Brain image segmentation method, device, computer equipment and readable storage medium

Abstract

The present application relates to a method, device, computer equipment and readable storage medium for segmenting brain images. The method includes: obtaining a brain CT image, which includes adjacent multi-layer tomographic images; inputting the brain CT image into a preset segmentation model to obtain an intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to sample images, segmentation gold standards and classification gold standards, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image. In this method, the classification loss is used to guide the training of the segmentation model, thereby greatly improving the accuracy of the trained segmentation model, thereby improving the accuracy of the brain CT image segmentation results. At the same time, the brain CT image is an adjacent multi-layer tomographic image, which can fully consider the contextual information between each layer of the image, and further improve the accuracy of the obtained brain CT image segmentation results.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a brain image segmentation method, device, computer equipment and readable storage medium. Background Art

[0002] Intracranial hemorrhage (ICH) refers to bleeding caused by ruptured blood vessels in the brain, which can damage brain cells and compress surrounding nerve tissues, causing other disorders, which has a great impact on the patient's health. In the current clinical medicine field, doctors usually look at the patient's CT images to determine the location of intracranial hemorrhage; with the continuous development of deep learning technology, there are deep learning-based intracranial hemorrhage segmentation methods to assist doctors in determining the location of intracranial hemorrhage, thereby improving doctors' work efficiency.

[0003] Before using a deep learning network to segment intracranial hemorrhage in CT images, the deep learning network needs to be trained. In traditional technology, each 2D image in the CT image is usually used as training data, and the deep learning network is trained based on the loss between the output result corresponding to the 2D image and the segmentation gold standard.

[0004] However, the deep learning network trained by traditional technology has low accuracy, and the accuracy of intracranial hemorrhage segmentation results is also low. Summary of the invention

[0005] Based on this, it is necessary to provide a brain image segmentation method, device, computer equipment and readable storage medium to address the problem of low accuracy of intracranial hemorrhage segmentation results obtained by deep learning networks in traditional technologies.

[0006] A brain image segmentation method, the method comprising:

[0007] Acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice tomographic images;

[0008] The brain CT image is input into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0009] In one embodiment, the segmentation model is trained by:

[0010] The sample image is input into the initial segmentation network to obtain the feature map and predicted segmentation probability map corresponding to the sample image; the predicted segmentation probability map is used to characterize the probability that each pixel point in the sample image belongs to an intracranial hemorrhage point;

[0011] The feature map is input into the classification network connected to the initial segmentation network to obtain the predicted classification result corresponding to the sample image;

[0012] The first loss between the predicted segmentation probability map and the segmentation gold standard, and the second loss between the predicted classification result and the classification gold standard are calculated, and the initial segmentation network is trained according to the first loss and the second loss to obtain a segmentation model.

[0013] In one embodiment, the segmentation gold standard includes an intracranial hemorrhage mask of the sample image, wherein the intracranial hemorrhage mask includes a hemorrhage location, a hemorrhage category, and a category of each pixel in the hemorrhage; the method further includes:

[0014] According to the predicted segmentation probability map, determine the prediction mask corresponding to the sample image;

[0015] Accordingly, the first loss between the predicted segmentation probability map and the segmentation gold standard is calculated, including:

[0016] According to the bleeding point categories in the prediction mask, the coincidence loss between the bleeding point positions in the prediction mask and the bleeding point positions in the intracranial bleeding point mask, and the category loss between the predicted category of each pixel in the bleeding point in the prediction mask and the category of each pixel in the bleeding point in the intracranial bleeding point mask are calculated;

[0017] The first loss is obtained by summing the coincidence loss and the category loss.

[0018] In one embodiment, after obtaining the segmentation model, the method further includes:

[0019] Input the sample image into the segmentation model to obtain the segmentation probability map;

[0020] The third loss between the segmentation probability map and the segmentation gold standard is calculated, and the segmentation model is optimized according to the third loss to obtain the optimized segmentation model.

[0021] In one embodiment, the brain CT image is input into a preset segmentation model, including:

[0022] The multi-layer tomographic images in the brain CT images are stitched together and the stitched images are input into the segmentation model.

[0023] In one embodiment, after stitching the multi-layer tomographic images in the brain CT image, the method further includes:

[0024] Setting different window widths and window positions for the stitched images to obtain a set of stitched images with different window widths and window positions;

[0025] Accordingly, the spliced ​​image is fed into the segmentation model, including:

[0026] The collection of stitched images is fed into the segmentation model.

[0027] In one embodiment, the intracranial hemorrhage segmentation result includes an intracranial hemorrhage probability map; the method further includes:

[0028] Based on the data source information of the brain CT images, the probability of the pixel points in the intracranial hemorrhage probability map is optimized to obtain an optimized intracranial hemorrhage probability map.

