A collaborative cross-segmentation system for intracerebral hemorrhage and surrounding edema based on automatically generated labels

Abstract

The present invention relates to a collaborative cross segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels. The system comprises: a preprocessing module, which preprocesses the collected 3D cranial CT images; a label image generation module, which automatically generates label images for the preprocessed 3D brain parenchyma CT images; a supervised label image generation module, which performs label fusion processing on the label images to obtain supervised label images; a training module, which uses the supervised label images to train a collaborative cross segmentation model for cerebral hemorrhage and surrounding edema based on deep learning; and a measurement module, which uses the deep learning segmentation results to perform morphological measurements and present the results. The collaborative cross segmentation model comprises two hemorrhage segmentation networks and a surrounding edema segmentation network with the same structure, the decoder of the hemorrhage segmentation network is cross-connected with the decoder of the surrounding edema segmentation network, and the decoder of the surrounding edema segmentation network is cross-connected with the decoder of the hemorrhage segmentation network.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a collaborative cross-segmentation system for cerebral hemorrhage and peripheral edema based on automatically generated labels. Background Art

[0002] Intracerebral hemorrhage refers to primary non-traumatic intracerebral hemorrhage, also known as spontaneous intracerebral hemorrhage, which is characterized by high morbidity, high disability rate and high mortality rate. Accurate positioning and volume measurement of intracranial hemorrhage (ICH) are the primary tasks for observing neurological dysfunction and craniocerebral injury, and perihematomal edema (PHE) is an important sign of secondary injury of intracranial hemorrhage. In clinical practice, computer tomography (CT) has become the preferred method for detecting intracerebral hemorrhage due to its low cost, short imaging time, good detection of bone details, and more feasible in unstable patients than magnetic resonance imaging (MRI). At present, most hospitals mainly measure clinical bleeding through manual segmentation and Tada's formula. Among them, the manual calculation of intracerebral hemorrhage volume by traditional Tada's formula is cumbersome, inefficient and has poor repeatability; in addition, the formula usually overestimates the actual intracerebral hemorrhage volume by 30%. Although manual segmentation of ICH can accurately estimate its volume, segmentation is cumbersome and operator-dependent, and the operation time limits its feasibility in emergency situations.

[0003] Therefore, there is a need for an intelligent auxiliary cranial CT image segmentation and display system that can automatically segment and measure cerebral hemorrhage and surrounding edema at the same time.

[0004] The above description of the background technology is only for facilitating an in-depth understanding of the technical solution of the present invention (such as the technical means used, the technical problems solved and the technical effects produced), and should not be regarded as an admission or any form of implication that the information constitutes prior art already known to those skilled in the art. Summary of the invention

[0005] In view of the defects existing in the prior art, the present invention proposes an intelligent auxiliary cranial CT image segmentation and display system which can automatically complete the segmentation and measurement of cerebral hemorrhage and surrounding edema at the same time. The system has strong universality and high intelligence, does not rely on manual or expert labels, and is not limited to CT cranial data from a single source or a single scanning protocol.

[0006] According to one embodiment of the present invention, a collaborative cross-segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels is provided, which includes the following modules: a preprocessing module, which preprocesses the acquired 3D cranial CT images; a label image generation module, which uses computer vision and machine learning algorithms to automatically generate cerebral hemorrhage label images and surrounding edema label images from the preprocessed 3D brain parenchyma CT images; a supervised label image generation module, which performs label fusion processing on the automatically generated cerebral hemorrhage label images and surrounding edema label images to obtain cerebral hemorrhage supervised label images and surrounding edema label images; a training module, which uses the cerebral hemorrhage supervised label images and surrounding edema supervised label images to generate a label image of the cerebral hemorrhage and the surrounding edema. The invention discloses a method for training a collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning using labeled images to simultaneously segment cerebral hemorrhage and surrounding edema; a measurement module uses the deep learning segmentation results of cerebral hemorrhage and surrounding edema to perform morphological measurement and present the results; wherein the collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning includes two hemorrhage segmentation networks and surrounding edema segmentation networks with the same structure; the hemorrhage segmentation network and the surrounding edema segmentation network include an encoder and a decoder respectively; the decoder of the hemorrhage segmentation network is cross-connected with the decoder of the surrounding edema segmentation network, and the decoder of the surrounding edema segmentation network is cross-connected with the decoder of the hemorrhage segmentation network.

[0007] Preferably, the encoder of the hemorrhage segmentation network and the encoder of the surrounding edema segmentation network can share or not share weights; the hemorrhage segmentation network and the surrounding edema segmentation network have the same input, and output the deep learning segmentation results of cerebral hemorrhage and the deep learning segmentation results of surrounding edema respectively; the hemorrhage segmentation network and the surrounding edema segmentation network respectively use the encoding-decoding 3D residual U-Net framework as the backbone network; the number of down-sampling layers in the encoders of the hemorrhage segmentation network and the surrounding edema segmentation network is equal to the number of up-sampling layers in the decoders of the hemorrhage segmentation network and the surrounding edema segmentation network; each layer or part of the layers of the decoder of the hemorrhage segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the surrounding edema segmentation network; each layer or part of the layers of the decoder of the surrounding edema segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the hemorrhage segmentation network.

[0008] Preferably, the deep learning-based collaborative cross-segmentation model of cerebral hemorrhage and peripheral edema further includes a channel attention module and a spatial attention module; the channel attention module is arranged at the cross connection between the decoder of the hemorrhage segmentation network and the decoder of the peripheral edema segmentation network and between the decoder of the peripheral edema segmentation network and the decoder of the hemorrhage segmentation network; the spatial attention module is arranged between the encoder and the decoder within the same segmentation network.

[0009] Preferably, in the preprocessing module: preprocessing the acquired 3D cranial CT images includes: data cleaning, format conversion, renaming, steering, zero voxel filling, registration, and window width and window position adjustment of the acquired 3D cranial CT images, and finally extracting the brain parenchyma image.

