the loss function value of the initial segmentation network to obtain the segmentation network.

Specifically, the computer device trains the initial segmentation network according to the value of the loss function of the initial segmentation network to obtain the segmentation network. Optionally, the computer device may determine, as the segmentation network, an initial segmentation network corresponding to a time when the value of the loss function of the initial segmentation network reaches a stable value or a minimum value. Optionally, the computer device may use five-fold cross validation in the training process of segmenting the network, and randomly divide the training set into 5 equal parts, and randomly select one part as the validation set in each fold cross validation process to determine the hyper-parameters of the model. And after the training is finished, fusing the verification results of the five models to determine the hyper-parameters of the test model. In the training process, the initial learning rate of the initial segmentation network can be set to be 0.01, and the learning rate is gradually reduced by adopting exponential decay along with the continuation of iteration, so that the model is more stable. Optionally, the computer device may use a normal distribution random initialization weight strategy to assign an initial value to the initial segmentation network. Optionally, the computer device may also optimize the loss function using a random gradient descent method with a Nesterov Momentum (parameter set to 0.9). Optionally, in this embodiment, the computer device may define the period of the training process (Epoch) as the iterative optimization over 250 Batch, set the maximum value to 1000, and stop the training when the learning rate is lower than 10 "6 or exceeds 1000 epochs. Alternatively, during the testing phase, the computer device may predict each test sample using a sliding window approach, where the overlap region is set to half the size of the input image, increasing the weight near the center region.

Optionally, in an embodiment, in the training process, the fat-suppressed magnetic resonance image of the sample may be subjected to resampling and contrast-limited adaptive histogram equalization preprocessing, the image sizes of the images subjected to the resampling and contrast-limited adaptive histogram equalization preprocessing are unified into a median Size of an image in a training data set, a Batch Size (Batch Size) of the training process may be determined according to a memory Size of a GPU of a computer device, and it is preferably satisfied that an input Size is as large as possible to ensure accuracy of a segmentation result, and it is second satisfied that the Batch Size is as small as 2 to prevent gradient noise caused by too few samples. Illustratively, taking training a breast segmentation network and a gland segmentation network as an example, the image input sizes of the breast segmentation and the gland segmentation network may be 40 × 256 × 192 and 64 × 96 × 224, respectively, and the batch size may be set to 2. To obtain features with sufficient information content, the size of the output feature map can be set to a minimum of 4 × 4 × 4; to limit the model size, an upper limit on the number of features may be set to 320. Optionally, the number of downsampling the sample fat suppression magnetic resonance image in the training process may be determined according to the size of the input sample fat suppression magnetic resonance image and the size of the feature map, and exemplarily, the number of downsampling may be 5 as an implementable manner.

In this embodiment, the computer device performs the block processing on the sample fat suppression magnetic resonance image, and can perform the contrast-limited adaptive histogram equalization processing on each sample image block as a unit to obtain the processed sample fat suppression magnetic resonance image, and since the contrast-limited adaptive histogram equalization processing is performed, the contrast of the sample fat suppression magnetic resonance image can be enhanced, and the noise point of the sample fat suppression magnetic resonance image can be reduced, so that the processed sample fat suppression magnetic resonance image is input into the initial segmentation network, and the sample mask image of the sample region-of-interest can be accurately obtained through the initial segmentation network, so that the value of the loss function of the initial segmentation network can be obtained according to the sample mask image and the gold standard image corresponding to the sample fat suppression magnetic resonance image, and further, the value of the loss function of the initial segmentation network can be obtained according to the value of the loss function of the initial segmentation network, the initial segmentation network is trained, the initial segmentation network can be accurately trained through a large number of sample fat suppression magnetic resonance images, and the accuracy of the obtained segmentation network is improved.

In one embodiment, as shown in fig. 6, taking a breast image as an example, an image segmentation method is provided, which is described by taking the method as an example applied to the computer device in fig. 1, and includes the following steps:

and S601, performing blocking processing on the fat-suppressed mammary gland image containing the breast, and performing contrast-limited adaptive histogram equalization processing by taking each image block as a unit to obtain the processed fat-suppressed mammary gland image.

