Animal bone structure segmentation method and system based on deep learning and storage medium

Abstract

The invention discloses an animal bone structure segmentation method and system based on deep learning and a storage medium, and the method comprises the steps: obtaining a bone structure segmentation data set, and carrying out the preprocessing of the bone structure segmentation data set; a deep neural network is constructed, and the deep neural network comprises an encoder module, a feature aggregation module, an attention module and a decoder module; training a deep neural network according to the preprocessed bone structure segmentation data set; and inputting the CT image data of the imaging target into the trained deep neural network to obtain a bone structure segmentation result of the imaging target. Therefore, for any CT image data, the trained deep neural network can be quickly utilized to obtain a corresponding bone structure segmentation result, so that animal bone structure automatic segmentation with high data processing efficiency and high segmentation precision is realized, and the CT image data can be utilized to evaluate osteoporosis, bone complexity and the like of an imaging target.

Description

Technical Field

[0001] The present invention relates to the field of animal detection technology, and in particular to an animal bone structure segmentation method based on deep learning, an animal bone structure segmentation system based on deep learning, and a computer-readable storage medium. Background Art

[0002] CT is one of the most commonly used imaging techniques in preclinical and clinical research. It provides detailed information about volume, texture, and abnormal changes in internal structures in a high-resolution form. However, since most CT imaging studies generate a large amount of data that needs to be processed, especially in high-resolution imaging studies, complex analysis of large amounts of CT data is more time-consuming than a single CT scan. Therefore, manual segmentation of various bone structures requires a lot of time. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems in the related art to a certain extent. To this end, the first object of the present invention is to propose an animal bone structure segmentation method based on deep learning, which can realize automatic segmentation of animal bone structure with fast data processing efficiency and high segmentation accuracy, so as to evaluate osteoporosis, bone complexity, etc. of the imaging target using CT image data.

[0004] The second object of the present invention is to propose an animal bone structure segmentation system based on deep learning.

[0005] A third object of the present invention is to provide a computer-readable storage medium.

[0006] To achieve the above-mentioned purpose, the first aspect of the present invention proposes an animal bone structure segmentation method based on deep learning, including: obtaining a bone structure segmentation dataset and preprocessing the bone structure segmentation dataset; constructing a deep neural network, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module; training the deep neural network according to the preprocessed bone structure segmentation dataset; inputting the CT image data of the imaging target into the trained deep neural network to obtain the bone structure segmentation result of the imaging target.

[0007] According to the animal bone structure segmentation method based on deep learning of an embodiment of the present invention, a bone structure segmentation data set is obtained, and the bone structure segmentation data set is preprocessed, and then a deep neural network is constructed, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module, and then the deep neural network is trained according to the preprocessed bone structure segmentation data set, and the CT image data of the imaging target is input into the trained deep neural network to obtain the bone structure segmentation result of the imaging target. Thus, for any CT image data, the corresponding bone structure segmentation result can be quickly obtained using the trained deep neural network, thereby realizing the automatic segmentation of animal bone structure with fast data processing efficiency and high segmentation accuracy, so as to facilitate the use of CT image data to evaluate osteoporosis, bone complexity, etc. of the imaging target.

[0008] In addition, the animal bone structure segmentation method based on deep learning according to the above embodiment of the present invention may also have the following additional technical features:

[0009] According to one embodiment of the present invention, the bone structure segmentation dataset is obtained by the following formula: D = {(x1, ..., x i , ..., x n ),(y1,...,y i , ..., y n )}, where x i is the animal CT image data, y i is the animal CT image data x i The corresponding annotation data.

[0010] According to one embodiment of the present invention, the preprocessing of the bone structure segmentation dataset includes: resampling the bone structure segmentation dataset; adjusting the window width and window position of the resampled bone structure segmentation dataset; and normalizing the adjusted bone structure segmentation dataset.

[0011] According to an embodiment of the present invention, the annotation data includes at least one of position data, shape data and size data corresponding to the bone structure.