[0010] Preferably, in the label image generation module, a threshold segmentation method and a cluster segmentation method are used to generate a cerebral hemorrhage label image respectively; generating a cerebral hemorrhage label image based on the threshold segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform global threshold segmentation on the smoothed image; generating a cerebral hemorrhage label image based on the cluster segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform fuzzy cluster segmentation on the smoothed image.

[0011] Preferably, in the label image generation module, the threshold segmentation method, the cluster segmentation method and the contralateral difference method are used to generate the peripheral edema label image respectively; the peripheral edema label image is generated based on the threshold segmentation method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then performing single threshold segmentation on the initial edema area; the peripheral edema label image is generated based on the cluster segmentation method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then performing fuzzy cluster segmentation on the initial edema area; the peripheral edema label image is generated based on the contralateral difference method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then reversing it based on the midline to extract the corresponding area of ​​the contralateral brain, and finally subtracting the two areas, and retaining the area with the difference within a certain range as the target edema area.

[0012] Preferably, the label fusion processing of the automatically generated cerebral hemorrhage label image and the surrounding edema label image includes: amplifying the automatically generated cerebral hemorrhage label image by selecting a certain expansion radius, and then performing label fusion processing and binarization processing on the automatically generated cerebral hemorrhage label image and the expanded cerebral hemorrhage label image; performing hole closing processing on the automatically generated peripheral edema label image by selecting a certain radius, and then performing label fusion processing and binarization processing on the automatically generated peripheral edema label image and the corresponding closed peripheral edema label image.

[0013] According to one embodiment of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored, and when the computer instructions are executed by a processor, the following steps are implemented: preprocessing the acquired 3D cranial CT images; automatically generating a brain hemorrhage label image and a surrounding edema label image from the preprocessed 3D brain parenchyma CT images using computer vision and machine learning algorithms; performing label fusion processing on the automatically generated brain hemorrhage label image and surrounding edema label image to obtain a brain hemorrhage and surrounding edema supervision label image; using the brain hemorrhage supervision label image and the surrounding edema supervision label image to train a deep learning-based collaborative cross-segmentation model of brain hemorrhage and surrounding edema to simultaneously segment the brain hemorrhage and surrounding edema; using the deep learning segmentation results of brain hemorrhage and surrounding edema, morphological measurements are performed and the results are presented; wherein the deep learning-based collaborative cross-segmentation model of brain hemorrhage and surrounding edema includes two hemorrhage segmentation networks and surrounding edema segmentation networks with the same structure; the hemorrhage segmentation network and the surrounding edema segmentation network respectively include an encoder and a decoder; the decoder of the hemorrhage segmentation network is cross-connected with the decoder of the surrounding edema segmentation network, and the decoder of the surrounding edema segmentation network is cross-connected with the decoder of the hemorrhage segmentation network.

[0014] Preferably, the encoder of the hemorrhage segmentation network and the encoder of the surrounding edema segmentation network can share or not share weights; the hemorrhage segmentation network and the surrounding edema segmentation network have the same input, and output the deep learning segmentation results of cerebral hemorrhage and the deep learning segmentation results of surrounding edema respectively; the hemorrhage segmentation network and the surrounding edema segmentation network respectively use the encoding-decoding 3D residual U-Net framework as the backbone network; the number of down-sampling layers in the encoders of the hemorrhage segmentation network and the surrounding edema segmentation network is equal to the number of up-sampling layers in the decoders of the hemorrhage segmentation network and the surrounding edema segmentation network; each layer or part of the layers of the decoder of the hemorrhage segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the surrounding edema segmentation network; each layer or part of the layers of the decoder of the surrounding edema segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the hemorrhage segmentation network.

[0015] Preferably, the deep learning-based collaborative cross-segmentation model of cerebral hemorrhage and peripheral edema further includes a channel attention module and a spatial attention module; the channel attention module is arranged at the cross connection between the decoder of the hemorrhage segmentation network and the decoder of the peripheral edema segmentation network and between the decoder of the peripheral edema segmentation network and the decoder of the hemorrhage segmentation network; the spatial attention module is arranged between the encoder and decoder of the same network.

[0016] Preferably, preprocessing the acquired 3D cranial CT images includes: data cleaning, format conversion, renaming, steering, zero voxel filling, registration, and window width and window position adjustment of the acquired 3D cranial CT images, and finally extracting brain parenchyma images.

[0017] Preferably, a threshold segmentation method and a clustering segmentation method are used to generate a cerebral hemorrhage label image respectively; generating a cerebral hemorrhage label image based on the threshold segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform global threshold segmentation on the smoothed image; generating a cerebral hemorrhage label image based on the clustering segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform fuzzy clustering segmentation on the smoothed image.

[0018] Preferably, the threshold segmentation method, the clustering segmentation method and the contralateral difference method are used to generate the peripheral edema label image respectively; the peripheral edema label image is generated based on the threshold segmentation method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then performing single threshold segmentation on the initial edema area; the peripheral edema label image is generated based on the clustering segmentation method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then performing fuzzy clustering segmentation on the initial edema area; the peripheral edema label image is generated based on the contralateral difference method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then reversing it based on the midline to extract the corresponding area of ​​the contralateral brain, and finally subtracting the two areas, and retaining the area with the difference within a certain range as the target edema area.

[0019] Preferably, the label fusion processing of the automatically generated cerebral hemorrhage label image and the surrounding edema label image includes: amplifying the automatically generated cerebral hemorrhage label image by selecting a certain expansion radius, and then performing label fusion processing and binarization processing on the automatically generated cerebral hemorrhage label image and the expanded cerebral hemorrhage label image; performing hole closing processing on the automatically generated peripheral edema label image by selecting a certain radius, and then performing label fusion processing and binarization processing on the automatically generated peripheral edema label image and the corresponding closed peripheral edema label image.