[0029] A brain image segmentation device, the device comprising:

[0030] An acquisition module, used for acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice images;

[0031] The segmentation module is used to input the brain CT image into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training based on the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0032] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0033] Acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice tomographic images;

[0034] The brain CT image is input into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0035] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0036] Acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice tomographic images;

[0037] The brain CT image is input into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0038] The above-mentioned brain image segmentation method, device, computer equipment and readable storage medium can obtain a brain CT image, which includes adjacent multi-layer tomographic images; the brain CT image is input into a preset segmentation model to obtain an intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image. In this method, the segmentation model is connected to the pooling layer and the fully connected layer to obtain the classification result and classification loss of the sample image, and the classification loss is used to guide the training of the segmentation model, which can greatly improve the accuracy of the trained segmentation model, thereby improving the accuracy of the brain CT image segmentation result. At the same time, the brain CT image input into the segmentation model is an adjacent multi-layer tomographic image, which can fully consider the contextual information between each layer of the image, and further improve the accuracy of the obtained brain CT image segmentation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is an internal structure diagram of a computer device in one embodiment;

[0040] Figure 2 is a schematic flow chart of a brain image segmentation method in one embodiment;

[0041] Figure 2a A schematic diagram of the network structure of a segmentation model and connected pooling layers and fully connected layers in one embodiment;

[0042] Figure 3 A schematic diagram of a flow chart of a segmentation model training process in one embodiment;

[0043] Figure 3a is a schematic diagram of a predicted segmentation probability map in one embodiment;

[0044] Figure 4 A schematic diagram of a flow chart of a segmentation model training process in another embodiment;

[0045] Figure 5 is a flow chart of a brain image segmentation method in another embodiment;

[0046] Figure 6 FIG. 4 is a structural block diagram of a brain image segmentation device in one embodiment. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] The brain image segmentation method provided in the embodiment of the present application can be applied to Figure 1 The computer device shown. The computer device includes a processor and a memory connected by a system bus, and a computer program is stored in the memory. When the processor executes the computer program, the steps of the following method embodiment can be executed. Optionally, the computer device may also include a communication interface, a display screen and an input device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory, and the non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal through a network connection. Optionally, the computer device can be a personal computer (personal computer, referred to as PC), a personal digital assistant, or other terminal devices, such as a tablet computer (portable android device, referred to as PAD), a mobile phone, etc., or a cloud or remote server. The embodiment of the present application does not limit the specific form of the computer device.

[0049] In one embodiment, Figure 2 As shown in FIG. 1 , a brain image segmentation method is provided, and the method is applied to Figure 1 The computer device in the example is used to illustrate that the present embodiment relates to a specific process in which a computer device segments an acquired brain CT image to obtain an intracranial hemorrhage segmentation result corresponding to the brain CT image. The method comprises the following steps:

[0050] S101, acquiring a brain CT image, where the brain CT image includes adjacent multi-slice images.

[0051] Among them, the CT image is a medical image obtained by electronic computer tomography, which usually includes multi-layer tomographic images. In this embodiment, the brain CT image is a medical image obtained by scanning the head of the subject by the CT device; after the CT device completes the scanning, the obtained brain CT image can be uploaded to the Picture Archiving and Communication Systems (PACS), and then the computer device can obtain the corresponding brain CT image from the PACS system.

[0052] Optionally, the computer device can obtain the brain CT images uploaded by the CT device from the PACS system in real time, or can obtain all the brain CT images uploaded by the CT device within the time period from the PACS system at a fixed time interval. Optionally, the above-mentioned brain CT images can be a tomographic image sequence obtained by scanning by the CT device, and the tomographic image sequence includes multi-layer tomographic images, and these tomographic images are adjacent to each other; or a preset number of adjacent multi-layer tomographic images are selected from the tomographic image sequence; the adjacent multi-layer tomographic images are subsequently processed, and the context information between each layer of the image can be considered to improve the accuracy of the subsequent segmentation results.

[0053] S102, inputting the brain CT image into a preset segmentation model to obtain an intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0054] Specifically, the computer device inputs the brain CT image obtained above into a preset segmentation model, and through the processing of the downsampling layer, convolution layer, and upsampling layer of the segmentation model, the intracranial hemorrhage segmentation result corresponding to the brain CT image can be obtained. Optionally, the intracranial hemorrhage segmentation result can be an intracranial hemorrhage point mask, that is, a binary image in which the intracranial hemorrhage point area and the background area are marked with different colors; it can also be an intracranial hemorrhage probability map, that is, a probability map of each pixel point in the brain CT image belonging to an intracranial hemorrhage point.

