A neural network model for classification and its applications

Abstract

An embodiment of this specification discloses a neural network model for classification and its application, belonging to the fields of medical imaging and computer technology. The model includes: an input module, a first module, a second module, and an output module; wherein, the input module is used to receive the image data to be processed; the first module is used to extract the first feature of the image data to be processed; the second module is used to extract the second feature of the image data to be processed; the output module is used to perform classification based on the first feature and the second feature of the image data to be processed, and obtain the classification result of the image data to be processed. The input module adopts the neural network model provided by the embodiment of this specification and can be applied to the imaging diagnosis of non-acute occlusion of the middle cerebral artery, providing a reference basis for clinical treatment.

Description

Technical Field

[0001] This specification relates to the fields of medical imaging and computer technology, and particularly to a neural network model for classification and its applications. Background Art

[0002] Convolutional neural network is a type of feedforward neural network that contains convolutional calculations and has a deep structure, and is one of the representative algorithms of deep learning. Convolutional neural networks are widely used in tasks such as image classification and segmentation.

[0003] Non-acute occlusion of intracranial arteries is an important cause of ischemic stroke, accounting for about 10% of all ischemic strokes, and the annual recurrence risk of stroke is 3.6% - 22.0%; among them, occlusion of the middle cerebral artery is clinically common, accounting for 79.6% of occlusive cerebrovascular diseases. At present, the main treatment method for symptomatic non-acute occlusion of intracranial arteries with occlusion time exceeding 24 hours is still drug treatment. Patients with ineffective drug treatment can also undergo extracranial-intracranial bypass surgery and endovascular treatment to reconstruct blood flow.

[0004] Since after the occlusion of the middle cerebral artery, it is impossible to compensate through the primary collateral circulation (Willis circle) and ophthalmic artery, the main compensatory pathway after the occlusion of the middle cerebral artery is the leptomeningeal artery collateral compensation. This compensation has a delay phenomenon, and it is difficult for imaging diagnosis, which further limits the effectiveness of endovascular treatment and increases the occurrence of complications.

[0005] Since the convolutional kernels and pooling kernels of convolutional neural networks both perform local feature extraction, due to the complexity of the compensation after the occlusion of the middle cerebral artery, the existing convolutional neural networks cannot meet the requirements of classifying the middle cerebral artery. Summary of the Invention

[0006] Embodiments of this specification provide a neural network model for classification and its applications to solve the following technical problems: Since the convolutional kernels and pooling kernels of convolutional neural networks both perform local feature extraction, due to the complexity of the compensation after the occlusion of the middle cerebral artery, the existing convolutional neural networks cannot meet the requirements of classifying the middle cerebral artery.

[0007] To solve the above technical problems, the embodiments of this specification are implemented as follows:

[0008] A neural network model for classification provided by an embodiment of this specification, the model includes:

[0009] An input module, a first module, a second module, and an output module;

[0010] Wherein,

[0011] The input module is used to receive the image data to be processed;

[0012] The first module is configured to extract the first feature of the image data to be processed;

[0013] The second module is configured to extract the second feature of the image data to be processed;

[0014] The output module is configured to perform classification based on the first feature and the second feature of the image data to be processed, and obtain the classification result of the image data to be processed.

[0015] Further, the model further includes a preprocessing module, and the preprocessing module is configured to preprocess the image data to be processed.

[0016] Where:

[0017] The preprocessing includes: after removing the skull from the image data to be processed, performing normalization processing to obtain the preprocessed image data for extracting the first feature and the second feature of the image data to be processed.

[0018] Further, the global pooling layer or the one-dimensional processing layer of the output module is configured to flatten the tensors of the first feature and the second feature of the image data to be processed or reconstruct the dimensions of the tensors, and flatten the multi-dimensional vector into a one-dimensional vector.

[0019] Further, it is characterized in that the first module includes a first convolution module and a second convolution module.

[0020] Further, it is characterized in that the first module and the second module are in a series relationship.

[0021] Further, the first layer of the output module is a global pooling layer or a one-dimensional processing layer.

[0022] Further, the output module further includes a fully connected layer.

[0023] Input module. An embodiment of the present specification further provides an application of a neural network model for classification. The neural network model provided by the embodiment of the present specification can be used for classifying image data. The image data is preferably CTA image data or MRA image data, and the classification is to perform middle cerebral artery classification on the image data.