[0020] The present invention adopts the above technical solution, which has the following beneficial effects:

[0021] The present invention combines computer vision and machine learning algorithms to automatically generate cerebral hemorrhage supervision label images and surrounding edema supervision label images, and trains a collaborative cross-segmentation model based on the cerebral hemorrhage supervision label images and the surrounding edema supervision label images, thereby obtaining accurate segmentation results of cerebral hemorrhage and surrounding edema, and performs morphological measurements based on the segmentation results, and performs three-dimensional skull, cerebral hemorrhage, and surrounding edema superposition (perspective) display. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The following will describe the exemplary embodiments of the present invention in more detail with reference to the accompanying drawings. For the sake of clarity, the same components in different drawings are indicated by the same reference numerals. It should be noted that the drawings are only for illustration and are not necessarily drawn to scale. In these drawings:

[0023] Figure 1 is a schematic diagram of the working principle of a collaborative cross-segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to an embodiment of the present invention;

[0024] Figure 2 is a flow chart of a process of preprocessing raw data by a collaborative cross segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to an embodiment of the present invention;

[0025] Figure 3(a) to Figure 3(h) is a schematic diagram of preprocessing results of a collaborative cross segmentation system for cerebral hemorrhage and peripheral edema based on automatically generated labels according to an embodiment of the present invention, taking two data as examples;

[0026] 4(a), 4(b) and 4(c) are schematic diagrams of a collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning according to an embodiment of the present invention;

[0027] Figure 5(a) to Figure 5(e) , Figure 6(a) to Figure 6(e) 7(a) and 7(b) are schematic diagrams of segmentation results of two data sets based on the collaborative cross segmentation system of cerebral hemorrhage and peripheral edema based on automatically generated labels according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The implementation scheme of the present invention is described in detail below. This implementation scheme is implemented on the premise of the technical scheme of the present invention, and a detailed implementation method and specific operation process are given, but the protection scope of the present invention is not limited to the implementation scheme described below.

[0029] Figure 1Schematic diagram of the working principle of the collaborative cross segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to an embodiment of the present invention. The working principle of the collaborative cross segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to the present invention is as follows, which is mainly divided into five modules.

[0030] The preprocessing module preprocesses the acquired 3D cranial CT images.

[0031] The label image generation module uses computer vision and machine learning algorithms to automatically generate cerebral hemorrhage label images and peripheral edema label images from the preprocessed 3D brain parenchyma CT images.

[0032] The supervised label image generation module performs label fusion processing on the automatically generated cerebral hemorrhage label image and surrounding edema label image to obtain the cerebral hemorrhage supervised label image and the surrounding edema label image.

[0033] The training module uses the supervised label images of cerebral hemorrhage and the supervised label images of surrounding edema to train a collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning, and then simultaneously segments cerebral hemorrhage and surrounding edema.

[0034] The measurement module uses the deep learning segmentation results of cerebral hemorrhage and surrounding edema to perform morphological measurements (e.g., volume, surface area, average width, thickness, specific surface area, etc.) to present the results.

[0035] The processing of each module of the collaborative cross-segmentation system for cerebral hemorrhage and peripheral edema based on automatically generated labels of the present invention is described in detail below.

[0036] Figure 2 The figure shows the process of preprocessing the original 3D cranial CT image by the collaborative cross segmentation system of cerebral hemorrhage and surrounding edema based on automatically generated labels according to the present invention.

[0037] like Figure 2 As shown, the collaborative cross-segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to the present invention can be applied to 3D cranial CT images, and its data format can include but is not limited to DICOM (Digital Imaging and Communications in Medicine) format and NIFTI (Neuroimaging Informatics Technology Initiative) format, and does not distinguish between scanning settings such as image resolution, layer thickness, peak voltage (Kilovolt Peak, KVP) (120, 130, 140KVP), that is, there is no restriction on the scanning protocol.

[0038] The preprocessing of the acquired 3D cranial CT images includes: data cleaning, format conversion, renaming, steering, zero voxel filling, registration, and window width and window position adjustment of the acquired 3D cranial CT images, and finally extracting the brain parenchyma image. Among them, steering and window width and window position adjustment are optional, and these two processing methods can be selected according to needs.

[0039] Among them, data cleaning is to exclude data that have undergone early treatment (for example, drainage) and have serious quality problems (for example, artifacts, virtual images), and the rest of the data can be used. Then, all the usable data are sequentially subjected to the following steps: format conversion, renaming, steering, zero voxel filling, registration, and window width and window position adjustment.

[0040] According to an embodiment of the present invention, if the available data is in DICOM format, it is converted into NIFTI format; if the available data is in NIFTI format, no format conversion is performed. The image size is unified by zero voxel filling, and the image size with a layer thickness less than 4 mm is uniformly changed to 512×512×150 voxels; the image size with a layer thickness greater than or equal to 4 mm is uniformly changed to 512×512×40 voxels. The registration of medical images is the process of converting different images of the same scene into the same coordinate system. This article uses the pre-processed selected image (i.e., the image in which the brain midline in the grayscale image coincides with the image geometric midline or the cross-sectional midline) as a reference, and the remaining images are used as moving images to complete the registration process so that the brain anatomical structure of each data corresponds to each other in the same coordinate space.

[0041] Figure 3(a) to Figure 3(h) The following is a schematic diagram of the preprocessing results of two data sets. Figure 3(a) and Figure 3(e) show two original 3D cranial CT images, Figure 3(b) and Figure 3(f) show the images obtained after registration, Figure 3(c) and Figure 3(g) show the images obtained after adjusting the window width and window position, and Figure 3(d) and Figure 3(h) show the final brain parenchyma images.

[0042] Hereinafter, the automatic generation of a label image using a preprocessed 3D brain parenchymal CT image will be described.