Specifically, the computer device performs a blocking process on the acquired fat-suppressed breast image including the breast, and performs a contrast-limited adaptive histogram equalization process on each image block unit, resulting in a processed fat-suppressed breast image. Here, the contrast-limited adaptive histogram equalization processing is performed on the acquired fat-suppressed breast image in units of blocks, and is a preprocessing operation added to the characteristics of the fat-suppressed breast image, such as a relatively high noise level, a low intensity contrast, and uneven fat suppression. Optionally, the computer device may perform blocking processing on the fat-suppressed breast image, perform histogram equalization with image blocks as units, determine inter-block voxel values by using linear difference values, and simultaneously implement contrast enhancement and noise suppression, thereby improving the accuracy of segmentation. Alternatively, the fat-suppressed breast image acquired by the computer device may be a fat-suppressed breast image including bilateral breasts, or a fat-suppressed breast image including a unilateral breast. In the present application, the fat-suppressed magnetic resonance image including the breast may be a fat-suppressed dynamic-enhancement magnetic resonance image. Alternatively, the computer device may acquire the fat-suppressed magnetic resonance image including the breast from a magnetic resonance apparatus, or may acquire the fat-suppressed magnetic resonance image including the breast from a PACS (Picture Archiving and Communication Systems) server.

And S602, inputting the processed fat-suppressed breast image into a preset breast segmentation network, and acquiring a breast mask image through the breast segmentation network.

Specifically, the computer device inputs the above-described processed fat-suppressed breast image into a preset breast segmentation network, through which a breast mask image is acquired. Wherein the convolution layer of the breast segmentation network comprises a filling convolution; the padding convolution is used to perform boundary padding on the fat-suppressed breast image and increase the size of an image matrix corresponding to the fat-suppressed breast image. It should be noted that, for a detailed description of the network structure of the breast segmentation network, reference is made to the above description of the segmentation network, and details of this embodiment are not repeated herein. Illustratively, the resulting breast mask image may be a breast mask image as exemplified in 7 a.

S603, acquiring a breast segmentation image according to the breast mask image and the fat-suppressed mammary gland image.

Specifically, the computer device acquires a breast segmentation image from the obtained breast mask image and the above-described fat-suppressed breast image. Alternatively, the computer device may perform a dot product of the breast mask image and the fat-suppressed breast image to obtain a breast segmentation image. Optionally, the computer device may also superimpose the breast mask image and the fat-suppressed breast image to obtain a breast segmentation image.

And S604, inputting the breast segmentation image into a preset gland segmentation network, and obtaining a gland mask image through the gland segmentation network.

Specifically, the computer device inputs the breast segmentation image into a preset gland segmentation network, and obtains a gland mask image through the gland segmentation network. Wherein the convolution layer of the gland segmentation network comprises a filling convolution; the padding convolution is used to perform boundary padding on the breast segmentation image and increase the size of an image matrix corresponding to the breast segmentation image. It should be noted that, for a specific description of the network structure of the gland segmentation network, reference is made to the description of the segmentation network, and details are not described herein again in this embodiment. Illustratively, the resulting gland mask image may be as exemplified in fig. 6 a. Optionally, before the computer device inputs the breast segmentation image into the gland segmentation network, the computer device may further perform blocking processing on the breast segmentation image, perform contrast-limited adaptive histogram equalization processing on each image block as a unit to obtain a processed breast segmentation image, and input the processed breast segmentation image into the gland segmentation network.

S605, acquiring a gland segmentation image according to the gland mask image and the fat-suppressed mammary gland image or according to the gland mask image and the breast segmentation image.

Specifically, the computer equipment acquires a gland segmentation image according to the obtained gland mask image and the fat-suppressed mammary gland image; or the computer equipment acquires a gland segmentation image according to the obtained gland mask image and the breast segmentation image. Optionally, the computer device may perform dot multiplication on the gland mask image and the fat-suppressed breast image to obtain a segmented gland image, or may superimpose the gland mask image and the fat-suppressed breast image to obtain a segmented gland image. Optionally, the computer device may perform dot multiplication on the gland mask image and the breast segmentation image to obtain a gland segmentation image, or may perform superposition on the gland mask image and the breast segmentation image to obtain a gland segmentation image. It can be understood that, since the obtained breast segmentation image may have a certain error, the obtaining of the gland segmentation image according to the gland mask image and the breast segmentation image may overlap the error, so that the accuracy of the obtained gland segmentation image is lower than that of the gland segmentation image obtained according to the gland mask image and the fat-suppressed breast image.