[0012] According to one embodiment of the present invention, the training of the deep neural network based on the preprocessed bone structure segmentation dataset includes: obtaining training samples from the preprocessed bone structure segmentation dataset; randomly selecting the center point of the training sample, and cropping an image block of a preset size from the center point; obtaining deep neural network training parameters; and training the deep neural network based on the image block of the preset size and the deep neural network training parameters until the deep neural network training is completed.

[0013] According to one embodiment of the present invention, the deep neural network includes at least one encoder module, at least one attention module and at least one decoder module.

[0014] According to one embodiment of the present invention, the encoder module adopts a residual connection structure, wherein the encoder module includes: a first convolution operation unit, a batch normalization operation unit, a ReLU activation function and a maximum pooling operation unit.

[0015] According to one embodiment of the present invention, the attention module includes: a linear interpolation operation unit, a second convolution operation unit, a ReLU activation function and a Sigmoid activation function.

[0016] According to one embodiment of the present invention, the decoder module adopts a residual connection structure, wherein the decoder module includes: a first convolution operation unit, a batch normalization operation unit and a ReLU activation function.

[0017] According to one embodiment of the present invention, the feature aggregation module includes: a first convolution operation unit, a second convolution operation unit, a batch normalization operation unit, a Leaky ReLU activation function, a Flatten operation unit, a fully connected layer unit, a layer normalization operation unit, a SiLU activation function, a forward SSM operation unit, a reverse SSM operation unit and a deformation operation unit.

[0018] To achieve the above-mentioned purpose, the animal bone structure segmentation system based on deep learning proposed in the second aspect of the embodiment of the present invention includes: a preprocessing unit, used to obtain a bone structure segmentation data set and preprocess the bone structure segmentation data set; a construction unit, used to construct a deep neural network, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module; a training unit, used to train the deep neural network according to the preprocessed bone structure segmentation data set; and a segmentation unit, used to input the CT image data of the imaging target into the trained deep neural network to obtain the bone structure segmentation result of the imaging target.

[0019] According to the animal bone structure segmentation system based on deep learning of the embodiment of the present invention, a bone structure segmentation data set is obtained through a preprocessing unit, and the bone structure segmentation data set is preprocessed, and then a deep neural network is constructed through a construction unit, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module, and then the deep neural network is trained according to the preprocessed bone structure segmentation data set through a training unit, and the CT image data of the imaging target is input into the trained deep neural network through the segmentation unit to obtain the bone structure segmentation result of the imaging target. Thus, for any CT image data, the corresponding bone structure segmentation result can be quickly obtained using the trained deep neural network, thereby realizing the automatic segmentation of animal bone structure with fast data processing efficiency and high segmentation accuracy, so as to facilitate the use of CT image data to evaluate osteoporosis, bone complexity, etc. of the imaging target.

[0020] To achieve the above-mentioned purpose, the computer-readable storage medium proposed in the third aspect of the embodiment of the present invention stores an animal bone structure segmentation program based on deep learning. When the animal bone structure segmentation program based on deep learning is executed by the processor, the animal bone structure segmentation method based on deep learning of the above-mentioned embodiment of the present invention is implemented.

[0021] According to the computer-readable storage medium of an embodiment of the present invention, by executing the deep learning-based animal bone structure segmentation program stored thereon, it is possible to achieve automatic segmentation of animal bone structure with fast data processing efficiency and high segmentation accuracy, so as to facilitate the use of CT image data to evaluate osteoporosis, bone complexity, etc. of the imaging target.