[0055] The above segmentation model can be a neural network model, such as a convolutional neural network, a recurrent neural network, etc. Optionally, the segmentation model can be a 3D-based V-Net model or a 2D-based U-Net model, and of course, it can also be other types of segmentation models. Among them, the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtypes of intracranial hemorrhage in the sample image; Optionally, the hemorrhage subtypes of intracranial hemorrhage can include 5 types of intracerebral hemorrhage, subdural hemorrhage, epidural hemorrhage, intraventricular hemorrhage and subarachnoid hemorrhage, and can also include other hemorrhage subtypes. In this embodiment, after the convolution layer of the segmentation model (used to extract the feature map of the sample image), a pooling layer and a fully connected layer are connected, the pooling layer is used to reduce the dimension of the feature map obtained by the convolution layer, and the fully connected layer is used to classify the feature map after the dimension reduction to obtain the classification result of the sample image, that is, the probability of belonging to the 5 hemorrhage subtypes in the above example (or the probability of belonging to other hemorrhage subtypes), and the classification loss between the classification result and the classification gold standard is calculated. At the same time, the above-mentioned feature map is also processed by the upsampling layer of the segmentation model to obtain the segmentation result of the sample image, and the segmentation loss between the segmentation result and the segmentation gold standard is calculated. The segmentation model is trained by comprehensively considering the classification loss and the segmentation loss to obtain a segmentation model with convergence of the training. Optionally, the above-mentioned pooling layer can be an average pooling layer, a maximum pooling layer, or other types of pooling layers, which is not limited in this embodiment.

[0056] Optionally, before the brain CT image is input into the segmentation model, the computer device may also preprocess the CT image, including but not limited to normalization so that the data distribution range is [0,1], random horizontal flipping or random angle image rotation, etc. Optionally, if the above intracranial hemorrhage segmentation result is an intracranial hemorrhage probability map, the computer device may also optimize the probability of the pixel points in the intracranial hemorrhage probability map based on the data source information of the brain CT image (such as the CT value of the pixel points in the CT image, etc.), and obtain an optimized intracranial hemorrhage probability map. Among them, a conditional random field (CRF or CRFs) structure can be cascaded after the upsampling layer of the segmentation model to receive the brain CT image (or the preprocessed brain CT image) and the intracranial hemorrhage probability map, and the intracranial hemorrhage probability map is optimized, which can solve the label mixing phenomenon that occurs in multi-category segmentation to a certain extent. The structure can have a better segmentation effect for boundary pixels, thereby correcting the segmentation category.

[0057] The network structure of the above segmentation model and the connected pooling layer and fully connected layer can be found in Figure 2aAs shown. It should be noted that, during the use of the segmentation model, only the brain CT images need to be segmented, and the classification process is no longer required, that is, the pooling layer and the fully connected layer only perform data processing during the training phase of the segmentation model to guide the training of the segmentation model; the pooling layer and the fully connected layer here are used to obtain the classification results of the sample images, and the segmentation model is trained by the calculated classification loss, that is, if the classification result error is large, it can be considered that the feature map extracted by the segmentation model has low accuracy. Therefore, under the joint training of the classification result and the segmentation result, the accuracy of the segmentation model is greatly improved, thereby improving the accuracy of the image segmentation result.

[0058] The brain image segmentation method provided in this embodiment is that the computer device inputs the acquired brain CT image into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image. The segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard. The classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image. In this method, the segmentation model connects the pooling layer and the fully connected layer to obtain the classification result and classification loss of the sample image. The classification loss is used to guide the training of the segmentation model, which can greatly improve the accuracy of the trained segmentation model, thereby improving the accuracy of the brain CT image segmentation result. At the same time, the brain CT image input into the segmentation model is an adjacent multi-layer tomographic image, which can fully consider the contextual information between each layer of the image, and further improve the accuracy of the obtained brain CT image segmentation result.

[0059] In one embodiment, the specific training process of the segmentation model is involved, optionally, as Figure 3 As shown, the above method also includes:

[0060] S201, input the sample image into the initial segmentation network to obtain a feature map and a predicted segmentation probability map corresponding to the sample image; the predicted segmentation probability map is used to characterize the probability that each pixel point in the sample image belongs to an intracranial hemorrhage point.