[0024] An embodiment of this specification provides a neural network model for classification, including: an input module, a first module, a second module, and an output module; wherein, the input module is used to receive image data to be processed; the first module is used to extract the first feature of the image data to be processed; the second module is used to extract the second feature of the image data to be processed; the output module is used to perform classification based on the first feature and the second feature of the image data to be processed, and obtain the classification result of the image data to be processed. The neural network model for classification provided by the input module in the embodiment of this specification can extract global features and meet the requirements of the middle cerebral artery classification. Description of the Drawings

[0025] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0026] Figure 1 It is a framework diagram of a neural network model for classification provided by an embodiment of this specification;

[0027] Figure 2 It is another framework diagram of a neural network model for classification provided by an embodiment of this specification;

[0028] Figure 3 It is a schematic diagram of the training process of a neural network model for classification provided by an embodiment of this specification;

[0029] Figure 4 It is a system structure diagram of a classification method provided by Embodiment 1 of this specification;

[0030] Figure 5 It is a specific schematic diagram of a classification method provided by Embodiment 1 of this specification;

[0031] Figure 6 It is a specific schematic diagram of another classification method provided by Embodiment 2 of this specification. Detailed Implementation Modes

[0032] To enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0033] Figure 1 This is a framework diagram of a neural network model for classification provided by the embodiments of this specification. As Figure 1 shown, a neural network model for classification provided by the embodiments of this specification includes:

[0034] An input module, a first module, a second module, and an output module; wherein, the input module is used to receive the image data to be processed; the first module is used to extract the first feature of the image data to be processed; the second module is used to extract the second feature of the image data to be processed; the output module is used to perform classification based on the first feature and the second feature of the image data to be processed, and obtain the classification result of the image data to be processed.

[0035] The input module 101 is used to receive the image data to be processed.

[0036] In the embodiments of this specification, the image data to be processed received by the input module 101 may specifically be CTA or MRA image data, or other image data, such as DSA, or CT perfusion / MRI perfusion imaging or cerebral angiography examination, or high-resolution magnetic resonance (HR-MRI).

[0037] The first module 103 is used to extract the first feature of the image data to be processed.

[0038] In the embodiments of this specification, the first feature of the image data to be processed is mainly the local feature of the image data to be processed, for example

[0039] In the embodiments of this specification, the first module includes a first convolution module and a second convolution module.

[0040] The second module 105 is used to extract the second feature of the image data to be processed.

[0041] In the embodiments of this specification, the second feature of the image data to be processed is mainly the global feature of the image data to be processed.

[0042] In the embodiments of this specification, the first module and the second module are in a series relationship.

[0043] An output module 107 is configured to perform classification based on the first feature and the second feature of the to-be-processed image data, and obtain a classification result of the to-be-processed image data.

[0044] In an embodiment of this specification, the first layer of the output module is a global pooling layer or a one-dimensional processing layer.

[0045] In an embodiment of this specification, the output module further includes a fully connected layer.

[0046] In an embodiment of this specification, the global pooling layer or the one-dimensional processing layer of the output module is configured to flatten the tensors of the first feature and the second feature of the to-be-processed image data or reconstruct the dimensions of the tensors, and flatten the multi-dimensional vectors into one-dimensional vectors. In an embodiment of this specification, the role of the global pooling layer or the one-dimensional processing layer of the first module is to input the local feature and the global feature of the pre-processed image data obtained in the foregoing steps into the output module, and then flatten the tensors of the local feature and the global feature of the pre-processed image data or reconstruct the dimensions of the tensors, and flatten the multi-dimensional vectors into one-dimensional vectors. In a specific implementation process, the flattening process (Flatten) can be performed using the framework of the open-source machine learning platform TensorFlow, or the dimensions of the tensors can be reconstructed using the view function in the framework of the open-source machine learning platform PyTorch. Of course, methods in other open-source machine learning platform frameworks that can flatten multi-dimensional vectors into one-dimensional vectors can also be regarded as the protection scope of this application.

[0047] Figure 2 This is a framework diagram of another neural network model for classification provided in an embodiment of this specification. As Figure 2 shown, in an embodiment of this specification, the neural network model for classification further includes a pre-processing module 209 in addition to the Figure 1 modules shown. The pre-processing module is configured to pre-process the to-be-processed image data.

[0048] Where:

[0049] The pre-processing steps include: after removing the skull from the to-be-processed image data, performing normalization processing to obtain pre-processed image data for extracting the first feature and the second feature of the to-be-processed image data.