[0022] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flowchart of an animal bone structure segmentation method based on deep learning according to an embodiment of the present invention;

[0024] Figure 2 is a flowchart of an animal bone structure segmentation method based on deep learning according to an embodiment of the present invention;

[0025] Figure 3 is a flowchart of an animal bone structure segmentation method based on deep learning according to another embodiment of the present invention;

[0026] Figure 4 is a schematic diagram of the structure of a deep neural network according to an embodiment of the present invention;

[0027] Figure 5is a schematic structural diagram of an encoder module according to an embodiment of the present invention;

[0028] Figure 6 is a schematic diagram of the structure of an attention module according to an embodiment of the present invention;

[0029] Figure 7 is a schematic structural diagram of a decoder module according to an embodiment of the present invention;

[0030] Figure 8 is a schematic diagram of the structure of a feature aggregation module according to an embodiment of the present invention;

[0031] Fig. 9 is a block diagram of an animal bone structure segmentation system based on deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0033] The following describes an animal bone structure segmentation method based on deep learning, an animal bone structure segmentation system based on deep learning, and a computer-readable storage medium according to an embodiment of the present invention with reference to the accompanying drawings.

[0034] Figure 1 4 is a flow chart of an animal bone structure segmentation method based on deep learning according to an embodiment of the present invention.

[0035] Specifically, in some embodiments of the present invention, Figure 1 As shown, the animal bone structure segmentation method based on deep learning includes:

[0036] S101, obtaining a bone structure segmentation dataset, and preprocessing the bone structure segmentation dataset.

[0037] It can be understood that in this embodiment of the present invention, a labeled bone structure segmentation dataset can be obtained in advance through, for example, research institutions, public datasets, etc., and the bone structure segmentation dataset can be divided into a training set, a validation set, and a test set according to any non-zero ratio, and then the bone structure segmentation dataset is preprocessed to obtain a preprocessed bone structure segmentation dataset.

[0038] S102, constructing a deep neural network, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module.

[0039] It can be understood that in this embodiment of the present invention, an encoder module, a feature aggregation module, an attention module and a decoder module can be used to construct a deep neural network, wherein the output end of the encoder module is connected to the attention module and the feature aggregation module respectively, the output end of the feature aggregation module is connected to the attention module, the output end of the attention module is connected to the decoder module, and finally the decoder module outputs the bone structure segmentation result.

[0040] S103, training a deep neural network according to the preprocessed bone structure segmentation dataset.

[0041] It can be understood that in this embodiment of the present invention, the preprocessed bone structure segmentation data set can be used to train a deep neural network, so that the trained deep neural network can directly perform automatic segmentation of animal bone structure with high quality.

[0042] S104, inputting the CT image data of the imaging target into the trained deep neural network to obtain the bone structure segmentation result of the imaging target.

[0043] It can be understood that in this embodiment of the present invention, after the deep neural network training is completed, the CT image data of the imaging target can be input into the trained deep neural network to obtain the corresponding bone structure segmentation result using the trained deep neural network, thereby achieving automatic segmentation of animal bone structure with fast data processing efficiency and high segmentation accuracy, so as to facilitate the use of CT image data to evaluate osteoporosis, bone complexity, etc. of the imaging target.

[0044] Furthermore, in some embodiments of the present invention, the bone structure segmentation dataset is obtained by the following formula:

[0045] D={(x1,...,x i , ..., x n ),(y1,...,y i , ..., y n )},

[0046] Among them, x i is the animal CT image data, y i is the animal CT image data x i The corresponding annotation data.

[0047] It should be noted that, in this embodiment of the present invention, the annotation data refers to annotation data of one or more structures of interest.

[0048] Optionally, in some embodiments of the present invention, the annotation data includes at least one of position data, shape data and size data corresponding to the bone structure.

[0049] For example, in this embodiment of the present invention, the standard data may include at least one of position data, shape data, and size data of trabecular bone, cortical bone, and any structure of interest (eg, bone gap).

[0050] Further, in some embodiments of the present invention, Figure 2 As shown, the bone structure segmentation dataset is preprocessed, including:

[0051] S201, resampling the bone structure segmentation dataset.

[0052] It is understandable that, in this embodiment of the present invention, each animal CT image data and its corresponding annotation data in the bone structure segmentation data can be resampled to obtain a resampled bone structure segmentation data set.

[0053] S202, adjusting the window width and window level of the resampled bone structure segmentation data set.