[0061] Specifically, the computer device can obtain a large number of sample images from the PACS system, including medical images with intracranial hemorrhage and normal medical images. It can also perform some preprocessing operations on the sample images, such as standardization to make the data distribution range [0,1], random horizontal flipping or random angle image rotation, etc., to enhance the generalization and robustness of the training network; at the same time, different window widths and window positions can be set for the sample images to obtain different window width and window position combinations so that sufficient prior information can be learned during the training process. Then the computer device inputs the sample image into the initial segmentation network (i.e., the segmentation model in the training process), and after being processed by the downsampling layer and convolution layer of the initial segmentation network, the feature map corresponding to the sample image can be obtained; after the feature map is processed by the upsampling layer, the predicted probability map corresponding to the sample image can be obtained. Among them, the predicted segmentation probability map is used to characterize the probability that each pixel in the sample image belongs to the intracranial hemorrhage point. It should be noted that, on the one hand, the probability that the pixel point here belongs to an intracranial hemorrhage point may include the probability of being an intracranial hemorrhage point and the probability of not being an intracranial hemorrhage point; on the other hand, the probability that the pixel point here belongs to an intracranial hemorrhage point may also be the probability of belonging to various hemorrhage subtypes, such as the probability of belonging to brain parenchymal hemorrhage, the probability of belonging to subdural hemorrhage, the probability of belonging to epidural hemorrhage, the probability of belonging to intraventricular hemorrhage, and the probability of belonging to subarachnoid hemorrhage, or the probability of belonging to other hemorrhage categories. For ease of understanding, assuming that a sample image includes 10 pixels, if the above-mentioned probability of belonging to an intracranial hemorrhage point is the probability of belonging to various hemorrhage subtypes, then the predicted segmentation probability map obtained is multi-channel; assuming that there are 5 types of hemorrhage subtypes, then the obtained probability map has 6 channels (including background categories), and the predicted segmentation probability map corresponding to each channel can be found in Figure 3a As shown in the figure, the number on each pixel in the figure is the probability that the point belongs to the current category; then the category represented by the channel with the largest probability is selected as the category of the pixel.

[0062] S202, inputting the feature map into the classification network connected to the initial segmentation network to obtain the predicted classification result corresponding to the sample image.

[0063] Specifically, the computer device inputs the feature map obtained above into a classification network connected to the initial segmentation network, and the classification network may include an average pooling layer and a fully connected layer to obtain a predicted classification result corresponding to the sample image. Optionally, on the one hand, the predicted classification result may be the probability that the sample image is an intracranial hemorrhage point image; on the other hand, the predicted classification result may be the probability that the sample image belongs to the above five hemorrhage subtypes. Exemplarily, assuming that the size of the feature map is C*M*N, the size of the feature map is reduced to C*1*1 by adding a pooling layer, where C represents the number of feature map channels; for the C*1*1 feature map, it is mapped to a multi-category classification target through a fully connected layer, and the classification probability of each category can be obtained by calculating sigmoid. For example, the obtained classification probability is [0.9, 0.1, 0.1, 0, 0], which means that the probability of belonging to the type of brain parenchymal hemorrhage is 0.9, the probability of belonging to the type of subdural hemorrhage is 0.1, the probability of belonging to the type of extradural hemorrhage is 0.1, the probability of belonging to the type of intraventricular hemorrhage is 0, and the probability of belonging to the type of subarachnoid hemorrhage is 0.

[0064] S203, calculating a first loss between the predicted segmentation probability map and the segmentation gold standard, and a second loss between the predicted classification result and the classification gold standard, and training the initial segmentation network according to the first loss and the second loss to obtain a segmentation model.

[0065] Specifically, the computer device can calculate the first loss between the predicted segmentation probability map and the segmentation gold standard, and the second loss between the predicted classification result and the classification gold standard, where the segmentation gold standard is the category of each pixel annotated by an experienced doctor, and the classification gold standard is the category of the annotated sample image. 1i The first loss is calculated by the relationship of , where i represents different categories and l 1i is the loss of this type, w 1i The second loss is calculated according to the relationship L2 = ∑w2jl2j, where j represents different categories and l represents the weight of the class. 2j is the loss of this type, w 2j is the weight of this class. Then the computer device adjusts the network parameters of the initial segmentation network according to the first loss and the second loss, and iterates the training until the initial segmentation network converges to obtain a segmentation model. Optionally, the computer device can adjust the network parameters of the initial segmentation network by directly summing, weighted summing, or averaging the first loss and the second loss. For example, the total loss is obtained according to the relationship L=W1L1+W2L2 to train the initial segmentation network, and W1 and W2 are the weights of the corresponding losses.

[0066] The brain image segmentation method provided in this embodiment is that the computer device inputs the sample image into the initial segmentation network, obtains the feature map and the predicted segmentation probability map corresponding to the sample image, and inputs the feature map into the classification network connected to the initial segmentation network to obtain the predicted classification result corresponding to the sample image; then the first loss between the predicted segmentation probability map and the segmentation gold standard and the second loss between the predicted classification result and the classification gold standard are calculated, and the initial segmentation network is trained according to the two losses to obtain a segmentation model. During the training process, the loss between the predicted classification result and the classification gold standard is used as a guide to guide the training of the segmentation model, and the loss between the output result of the initial segmentation network (i.e., the predicted segmentation probability map) and the segmentation gold standard is considered, thereby greatly improving the accuracy of the segmentation model obtained by training, thereby improving the accuracy of the obtained brain CT image segmentation result.