[0050] Since there are irrelevant tissues such as the skull in the image data to be processed, in order to ensure the accuracy of subsequent classification, it is necessary to remove the skull from the image data to be processed. In the specific implementation process, removing the skull from the image data to be processed specifically includes: through threshold segmentation, according to the first threshold, extracting the skull from the first image to obtain a skull mask image, and dividing the brain into two parts: inside the skull and outside the skull. In practical applications, the threshold for extracting the skull is >100. Further, pixel points below the second threshold belong to the inside of the skull. The skull is removed from the skull mask image to obtain a tissue mask image after removing the skull. In the specific implementation process, the second threshold can be 80. Removing the skull from the image data to be processed can also adopt other methods for removing the skull, and the specific method of skull removal does not constitute a limitation to this application.

[0051] In the embodiments of this specification, the normalization process includes one or several of coordinate centralization, x-shearing normalization, scaling normalization, or rotation normalization. The normalization process can also adopt other methods, and the specific method of normalization does not constitute a limitation to this application.

[0052] Using the neural network model for classification provided in the embodiments of this specification can achieve the extraction of global features and meet the requirements of the middle cerebral artery classification.

[0053] In the embodiments of this specification, the process of training the neural network model for classification adopts the Figure 3 process shown. Figure 3 is a schematic diagram of the training process of a neural network model for classification provided in the embodiments of this specification. As Figure 3 shown, the training of the neural network model for classification includes:

[0054] Step S301: Divide the obtained learning sample set data into training set data, tuning set data, and test set data.

[0055] In the embodiments of this specification, the learning sample set data is brain image data, specifically CTA or MRA image data, or other image data, such as DSA, or CT perfusion / magnetic resonance perfusion imaging, or cerebral angiography, or high-resolution magnetic resonance (HR-MRI). The learning sample set data is labeled to determine the classification of the middle cerebral artery occlusion in the learning sample set data. The labeled learning sample set data is randomly divided into training set data, tuning set data, and test set data according to a ratio of 5:2:3.

[0056] Step S303: Perform post-normalization processing on the learning sample set data to obtain normalized learning sample set data.

[0057] Normalize the learning sample set data. After resampling, a matrix of 512*512*256 is obtained, which corresponds to the categories of the corresponding middle cerebral artery occlusion classification.

[0058] Step S305: Train the normalized learning sample set data to obtain a neural network model for classification.

[0059] Based on the constructed neural network, select the optimal hyperparameters. Pair the image matrices in the training set data with the corresponding classification categories and input them into the established neural network. First, warm up the model with a lower learning rate, and then gradually increase the learning rate. During the training process, the cross-entropy cost function can be used as the loss function to train the model. During the training process, if the loss function on the tuning data no longer decreases, stop the training to prevent overfitting. For each model structure, use different hyperparameters, and train multiple models under the same initialization conditions. Take the average value of the loss function of the model on the tuning set data as the evaluation index for this set of hyperparameters. Select the hyperparameters with the smallest average loss function as the optimal hyperparameters for each model structure.

[0060] Subsequently, use the test set data to test the neural network model for classification corresponding to the above optimal hyperparameters, and select the optimal neural network model for classification for subsequent classification.

[0061] It should be noted that in the case of insufficient test set data, the cross-validation method can also be used to select the optimal neural network model for classification.

[0062] To further understand the neural network in the training process of the neural network model for classification provided in the embodiments of this specification, the output of the neural network will be described below in combination with specific embodiments. Taking the MCA classification as an example,

[0063] For an image data, the output of the neural network is a vector of length 9, [p1, p2,..., p9], where pi represents the probability that the image data belongs to the i-th type of classification, and p1 + p2 +... + p9 = 1. If p2 = 0.5, p3 = 0.5, and the rest are all 0, then the probability that the image belongs to type 2 [MCA occlusion, M1 main trunk developed, M1 distal bifurcation developed, M2 developed] or type 3 [MCA occlusion, M1 main trunk developed, M1 distal bifurcation developed, M2 not developed] is both 0.5, indicating that M2 may be partially developed.

[0064] The embodiments of this specification also provide an application of a neural network model for classification. The neural network model provided in the embodiments of this specification can be used for the classification of image data. The image data is preferably CTA image data or MRA image data, and the classification is to perform middle cerebral artery classification on the image data.