And S606, acquiring the breast density corresponding to the fat-suppressed breast image according to the breast segmentation image and the gland segmentation image.

Specifically, the computer device obtains a breast density corresponding to the fat-suppressed breast image from the obtained breast segmentation image and the obtained gland segmentation image. Alternatively, since the obtained breast segmentation image and the gland segmentation image are both three-dimensional images, the computer device may obtain the volume of the breast and the volumes of bilateral glands according to the voxel intervals of the breast segmentation image and the gland segmentation image and the resolutions of the breast segmentation image and the gland segmentation image, so as to obtain the breast density. Alternatively, the computer device may multiply the number of voxels of the breast-segmented image and the gland-segmented image and the resolution of the breast-segmented image and the gland-segmented image to obtain the volume of the breast and the volume of the bilateral glands.

In the image segmentation method, the fat-suppressed breast image containing the breast is subjected to the blocking processing, the contrast-limited adaptive histogram equalization processing is carried out by taking each image block as a unit, the contrast of the fat-suppressed breast image can be enhanced and the noise points of the fat-suppressed breast image can be reduced by the contrast-limited adaptive histogram equalization processing, so that the processed fat-suppressed breast image is input into a preset segmentation network, the processed fat-suppressed breast image can be accurately segmented by the segmentation network, the breast mask image can be accurately obtained, the accuracy of the obtained breast mask image is improved, and the accuracy of the obtained gland mask image is also improved, so that the accuracy of the obtained breast segmentation image and the gland segmentation image is improved, and the image can be segmented according to the breast segmentation image and the gland, the mammary gland density corresponding to the fat-suppressed mammary gland image is accurately obtained, and the accuracy of obtaining the mammary gland density corresponding to the fat-suppressed mammary gland image is improved.

In one embodiment, the effect of the image segmentation method proposed in the present application is evaluated, and table 1 shows five evaluation index results of the fat-suppressed breast image segmentation using the segmentation network in comparison with the gold standard, where the results are all expressed in the form of a mean ± standard deviation.

TABLE 1 evaluation of the segmentation Performance

As can be seen from Table 1, the DSC of the breast segmentation and the glandular segmentation are 0.968 + -0.017 and 0.877 + -0.081 respectively, and the ASD are 0.201 + -0.082 mm and 0.310 + -0.041 mm respectively, which shows that the segmentation model can accurately segment the target region and accurately detect the boundary. In general, a deep learning method for medical images tends to produce a high False Positive Rate (FPR), resulting in over-segmentation. For breast and gland segmentation, segmentation results of the segmentation network provided by the application have high sensitivity and specificity, and the segmentation network has higher True Positive Rate (TPR) and lower false Positive Rate, so that excessive segmentation can be effectively reduced. Fig. 7 shows an example of the segmentation result, with a DSC of 0.961 and an ASD of 0.175 mm for the breast segmentation; the DSC for the glandular division was 0.922 mm and the ASD was 0.296 mm. The correlation between the segmentation result and the gold standard can be shown in fig. 8, where the x-axis in fig. 8 represents the physical actual volume of the gold standard, and the y-axis represents the physical actual volume of the segmentation result. The correlation coefficients r of the breast segmentation and the gland segmentation were 0.995(p-value <0.001) and 0.971(p-value <0.001), respectively, indicating that the segmentation results are in strong agreement with the gold standard.

It should be understood that although the various steps in the flow charts of fig. 2-6 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2-6 may include multiple steps or phases, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed in turn or alternately with other steps or at least some of the other steps or phases.

In one embodiment, as shown in fig. 9, there is provided an image segmentation apparatus including: first acquisition module, first segmentation module and second acquisition module, wherein:

the first acquisition module is used for carrying out blocking processing on the fat-suppressed magnetic resonance image and carrying out contrast-limiting adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat-suppressed magnetic resonance image; the fat-suppressed magnetic resonance image includes a region of interest.

And the first segmentation module is used for inputting the processed fat suppression magnetic resonance image into a preset segmentation network and obtaining a mask image of the region of interest through the segmentation network.

The image segmentation apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.

On the basis of the foregoing embodiment, optionally, the split network includes: an encoding unit and a decoding unit; the first division module includes: a down-sampling unit and an up-sampling unit, wherein:

and the downsampling unit is used for inputting the processed fat suppression magnetic resonance image into the encoding unit, performing boundary filling on the processed fat suppression magnetic resonance image by adopting filling convolution, and downsampling the fat suppression magnetic resonance image after the boundary filling to obtain a downsampling feature map corresponding to the processed fat suppression magnetic resonance image.