[0054] It is understandable that, in this embodiment of the present invention, the adjusted bone structure segmentation dataset can be obtained by adjusting the window width and window position of each animal CT image data in the resampled bone structure segmentation dataset to a window width and window position suitable for bones.

[0055] S203, performing a normalization operation on the adjusted bone structure segmentation dataset.

[0056] It can be understood that, in this embodiment of the present invention, the normalization operation can be performed on each animal CT image data in the adjusted bone structure segmentation data set by the following normalization formula:

[0057]

[0058] Among them, mean(x i ) represents x i The mean value, std(x i ) represents x i The standard deviation of .

[0059] Thus, the preprocessed bone segmentation dataset is obtained as shown below:

[0060] D′={(x′1,...,x′ i , ..., x′ n ),(y′1,...,y′ i , ..., y′ n )},

[0061] Among them, x′ n is the animal CT image data after normalization, y′ nis the animal CT image data x after normalization operation i The corresponding annotation data.

[0062] Further, in some embodiments of the present invention, Figure 3 As shown, the deep neural network is trained based on the preprocessed bone structure segmentation dataset, including:

[0063] S301, obtaining training samples from the preprocessed bone structure segmentation dataset.

[0064] It is understandable that in this embodiment of the present invention, the training samples (x′) for training the deep neural network can be selected from the preprocessed bone structure segmentation dataset. i , y′ i ).

[0065] S302, randomly selecting a center point of a training sample, and cutting out an image block of a preset size from the center point.

[0066] It can be understood that in this embodiment of the present invention, by randomly selecting a center point in the training sample and cutting out an image block (px′) of a preset size from the center point, i ,py′ i ), and then, px′ i As the input of the deep neural network, py′ i As the target output of the deep neural network.

[0067] S303, obtaining deep neural network training parameters.

[0068] It can be understood that the deep neural network training parameters may include a loss function, an optimization algorithm, an initial learning rate, and a learning rate adjustment strategy, etc. Specifically, in this embodiment of the present invention, Dice and Hausdorff distances can be set as loss functions, the Adam optimization algorithm can be set as the optimization algorithm, the initial learning rate can be set to 0.01, and the learning rate adjustment strategy can be set to cosine annealing.

[0069] S304: Training the deep neural network according to the image blocks of a preset size and the neural network training parameters until the deep neural network training is completed.

[0070] It can be understood that in this embodiment of the present invention, the deep neural network can be trained with the help of image blocks of preset sizes and neural network training parameters until the deep neural network training is completed, thereby obtaining a trained deep neural network.

[0071] Further, in some embodiments of the present invention, the deep neural network includes at least one encoder module, at least one attention module and at least one decoder module.

[0072] It can be understood that in this embodiment of the present invention, Figure 4 As shown, the number of encoder modules is 3, including: encoder module 1, encoder module 2 and encoder module 3, the number of attention modules is 2, including: attention module 1 and attention module 2, and the number of decoder modules is 2, including: decoder module 1 and decoder module 2.

[0073] Specifically, in the above-mentioned embodiments of the present invention, Figure 4 As shown, the animal CT image data is inputted by the input end of the encoder module 1, the output end of the encoder module 1 is respectively connected to the input ends of the encoder module 2 and the attention module 2, the output end of the encoder module 2 is respectively connected to the input ends of the encoder module 3 and the attention module 2, the output end of the encoder module 3 is connected to the input end of the feature aggregation module, the output end of the feature aggregation module is connected to the input end of the attention module 1, the output end of the attention module 1 is connected to the input end of the decoder module 1, the output end of the decoder module 1 is connected to the input end of the attention module 2, the output end of the attention module 2 is connected to the input end of the decoder module 2, and the output end of the decoder module 2 outputs the bone structure segmentation result corresponding to the animal CT image data.

[0074] In addition, in the above-mentioned embodiments of the present invention, the specific structure of the deep neural network is not comprehensively and specifically listed, but only briefly summarized. Those skilled in the art can make corresponding settings for the number of encoder modules, the number of attention modules, and the number of decoder modules in the deep neural network according to specific actual conditions.