[0043] After preprocessing the original 3D cranial CT image (i.e., obtaining the brain parenchyma image), computer vision and machine learning algorithms are used to automatically generate a cerebral hemorrhage label image and a peripheral hematoma label image from the preprocessed 3D brain parenchyma CT image.

[0044] According to the image characteristics of cerebral hemorrhage, since the high-density area of ​​cerebral hemorrhage (i.e., blood coagulation) can contrast with the adjacent tissue, methods such as threshold segmentation method and clustering segmentation method can be used to locate and segment the boundaries of cerebral hemorrhage. The following uses the threshold segmentation method and clustering segmentation method as examples to describe the location and boundary segmentation of cerebral hemorrhage.

[0045] The generation of the cerebral hemorrhage label image based on the threshold segmentation method is to perform single threshold segmentation based on the CT value range of general cerebral hemorrhage, thereby dividing the CT image into two categories: cerebral hemorrhage and non-cerebral hemorrhage areas. In the embodiment of the present invention, the pre-processed 3D brain parenchyma CT image can be directly subjected to single threshold segmentation; or the pre-processed 3D brain parenchyma CT image can be first subjected to Gaussian smoothing, where the maximum range value of the smoothing coefficient sigma is (0, 5], preferably [1, 4], and is selected as 1.5 in this embodiment; then the smoothed image is subjected to global threshold segmentation, where the fixed range of the threshold is [40, 100] CT values.

[0046] The generation of a brain hemorrhage label image based on a clustering segmentation method is to generate multiple classifications based on a preprocessed 3D brain parenchyma CT image, and to continuously correct the cluster center by repeatedly calculating the degree of membership of each voxel point in the image, so that the objective function reaches the optimal solution. Among them, the last category is brain hemorrhage. In an embodiment of the present invention, the preprocessed 3D brain parenchyma CT image is first subjected to Gaussian smoothing, and the maximum range value of the smoothing coefficient is (0, 5], preferably [1, 3], and is selected as 1.5 in this embodiment; the smoothed image is subjected to fuzzy clustering segmentation, where the maximum range of the number of clusters is [3, 10], preferably [3, 8], and is selected as 5 in this embodiment.

[0047] The implementation scheme of the present invention utilizes a threshold segmentation method and a cluster segmentation method that are closely related to the difference in 3D cranial CT values ​​to locate and segment the boundaries of cerebral hemorrhage, but the present invention is not limited to the above threshold segmentation method and cluster segmentation method.

[0048] By using the above-mentioned computer vision and machine learning algorithms, such as the threshold segmentation method and the clustering segmentation method, the cerebral hemorrhage image can be segmented. Different from the characteristics of the cerebral hemorrhage image, there is a large amount of overlap between the peripheral edema image and the adjacent cerebrospinal fluid and white matter grayscale image ranges, resulting in unclear boundaries of the peripheral edema image. Therefore, it is difficult to segment the peripheral edema area. The embodiment of the present invention transforms the principles and experience of judging peripheral edema into the peripheral edema segmentation algorithm, which increases the interpretability and reliability of the results. The peripheral edema segmentation algorithm may include but is not limited to threshold segmentation, clustering segmentation, and contralateral difference algorithm.

[0049] The following describes the positioning and boundary segmentation of peripheral edema by taking the threshold segmentation method, clustering segmentation method and contralateral difference method as examples. The embodiment of the present invention positions and segment the boundary of peripheral edema based on the cerebral hemorrhage label image of the same 3D cranial CT image.

[0050] The method of generating a peripheral edema label image based on the threshold segmentation method is to expand the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and determine the initial edema area by dot multiplication with the preprocessed 3D brain parenchyma CT image, and then perform single threshold segmentation on the initial edema area according to the threshold range of edema around the general blood clot, and divide it into two categories: edema and non-edema areas. In the present invention, the maximum range value of the radius of the outward expansion of cerebral hemorrhage is (0, 50] voxels, preferably [10, 30] voxels, and 20 voxels are selected in this embodiment; the initial edema area after dot multiplication is threshold segmented, and here, the fixed range of the threshold is [5, 33] CT value.

[0051] The method of generating a peripheral edema label image based on the cluster segmentation method is to expand the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and determine the initial area of ​​edema by dot multiplication with the preprocessed 3D brain parenchyma CT image, and then generate multiple classifications according to the initial area of ​​edema, and continuously correct the cluster center by repeatedly calculating the membership degree of each voxel point in the image, so that the objective function reaches the optimal solution. Among them, the last category is peripheral edema. In the present invention, the maximum range value of the radius of outward expansion of cerebral hemorrhage is (0, 50] voxels, preferably [10, 30] voxels, and is selected as 20 in this embodiment; the initial area of ​​edema is fuzzy clustered, and the maximum range of the number of clusters is [3, 10], preferably [6, 10], and is selected as 8 in this embodiment.

[0052] The contralateral difference method generates a peripheral edema label image by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then reversing it based on the midline to extract the corresponding area of ​​the contralateral brain, and finally subtracting the two areas, and the area with the difference within the specified range is the target edema area. In an embodiment of the present invention, the maximum range of the radius of the outward expansion of cerebral hemorrhage is (0, 50] voxels, and the preferred range is [10, 30] voxels, which is selected as 20 voxels in this embodiment; all voxels with a difference greater than 0 between the two areas are retained.

[0053] The implementation scheme of the present invention utilizes a threshold segmentation method, a clustering segmentation method, and a contralateral difference method to locate and segment the boundaries of peripheral edema according to the image features of peripheral edema, but the present invention is not limited to the above-mentioned threshold segmentation method, clustering segmentation method, and contralateral difference method.