And the up-sampling unit is used for inputting the down-sampling feature map into the decoding unit, performing boundary filling on the down-sampling feature map by adopting filling convolution, and performing up-sampling on the down-sampling feature map after the boundary filling to obtain a mask image of the interesting region.

On the basis of the foregoing embodiment, optionally, the encoding unit includes a plurality of convolution blocks, each convolution block is composed of a convolution layer and a pooling layer, and each convolution layer connects a normalization layer and an activation function layer of the partition network; the convolutional layer comprises a padding convolution; the filling convolution is used for filling the boundary of the fat suppression magnetic resonance image and increasing the size of an image matrix corresponding to the fat suppression magnetic resonance image; the decoding unit comprises a plurality of deconvolution blocks, each deconvolution block consists of a convolution layer and a deconvolution layer, and each convolution layer is connected with a normalization layer and an activation function layer; wherein the number of the rolling blocks is the same as the number of the reverse rolling blocks; the function adopted by the normalization layer is an example normalization function; the function adopted by the activation function layer is a band leakage correction linear unit function.

On the basis of the above embodiment, optionally, the split network further includes a connection unit; the coding unit is connected with the decoding unit through the connecting unit; the upsampling unit is specifically configured to connect the downsampling feature map output by the first convolution block and the upsampling feature map output by the first deconvolution block through the connection unit to obtain a connected feature map; the first reverse volume block is a reverse volume block corresponding to the first volume block; inputting the connected feature map into a second deconvolution block for up-sampling to obtain a mask image of the region of interest; the second deconvolution block is the next deconvolution block adjacent to the first deconvolution block.

On the basis of the above embodiment, optionally, the number of the rolling blocks and the number of the deconvolution blocks are determined according to the size of the fat-suppressed magnetic resonance image, the size of the up-sampling feature map, and the size of the down-sampling feature map.

On the basis of the foregoing embodiment, optionally, the apparatus further includes: a resampling module, wherein:

and the resampling module is used for resampling the fat suppression magnetic resonance image according to the size of the fat suppression magnetic resonance image and the space between each voxel of the fat suppression magnetic resonance image to obtain a resampled magnetic resonance image.

On the basis of the foregoing embodiment, optionally, the apparatus further includes: the third obtains module, processing module, second segmentation module, fourth and obtains module and training module, wherein:

the third acquisition module is used for acquiring a sample fat suppression magnetic resonance image and a gold standard image corresponding to the sample fat suppression magnetic resonance image; the sample fat suppression magnetic resonance image includes a sample region of interest.

And the processing module is used for carrying out blocking processing on the sample fat suppression magnetic resonance image, carrying out contrast ratio limiting adaptive histogram equalization processing by taking each sample image block as a unit, and obtaining the processed sample fat suppression magnetic resonance image.

And the second segmentation module is used for inputting the processed sample fat suppression magnetic resonance image into the initial segmentation network and obtaining a sample mask image of the sample region of interest through the initial segmentation network.

The fourth acquisition module is used for acquiring a loss function value of the initial segmentation network according to the sample mask image and the gold standard image; the loss function of the initial segmentation network comprises a Dice loss function and a cross entropy loss function; the value of the loss function of the initial segmentation network is the sum of the value of the Dice loss function and the value of the cross entropy loss function.

And the training module is used for training the initial segmentation network according to the value of the loss function of the initial segmentation network to obtain the segmentation network.

For specific limitations of the image segmentation apparatus, reference may be made to the above limitations of the image segmentation method, which are not described herein again. The respective modules in the image segmentation apparatus can be wholly or partially implemented by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor calls and executes operations corresponding to the modules.

In one embodiment, as shown in fig. 10, there is provided an image segmentation apparatus including: the first module of acquireing, first segmentation module, second acquire module, second segmentation module, third acquire module and fourth acquire the module, wherein:

the first acquisition module is used for carrying out blocking processing on the fat-suppressed breast image containing the breast, and carrying out contrast-limited adaptive histogram equalization processing by taking each image block as a unit to obtain the processed fat-suppressed breast image.