[0075] Further, in some embodiments of the present invention, Figure 5 As shown, the encoder module adopts a residual connection structure, wherein the encoder module includes: a first convolution operation unit, a batch normalization operation unit, a ReLU activation function and a maximum pooling operation unit.

[0076] Specifically, in this embodiment of the present invention, if Figure 5 As shown in the figure, the encoder module adopts a residual connection structure, wherein the animal CT image data is input through the input end of the encoder module. At this time, the feature data (i.e., the data transmitted in the deep neural network) passes through two first convolution operation units, two batch normalization operation units and a ReLU activation function, and is added to the input of the residual connection, and then passes through a ReLU activation function and a maximum pooling operation unit to obtain the feature data finally output by the encoder.

[0077] Optionally, in the above embodiment of the present invention, the first convolution operation unit is a 3×3×3 convolution operation unit.

[0078] In addition, in the above-mentioned embodiments of the present invention, the specific structure of the encoder module is not comprehensively and specifically listed, but only briefly summarized. Those skilled in the art can make corresponding settings for the number of first convolution operation units, batch normalization operation units, ReLU activation functions, and maximum pooling operation units in the encoder module according to specific actual conditions.

[0079] Further, in some embodiments of the present invention, Figure 6 As shown, the attention module includes: a linear interpolation operation unit, a second convolution operation unit, a ReLU activation function and a Sigmoid activation function.

[0080] Specifically, in this embodiment of the present invention, if Figure 6 As shown, the feature data output by the decoder module or the feature aggregation module is output after passing through a linear interpolation operation unit and a second convolution operation unit, and the feature data output by the encoder module is output after passing through a second convolution unit, the two output feature data are summed, and are successively output after passing through a ReLU activation function, a second convolution operation unit and a Sigmoid activation function, and the feature data after the Sigmoid activation function is multiplied by the feature data output from the encoder module to obtain the feature data finally output by the attention module.

[0081] Optionally, in the above embodiment of the present invention, the second convolution operation unit is a 1×1×1 convolution operation unit.

[0082] It should be understood that in some embodiments of the present invention, by using a second convolution operation unit to construct an attention module located between the encoder module and the decoder module, the difference between the decoder input features and the encoder output features can be effectively reduced, thereby improving the accuracy of the bone structure segmentation results of the CT image data.

[0083] In addition, in the above-mentioned embodiments of the present invention, the specific structure of the attention module is not comprehensively and specifically listed, but only briefly summarized. Those skilled in the art can make corresponding settings for the number of linear interpolation operation units, the number of second convolution operation units, the number of ReLU activation functions, and the number of Sigmoid activation functions in the attention module according to specific actual conditions.

[0084] Further, in some embodiments of the present invention, Figure 7 As shown, the decoder module adopts a residual connection structure, wherein the decoder module includes: a first convolution operation unit, a batch normalization operation unit and a ReLU activation function.

[0085] Specifically, in this embodiment of the present invention, if Figure 7 As shown, the feature data first passes through two first convolution operation units, two batch normalization operation units and a ReLU activation function, and then is added to the input feature data of the residual connection, and finally passes through a ReLU activation function to obtain the bone structure segmentation result finally output by the decoder module.

[0086] It should be understood that, in some embodiments of the present invention, by constructing a decoder module using a first convolution operation unit with a residual, the bone structure segmentation result can be gradually restored, the information of the corresponding level in the encoder can be effectively utilized, and the decoding effect of the feature data can be improved.

[0087] In addition, in the above-mentioned embodiments of the present invention, the specific structure of the decoder module is not comprehensively and specifically listed, but only briefly summarized. Those skilled in the art can make corresponding settings for the number of first convolution operation units, the number of batch normalization operation units, and the number of ReLU activation functions in the decoder module according to specific actual conditions.

[0088] Further, in some embodiments of the present invention, Figure 8 As shown, the feature aggregation module includes: a first convolution operation unit, a second convolution operation unit, a batch normalization operation unit, a Leaky ReLU activation function, a Flatten operation unit, a fully connected layer unit, a layer normalization operation unit, a SiLU activation function, a forward SSM operation unit, a reverse SSM operation unit and a deformation operation unit.