[0054] In the generation of the above-mentioned cerebral hemorrhage and surrounding edema label images, computer vision and machine learning algorithms (including cluster segmentation methods, threshold segmentation methods, and contralateral difference methods, etc.) can be used to quickly generate automatic labels, eliminating the reliance on manual annotation. In order to obtain a label image close to the "gold standard" from a large number of cerebral hemorrhage and surrounding edema label images obtained by different algorithms, the present invention performs label fusion processing on the automatically generated cerebral hemorrhage label images and surrounding edema label images to obtain cerebral hemorrhage supervision label images and peripheral edema supervision label images. In the following, the process of performing label fusion processing on the automatically generated label images to obtain supervision label images will be described.

[0055] There are certain differences between the cerebral hemorrhage label images generated by computer vision and machine learning algorithms (including clustering segmentation method and threshold segmentation method). In order to extract label images close to the "gold standard" from a large number of cerebral hemorrhage segmentation result images obtained by different algorithms, firstly, a certain expansion radius is selected in the cerebral hemorrhage label images of different algorithms to amplify the pixels in the cerebral hemorrhage label images, so as to obtain the corresponding expanded cerebral hemorrhage label images. Here, the maximum range value of the expansion coefficient is (0, 25] voxels, preferably [2, 20] voxels, and in this embodiment, [3, 15] voxels are selected to ensure the improvement of the true positive (TP) result and facilitate the subsequent extraction of the supervised label image. Then, for the cerebral hemorrhage label images generated by different algorithms and the corresponding cerebral hemorrhage label images after expansion, the cerebral hemorrhage label images generated by different algorithms and the corresponding cerebral hemorrhage label images after expansion are synthesized into a probability map by a label fusion or label averaging method, and then binarized by selecting a certain threshold. The maximum range value of the threshold is (0, 1), preferably [0.6, 0.9], and in this embodiment, 0.8 is selected, and finally the supervised label image of the cerebral hemorrhage corresponding to each data is obtained for subsequent network training. The label fusion method may include, for example, Simultaneous Truth and Performance Level Estimation (STAPLE), Majority Voting algorithm, etc.

[0056] The peripheral edema label image generated by the computer vision and machine learning algorithm (including cluster segmentation method, threshold segmentation method and contralateral difference segmentation method) shows that the segmented edema area has certain holes and is discontinuous in some edge areas. In order to make the label image generated by the automatic segmentation of peripheral edema closer to the actual situation, the hole closure processing of the peripheral edema label image is performed by selecting a certain radius, and the maximum range value of the closure radius coefficient is (0, 15] voxels, preferably (0, 10] voxels, and in this embodiment, 5 voxels are selected to obtain the corresponding closed peripheral edema label image.

[0057] For the peripheral edema label images generated by different algorithms and the corresponding closed peripheral edema label images, a large number of non-manual peripheral edema label images generated by automatic segmentation are feature extracted and probability maps are generated through label fusion or label averaging methods, and then binarized by selecting a suitable threshold. The maximum range value of the threshold is (0, 1), preferably [0.4, 0.7]. In this embodiment, 0.6 is selected to finally obtain the peripheral edema supervision label image corresponding to each data set for subsequent network training.

[0058] In the following, the automatically generated supervised label images of cerebral hemorrhage and surrounding edema will be used to train a deep learning-based collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema, and then perform (semantic) segmentation of cerebral hemorrhage and surrounding edema simultaneously.

[0059] Figure 4(a) to Figure 4(c) A schematic diagram of the deep learning-based collaborative cross-segmentation model for cerebral hemorrhage and surrounding edema is shown.

[0060] Cerebral hemorrhage and surrounding edema are closely related in position and morphology, and there is a fusion transition edge between the two in CT images, and there is certain shared semantic information, so the segmentation of cerebral hemorrhage and surrounding edema can be considered as a set of paired tasks. In order to better utilize the shared information between the two and realize the automatic segmentation of cerebral hemorrhage and surrounding edema at the same time, the present invention proposes a collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on the principle of Siamese networks, and establishes a coupling architecture for two neural networks with the same structure.

[0061] In an embodiment of the present invention, a collaborative cross segmentation model based on a 3D residual U-Net decoding-encoding structure is constructed. Referring to FIG4(a), the collaborative cross segmentation model may include a hemorrhage segmentation network and a surrounding edema segmentation network, each neural network uses an encoding-decoding 3D residual U-Net framework as a backbone network, and is provided with a spatial attention and a channel attention module for transmitting information within or between networks.

[0062] The encoder parts of the two neural networks can share the same weights or be independent of each other and not share weights. The two neural networks input the same labeled image and output the deep learning segmentation results of cerebral hemorrhage and the deep learning segmentation results of peripheral edema respectively.

[0063] As shown in FIG4(a), the hemorrhage segmentation network and the surrounding edema segmentation network include an encoder and a decoder, respectively. The number of down-sampling layers in the encoder of the hemorrhage segmentation network and the surrounding edema segmentation network is equal to the number of up-sampling layers in the decoder of the hemorrhage segmentation network and the surrounding edema segmentation network. For example, the encoder of the hemorrhage segmentation network may include the first layer to the nth layer, and the decoder may include the first layer to the n-1th layer; the encoder of the surrounding edema segmentation network may include the first layer to the nth layer, and the decoder may include the first layer to the n-1th layer, where n is an integer, and n≥5.

[0064] Except for the bottom layer (nth layer) of the encoder, each layer of the encoder is followed by a downsampling module, as shown by the arrow in Figure 4(a), which downsamples the feature map generated by the encoder and reduces the size of the feature map. Except for the bottom layer of the encoder, the output of each layer of the encoder is simultaneously passed to the downsampling module and the spatial attention module of the corresponding level. Before each layer of the decoder, an upsampling module is set to upsample the feature map and increase the size of the feature map. The output of the upsampling module is used as the input of the channel attention module, and the output of the upsampling module is spliced ​​with the output of the spatial attention module on the channel. After the last upsampling, a feature map of the same size as the input image is obtained. The output results of the last few layers of the decoder will be superimposed, and the probability map of the final hemorrhage or edema label (probability between 0-1) is output through a layer of convolution and Sigmoid activation function, and the final segmentation result is obtained by binarization with a threshold of 0.5.