[0067] In one embodiment, the above-mentioned segmentation gold standard includes an intracranial hemorrhage point mask of the sample image, and the intracranial hemorrhage point mask includes the bleeding point location, bleeding point category and the category of each pixel in the bleeding point. Then, optionally, the training process of the above-mentioned segmentation model may also include: determining the predicted mask corresponding to the sample image based on the predicted segmentation probability map.

[0068] Specifically, for each sample image, if the predicted segmentation probability map obtained is multi-channel, then for each channel, it represents a type (one of which is the background), then the category represented by the channel with the highest probability is used as the category of the pixel. Therefore, based on the predicted segmentation probability map, the computer device can determine which pixels belong to intracranial hemorrhage points and the categories of the pixels, and use the collection area of ​​these pixels as the intracranial hemorrhage point area, and the collection area of ​​other pixels as the background area, that is, to obtain the prediction mask corresponding to the sample image. The prediction mask can also include the bleeding point location, bleeding point category and the category of each pixel in the bleeding point.

[0069] On this basis, the computer device can calculate the overlap loss (i.e., overall loss) between the bleeding point position in the prediction mask and the bleeding point position in the intracranial bleeding point mask, and the category loss (i.e., pixel-level loss) between the predicted category of each pixel in the bleeding point of the prediction mask and the category of each pixel in the bleeding point of the intracranial bleeding point mask according to the bleeding point category in the prediction mask. Exemplarily, assuming that the sample image includes bleeding points of two bleeding subtypes, for each bleeding point, it is necessary to calculate the overlap loss and category loss between the bleeding points of the same category as the bleeding point in the intracranial bleeding point mask. Optionally, the overlap loss can be calculated according to the Dice function, and the category loss can be calculated according to the cross entropy function, which is not limited in this embodiment. Then, the computer device performs a summation operation on the overlap loss and the category loss, such as direct summation, weighted summation, or average summation, to obtain the first loss. By optimizing the interaction of the overall loss and the pixel-level loss, the accuracy of the segmentation model and the accuracy of the processing results can be further improved.

[0070] In the above embodiment, the classification loss is used to guide the segmentation model to train. After the training process is completed, in order to further improve the accuracy of the segmentation model, the computer device can also train the segmentation model separately. Optionally, Figure 4 As shown, the training process of the above segmentation model may also include:

[0071] S301, input the sample image into the segmentation model to obtain a segmentation probability map.

[0072] S302, calculating a third loss between the segmentation probability map and the segmentation gold standard, optimizing the segmentation model according to the third loss, and obtaining an optimized segmentation model.

[0073] Specifically, the specific process of the computer device inputting the sample image into the segmentation model to obtain the segmentation probability map can be referred to the description of the above embodiment, which will not be repeated here. Then the computer device only calculates the third loss between the segmentation probability map and the above segmentation gold standard, and no longer goes through the processing process of the classification network. According to the third loss, the network parameters of the segmentation model are further adjusted to optimize the segmentation model to obtain the optimized segmentation model. In this embodiment, the segmentation model is optimized again by the third loss (i.e., a separate segmentation loss), which can further improve the accuracy of the segmentation model, thereby improving the accuracy of the brain CT image segmentation results obtained by subsequent processing using the segmentation model.

[0074] In one embodiment, since the above-mentioned brain CT image includes adjacent multi-layer tomographic images, if the multi-layer tomographic images are used as 3D images and input into the segmentation model for processing, since the layer spacing of the usual CT image is relatively thick, when it is used as a 3D image, the adjacent layers need to be resampled to obtain new images to reduce the layer spacing, which will destroy the source information to a certain extent and also bring uncertainty to the network processing results; at the same time, a larger video memory is required when using 3D images for training. Therefore, in this embodiment, the computer device can splice the multi-layer tomographic images in the above-mentioned brain CT image, such as first splicing them into a whole image, and then inputting the spliced ​​images into the segmentation model for segmentation processing to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image. Therefore, there is no need to resample the original CT image, which can reduce the damage to the CT image source information; at the same time, splicing the multi-layer tomographic images can also reduce the required computing video memory to a certain extent.

[0075] In one embodiment, after obtaining the above-mentioned stitched images, the computer device can also set different window widths and window positions for the stitched images to obtain a set of stitched images with different window widths and window positions, and input the set of stitched images into the above-mentioned segmentation model. Optionally, different window widths and window positions can be set to window width 80, window center 40; window width 200, window center 80; window width 380, window center 40, etc. Setting different window widths and window positions helps the segmentation model distinguish different types of bleeding in different parts and tissues.