[0089] Specifically, in this embodiment of the present invention, if Figure 8 As shown, the feature aggregation module has two branches, and the feature data is first processed by the two branches respectively, wherein one branch is composed of N first convolution operation units, a batch normalization operation unit and a Leaky ReLU activation function, and the other branch is composed of M second convolution units, a batch normalization operation unit and a Leaky ReLU activation function. Then, the processing results of the two branches are added and passed through the Flatten operation unit, and the output of the Flatten operation unit is processed again by the two branches respectively, wherein one branch is composed of a fully connected layer unit, a layer normalization operation unit, a SiLU activation function and a forward SSM operation unit, and the other branch is composed of a fully connected layer unit, a layer normalization operation unit, a SiLU activation function and a reverse SSM operation unit. Finally, the processing results of the two branches are added and passed through the fully connected layer unit and the deformation operation unit in sequence to obtain the feature data after feature aggregation finally output by the feature aggregation module.

[0090] It should be understood that in some embodiments of the present invention, by using the SSM operation to construct a feature aggregation module, the ability of the deep neural network for automatic segmentation of bone structure to aggregate features at various levels and capture the global view is enhanced.

[0091] In addition, in the above-mentioned embodiments of the present invention, the specific structure of the feature aggregation module is not comprehensively and specifically listed, but only briefly summarized. Those skilled in the art can make corresponding settings for the number of first convolution operation units, the number of second convolution operation units, the number of batch normalization operation units, the number of Leaky ReLU activation functions, the number of Flatten operation units, the number of fully connected layer units, the number of layer normalization operation units, the number of SiLU activation functions, the number of forward SSM operation units, the number of reverse SSM operation units, and the number of deformation operation units in the feature aggregation module according to specific actual conditions.

[0092] In summary, according to the animal bone structure segmentation method based on deep learning in an embodiment of the present invention, a bone structure segmentation data set is obtained, and the bone structure segmentation data set is preprocessed, and then a deep neural network is constructed, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module, and then the deep neural network is trained according to the preprocessed bone structure segmentation data set, and the CT image data of the imaging target is input into the trained deep neural network to obtain the bone structure segmentation result of the imaging target. Therefore, for any CT image data, the corresponding bone structure segmentation result can be quickly obtained using the trained deep neural network, thereby realizing the automatic segmentation of animal bone structure with fast data processing efficiency and high segmentation accuracy, so as to facilitate the use of CT image data to evaluate osteoporosis, bone complexity, etc. of the imaging target.

[0093] Fig. 9 is a block diagram of an animal bone structure segmentation system based on deep learning according to an embodiment of the present invention.

[0094] Specifically, in some embodiments of the present invention, Fig. 9 As shown, the animal bone structure segmentation system 100 based on deep learning includes: a preprocessing unit 10, a construction unit 20, a training unit 30 and a segmentation unit 40.

[0095] Among them, the preprocessing unit 10 is used to obtain a bone structure segmentation dataset and preprocess the bone structure segmentation dataset; the construction unit 20 is used to construct a deep neural network, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module; the training unit 30 is used to train the deep neural network according to the preprocessed bone structure segmentation dataset; and the segmentation unit 30 is used to input the CT image data of the imaging target into the trained deep neural network to obtain the bone structure segmentation result of the imaging target.

[0096] Furthermore, in some embodiments of the present invention, the bone structure segmentation dataset is obtained by the following formula:

[0097] D={(x1,...,x i , ..., x n ),(y1,...,y i , ..., y n )},

[0098] Among them, x i is the animal CT image data, y i is the animal CT image data x i The corresponding annotation data.

[0099] Furthermore, in some embodiments of the present invention, the preprocessing unit 10 is also used to resample the bone structure segmentation dataset; adjust the window width and window position of the resampled bone structure segmentation dataset; and perform normalization operation on the adjusted bone structure segmentation dataset.