[0065] Each layer of the encoder of the hemorrhage segmentation network except the bottom layer is connected to the corresponding layer of its decoder through the spatial attention module, and each layer of the encoder of the peripheral edema segmentation network except the bottom layer is also connected to the corresponding layer of its decoder through the spatial attention module. For example, as shown in Figure 4(a), the first layer of the encoder of the hemorrhage segmentation network can be connected to the first layer of its decoder, the second layer of the encoder of the hemorrhage segmentation network can be connected to the second layer of its decoder, ..., the n-1th layer of the encoder of the hemorrhage segmentation network can be connected to the n-1th layer of its decoder. Similarly, the first layer of the encoder of the peripheral edema segmentation network can be connected to the first layer of its decoder, the second layer of the encoder of the peripheral edema segmentation network can be connected to the second layer of its decoder, ..., the n-1th layer of the encoder of the peripheral edema segmentation network can be connected to the n-1th layer of its decoder.

[0066] Specifically, as shown in Figure 4(b), the output of a certain layer in the encoder passes through the spatial attention module to obtain a spatially weighted feature map, which is then spliced ​​with the output of the corresponding upsampling module in the decoder of this network on the channel and input to the decoder of this network. In the spatial attention module, a weight is calculated for each point of the input feature map, and then the weight is multiplied by the input feature map, thereby spatially suppressing information in irrelevant areas and focusing mainly on areas related to hemorrhage or edema segmentation. The spatial attention module can be, but is not limited to, an attention mechanism module based on a convolutional module, an attention gate, or a Transformer module.

[0067] Further, except for the top layer (first layer), each layer or part of the layers of the decoder of the hemorrhage segmentation network can be cross-connected with the corresponding layer of the decoder of the surrounding edema segmentation network through the channel attention module. For example, as shown in Figure 4 (a), the input of the second layer of the decoder of the hemorrhage segmentation network can be cross-connected with the output of the second layer of the decoder of the surrounding edema segmentation network, and the input of the second layer of the decoder of the surrounding edema segmentation network can be cross-connected with the output of the second layer of the decoder of the hemorrhage segmentation network; the input of the third layer of the decoder of the hemorrhage segmentation network can be cross-connected with the output of the third layer of the decoder of the surrounding edema segmentation network, and the input of the third layer of the decoder of the surrounding edema segmentation network can be cross-connected with the output of the third layer of the decoder of the hemorrhage segmentation network.

[0068] Specifically, as shown in Figure 4(c), the output of the upsampling module of a certain layer in the decoder passes through the channel attention module to generate a weight coefficient for each channel of the output of the corresponding layer of the other network decoder, and then multiply it with the feature map of each channel to achieve weighting of different channels, highlight the importance of the channel related to the segmentation task of another network, and suppress the information transmission of irrelevant channels. The channel attention module can be, but is not limited to, an attention mechanism module based on a convolutional module, an attention gate, or a Transformer module.

[0069] In addition, the collaborative cross-segmentation model uses the Dice Similarity coefficient loss function, the initial convolution kernel weights are obtained from the Gaussian distribution, and then the model is optimized using the stochastic gradient descent optimizer.

[0070] During the experiment, different numbers of crossover layers and positions were set to obtain the best experimental results. The following Table 1 shows the comparative experimental results of different crossover modes.

[0071] [Table 1]

[0072]

[0073] Among them, the top layer and bottom layer in Table 1 refer to the topmost or bottommost layer in the decoder with crossover. DSC (Dice Similarity Coefficient) indicates the degree of spatial overlap between the segmented cerebral hemorrhage and surrounding edema, 0 indicates no overlap, and 1 indicates complete pixel-by-pixel consistency. Recall indicates the proportion of true positive samples to all samples, that is, the proportion of positive samples recalled, and its value is between 0 and 1. Precision indicates the proportion of true positive samples in the segmented image output by the current network to the network segmentation result map, and its value is between 0 and 1.

[0074] It can be seen from the experimental results that when different cross patterns are selected in the collaborative cross segmentation model, the pattern without crossover of the bottom two layers can achieve better results in the segmentation of overall cerebral hemorrhage and surrounding edema.

[0075] Figure 5(a) to Figure 5(e) , Figure 6(a) to Figure 6(e) 7(a) and 7(b) are schematic diagrams of segmentation results of two data sets based on the collaborative cross segmentation system of cerebral hemorrhage and peripheral edema based on automatically generated labels according to an embodiment of the present invention.

[0076] The collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning can output the segmentation results of cerebral hemorrhage and surrounding edema, so as to intuitively display the location of cerebral hemorrhage and surrounding edema, thereby performing a three-dimensional overlay (perspective) display of the skull, cerebral hemorrhage, and surrounding edema.

[0077] Among them, Figure 5(a) shows an original two-dimensional CT image of cerebral hemorrhage; Figure 5(b) and Figure 5(c) respectively show the two-dimensional network segmentation result and manual segmentation result of cerebral hemorrhage; Figure 5(d) and Figure 5(e) respectively show the two-dimensional network segmentation result and manual segmentation result of surrounding edema. Figure 6(a) shows another original two-dimensional CT image of cerebral hemorrhage; Figure 6(b) and Figure 6(c) respectively show the two-dimensional network segmentation result and manual segmentation result of cerebral hemorrhage; Figure 6(d) and Figure 6(e) respectively show the two-dimensional network segmentation result and manual segmentation result of surrounding edema. Figure 7(a) and Figure 7(b) respectively show two examples of three-dimensional network segmentation results of cerebral hemorrhage and surrounding edema.

