Cerebral hemorrhage surgery robot and electronic device

Abstract

Embodiments of the present application provide a brain hemorrhage surgery robot and an electronic device. The brain hemorrhage surgery robot comprises a control host and a mechanical arm, and a preprocessing module, an intracranial hematoma prediction model and a post-processing module are arranged on the control host. The preprocessing module is used for preprocessing a to-be-processed brain axial tomography grayscale image to convert it into an input image of the intracranial hematoma segmentation module. The intracranial hematoma prediction model is used for feature extraction and voxel classification on the input image to obtain a prediction result of intracranial hematoma. The post-processing module is used for three-dimensional reconstruction of intracranial hematoma according to the prediction result of intracranial hematoma to obtain an intracranial hematoma model to control the action of the mechanical arm.

Description

Technical Field

[0001] The present invention relates to the field of robotics, and more particularly to a surgical robot and electronic device for treating cerebral hemorrhage. Background Technology

[0002] Currently, intracranial hematoma reconstruction techniques encompass threshold segmentation, region-growing-based methods, and level-set-based methods.

[0003] Thresholding segmentation aims to effectively separate hematoma areas from images by precisely setting grayscale thresholds. However, this method is highly dependent on manual operation, requiring frequent threshold adjustments to ensure optimal separation results. Furthermore, given that the grayscale values ​​of bones in images are similar to those of hematomas, this method may mistakenly include bones while separating hematomas, necessitating manual bone removal.

[0004] Region-growing methods start from a specific point in the image and gradually absorb surrounding points with similar gray levels to form and continuously expand the hematoma region until no more new points matching the criteria can be added. This method requires the user to manually locate the hematoma and specify the starting point for growth. For complex cases with multiple hemorrhages, multiple operations may be required to complete the reconstruction.

[0005] The level set-based method requires the user to set an initial contour and then gradually adjust the contour shape and position through the evolution and iteration of the level set function to ultimately achieve accurate segmentation of the target region. However, this method has high computational complexity and relatively slow iteration speed, which may affect the overall reconstruction efficiency.

[0006] In addition, all three methods mentioned above focus on the reconstruction of hematoma within the entire intracranial cavity, failing to achieve more refined granular control, which to some extent limits their application in complex cases. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a surgical robot and electronic device for cerebral hemorrhage surgery, to at least partially solve the above-mentioned problems.

[0008] According to a first aspect of the present invention, a surgical robot for cerebral hemorrhage is provided, comprising: a control host and a robotic arm. The control host is equipped with a preprocessing module, an intracranial hematoma prediction model, and a postprocessing module. The preprocessing module is used to preprocess a grayscale image of a brain axial tomographic scan to convert it into an input image for the intracranial hematoma segmentation module. The intracranial hematoma prediction model is used to extract features and classify voxels from the input image to obtain a prediction result of the intracranial hematoma. The postprocessing module is used to perform three-dimensional reconstruction of the intracranial hematoma based on the prediction result to obtain an intracranial hematoma model, thereby controlling the movement of the robotic arm.

[0009] Optionally, when the preprocessing module preprocesses the axial tomographic grayscale image of the brain to be processed to convert it into the input image of the intracranial hematoma segmentation module, it performs the following steps:

[0010] Based on the total number of input channels in the intracranial hematoma segmentation module, the brain axial tomographic grayscale image to be processed is divided into blocks to obtain multiple brain axial tomographic grayscale image sub-blocks, such that the total number of voxels included in each brain axial tomographic grayscale image sub-block is equal to the total number of input channels.

[0011] Center enhancement processing is performed on each sub-block of the brain axial tomographic grayscale image to obtain an influence matrix.

[0012] Each brain axial tomographic grayscale image sub-block is enhanced according to the corresponding influence matrix to obtain an enhanced brain axial tomographic grayscale image sub-block, and the fused brain axial tomographic grayscale image sub-block is used as the input image.

[0013] Optionally, when the preprocessing module divides the brain axial tomographic grayscale image to be processed into multiple brain axial tomographic grayscale image sub-blocks based on the total number of input channels in the intracranial hematoma segmentation module, the following steps are performed:

[0014] Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the X direction and the planned size of a single brain axial tomographic grayscale image sub-block in the X direction, calculate the X-direction segmentation parameters when processing the brain axial tomographic grayscale image to be processed into blocks along the X direction.

[0015] Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the Y direction and the planned size of a single brain axial tomographic grayscale image sub-block in the Y direction, calculate the Y-direction segmentation parameters when processing the brain axial tomographic grayscale image to be processed into blocks along the Y direction.

[0016] Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the Z direction and the planned size of a single brain axial tomographic grayscale image sub-block in the Z direction, calculate the Z-direction segmentation parameters when segmenting the brain axial tomographic grayscale image to be processed along the Z direction.

[0017] Based on the X-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the X direction to obtain several X-direction tomographic grayscale image sub-blocks.

[0018] Based on the Y-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the Y direction to obtain several Y-direction tomographic grayscale image sub-blocks.

[0019] Based on the Z-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the Z-direction to obtain several Z-direction tomographic grayscale image sub-blocks.

[0020] Based on the aforementioned X-axis tomographic grayscale image sub-blocks, Y-axis tomographic grayscale image sub-blocks, and Z-axis tomographic grayscale image sub-blocks, the aforementioned multiple brain axial tomographic grayscale image sub-blocks are generated.

[0021] Optionally, when the preprocessing module performs center enhancement processing on each sub-block of the brain axial tomographic grayscale image to obtain the influence matrix, it performs the following steps:

[0022] The dimension for center enhancement processing of each brain axial tomographic grayscale image sub-block is determined, and the direction of center enhancement processing is determined based on the number of said dimensions;

[0023] Based on the direction of the center enhancement processing, several voxels in the central region of each brain axial tomographic grayscale image sub-block are subjected to influence enhancement processing along the dimension to obtain an influence matrix diagram.

[0024] Optionally, when the preprocessing module enhances each sub-block of the brain axial tomographic grayscale image according to the corresponding influence matrix to obtain the enhanced sub-block of the brain axial tomographic grayscale image, it performs the following steps:

[0025] The gray values ​​of several voxels in the central region of each brain axial tomographic grayscale image sub-block are positively adjusted according to the corresponding influence matrix diagram to obtain an enhanced brain axial tomographic grayscale image sub-block.

[0026] Optionally, the intracranial hematoma prediction model is trained according to the following steps to extract features and classify voxels from the input image to obtain the prediction result of the intracranial hematoma:

[0027] Obtain axial tomographic grayscale image samples of the brain and corresponding intracranial hematoma labels. The intracranial hematoma labels indicate that the category of the axial tomographic grayscale image of the brain is background, brain parenchymal hemorrhage, or intraventricular hemorrhage.

[0028] The brain axial tomographic grayscale image samples were preprocessed to obtain preprocessed image samples.

[0029] The preprocessed image samples are input into the specified neural network model for feature extraction and voxel classification to calculate the predicted probability of each intracranial hematoma label.

[0030] The loss value of the neural network model is calculated based on the predicted probability of the intracranial hematoma label.

[0031] The network parameters of the neural network model are adjusted based on the loss value, so that the neural network model when the loss value reaches a set loss threshold is used as the intracranial hematoma prediction model.