[0100] Furthermore, in some embodiments of the present invention, the annotation data includes at least one of position data, shape data and size data corresponding to the bone structure.

[0101] Furthermore, in some embodiments of the present invention, the training unit 30 is also used to obtain training samples from the preprocessed bone structure segmentation data set; randomly select the center point of the training sample, and crop an image block of a preset size from the center point; obtain deep neural network training parameters; and train the deep neural network according to the image block of the preset size and the deep neural network training parameters until the deep neural network training is completed.

[0102] Further, in some embodiments of the present invention, the deep neural network includes at least one encoder module, at least one attention module and at least one decoder module.

[0103] Furthermore, in some embodiments of the present invention, the encoder module adopts a residual connection structure, wherein the encoder module includes: a first convolution operation unit, a batch normalization operation unit, a ReLU activation function and a maximum pooling operation unit.

[0104] Furthermore, in some embodiments of the present invention, the attention module includes: a linear interpolation operation unit, a second convolution operation unit, a ReLU activation function and a Sigmoid activation function.

[0105] Furthermore, in some embodiments of the present invention, the decoder module adopts a residual connection structure, wherein the decoder module includes: a first convolution operation unit, a batch normalization operation unit and a ReLU activation function.

[0106] Furthermore, in some embodiments of the present invention, the feature aggregation module includes: a first convolution operation unit, a second convolution operation unit, a batch normalization operation unit, a Leaky ReLU activation function, a Flatten operation unit, a fully connected layer unit, a layer normalization operation unit, a SiLU activation function, a forward SSM operation unit, a reverse SSM operation unit and a deformation operation unit.

[0107] It should be understood that the specific implementation of the animal bone structure segmentation system based on deep learning in the embodiment of the present invention corresponds one to one with the specific implementation of the animal bone structure segmentation method based on deep learning in the aforementioned embodiment of the present invention, and will not be repeated here to reduce redundancy.

[0108] In summary, according to the animal bone structure segmentation system based on deep learning of an embodiment of the present invention, a bone structure segmentation data set is obtained through a preprocessing unit, and the bone structure segmentation data set is preprocessed, and then a deep neural network is constructed through a construction unit, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module, and then the deep neural network is trained according to the preprocessed bone structure segmentation data set through a training unit, and the CT image data of the imaging target is input into the trained deep neural network through a segmentation unit to obtain the bone structure segmentation result of the imaging target. Therefore, for any CT image data, the corresponding bone structure segmentation result can be quickly obtained using the trained deep neural network, thereby realizing automatic segmentation of animal bone structure with fast data processing efficiency and high segmentation accuracy, so as to facilitate the use of CT image data to evaluate osteoporosis, bone complexity, etc. of the imaging target.

[0109] Based on the animal bone structure segmentation method based on deep learning of the aforementioned embodiment of the present invention, the embodiment of the present invention further proposes a computer-readable storage medium, on which a deep learning-based animal bone structure segmentation program is stored. When the deep learning-based animal bone structure segmentation program is executed by a processor, the animal bone structure segmentation method based on deep learning of the aforementioned embodiment of the present invention is implemented.

[0110] It should be understood that the specific implementation of the computer-readable storage medium of the embodiment of the present invention can refer to the specific implementation of the animal bone structure segmentation method based on deep learning in the aforementioned embodiment of the present invention. In order to reduce redundancy, it will not be repeated here.

[0111] In summary, according to the computer-readable storage medium of an embodiment of the present invention, by executing the deep learning-based animal bone structure segmentation program stored thereon, it is possible to achieve automatic segmentation of animal bone structures with fast data processing efficiency and high segmentation accuracy, so as to facilitate the use of CT image data to evaluate osteoporosis, bone complexity, etc. of the imaging target.

[0112] In addition, this technology can be applied to various medical researches, such as preclinical imaging, clinical imaging, etc. The present invention can be used not only for automatic segmentation of animal bone structures, but also for automatic segmentation of human bone structures.