[0078] After the segmentation results are obtained, morphological measurements can also be performed, such as volume, surface area, average width, thickness, specific surface area, etc. The volume is calculated by multiplying the number of voxels with a label value of 1 by the volume of a single voxel; the surface area is calculated by reconstructing the surface of the segmentation result according to the Marching cube face drawing algorithm to obtain a three-dimensional model surface composed of several triangular facets. The area of ​​each triangular facet can be calculated according to the Heron formula. The areas of all triangular facets can be summed to obtain the total surface area. The remaining index values ​​are calculated based on volume and surface area.

[0079] Therefore, the collaborative cross segmentation system of cerebral hemorrhage and surrounding edema based on automatically generated labels of the present invention combines computer vision and machine learning algorithms to automatically generate cerebral hemorrhage supervision label images and surrounding edema supervision label images, and trains the collaborative cross segmentation model based on cerebral hemorrhage supervision label images and surrounding edema supervision label images, thereby obtaining accurate cerebral hemorrhage and surrounding edema segmentation results, and performing three-dimensional skull, cerebral hemorrhage, surrounding edema superposition (perspective) display. The system has strong universality and high intelligence, is independent of artificial or expert labels, and is not limited to CT cranial brain data from a single source or a single scanning protocol.

[0080] Although for the sake of clarity, the exemplary method of the present invention described above is represented as a series of operations, it is not intended to limit the order of execution of the steps, and each step can be performed simultaneously or in a desired different order. In order to implement the method according to the present invention, the steps shown may further include other steps, may include the remaining steps except some steps, or may include other additional steps except some steps.

[0081] The various embodiments of the invention are not an exhaustive list of all possible combinations, but are intended to describe representative aspects of the invention, and what is described in various embodiments may be applied independently or in combinations of two or more.

[0082] In addition, various embodiments of the present invention may be implemented by hardware, firmware, software or a combination thereof. The hardware may be implemented by one or more of a graphics processor (GPU) or other application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a general purpose processor, a controller, a microcontroller, a microprocessor, etc.

[0083] The scope of the present invention is intended to include software or machine-executable instructions (e.g., operating systems, applications, firmware, programs, etc.) that enable operations according to various embodiments to be performed on a device or computer, as well as non-volatile computer-readable media executable on a device or computer having such software or instructions, etc. stored thereon.

[0084] The description presented in the above exemplary embodiments is only used to illustrate the technical solution of the present invention, and is not intended to be exhaustive, nor is it intended to limit the present invention to the precise form described. Obviously, it is possible for a person of ordinary skill in the art to make many changes and variations based on the above teachings. The exemplary embodiments are selected and described to explain the specific principles of the present invention and its practical application, so that other technicians in the field can easily understand, implement and use the various exemplary embodiments of the present invention and its various selected forms and modified forms. The scope of protection of the present invention is intended to be limited by the attached claims and their equivalent forms.

Claims

1. A collaborative cross-segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels, characterized by Includes the following modules: A preprocessing module, which preprocesses the acquired 3D cranial CT images; A label image generation module, which uses computer vision and machine learning algorithms to automatically generate cerebral hemorrhage label images and peripheral edema label images from preprocessed 3D brain parenchyma CT images; A supervised label image generation module performs label fusion processing on the automatically generated cerebral hemorrhage label image and the surrounding edema label image to obtain a cerebral hemorrhage supervised label image and a surrounding edema supervised label image; A training module, which uses the supervised label image of cerebral hemorrhage and the supervised label image of surrounding edema to train a collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning, so as to simultaneously segment cerebral hemorrhage and surrounding edema; The measurement module uses the deep learning segmentation results of cerebral hemorrhage and surrounding edema to perform morphological measurements and present the results; Among them, the collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning includes two hemorrhage segmentation networks and surrounding edema segmentation networks with the same structure; The hemorrhage segmentation network and the surrounding edema segmentation network have the same input and output the deep learning segmentation results of cerebral hemorrhage and the deep learning segmentation results of surrounding edema respectively; The hemorrhage segmentation network and the surrounding edema segmentation network use the encoder-decoder 3D residual U-Net framework as the backbone network respectively; The number of down-sampling layers in the encoder of the hemorrhage segmentation network and the surrounding edema segmentation network is equal to the number of up-sampling layers in the decoder of the hemorrhage segmentation network and the surrounding edema segmentation network; Each layer of the decoder of the hemorrhage segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the surrounding edema segmentation network; Each layer of the decoder of the peripheral edema segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the hemorrhage segmentation network; The deep learning-based collaborative cross-segmentation model for ICH and peripheral edema further includes a channel attention module and a spatial attention module; The channel attention module is arranged between the decoder of the hemorrhage segmentation network and the decoder of the surrounding edema segmentation network; The spatial attention module is arranged between the encoder and the decoder in the same segmentation network; The label fusion processing of the automatically generated cerebral hemorrhage label image and the surrounding edema label image includes: The automatically generated cerebral hemorrhage label image is enlarged by selecting a certain expansion radius, and then the automatically generated cerebral hemorrhage label image and the expanded cerebral hemorrhage label image are subjected to label fusion processing and binarization processing; The automatically generated peripheral edema label image is subjected to hole closing processing by selecting a certain radius, and then the automatically generated peripheral edema label image and the corresponding peripheral edema label image after closure are subjected to label fusion processing and binarization processing.

2. The collaborative cross-segmentation system for cerebral hemorrhage and peripheral edema based on automatically generated labels according to claim 1, characterized in that: The encoder of the hemorrhage segmentation network and the encoder of the surrounding edema segmentation network share or do not share weights.

3. The collaborative cross-segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to claim 1, characterized in that: In the preprocessing module: The preprocessing of the acquired 3D cranial CT images includes: data cleaning, format conversion, renaming, steering, zero voxel filling, registration, and window width and window position adjustment of the acquired 3D cranial CT images, and finally extracting the preprocessed 3D brain parenchyma CT images.