[0032] Optionally, during the training of the intracranial hematoma prediction model, the brain axial tomographic grayscale image samples are preprocessed to obtain preprocessed image samples, including:

[0033] Spatial consistency correction and grayscale value distribution standardization are performed on the voxels of the brain axial tomographic grayscale image samples to obtain preprocessed image samples.

[0034] Optionally, during the training of the intracranial hematoma prediction model, the preprocessed image samples are input into the designated neural network model for feature extraction and voxel classification to calculate the prediction probability of each intracranial hematoma label, including:

[0035] The preprocessed image sample is input into the feature extraction module of the specified neural network model to extract features and obtain feature vector samples.

[0036] The feature vector samples are input into the output module of the neural network model for three-class classification prediction to obtain three-channel four-dimensional output data. The first dimension of the three-channel four-dimensional data includes the length information of voxels in the preprocessed image samples, the second dimension includes the width information of voxels in the preprocessed image samples, the third dimension includes the width information of voxels in the preprocessed image samples, and the fourth dimension includes three channels, with each channel representing a predicted probability sample of an intracranial hematoma label.

[0037] Optionally, when the preprocessing module preprocesses the axial tomographic grayscale image of the brain to be processed to convert it into the input image of the intracranial hematoma segmentation module, it preprocesses each sub-block of the axial tomographic grayscale image of the brain to convert it into the input image of the intracranial hematoma segmentation module, and one sub-block of the axial tomographic grayscale image of the brain corresponds to one input image.

[0038] The intracranial hematoma prediction model performs feature extraction and voxel classification on the input image according to the following steps to obtain the prediction result of intracranial hematoma. One input image corresponds to one prediction result of intracranial hematoma:

[0039] The feature extraction module in the intracranial hematoma prediction model extracts features from the input image to obtain a feature vector.

[0040] The feature extraction module in the intracranial hematoma prediction model performs three-class classification prediction based on the feature vector to obtain three-channel four-dimensional output data. The first dimension of the three-channel four-dimensional data includes the length information of voxels in the input image, the second dimension includes the width information of voxels in the input image, the third dimension includes the height information of voxels in the input image, and the fourth dimension includes three channels, each channel representing the prediction probability of an intracranial hematoma type, namely background, parenchymal hemorrhage, and intraventricular hemorrhage. The intracranial hematoma type corresponding to the channel with the highest prediction probability is used as the prediction result of the intracranial hematoma corresponding to the input image.

[0041] Optionally, the preprocessing module is further configured to flip the input image along at least one of the X, Y, and Z directions of the to-be-processed brain axis tomographic grayscale image to obtain a flipped input image, so that the intracranial hematoma prediction model can extract features from the flipped input image to obtain a prediction sub-result of intracranial hematoma, and the prediction sub-results of intracranial hematoma corresponding to all the flipped input features are combined to obtain the prediction result of intracranial hematoma corresponding to the input image.

[0042] Optionally, when the post-processing module performs three-dimensional reconstruction of the intracranial hematoma based on the predicted results to obtain an intracranial hematoma model, it performs the following steps:

[0043] The prediction results of intracranial hematoma corresponding to all input images are stitched together to obtain the global prediction result of intracranial hematoma corresponding to the brain axial tomographic grayscale image to be processed.

[0044] The influence matrix diagrams corresponding to all sub-blocks of the brain axial tomographic grayscale image are stitched together to obtain the global influence matrix diagram corresponding to the brain axial tomographic grayscale image to be processed.

[0045] Based on the global influence matrix of the brain axial tomographic grayscale image to be processed, the global prediction result of the corresponding intracranial hematoma is adjusted in three dimensions to obtain the output prediction result of the intracranial hematoma.

[0046] The output prediction result of the intracranial hematoma is restored into a three-dimensional grayscale image;

[0047] Isosurfaces are extracted from the three-dimensional grayscale image, and all extracted isosurfaces are stitched together to generate an intracranial hematoma model.

[0048] According to a second aspect of the present invention, an electronic device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, the executable instruction causing the processor to perform the following steps:

[0049] The axial tomographic grayscale image of the brain to be processed is preprocessed to convert it into the input image of the intracranial hematoma segmentation module;

[0050] Feature extraction and voxel classification are performed on the input image to obtain the prediction result of intracranial hematoma;

[0051] Based on the predicted intracranial hematoma, a three-dimensional reconstruction of the intracranial hematoma is performed to obtain an intracranial hematoma model, which is then used to control the movement of the robotic arm.

[0052] In this embodiment of the invention, the brain hemorrhage surgical robot automatically processes the axial tomographic grayscale image of the brain through a preprocessing module and extracts features using an intracranial hematoma prediction model. This eliminates the need for frequent manual threshold adjustments, significantly reducing human intervention and increasing automation. Simultaneously, the prediction model, through techniques such as length learning, can more accurately identify hematoma regions, effectively avoiding missegmentation of similar grayscale regions such as bones. Furthermore, the robot of this invention, through the global feature extraction capability of the intracranial hematoma prediction model, can identify and segment all hematoma regions at once, eliminating the need for multiple manual localizations by the user, significantly improving the efficiency and accuracy in handling complex cases. Moreover, the robot of this invention employs efficient prediction models (such as length learning networks). These models are typically optimized to achieve fast processing speeds while maintaining high accuracy, thus overcoming the computational efficiency problem of level set methods. Finally, the post-processing module of this invention can perform high-precision 3D reconstruction based on the prediction results, generating a detailed intracranial hematoma model. This allows for separate reconstruction of intracranial hematoma models for intracranial parenchymal hemorrhage and intraventricular hemorrhage, which is particularly important in complex cases where such fine-grained control is crucial. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0054] Figure 1 This is a schematic diagram of a surgical robot for cerebral hemorrhage according to an embodiment of this application.

[0055] Figure 2 This is a schematic diagram of a method for establishing an intracranial hematoma model provided in this embodiment.

[0056] Figure 3 This is a schematic diagram illustrating the prediction results of intracranial hematoma according to an embodiment of the present invention.

[0057] Figure 4 This is a schematic diagram of an intracranial hematoma model according to an embodiment of the present invention.

[0058] Figure 5 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0059] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art should fall within the protection scope of the present invention.

[0060] It should be understood that the terms "first," "second," and "third," etc., in the claims, specification, and drawings of this disclosure are used to distinguish different objects, not to describe a specific order. The terms "comprising" and "including" as used in the specification and claims of this disclosure indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or sets thereof.

[0061] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0062] Figure 1 This is a schematic diagram of a surgical robot for treating cerebral hemorrhage, according to an embodiment of this application. Figure 1 As shown, it includes a control host and a robotic arm. The control host is equipped with a preprocessing module, an intracranial hematoma prediction model, and a post-processing module. The preprocessing module is used to preprocess the axial tomographic grayscale image of the brain to be processed, so as to convert it into the input image of the intracranial hematoma segmentation module. The intracranial hematoma prediction model is used to extract features and classify voxels from the input image to obtain the prediction result of the intracranial hematoma. The post-processing module is used to perform three-dimensional reconstruction of the intracranial hematoma based on the prediction result of the intracranial hematoma to obtain the intracranial hematoma model, so as to control the movement of the robotic arm.