[0076] In order to better understand the specific process of the above-mentioned brain image segmentation method, the method is introduced in the following as an overall embodiment. Figure 5 As shown, the method includes:

[0077] S401, inputting the sample image into the initial segmentation network to obtain a feature map and a predicted segmentation probability map corresponding to the sample image;

[0078] S402, determining a prediction mask corresponding to the sample image according to the prediction segmentation probability map;

[0079] S403, inputting the feature map into the classification network connected to the initial segmentation network to obtain a predicted classification result corresponding to the sample image;

[0080] S404, according to the bleeding point categories in the prediction mask, calculating the coincidence loss between the bleeding point positions in the prediction mask and the bleeding point positions in the intracranial bleeding point mask, and the category loss between the predicted category of each pixel in the bleeding point in the prediction mask and the category of each pixel in the bleeding point in the intracranial bleeding point mask; summing the coincidence loss and the category loss to obtain a first loss;

[0081] S405, calculating a second loss between the predicted classification result and the classification gold standard, and training the initial segmentation network according to the first loss and the second loss to obtain a segmentation model;

[0082] S406, inputting the sample image into the segmentation model to obtain a segmentation probability map; calculating a third loss between the segmentation probability map and the segmentation gold standard, optimizing the segmentation model according to the third loss, and obtaining an optimized segmentation model;

[0083] The above steps implement the training process of the segmentation model. The following steps introduce the use process of the segmentation model:

[0084] S407, stitching the multi-layer tomographic images in the brain CT image, and setting different window widths and window positions for the stitched images to obtain a set of stitched images with different window widths and window positions;

[0085] S408, inputting the spliced ​​image set into the optimized segmentation model to obtain an intracranial hemorrhage probability map corresponding to the brain CT image;

[0086] S409, based on the data source information of the brain CT image, the probability of the pixel points in the intracranial hemorrhage probability map is optimized to obtain an optimized intracranial hemorrhage probability map.

[0087] The implementation process of each step can refer to the description of the above embodiment. The implementation principle and technical effect are similar and will not be repeated here.

[0088] It should be understood that although Figure 2-Figure 5 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2-Figure 5 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0089] In one embodiment, Figure 6 As shown, a brain image segmentation device is provided, including: an acquisition module 11 and a segmentation module 12.

[0090] Specifically, the acquisition module 11 is used to acquire a brain CT image, which includes adjacent multi-slice tomographic images;

[0091] The segmentation module 12 is used to input the brain CT image into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training based on the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0092] The brain image segmentation device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0093] In one embodiment, the above-mentioned device also includes a training module, which is used to input the sample image into the initial segmentation network to obtain a feature map and a predicted segmentation probability map corresponding to the sample image; the predicted segmentation probability map is used to characterize the probability that each pixel point in the sample image belongs to an intracranial hemorrhage point; the feature map is input into the classification network connected to the initial segmentation network to obtain the predicted classification result corresponding to the sample image; the first loss between the predicted segmentation probability map and the segmentation gold standard, and the second loss between the predicted classification result and the classification gold standard are calculated, and the initial segmentation network is trained according to the first loss and the second loss to obtain a segmentation model.

[0094] In one embodiment, the segmentation gold standard includes an intracranial hemorrhage point mask of the sample image, and the intracranial hemorrhage point mask includes the bleeding point position, the bleeding point category and the category of each pixel in the bleeding point; the training module is also used to determine the prediction mask corresponding to the sample image according to the predicted segmentation probability map; according to the bleeding point category in the prediction mask, calculate the coincidence loss between the bleeding point position in the prediction mask and the bleeding point position in the intracranial hemorrhage point mask, and the category loss between the predicted category of each pixel in the bleeding point in the prediction mask and the category of each pixel in the bleeding point in the intracranial hemorrhage point mask; sum the coincidence loss and the category loss to obtain a first loss.

[0095] In one embodiment, the training module is further used to input the sample image into the segmentation model to obtain a segmentation probability map; calculate the third loss between the segmentation probability map and the segmentation gold standard, optimize the segmentation model according to the third loss, and obtain an optimized segmentation model.

[0096] In one embodiment, the segmentation module 12 is specifically used to splice the multi-layer tomographic images in the brain CT image and input the spliced ​​images into the segmentation model.

[0097] In one embodiment, the segmentation module 12 is further used to set different window widths and window levels for the spliced ​​images to obtain a set of spliced ​​images with different window widths and window levels; and input the spliced ​​image set into the segmentation model.

[0098] In one embodiment, the intracranial hemorrhage segmentation result includes an intracranial hemorrhage probability map; the above-mentioned device also includes an optimization module, which is used to optimize the probability of pixel points in the intracranial hemorrhage probability map based on the data source information of the brain CT image to obtain an optimized intracranial hemorrhage probability map.