[0113] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0114] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0115] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0116] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “clockwise”, “counterclockwise”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0117] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0118] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0119] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0120] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for segmenting animal bone structure based on deep learning, characterized in that: The method comprises: Acquire a bone structure segmentation dataset, and preprocess the bone structure segmentation dataset; Constructing a deep neural network, wherein the deep neural network includes an encoder module, a feature aggregation module, an attention module and a decoder module; Training the deep neural network according to the preprocessed bone structure segmentation dataset; The CT image data of the imaging target is input into the trained deep neural network to obtain the bone structure segmentation result of the imaging target.

2. The animal bone structure segmentation method based on deep learning according to claim 1, characterized in that: The bone structure segmentation dataset is obtained by the following formula: D={(x1,...,x i ,...,x n ),(y1,...,y i ,...,y n )}, Among them, x i is the animal CT image data, y i is the animal CT image data x i The corresponding annotation data.

3. The animal bone structure segmentation method based on deep learning according to claim 1, characterized in that: The preprocessing of the bone structure segmentation data set includes: Resampling the bone structure segmentation dataset; Adjust the window width and window position of the resampled bone structure segmentation dataset; The adjusted bone structure segmentation dataset is normalized.

4. The animal bone structure segmentation method based on deep learning according to claim 2 or 3, characterized in that: The animal CT image data x i The corresponding annotation data includes at least one of position data, shape data and size data corresponding to the bone structure.

5. The animal bone structure segmentation method based on deep learning according to claim 1, characterized in that: The step of training the deep neural network according to the preprocessed bone structure segmentation data set includes: Acquire training samples from the preprocessed bone structure segmentation dataset; Randomly selecting a center point of the training sample, and cutting out an image block of a preset size from the center point; Get deep neural network training parameters; The deep neural network is trained according to the image block of the preset size and the deep neural network training parameters until the deep neural network training is completed.

6. The animal bone structure segmentation method based on deep learning according to claim 1, characterized in that: The deep neural network includes at least one encoder module, at least one attention module and at least one decoder module.

7. The animal bone structure segmentation method based on deep learning according to claim 6, characterized in that: The encoder module adopts a residual connection structure, wherein the encoder module includes: a first convolution operation unit, a batch normalization operation unit, a ReLU activation function and a maximum pooling operation unit.

8. The animal bone structure segmentation method based on deep learning according to claim 6, characterized in that: The attention module includes: a linear interpolation operation unit, a second convolution operation unit, a ReLU activation function and a Sigmoid activation function.

9. The animal bone structure segmentation method based on deep learning according to claim 6, characterized in that: The decoder module adopts a residual connection structure, wherein the decoder module includes: a first convolution operation unit, a batch normalization operation unit and a ReLU activation function.

10. The animal bone structure segmentation method based on deep learning according to claim 1, characterized in that: The feature aggregation module includes: a first convolution operation unit, a second convolution operation unit, a batch normalization operation unit, a Leaky ReLU activation function, a Flatten operation unit, a fully connected layer unit, a layer normalization operation unit, a SiLU activation function, a forward SSM operation unit, a reverse SSM operation unit and a deformation operation unit.

11. An animal bone structure segmentation system based on deep learning, characterized in that: The system comprises: A preprocessing unit, used for acquiring a bone structure segmentation data set and preprocessing the bone structure segmentation data set; A construction unit, configured to construct a deep neural network, wherein the deep neural network comprises an encoder module, a feature aggregation module, an attention module, and a decoder module; A training unit, configured to train the deep neural network according to the preprocessed bone structure segmentation dataset; and The segmentation unit is used to input the CT image data of the imaging target into the trained deep neural network to obtain the bone structure segmentation result of the imaging target.

12. A computer-readable storage medium, characterized in that: An animal bone structure segmentation program based on deep learning is stored thereon, and when the animal bone structure segmentation program based on deep learning is executed by the processor, the animal bone structure segmentation method based on deep learning according to any one of claims 1-10 is implemented.

Images