[0063] In this embodiment of the invention, the brain hemorrhage surgical robot automatically processes the axial tomographic grayscale image of the brain through a preprocessing module and extracts features using an intracranial hematoma prediction model. This eliminates the need for frequent manual threshold adjustments, significantly reducing human intervention and increasing automation. Simultaneously, the prediction model, through techniques such as length learning, can more accurately identify hematoma regions, effectively avoiding missegmentation of similar grayscale regions such as bones. Furthermore, the robot of this invention, through the global feature extraction capability of the intracranial hematoma prediction model, can identify and segment all hematoma regions at once, eliminating the need for multiple manual localizations by the user, significantly improving the efficiency and accuracy in handling complex cases. Moreover, the robot of this invention employs efficient prediction models (such as length learning networks). These models are typically optimized to achieve fast processing speeds while maintaining high accuracy, thus overcoming the computational efficiency problem of level set methods. Finally, the post-processing module of this invention can perform high-precision 3D reconstruction based on the prediction results, generating a detailed intracranial hematoma model. This allows for separate reconstruction of intracranial hematoma models for intracranial parenchymal hemorrhage and intraventricular hemorrhage, which is particularly important in complex cases where such fine-grained control is crucial.

[0064] Optionally, the brain axial tomographic grayscale image to be processed may be, but is not limited to, CT (Computed Tomography) or CTA (Computed Tomography Angiography) images.

[0065] Optionally, when the preprocessing module preprocesses the axial tomographic grayscale image of the brain to be processed to convert it into the input image of the intracranial hematoma segmentation module, it performs the following steps:

[0066] Based on the total number of input channels in the intracranial hematoma segmentation module, the brain axial tomographic grayscale image to be processed is divided into blocks to obtain multiple brain axial tomographic grayscale image sub-blocks, such that the total number of voxels included in each brain axial tomographic grayscale image sub-block is equal to the total number of input channels.

[0067] Center enhancement processing is performed on each sub-block of the brain axial tomographic grayscale image to obtain an influence matrix.

[0068] Each brain axial tomographic grayscale image sub-block is enhanced according to the corresponding influence matrix to obtain an enhanced brain axial tomographic grayscale image sub-block, and the fused brain axial tomographic grayscale image sub-block is used as the input image.

[0069] Therefore, by employing steps such as block processing, center enhancement processing, and enhancement based on the influence matrix diagram, the following technical advantages are achieved:

[0070] 1. Improve data processing efficiency and adaptability

[0071] Block processing: This technical solution first divides the brain axial tomographic grayscale image to be processed into blocks based on the total number of input channels in the intracranial hematoma segmentation module. The advantage of this is that it can divide large images into multiple smaller, more easily processed blocks. This divide-and-conquer strategy can significantly improve data processing efficiency, especially when processing high-resolution images. Furthermore, block processing enhances the system's adaptability, enabling it to flexibly handle images of different sizes.

[0072] 2. Enhance information on key areas

[0073] Center Enhancement Processing: Center enhancement processing is performed on each sub-block of the brain axial tomographic grayscale image to obtain an influence matrix. The purpose of this step is to highlight key regions in the image, especially those that may be associated with intracranial hematoma. By enhancing the information in these regions, the accuracy of subsequent segmentation can be improved. The influence matrix, as a weighting tool, can guide subsequent enhancement processing, ensuring that important information receives sufficient attention.

[0074] 3. Improve segmentation accuracy and robustness

[0075] Enhancement based on influence matrix: Enhancement processing is performed on the influence matrix corresponding to each sub-block of the brain axial tomographic grayscale image, resulting in enhanced sub-blocks. These enhanced sub-blocks serve as input images for the intracranial hematoma segmentation module, providing richer and more accurate information. Because the information in key regions is enhanced, the segmentation module can identify hematomas more accurately. Simultaneously, this enhancement process improves the system's robustness, making it more stable in the face of noise, artifacts, and other interference factors.

[0076] 4. Achieve refined segmentation

[0077] Fusion of Enhanced Sub-blocks: Finally, all enhanced axial tomographic grayscale image sub-blocks of the brain are fused together and used as the input image for the intracranial hematoma segmentation module. This process achieves refined segmentation of the original image. Because each sub-block undergoes targeted enhancement processing, the segmentation accuracy of the entire image is significantly improved. This refined segmentation is crucial for accurately identifying the morphology, location, and relationship of intracranial hematomas to other tissues.

[0078] Optionally, when the preprocessing module divides the brain axial tomographic grayscale image to be processed into multiple brain axial tomographic grayscale image sub-blocks based on the total number of input channels in the intracranial hematoma segmentation module, the following steps are performed:

[0079] Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the X direction and the planned size of a single brain axial tomographic grayscale image sub-block in the X direction, calculate the X-direction segmentation parameters when processing the brain axial tomographic grayscale image to be processed into blocks along the X direction.

[0080] Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the Y direction and the planned size of a single brain axial tomographic grayscale image sub-block in the Y direction, calculate the Y-direction segmentation parameters when processing the brain axial tomographic grayscale image to be processed into blocks along the Y direction.

[0081] Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the Z direction and the planned size of a single brain axial tomographic grayscale image sub-block in the Z direction, calculate the Z-direction segmentation parameters when segmenting the brain axial tomographic grayscale image to be processed along the Z direction.

[0082] Based on the X-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the X direction to obtain several X-direction tomographic grayscale image sub-blocks.

[0083] Based on the Y-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the Y direction to obtain several Y-direction tomographic grayscale image sub-blocks.

[0084] Based on the Z-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the Z-direction to obtain several Z-direction tomographic grayscale image sub-blocks.

[0085] Based on the aforementioned X-axis tomographic grayscale image sub-blocks, Y-axis tomographic grayscale image sub-blocks, and Z-axis tomographic grayscale image sub-blocks, the aforementioned multiple brain axial tomographic grayscale image sub-blocks are generated.

[0086] Therefore, by accurately calculating the block parameters and performing block processing along the X, Y, and Z directions, the following technical advantages are available:

[0087] 1. Precise block division improves processing efficiency.

[0088] This technical solution divides the brain axial tomographic grayscale image into blocks by precisely calculating the block parameters in the X, Y, and Z directions. This precise block division ensures that each sub-block is of appropriate size—neither too large, leading to low processing efficiency, nor too small, resulting in information loss or increased processing complexity. Therefore, this block division significantly improves image processing efficiency.

[0089] 2. Highly adaptable and capable of flexibly handling images of different sizes.

[0090] Because this technical solution calculates the block parameters based on the difference between the size of the image to be processed and the planned size of a single sub-block, it can flexibly adapt to images of different sizes. Regardless of how the image size changes, the appropriate sub-block size can be obtained by adjusting the block parameters, thereby ensuring the accuracy and stability of image processing.

[0091] 3. Maintain image integrity and avoid information loss.

[0092] In traditional image segmentation, improper segmentation can lead to the loss or duplication of image information. This new technique, however, precisely calculates segmentation parameters and performs segmentation along the X, Y, and Z directions, ensuring that each sub-block contains complete image information and preventing loss or duplication. This is crucial for subsequent image processing and segmentation.