And the first segmentation module is used for inputting the processed fat-suppressed breast image into a preset breast segmentation network and acquiring a breast mask image through the breast segmentation network.

The second acquisition module is used for acquiring a breast segmentation image according to the breast mask image and the fat-suppressed mammary gland image;

and the second segmentation module is used for inputting the breast segmentation image into a preset gland segmentation network and obtaining a gland mask image through the gland segmentation network.

The third acquisition module is used for acquiring a gland segmentation image according to the gland mask image and the fat-suppressed mammary gland image or according to the gland mask image and the breast segmentation image;

In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:

partitioning the fat suppression magnetic resonance image, and performing contrast-limited adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat suppression magnetic resonance image; the fat suppressed magnetic resonance image includes a region of interest;

The implementation principle and technical effect of the computer device provided by the above embodiment are similar to those of the above method embodiment, and are not described herein again.

In one embodiment, a computer-readable storage medium is provided, having stored thereon a computer program which, when executed by a processor, performs the steps of:

The implementation principle and technical effect of the computer-readable storage medium provided by the above embodiments are similar to those of the above method embodiments, and are not described herein again.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, or the like. Volatile Memory can include Random Access Memory (RAM) or external cache Memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), for example.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent application shall be subject to the appended claims.

Claims

1. A method of image segmentation, the method comprising:

partitioning the fat suppression magnetic resonance image, and performing contrast-limiting adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat suppression magnetic resonance image; the fat-suppressed magnetic resonance image comprises a region of interest;

2. The method of claim 1, wherein splitting the network comprises: an encoding unit and a decoding unit; the step of inputting the processed fat suppression magnetic resonance image into a preset segmentation network, and obtaining a mask image of the region of interest through the segmentation network includes:

3. The method of claim 2, wherein the coding unit comprises a plurality of convolutional blocks, each convolutional block consisting of a convolutional layer and a pooling layer, and each convolutional layer connecting a normalization layer and an activation function layer of the partition network; the convolutional layer comprises the filler convolution; the filling convolution is used for filling the boundary of the fat suppression magnetic resonance image and increasing the size of an image matrix corresponding to the fat suppression magnetic resonance image; the decoding unit comprises a plurality of deconvolution blocks, each deconvolution block consists of a convolution layer and a deconvolution layer, and each convolution layer is connected with the normalization layer and the activation function layer; wherein the number of the volume blocks is the same as the number of the anti-volume blocks; the function adopted by the normalization layer is an example normalization function; the function adopted by the activation function layer is a linear unit function with leakage correction.

4. The method of claim 3, wherein the split network further comprises a connection unit; the coding unit is connected with the decoding unit through the connecting unit; the inputting the downsampling feature map into the decoding unit, and upsampling the downsampling feature map to obtain a mask image of the region of interest includes:

connecting the down-sampling feature map output by the first convolution block with the up-sampling feature map output by the first anti-convolution block through the connecting unit to obtain a connected feature map; the first reverse volume block is a reverse volume block corresponding to the first volume block;

5. The method of claim 2, wherein the number of the convolution blocks and the de-convolution blocks is determined according to a size of the fat-suppressed magnetic resonance image, a size of the up-sampled feature map, and a size of the down-sampled feature map.

6. The method according to claim 1, wherein before the fat-suppressed magnetic resonance image is subjected to the blocking processing and the contrast-limited adaptive histogram equalization processing is performed on a per image block basis to obtain the processed fat-suppressed magnetic resonance image, the method further comprises:

7. The method according to any one of claims 1 to 6, wherein the training process of the split network comprises:

acquiring a sample fat suppression magnetic resonance image and a gold standard image corresponding to the sample fat suppression magnetic resonance image; the sample fat suppression magnetic resonance image comprises a sample region of interest;

partitioning the sample fat suppression magnetic resonance image, and carrying out contrast limiting adaptive histogram equalization processing by taking each sample image block as a unit to obtain a processed sample fat suppression magnetic resonance image;

8. A method of image segmentation, the method comprising:

carrying out blocking processing on the fat-suppressed mammary gland image containing the breast, and carrying out contrast-limited adaptive histogram equalization processing by taking each image block as a unit to obtain a processed fat-suppressed mammary gland image;

inputting the processed fat-suppressed breast image into a preset breast segmentation network, and acquiring a breast mask image through the breast segmentation network;

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 8.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 8.