[0099] For the specific definition of the brain image segmentation device, please refer to the definition of the brain image segmentation method above, which will not be repeated here. Each module in the above-mentioned brain image segmentation device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0100] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 1 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for segmenting brain images is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0101] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0102] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0103] Acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice tomographic images;

[0104] The brain CT image is input into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0105] The implementation principle and technical effects of the computer device provided in this embodiment are similar to those of the above method embodiments, and will not be repeated here.

[0106] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0107] The sample image is input into the initial segmentation network to obtain the feature map and predicted segmentation probability map corresponding to the sample image; the predicted segmentation probability map is used to characterize the probability that each pixel point in the sample image belongs to an intracranial hemorrhage point;

[0108] The feature map is input into the classification network connected to the initial segmentation network to obtain the predicted classification result corresponding to the sample image;

[0109] The first loss between the predicted segmentation probability map and the segmentation gold standard, and the second loss between the predicted classification result and the classification gold standard are calculated, and the initial segmentation network is trained according to the first loss and the second loss to obtain a segmentation model.

[0110] In one embodiment, the segmentation gold standard includes an intracranial hemorrhage mask of the sample image, and the intracranial hemorrhage mask includes a hemorrhage location, a hemorrhage category, and a label category of each pixel in the hemorrhage; when the processor executes the computer program, the following steps are also implemented:

[0111] According to the predicted segmentation probability map, determine the prediction mask corresponding to the sample image;

[0112] Accordingly, when the processor executes the computer program, the following steps are also implemented:

[0113] According to the bleeding point categories in the prediction mask, the coincidence loss between the bleeding point positions in the prediction mask and the bleeding point positions in the intracranial bleeding point mask, and the category loss between the predicted category of each pixel in the bleeding point in the prediction mask and the category of each pixel in the bleeding point in the intracranial bleeding point mask are calculated;

[0114] The first loss is obtained by summing the coincidence loss and the category loss.

[0115] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0116] Input the sample image into the segmentation model to obtain the segmentation probability map;

[0117] The third loss between the segmentation probability map and the segmentation gold standard is calculated, and the segmentation model is optimized according to the third loss to obtain the optimized segmentation model.

[0118] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0119] The multi-layer tomographic images in the brain CT images are stitched together and the stitched images are input into the segmentation model.

[0120] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0121] Setting different window widths and window positions for the stitched images to obtain a set of stitched images with different window widths and window positions;

[0122] Accordingly, when the processor executes the computer program, the following steps are also implemented:

[0123] The collection of stitched images is fed into the segmentation model.

[0124] In one embodiment, the intracranial hemorrhage segmentation result includes an intracranial hemorrhage probability map; when the processor executes the computer program, the following steps are also implemented:

[0125] Based on the data source information of the brain CT images, the probability of the pixel points in the intracranial hemorrhage probability map is optimized to obtain an optimized intracranial hemorrhage probability map.

[0126] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0127] Acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice tomographic images;

[0128] The brain CT image is input into a preset segmentation model to obtain the intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to the sample image, the segmentation gold standard and the classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image.

[0129] The computer-readable storage medium provided in this embodiment has similar implementation principles and technical effects to those of the above method embodiments, and will not be described in detail here.

[0130] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0131] The sample image is input into the initial segmentation network to obtain the feature map and predicted segmentation probability map corresponding to the sample image; the predicted segmentation probability map is used to characterize the probability that each pixel point in the sample image belongs to an intracranial hemorrhage point;

[0132] The feature map is input into the classification network connected to the initial segmentation network to obtain the predicted classification result corresponding to the sample image;

[0133] The first loss between the predicted segmentation probability map and the segmentation gold standard, and the second loss between the predicted classification result and the classification gold standard are calculated, and the initial segmentation network is trained according to the first loss and the second loss to obtain a segmentation model.

[0134] In one embodiment, the segmentation gold standard includes an intracranial hemorrhage mask of the sample image, the intracranial hemorrhage mask including the hemorrhage location, the hemorrhage category, and the labeled category of each pixel in the hemorrhage; when the computer program is executed by the processor, the following steps are also implemented:

[0135] According to the predicted segmentation probability map, determine the prediction mask corresponding to the sample image;

[0136] Accordingly, when the computer program is executed by the processor, the following steps are also implemented:

[0137] According to the bleeding point categories in the prediction mask, the coincidence loss between the bleeding point positions in the prediction mask and the bleeding point positions in the intracranial bleeding point mask, and the category loss between the predicted category of each pixel in the bleeding point in the prediction mask and the category of each pixel in the bleeding point in the intracranial bleeding point mask are calculated;

[0138] The first loss is obtained by summing the coincidence loss and the category loss.

[0139] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0140] Input the sample image into the segmentation model to obtain the segmentation probability map;

[0141] The third loss between the segmentation probability map and the segmentation gold standard is calculated, and the segmentation model is optimized according to the third loss to obtain the optimized segmentation model.