[0093] 4. Facilitates subsequent processing

[0094] Multiple brain axial tomographic grayscale image sub-blocks obtained after block processing can be used as input data for subsequent processing. These sub-blocks not only contain complete image information, but also have small size and moderate complexity, which facilitates subsequent image processing, feature extraction, and segmentation.

[0095] Optionally, when the preprocessing module performs center enhancement processing on each sub-block of the brain axial tomographic grayscale image to obtain the influence matrix, it performs the following steps:

[0096] The dimension for center enhancement processing of each brain axial tomographic grayscale image sub-block is determined, and the direction of center enhancement processing is determined based on the number of said dimensions;

[0097] Based on the direction of the center enhancement processing, several voxels in the central region of each brain axial tomographic grayscale image sub-block are subjected to influence enhancement processing along the dimension to obtain an influence matrix diagram.

[0098] Therefore, the central enhancement processing technology offers the following technical advantages:

[0099] 1. Increase the prominence of information in key areas.

[0100] Center enhancement processing: This technique significantly highlights information in the central region of each brain axial tomographic grayscale image sub-block by performing center enhancement processing. In medical image processing, the central region often contains the most critical information, such as lesions and hematomas. Center enhancement processing makes this information easier to identify and extract in subsequent processing.

[0101] 2. Improve segmentation accuracy

[0102] Influence Matrix: Center enhancement not only amplifies the information in the central region but also generates an influence matrix. This matrix reflects the degree of influence of each voxel on the central region, providing an important reference for subsequent segmentation. By utilizing the influence matrix, lesion boundaries can be located more accurately, improving segmentation precision.

[0103] 3. Enhance the robustness of the algorithm

[0104] Multi-dimensional processing: The technical solution mentions that the dimension and direction of the central enhancement processing can be determined as needed. This multi-dimensional processing approach can adapt to lesions of different shapes and orientations, enhancing the robustness of the algorithm. Even if the morphology and location of the lesion change, this technology can ensure accurate information extraction by adjusting the processing dimension and direction.

[0105] 4. Improve processing efficiency

[0106] Targeted Enhancement: Center-based enhancement only strengthens the central region of the sub-block, avoiding indiscriminate processing of the entire sub-block. This targeted enhancement method reduces computation and improves processing efficiency. Furthermore, because only key areas are enhanced, subsequent processing becomes more focused and efficient.

[0107] 5. Facilitates subsequent processing

[0108] Preprocessing optimization: Center enhancement, as one of the preprocessing steps, provides optimized input data for subsequent feature extraction, segmentation, and other processing. This optimized data has a higher signal-to-noise ratio and better separability, which helps subsequent processing algorithms to perform better.

[0109] Optionally, when the preprocessing module enhances each sub-block of the brain axial tomographic grayscale image according to the corresponding influence matrix to obtain the enhanced sub-block of the brain axial tomographic grayscale image, it performs the following steps:

[0110] The gray values ​​of several voxels in the central region of each brain axial tomographic grayscale image sub-block are positively adjusted according to the corresponding influence matrix diagram to obtain an enhanced brain axial tomographic grayscale image sub-block.

[0111] Therefore, the technical solution of enhancing sub-blocks of brain axial tomographic grayscale images according to the influence matrix diagram has the following technical advantages:

[0112] 1. Highlight key information

[0113] By positively adjusting the voxel grayscale values ​​of the central region of a sub-block in a brain axial tomographic grayscale image using an influence matrix, information from these key areas can be significantly highlighted. In medical image processing, particularly for the segmentation of intracranial hematomas, the central region often contains the most important lesion information. Enhancing these regions makes it easier for doctors or subsequent processing algorithms to identify and locate the lesion.

[0114] 2. Improve segmentation accuracy

[0115] The enhanced axial tomographic grayscale image sub-blocks of the brain exhibit higher signal-to-noise ratios and contrasts, resulting in clearer boundaries between the lesion and surrounding tissues. This translates to more accurate lesion boundary identification in subsequent segmentation algorithms, thereby improving segmentation precision. This is crucial for the precise operation of robotic brain hemorrhage surgery.

[0116] 3. Adaptive Enhancement

[0117] The influence matrix is ​​generated based on the specific circumstances of each sub-block, making the enhancement process adaptive. This adaptive enhancement ensures appropriate enhancement effects between different sub-blocks and between different regions within a sub-block. This avoids over-enhancement or under-enhancement, thereby improving the overall processing quality.

[0118] 4. Maintain image continuity

[0119] Although only the central region of the sub-block was enhanced, the enhancement was smooth and coherent thanks to the guidance of the influence matrix. This avoided abrupt boundaries or artifacts during the enhancement process, thus maintaining the overall coherence and realism of the image.

[0120] 5. Provide optimized input for subsequent processing

[0121] The enhanced axial tomographic grayscale image sub-blocks of the brain, used as input data for subsequent processing, possess higher quality and richer information. This means that subsequent processing steps such as feature extraction, classification, and recognition can more accurately extract the required information, thereby improving the performance and accuracy of the entire processing flow.

[0122] Optionally, the intracranial hematoma prediction model is trained according to the following steps to extract features and classify voxels from the input image to obtain the prediction result of the intracranial hematoma:

[0123] Obtain axial tomographic grayscale image samples of the brain and corresponding intracranial hematoma labels. The intracranial hematoma labels indicate that the category of the axial tomographic grayscale image of the brain is background, brain parenchymal hemorrhage, or intraventricular hemorrhage.

[0124] The brain axial tomographic grayscale image samples were preprocessed to obtain preprocessed image samples.

[0125] The preprocessed image samples are input into the specified neural network model for feature extraction and voxel classification to calculate the predicted probability of each intracranial hematoma label.

[0126] The loss value of the neural network model is calculated based on the predicted probability of the intracranial hematoma label.

[0127] The network parameters of the neural network model are adjusted based on the loss value, so that the neural network model when the loss value reaches a set loss threshold is used as the intracranial hematoma prediction model.

[0128] In summary, the above training process for the intracranial hematoma prediction model provides strong support for the accurate prediction and auxiliary diagnosis of cerebral hemorrhage using surgical robots. Its technical advantages are as follows:

[0129] 1. Improve forecast accuracy

[0130] Samples and Labels: By acquiring axial tomographic grayscale image samples of the brain and corresponding intracranial hematoma labels, the training process can clearly distinguish between different categories such as background, brain parenchymal hemorrhage, and intraventricular hemorrhage. This clear classification helps improve the accuracy of the model in prediction and ensures the correct identification of intracranial hematomas.

[0131] Feature extraction: By using a specified neural network model to extract features from preprocessed image samples, key information in the image can be deeply explored, such as the shape, size, and location of hematomas, thereby further improving the accuracy of prediction.

[0132] 2. Optimize model performance

[0133] Loss calculation: The loss value of the neural network model is calculated based on the predicted probability of the intracranial hematoma label, which quantifies the error of the model in the prediction process. This quantification helps to evaluate the model's performance and provides a basis for subsequent network parameter adjustments.