4. The collaborative cross-segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to claim 1, characterized in that: In the label image generation module, the threshold segmentation method and the clustering segmentation method are used to generate the cerebral hemorrhage label image; The method of generating a label image of cerebral hemorrhage based on the threshold segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform global threshold segmentation on the smoothed image; Generating a cerebral hemorrhage label image based on a clustering segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform fuzzy clustering segmentation on the smoothed image.

5. The collaborative cross-segmentation system for cerebral hemorrhage and surrounding edema based on automatically generated labels according to claim 4, characterized in that: In the label image generation module, the threshold segmentation method, cluster segmentation method and contralateral difference method are used to generate the peripheral edema label image; The method of generating the peripheral edema label image based on the threshold segmentation method is to expand the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiply it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then perform single threshold segmentation on the initial edema area; The peripheral edema label image is generated based on the clustering segmentation method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then performing fuzzy clustering segmentation on the initial edema area; The peripheral edema label image is generated based on the contralateral difference method. The cerebral hemorrhage label image of the same 3D cranial CT image is expanded outward according to a certain radius, and the initial edema area is determined by dot multiplication with the preprocessed 3D brain parenchyma CT image. Then, the corresponding area of ​​the contralateral brain is extracted by mirror reversal based on the midline. Finally, the two areas are subtracted, and the area with the difference within a certain range is taken as the target edema area.

6. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the following steps are implemented: Preprocessing the acquired 3D cranial CT images; Computer vision and machine learning algorithms are used to automatically generate brain hemorrhage label images and surrounding edema label images from preprocessed 3D brain parenchyma CT images; Perform label fusion processing on the automatically generated cerebral hemorrhage label image and surrounding edema label image to obtain a cerebral hemorrhage supervision label image and a surrounding edema supervision label image; The supervised label images of cerebral hemorrhage and the supervised label images of surrounding edema are used to train a collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning, so as to segment cerebral hemorrhage and surrounding edema simultaneously. Using the deep learning segmentation results of cerebral hemorrhage and surrounding edema, morphological measurements were performed and the results were presented; Among them, the collaborative cross-segmentation model of cerebral hemorrhage and surrounding edema based on deep learning includes two hemorrhage segmentation networks and surrounding edema segmentation networks with the same structure; The hemorrhage segmentation network and the surrounding edema segmentation network have the same input and output the deep learning segmentation results of cerebral hemorrhage and the deep learning segmentation results of surrounding edema respectively; The hemorrhage segmentation network and the surrounding edema segmentation network use the encoder-decoder 3D residual U-Net framework as the backbone network respectively; The number of down-sampling layers in the encoder of the hemorrhage segmentation network and the surrounding edema segmentation network is equal to the number of up-sampling layers in the decoder of the hemorrhage segmentation network and the surrounding edema segmentation network; Each layer of the decoder of the hemorrhage segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the surrounding edema segmentation network; Each layer of the decoder of the peripheral edema segmentation network except the top layer is cross-connected with the corresponding layer of the decoder of the hemorrhage segmentation network; The deep learning-based collaborative cross-segmentation model for ICH and peripheral edema further includes a channel attention module and a spatial attention module; The channel attention module is arranged between the decoder of the hemorrhage segmentation network and the decoder of the surrounding edema segmentation network; The spatial attention module is arranged between the encoder and the decoder in the same segmentation network; The label fusion processing of the automatically generated cerebral hemorrhage label image and the surrounding edema label image includes: The automatically generated cerebral hemorrhage label image is enlarged by selecting a certain expansion radius, and then the automatically generated cerebral hemorrhage label image and the expanded cerebral hemorrhage label image are subjected to label fusion processing and binarization processing; The automatically generated peripheral edema label image is subjected to hole closing processing by selecting a certain radius, and then the automatically generated peripheral edema label image and the corresponding peripheral edema label image after closure are subjected to label fusion processing and binarization processing.

7. The computer-readable storage medium according to claim 6, wherein: The encoder of the hemorrhage segmentation network and the encoder of the surrounding edema segmentation network share or do not share weights.

8. The computer-readable storage medium according to claim 6, wherein: The preprocessing of the acquired 3D cranial CT images includes: data cleaning, format conversion, renaming, steering, zero voxel filling, registration, and window width and window position adjustment of the acquired 3D cranial CT images, and finally extracting the preprocessed 3D brain parenchyma CT images.

9. The computer-readable storage medium according to claim 6, wherein: The threshold segmentation method and clustering segmentation method are used to generate the cerebral hemorrhage label image; The method of generating a label image of cerebral hemorrhage based on the threshold segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform global threshold segmentation on the smoothed image; Generating a cerebral hemorrhage label image based on a clustering segmentation method is to smooth the preprocessed 3D brain parenchyma CT image, and then perform fuzzy clustering segmentation on the smoothed image.

10. The computer-readable storage medium according to claim 9, wherein: The peripheral edema label images were generated using the threshold segmentation method, clustering segmentation method and contralateral difference method respectively; The method of generating the peripheral edema label image based on the threshold segmentation method is to expand the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiply it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then perform single threshold segmentation on the initial edema area; The peripheral edema label image is generated based on the clustering segmentation method by expanding the cerebral hemorrhage label image of the same 3D cranial CT image outward according to a certain radius, and multiplying it with the preprocessed 3D brain parenchyma CT image to determine the initial edema area, and then performing fuzzy clustering segmentation on the initial edema area; The peripheral edema label image is generated based on the contralateral difference method. The cerebral hemorrhage label image of the same 3D cranial CT image is expanded outward according to a certain radius, and the initial edema area is determined by dot multiplication with the preprocessed 3D brain parenchyma CT image. Then, the corresponding area of ​​the contralateral brain is extracted by reversing based on the midline. Finally, the two areas are subtracted, and the area with the difference within a certain range is taken as the target edema area.

Images