[0142] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0143] The multi-layer tomographic images in the brain CT images are stitched together and the stitched images are input into the segmentation model.

[0144] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0145] Setting different window widths and window positions for the stitched images to obtain a set of stitched images with different window widths and window positions;

[0146] Accordingly, when the computer program is executed by the processor, the following steps are also implemented:

[0147] The collection of stitched images is fed into the segmentation model.

[0148] In one embodiment, the intracranial hemorrhage segmentation result includes an intracranial hemorrhage probability map; when the computer program is executed by the processor, the following steps are further implemented:

[0149] Based on the data source information of the brain CT images, the probability of the pixel points in the intracranial hemorrhage probability map is optimized to obtain an optimized intracranial hemorrhage probability map.

[0150] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0151] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0152] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A brain image segmentation method, characterized in that: The method comprises: Acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice images; Inputting the brain CT image into a preset segmentation model to obtain an intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to a sample image, a segmentation gold standard, and a classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image; The training method of the segmentation model includes: calculating a first loss between the predicted segmentation probability map and the segmentation gold standard, and a second loss between the predicted classification result corresponding to the sample image and the classification gold standard, and training an initial segmentation network according to the first loss and the second loss to obtain the segmentation model; the first loss is obtained by summing the coincidence loss and the category loss, the coincidence loss is determined according to the position of the bleeding point, and the category loss is determined by the category of each pixel in the bleeding point.

2. The method according to claim 1, characterized in that The training method of the segmentation model also includes: Inputting the sample image into the initial segmentation network to obtain a feature map corresponding to the sample image and the predicted segmentation probability map; the predicted segmentation probability map is used to characterize the probability that each pixel point in the sample image belongs to an intracranial hemorrhage point; The feature map is input into a classification network connected to the initial segmentation network to obtain a predicted classification result corresponding to the sample image.

3. The method according to claim 2, characterized in that The segmentation gold standard includes an intracranial hemorrhage mask of the sample image, and the intracranial hemorrhage mask includes a hemorrhage location, a hemorrhage category, and a category of each pixel in the hemorrhage; the method further includes: Determining a prediction mask corresponding to the sample image according to the predicted segmentation probability map; Accordingly, the calculating a first loss between the predicted segmentation probability map and the segmentation gold standard includes: According to the bleeding point categories in the prediction mask, calculating the coincidence loss between the bleeding point positions in the prediction mask and the bleeding point positions in the intracranial bleeding point mask, and the category loss between the prediction category of each pixel in the bleeding point of the prediction mask and the category of each pixel in the bleeding point of the intracranial bleeding point mask; The overlap loss and the category loss are summed to obtain the first loss.

4. The method according to claim 2 or 3, characterized in that: After obtaining the segmentation model, the method further includes: Inputting the sample image into the segmentation model to obtain a segmentation probability map; A third loss between the segmentation probability map and the segmentation gold standard is calculated, and the segmentation model is optimized according to the third loss to obtain an optimized segmentation model.

5. The method according to claim 1, characterized in that The step of inputting the brain CT image into a preset segmentation model comprises: The multi-layer tomographic images in the brain CT image are spliced, and the spliced ​​images are input into the segmentation model.

6. The method according to claim 5, characterized in that After splicing the multi-layer tomographic images in the brain CT image, the method further includes: Setting different window widths and window positions for the stitched images to obtain a set of stitched images with different window widths and window positions; Accordingly, the step of inputting the spliced ​​image into the segmentation model includes: The set of stitched images is input into the segmentation model.

7. The method according to claim 1, characterized in that The intracranial hemorrhage segmentation result includes an intracranial hemorrhage probability map; the method further includes: Based on the data source information of the brain CT image, the probability of the pixel points in the intracranial hemorrhage probability map is optimized to obtain an optimized intracranial hemorrhage probability map.

8. A brain image segmentation device, characterized in that: The device comprises: An acquisition module, used for acquiring a brain CT image, wherein the brain CT image includes adjacent multi-slice images; A segmentation module, used for inputting the brain CT image into a preset segmentation model to obtain an intracranial hemorrhage segmentation result corresponding to the brain CT image; wherein the segmentation model is obtained by training according to a sample image, a segmentation gold standard and a classification gold standard, and the classification gold standard is used to characterize the hemorrhage subtype of intracranial hemorrhage in the sample image; A training module is used to calculate a first loss between the predicted segmentation probability map and the segmentation gold standard, and a second loss between the predicted classification result corresponding to the sample image and the classification gold standard, and train the initial segmentation network according to the first loss and the second loss to obtain the segmentation model; the first loss is obtained by summing the coincidence loss and the category loss, the coincidence loss is determined according to the position of the bleeding point, and the category loss is determined by the category of each pixel in the bleeding point.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images