[0134] Parameter tuning: Adjusting the network parameters of a neural network model based on the loss value can gradually optimize the model's performance. Through continuous iterative training and parameter tuning, the model's loss value can be gradually reduced to reach a set loss threshold, thereby achieving better prediction results.

[0135] 3. Improve training efficiency

[0136] Preprocessing steps: Preprocessing the axial tomographic grayscale image samples of the brain, such as through block segmentation and center enhancement, can reduce noise and redundant information in the image, thereby improving image quality. This helps to accelerate the training speed of neural networks and improve training efficiency.

[0137] Iterative training: By continuously training and adjusting parameters, the model can gradually converge to the optimal solution. This iterative training method can make full use of the information in the dataset, improving the model's generalization ability and predictive performance.

[0138] 4. Enhance the robustness of the model

[0139] Multi-sample training: Training the model using a large number of axial tomographic grayscale image samples allows it to learn more hematoma features under different conditions. This helps enhance the model's robustness, enabling it to provide accurate predictions when faced with intracranial hematomas of different shapes, sizes, and locations.

[0140] Loss threshold setting: By setting a loss threshold, the training process of the model can be controlled to ensure that the model stops training after reaching a certain performance level. This helps to avoid overfitting or underfitting, thereby maintaining the stability and robustness of the model.

[0141] During model training, the ReLU activation function originally used in the existing technology is replaced with Leaky ReLU:

[0142] The expression for the leaky ReLU function is:

[0143]

[0144] in, Choosing 0.01, compared to the ReLU function, the leaky ReLU function no longer outputs 0 for negative numbers, but instead outputs a very small negative number, which can solve the problem that some neurons in the model will not be activated during training, resulting in the weights not being updated.

[0145] In addition, for the loss function, the sum of the Dice loss function and the cross-entropy loss function is selected as the loss function:

[0146] The Dice loss function is:

[0147] in Indicates the first The class's true label (also known as ground truth). Indicates the prediction of the first The result of the class.

[0148] The cross-entropy loss function is:

[0149] in Indicates the first in the real label The value of the class can be either 0 or 1. Indicates the prediction of the first The probability of a class.

[0150] Therefore, combining the Dice loss function and the cross-entropy loss function can fully leverage the advantages of both to further improve the model's performance, specifically offering the following technical benefits:

[0151] (1) By simultaneously optimizing Dice loss and cross-entropy loss, the model can focus on the overlap between the predicted results and the true labels, while also considering the accuracy of the overall predicted distribution, thus achieving more comprehensive optimization.

[0152] (2) Improve model stability: The combination of the robustness of the Dice loss function to imbalanced data and the fast convergence of the cross-entropy loss function helps to improve the stability of the model during training and reduce the risk of overfitting or underfitting.

[0153] (3) Adapting to different task requirements: Different tasks may have different requirements for model performance. Combining Dice loss and cross-entropy loss can make the model more flexible and adaptable to the needs of different tasks. For example, in image segmentation tasks, the segmentation accuracy and classification effect can be balanced by adjusting the weights of the two.

[0154] Optionally, during the training of the intracranial hematoma prediction model, the brain axial tomographic grayscale image samples are preprocessed to obtain preprocessed image samples, including:

[0155] Spatial consistency correction and grayscale value distribution standardization are performed on the voxels of the brain axial tomographic grayscale image samples to obtain preprocessed image samples.

[0156] Therefore, preprocessing brain axial tomographic grayscale image samples (including spatial consistency correction and grayscale value distribution standardization) has the following technical advantages in training intracranial hematoma prediction models:

[0157] 1. Improve image quality

[0158] Spatial consistency correction: Due to factors such as imaging equipment, scanning parameters, or patient positioning, voxels in axial tomographic grayscale images of the brain may exhibit spatial inconsistencies. Spatial consistency correction can eliminate these inconsistencies, making the voxels in the image more accurate and consistent in space. This helps improve image quality and provides a more reliable basis for subsequent feature extraction and prediction.

[0159] Gray-scale distribution standardization: The gray-scale distribution of different image samples may vary, which can affect the model's feature recognition and extraction. Gray-scale distribution standardization unifies the gray-scale values ​​of different image samples to the same range, eliminating differences in gray-scale distribution. This helps the model better learn the features in the image and improves prediction accuracy.

[0160] 2. Enhance the model's generalization ability

[0161] Reducing data bias: Preprocessing steps can reduce data bias in image samples, allowing the model to focus more on the essential features of the image during training, rather than being interfered with by irrelevant factors. This helps enhance the model's generalization ability, enabling it to provide accurate predictions when faced with image samples from different sources and of different qualities.

[0162] 3. Improve training efficiency

[0163] Accelerating model convergence: Preprocessing steps can eliminate noise and redundant information in image samples, allowing the model to focus more on learning useful features during training. This helps accelerate model convergence, reduce training time, and improve training efficiency.

[0164] 4. Facilitates subsequent processing

[0165] Unifying image formats: The preprocessing step can unify image samples of different formats and resolutions to the same format and resolution, facilitating subsequent feature extraction, classification, and recognition processes. This helps simplify the entire processing workflow and improves processing efficiency and accuracy.

[0166] When performing grayscale value distribution standardization, tissue density values ​​(such as HU values, Hounsfield Units) can also be standardized. For example, since CT images often contain isolated values ​​that are abnormally large or small, narrowing the range of HU values ​​to the [0.5, 99.5] percentile range can effectively remove interference. Therefore, the correspondence between [0.5, 99.5] and grayscale values ​​is: 0.5 percentile grayscale value... 99.5 percentile gray value .

[0167] Therefore, after grayscale value distribution standardization, the grayscale value of each voxel satisfies the following relationship:

[0168]

[0169] Where M represents a grayscale brain axial tomographic image sample with p voxels, and N represents a grayscale brain axial tomographic image sample after grayscale value distribution normalization, with voxels denoted as p. , This represents the standard deviation of voxel grayscale values ​​in the brain axial tomographic grayscale image samples. This represents the mean value of voxel grayscale values ​​in the brain axial tomographic grayscale image samples.

[0170] Optionally, during the training of the intracranial hematoma prediction model, the preprocessed image samples are input into the designated neural network model for feature extraction and voxel classification to calculate the prediction probability of each intracranial hematoma label, including:

[0171] The preprocessed image sample is input into the feature extraction module of the specified neural network model to extract features and obtain feature vector samples.

[0172] The feature vector samples are input into the output module of the neural network model for three-class classification prediction to obtain three-channel four-dimensional output data. The first dimension of the three-channel four-dimensional data includes the length information of the voxels in the preprocessed image samples, the second dimension includes the width information of the voxels in the preprocessed image samples, the third dimension includes the height information of the voxels in the preprocessed image samples, and the fourth dimension includes three channels, with each channel representing a predicted probability sample of an intracranial hematoma label.

[0173] Therefore, when training an intracranial hematoma prediction model, inputting preprocessed image samples into a designated neural network model for feature extraction and prediction probability calculation has the following technical advantages:

[0174] 1. Efficient Feature Extraction

[0175] Feature Extraction Module: The feature extraction module within the neural network model efficiently extracts features from preprocessed image samples, yielding feature vector samples. This step delves into key information within the image, such as the shape, size, location of the hematoma, and its contrast with surrounding tissues, providing strong support for subsequent predictions.

[0176] 2. Accurate prediction

[0177] Three-class classification prediction: By inputting feature vector samples into the output module of the neural network model for three-class classification prediction, three-channel four-dimensional output data can be obtained. This design ensures that the model makes independent predictions for each voxel and provides predicted probabilities for three intracranial hematoma labels (background, parenchymal hemorrhage, and intraventricular hemorrhage). This helps improve the accuracy of prediction and ensures accurate identification of intracranial hematomas.

[0178] 3. Abundant output information

[0179] Three-dimensional data structure: The output three-channel four-dimensional data contains rich information. The first dimension includes voxel length information, the second dimension includes voxel width information in the preprocessed image sample, the third dimension includes voxel height information in the preprocessed image sample, and the fourth dimension includes three channels, each representing a predicted probability sample of an intracranial hematoma label. This data structure can comprehensively reflect the situation of each voxel in the image, providing detailed reference for doctors' diagnosis and treatment.

[0180] 4. Improve model performance

[0181] Independent prediction and comprehensive judgment: The model makes independent predictions for each voxel and then combines the predictions of all voxels to make a final judgment. This design can fully utilize the information in the image, improving the model's performance and accuracy. Furthermore, by comparing the prediction probabilities of different voxels, the model can also more accurately assess the distribution and extent of intracranial hematomas.

[0182] 5. Facilitates subsequent processing

[0183] Unified Data Format: The output three-channel four-dimensional data has a unified data format, facilitating subsequent processing and analysis. This helps simplify the entire processing workflow and improve processing efficiency and accuracy. Simultaneously, this data format also facilitates integration and interoperability with other medical image processing systems, enabling data sharing and interoperability.

[0184] Optionally, when the preprocessing module preprocesses the axial tomographic grayscale image of the brain to be processed to convert it into the input image of the intracranial hematoma segmentation module, it preprocesses each sub-block of the axial tomographic grayscale image of the brain to convert it into the input image of the intracranial hematoma segmentation module, and one sub-block of the axial tomographic grayscale image of the brain corresponds to one input image.

[0185] The intracranial hematoma prediction model performs feature extraction and voxel classification on the input image according to the following steps to obtain the prediction result of intracranial hematoma. One input image corresponds to one prediction result of intracranial hematoma:

[0186] The feature extraction module in the intracranial hematoma prediction model extracts features from the input image to obtain a feature vector.

[0187] The feature extraction module in the intracranial hematoma prediction model performs three-class classification prediction based on the feature vector to obtain three-channel four-dimensional output data. The first dimension of the three-channel four-dimensional data includes the length information of voxels in the input image, the second dimension includes the width information of voxels in the input image, the third dimension includes the height information of voxels in the input image, and the fourth dimension includes three channels, each channel representing the prediction probability of an intracranial hematoma type, namely background, parenchymal hemorrhage, and intraventricular hemorrhage. The intracranial hematoma type corresponding to the channel with the highest prediction probability is used as the prediction result of the intracranial hematoma corresponding to the input image.

[0188] Therefore, the above-mentioned scheme for the preprocessing module and intracranial hematoma prediction model in processing axial tomographic grayscale images of the brain and obtaining prediction results has the following technical advantages:

[0189] (1) Block processing improves efficiency:

[0190] The preprocessing module divides the axial tomographic grayscale image of the brain into multiple sub-blocks and performs independent preprocessing on each sub-block. This block-based processing method reduces computation and improves processing efficiency, especially when processing large-scale image data.

[0191] (2) Refined pretreatment:

[0192] Performing independent preprocessing on each sub-block allows for more refined processing tailored to the characteristics of each sub-block, such as local enhancement and noise suppression, thereby improving image quality and the accuracy of subsequent processing.

[0193] (3) The efficiency of the feature extraction module:

[0194] The feature extraction module in the intracranial hematoma prediction model can efficiently extract features from the input image, obtaining feature vectors containing key information. This helps reduce redundant information and improve the accuracy of subsequent predictions.

[0195] (4) The comprehensiveness of three-class classification prediction:

[0196] The model performs a three-class classification prediction, outputting three-channel four-dimensional data that comprehensively considers three scenarios: background hemorrhage, intracranial hemorrhage, and intraventricular hemorrhage. This comprehensive prediction method ensures accurate identification of intracranial hematomas and avoids missed or misdiagnosis.

[0197] (5) Clarity of prediction results:

[0198] The model compares the predicted probabilities of the three channels and selects the intracranial hematoma type corresponding to the channel with the highest predicted probability as the final prediction result. This explicit prediction method facilitates decision-making and judgment by doctors or subsequent processing algorithms.

[0199] (6) Structuring of output data:

[0200] The output three-channel four-dimensional data has a clear structure, containing voxel length, width, and height information, as well as the predicted probabilities for each of the three channels. This structured data facilitates subsequent analysis and processing, such as visualization and statistical analysis.

[0201] Optionally, the preprocessing module is further configured to flip the input image along at least one of the X, Y, and Z directions of the to-be-processed brain axis tomographic grayscale image to obtain a flipped input image, so that the intracranial hematoma prediction model can extract features from the flipped input image to obtain a prediction sub-result of intracranial hematoma, and the prediction sub-results of intracranial hematoma corresponding to all the flipped input features are combined to obtain the prediction result of intracranial hematoma corresponding to the input image.

[0202] Therefore, the scheme of using the preprocessing module to flip the input image has the following technical advantages:

[0203] (1) Improve the robustness of the prediction model:

[0204] By flipping the input image in the X, Y, and Z directions (or at least one of these directions), images from different perspectives can be obtained. This allows the intracranial hematoma prediction model to access more diverse data during training, thereby improving its ability to recognize hematomas from different perspectives and morphologies, and enhancing the model's robustness.

[0205] (2) Reduce prediction bias:

[0206] Since image flipping can produce a mirror effect, for some hematomas that are symmetrical or nearly symmetrical, the flipped image may help the model capture their features more accurately, thereby reducing prediction bias caused by differences in viewpoint or shape.

[0207] (3) Increase data diversity:

[0208] The flipping operation essentially provides the model with more training samples without altering the essential characteristics of the hematoma. This data augmentation technique helps increase the diversity of training data, enabling the model to better learn the general characteristics of hematomas, rather than overfitting to specific training samples.

[0209] (4) Improve forecast accuracy:

[0210] By combining the prediction sub-results from multiple flipped input images, more comprehensive prediction information can be obtained. This helps reduce prediction errors from a single viewpoint and improves the overall accuracy of the prediction. Especially when dealing with complex or difficult-to-diagnose hematomas, multi-view prediction results can provide doctors with more reliable references.

[0211] (5) Facilitates model optimization and debugging:

[0212] Flipping, as a simple data transformation method, facilitates debugging and verification during model training and optimization. By observing the prediction results under different flipping directions, it is easier to identify problems and shortcomings in the model, thus enabling targeted improvements and optimizations.

[0213] Optionally, when the post-processing module performs three-dimensional reconstruction of the intracranial hematoma based on the predicted results to obtain an intracranial hematoma model, it performs the following steps:

[0214] The prediction results of intracranial hematoma corresponding to all input images are stitched together to obtain the global prediction result of intracranial hematoma corresponding to the brain axial tomographic grayscale image to be processed.

[0215] The influence matrix diagrams corresponding to all sub-blocks of the brain axial tomographic grayscale image are stitched together to obtain the global influence matrix diagram corresponding to the brain axial tomographic grayscale image to be processed.

[0216] Based on the global influence matrix of the brain axial tomographic grayscale image to be processed, the global prediction result of the corresponding intracranial hematoma is adjusted in three dimensions to obtain the output prediction result of the intracranial hematoma.

[0217] The output prediction result of the intracranial hematoma is restored into a three-dimensional grayscale image;

[0218] Isosurfaces are extracted from the three-dimensional grayscale image, and all extracted isosurfaces are stitched together to generate an intracranial hematoma model.

[0219] Therefore, the technical solution of using the post-processing module to perform three-dimensional reconstruction of the intracranial hematoma based on the prediction results to obtain an intracranial hematoma model has the following technical advantages:

[0220] (1) Accuracy of global prediction results:

[0221] By stitching together the intracranial hematoma prediction results corresponding to all input images, the post-processing module can generate a global prediction result for the intracranial hematoma corresponding to the axial tomographic grayscale image of the brain to be processed. This global approach helps reduce local prediction errors and improve the overall prediction accuracy.

[0222] (2) Effective use of the influence matrix:

[0223] The influence matrix reflects the degree of influence of each voxel on the prediction result. By stitching together the influence matrices corresponding to all sub-blocks of the brain axial tomographic grayscale image, the post-processing module can obtain a global influence matrix, thereby more accurately assessing the contribution of different regions to the prediction of intracranial hematoma.

[0224] (3) Accuracy of three-dimensional weighted adjustment:

[0225] Based on the global influence matrix, the post-processing module performs a three-dimensional weighted adjustment on the global prediction results of intracranial hematoma. This adjustment method can fully consider the influence of different regions, making the final output prediction results more accurate and reliable.

[0226] (4) Restoration of three-dimensional grayscale images:

[0227] Reconstructing the predicted results of intracranial hematoma into a 3D grayscale image helps doctors understand and analyze the predictions more intuitively. This reconstruction method also facilitates subsequent isosurface extraction and model generation.

[0228] (5) Generation of an intracranial hematoma model:

[0229] By extracting isosurfaces from a 3D grayscale image and stitching together all the extracted isosurfaces, the post-processing module can generate an intracranial hematoma model. This model can accurately reflect the morphology and location of the intracranial hematoma, providing strong support for doctors' diagnosis and treatment.

[0230] (6) Surgical planning assistance:

[0231] Intracranial hematoma models can also provide important reference information for surgical planning. Surgeons can develop more reasonable surgical plans based on the shape and location of the model, thereby improving the success rate and safety of the operation.

[0232] Figure 2 This is a schematic diagram of a method for establishing an intracranial hematoma model provided in this embodiment. Figure 3 This is a schematic diagram illustrating the prediction results of intracranial hematoma according to an embodiment of the present invention. Figure 4 This is a schematic diagram of an intracranial hematoma model according to an embodiment of the present invention.

[0233] like Figure 2-4 As shown, the method includes:

[0234] The axial tomographic grayscale image of the brain to be processed is preprocessed to convert it into the input image of the intracranial hematoma segmentation module;

[0235] Feature extraction and voxel classification are performed on the input image to obtain the prediction result of intracranial hematoma;

[0236] Based on the predicted intracranial hematoma, a three-dimensional reconstruction of the intracranial hematoma was performed to obtain an intracranial hematoma model.

[0237] For an exemplary explanation of the above steps in this method for establishing an intracranial hematoma model, please refer to the above... Figure 1 Description of the embodiments.

[0238] This application embodiment also provides an electronic device, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; the memory is used to store at least one executable instruction, the executable instruction causing the processor to perform the following steps:

[0239] The axial tomographic grayscale image of the brain to be processed is preprocessed to convert it into the input image of the intracranial hematoma segmentation module;

[0240] Feature extraction and voxel classification are performed on the input image to obtain the prediction result of intracranial hematoma;

[0241] Based on the predicted intracranial hematoma, a three-dimensional reconstruction of the intracranial hematoma is performed to obtain an intracranial hematoma model, which is then used to control the movement of the robotic arm.

[0242] For an exemplary explanation of the above steps in this electronic device, please refer to the above... Figure 1 Description of the embodiments.

[0243] The aforementioned intracranial hematoma prediction model is, for example, but not limited to, the nnUNet model. To increase model portability, the intracranial hematoma prediction model can be trained in a PyTorch environment. By converting the intracranial hematoma prediction model to ONNX format and porting it to the C++ platform, it can be applied to multiple processing platforms, improving the portability of the preset model. Among them, ONNX (Open Neural Network Exchange) is a multi-framework shared, open protocol neural network exchange format that defines a set of standard formats that are independent of environment and platform, used to enhance the interoperability of various AI models.

[0244] Reference Figure 5 The diagram shows a schematic of an electronic device according to another embodiment of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the electronic device.

[0245] like Figure 5 As shown, the electronic device may include: a processor 402, a communications interface 404, a memory 406 storing a program 410, and a communications bus 408.

[0246] The processor, communication interface, and memory communicate with each other via a communication bus. The communication interface is used to communicate with other electronic devices or servers. The processor executes programs, specifically the steps described in the method embodiments above. Specifically, the program may include program code, which includes computer operation instructions.

[0247] The processor may be a CPU, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a smart device may be of the same type, such as one or more CPUs; or they may be of different types, such as one or more CPUs and one or more ASICs.

[0248] Memory is used to store programs. Memory may include high-speed RAM, and may also include non-volatile memory, such as at least one disk drive.

[0249] Specifically, the program can be used to enable the processor to perform the following operations: segmenting the hematoma image using a sliding window to obtain a segmented image; calculating and obtaining the Gaussian weight map corresponding to the segmented image; multiplying the segmented image and the corresponding Gaussian weight map by voxels to obtain a weighted map of the segmented image; inputting the weighted map of the segmented image into a preset model for processing to obtain a probability map of the segmented image; and fusing the probability map of the segmented image and the corresponding Gaussian weight map to obtain the reconstructed hematoma image.

[0250] The above embodiments are only used to illustrate the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the patent protection scope of the embodiments of the present invention should be defined by the claims. The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or physical entities, or by products with certain functions.

[0251] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this invention, the functions of each unit can be implemented in one or more software and / or hardware components.

[0252] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0253] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0254] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0255] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0256] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, a network interface, and memory. Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0257] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0258] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0259] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0260] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific transactions or implement specific abstract data types. This invention can also be practiced in distributed computing environments where transactions are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0261] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

Claims

1. A surgical robot for cerebral hemorrhage, characterized in that, include: The system includes a control host and a robotic arm. The control host is equipped with a preprocessing module, an intracranial hematoma prediction model, and a post-processing module. The preprocessing module preprocesses the axial tomographic grayscale image of the brain to be processed, converting it into an input image for the intracranial hematoma segmentation module. The intracranial hematoma prediction model extracts features and performs voxel classification on the input image to obtain a prediction result of the intracranial hematoma. The post-processing module performs three-dimensional reconstruction of the intracranial hematoma based on the prediction result to obtain an intracranial hematoma model, thereby controlling the movement of the robotic arm. When the preprocessing module preprocesses the axial tomographic grayscale image of the brain to be processed and converts it into the input image of the intracranial hematoma segmentation module, it performs the following steps: Based on the total number of input channels in the intracranial hematoma segmentation module, the brain axial tomographic grayscale image to be processed is divided into blocks to obtain multiple brain axial tomographic grayscale image sub-blocks, such that the total number of voxels included in each brain axial tomographic grayscale image sub-block is equal to the total number of input channels. Center enhancement processing is performed on each sub-block of the brain axial tomographic grayscale image to obtain an influence matrix. Each brain axial tomographic grayscale image sub-block is enhanced according to the corresponding influence matrix to obtain enhanced brain axial tomographic grayscale image sub-blocks. All enhanced brain axial tomographic grayscale image sub-blocks are fused together as the input image for the intracranial hematoma segmentation module. The intracranial hematoma prediction model is trained according to the following steps to extract features and classify voxels from the input image to obtain the prediction result of intracranial hematoma: Obtain axial tomographic grayscale image samples of the brain and corresponding intracranial hematoma labels. The intracranial hematoma labels indicate that the category of the axial tomographic grayscale image of the brain is background, brain parenchymal hemorrhage, or intraventricular hemorrhage. The brain axial tomographic grayscale image samples were preprocessed to obtain preprocessed image samples. The preprocessed image samples are input into a neural network model for feature extraction and voxel classification to calculate the predicted probability of each intracranial hematoma label. The loss value of the neural network model is calculated based on the predicted probability of the intracranial hematoma label. The network parameters of the neural network model are adjusted based on the loss value, so that the neural network model when the loss value reaches a set loss threshold is used as the intracranial hematoma prediction model.

2. The surgical robot for cerebral hemorrhage according to claim 1, characterized in that, The preprocessing module, based on the total number of input channels in the intracranial hematoma segmentation module, divides the brain axial tomographic grayscale image to be processed into blocks. When multiple brain axial tomographic grayscale image sub-blocks are obtained, the following steps are performed: Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the X direction and the planned size of a single brain axial tomographic grayscale image sub-block in the X direction, calculate the X-direction segmentation parameters when processing the brain axial tomographic grayscale image to be processed into blocks along the X direction. Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the Y direction and the planned size of a single brain axial tomographic grayscale image sub-block in the Y direction, calculate the Y-direction segmentation parameters when processing the brain axial tomographic grayscale image to be processed into blocks along the Y direction. Based on the difference between the size of the brain axial tomographic grayscale image to be processed in the Z direction and the planned size of a single brain axial tomographic grayscale image sub-block in the Z direction, calculate the Z-direction segmentation parameters when segmenting the brain axial tomographic grayscale image to be processed along the Z direction. Based on the X-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the X direction to obtain several X-direction tomographic grayscale image sub-images. Based on the Y-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the Y direction to obtain several Y-direction tomographic grayscale image sub-blocks. Based on the Z-direction segmentation parameters, the brain axial tomographic grayscale image to be processed is segmented along the Z-direction to obtain several Z-direction tomographic grayscale image sub-blocks. Based on the aforementioned X-axis tomographic grayscale image sub-blocks, Y-axis tomographic grayscale image sub-blocks, and Z-axis tomographic grayscale image sub-blocks, the aforementioned multiple brain axial tomographic grayscale image sub-blocks are generated.

3. The surgical robot for cerebral hemorrhage according to claim 1, characterized in that, When the preprocessing module performs center enhancement processing on each sub-block of the brain axial tomographic grayscale image to obtain the influence matrix, it performs the following steps: The dimension for center enhancement processing of each brain axial tomographic grayscale image sub-block is determined, and the direction of center enhancement processing is determined based on the number of said dimensions; Based on the direction of the center enhancement processing, several voxels in the central region of each brain axial tomographic grayscale image sub-block are subjected to influence enhancement processing along the dimension to obtain an influence matrix diagram.

4. The surgical robot for cerebral hemorrhage according to claim 1, characterized in that, The preprocessing module enhances each sub-block of the brain axial tomographic grayscale image according to the corresponding influence matrix diagram to obtain the enhanced brain axial tomographic grayscale image sub-block. The following steps are performed: The gray values ​​of several voxels in the central region of each brain axial tomographic grayscale image sub-block are positively adjusted according to the corresponding influence matrix diagram to obtain an enhanced brain axial tomographic grayscale image sub-block.

5. The surgical robot for cerebral hemorrhage according to claim 1, characterized in that, During the training of the intracranial hematoma prediction model, the axial tomographic grayscale image samples of the brain are preprocessed to obtain preprocessed image samples, including: Spatial consistency correction and grayscale value distribution standardization are performed on the voxels of the brain axial tomographic grayscale image samples to obtain preprocessed image samples.

6. The surgical robot for cerebral hemorrhage according to claim 1, characterized in that, During the training of the intracranial hematoma prediction model, the preprocessed image samples are input into the neural network model for feature extraction and voxel classification to calculate the prediction probability of each intracranial hematoma label, including: The preprocessed image sample is input into the feature extraction module of the neural network model to extract features and obtain feature vector samples. The feature vector samples are input into the output module of the neural network model for three-class classification prediction to obtain three-channel four-dimensional output data. The first dimension of the three-channel four-dimensional data includes the length information of the voxels in the preprocessed image samples, the second dimension includes the width information of the voxels in the preprocessed image samples, the third dimension includes the height information of the voxels in the preprocessed image samples, and the fourth dimension includes three channels, with each channel representing a predicted probability sample of an intracranial hematoma label.

7. The surgical robot for cerebral hemorrhage according to claim 1, characterized in that, When the preprocessing module preprocesses the axial tomographic grayscale image of the brain to be processed and converts it into the input image of the intracranial hematoma segmentation module, it preprocesses each sub-block of the axial tomographic grayscale image of the brain to convert it into the input image of the intracranial hematoma segmentation module. One sub-block of the axial tomographic grayscale image of the brain corresponds to one input image. The intracranial hematoma prediction model performs feature extraction and voxel classification on the input image according to the following steps to obtain the prediction result of intracranial hematoma. One input image corresponds to one prediction result of intracranial hematoma: The feature extraction module in the intracranial hematoma prediction model extracts features from the input image to obtain a feature vector. The feature extraction module in the intracranial hematoma prediction model performs three-class classification prediction based on the feature vector to obtain three-channel four-dimensional output data. The first dimension of the three-channel four-dimensional data includes the length information of voxels in the input image, the second dimension includes the width information of voxels in the input image, the third dimension includes the height information of voxels in the input image, and the fourth dimension includes three channels, each channel representing the prediction probability of an intracranial hematoma type, namely background, parenchymal hemorrhage, and intraventricular hemorrhage. The intracranial hematoma type corresponding to the channel with the highest prediction probability is used as the prediction result of the intracranial hematoma corresponding to the input image.

Images