[0065] To facilitate the understanding of the application of the neural network model for classification, the following will be described in conjunction with specific application embodiments.

[0066] Taking CTA or MRA image data as an example, the following specific embodiments will specifically illustrate the entire operation process. In the embodiments of this specification, the first module and the second module are in a series relationship. Figure 4 This is a system structure diagram of a classification method provided in Embodiment 2 of this specification. As Figure 4 shown, the features extracted by the first convolutional module are input into the second module. After being processed by the second module, they are input into the second convolutional module for further convolution and pooling operations, so as to be used for subsequent output modules to perform classification.

[0067] For a further understanding of Figure 5 the specific implementation process of the system structure diagram of the classification method shown, taking CTA or MRA image data as an example, the following specific embodiments will specifically illustrate the entire operation process. Figure 5 This is a specific schematic diagram of a classification method provided in Embodiment 1 of this specification.

[0068] Embodiment 1

[0069] After the CTA or MRA image data is normalized and resampled, it becomes a matrix of 512*512*256, that is, the image is 256 frames of tomograms, and each tomogram is a grayscale image of 512*512. Therefore, the number of channels is 1. Under the framework of the open-source machine learning platform TensorFlow or under the framework of the open-source machine learning platform PyTorch, the input of the network is a matrix of 512*512*256*1. The first convolution module (within the dashed box) consists of three convolution blocks. Each convolution block has a convolution operation and a pooling operation. Generally, a 3*3*3 convolution kernel is used. After the convolution operation, operations such as dropout (random inactivation), normalization, and activation operations can be performed. The pooling operation can be max pool (maximum pool) or average pool, etc. In the current embodiment, the stride of the pooling operation is 2. The first convolution block uses 32 convolution kernels and obtains 32 feature maps. After the pooling operation, the image size is reduced from 512*512*256 to 256*256*128; the second convolution block uses 64 convolution kernels and obtains 64 feature maps. After the pooling operation, the image size is reduced from 256*256*128 to 128*128*64; the third convolution block uses 128 convolution kernels, obtains features, and obtains 128 feature maps. After the pooling operation, the image size is reduced from 128*128*64 to 64*64*32, that is, the size of the first feature map is 64*64*32, and the number of channels is 128. In the embodiment of this specification, the first feature may be a local feature. It should be particularly noted that the framework of the open-source machine learning platform may also be other frameworks, such as caffe. The specific type of the framework of the open-source machine learning platform does not constitute a limitation to this application.

[0070] In the embodiment of this specification, the first feature of the first convolution module in the first module is input into the second module to obtain a second feature. Since the first module and the second module are in a series relationship, the size of the second feature map obtained by the second module is the same as the size of the feature map of the first convolution module. In the embodiment of this specification, the second module consists of two convolution blocks, both of which use 128 convolution kernels. The first feature of the first convolution module is input into the second module, and a second feature is obtained, and a second feature map with a size of 64*64*32 is obtained. The number of channels of this second feature map is 128. In the embodiment of this specification, the second feature may be a global feature.

[0071] In the embodiment of this specification, the second convolution module consists of only one convolution block. The second convolution module uses 64 convolution kernels. After the second convolution module performs a convolution plus pooling operation, the second feature map with a size of 64*64*32 in the second module is output as a third feature map with a size of 32*32*16, and the number of channels of the third feature map is 64.

[0072] In one embodiment of the present specification, the first convolutional module consists of three convolutional blocks, and each convolutional block has a convolutional operation and a pooling operation. In the convolutional layer, the size of the convolutional kernel is 3*3*3, the stride of the convolutional operation is 1, the number of convolutional kernels in the first convolutional block is 32, the number of convolutional kernels in the second convolutional block is 64, the number of convolutional kernels in the third convolutional block is 128, and the stride of the pooling operation is 2. The second convolutional module consists of one convolutional block, and the number of convolutional kernels therein is 128. In the second module, there are two convolutional blocks, and the number of convolutional kernels in each convolutional block is 128.

[0073] Embodiment 2

[0074] The system structure diagram of the classification method in Embodiment 2 is the same as that in Embodiment 1, and the specific implementation process is as Figure 6 shown. Figure 6 This is a specific schematic diagram of another classification method provided in Embodiment 2 of the present specification.

[0075] The input module, the first convolutional module, the second module, and the second convolutional module are all the same as those in Embodiment 1. The first layer of the output module flattens the output of the previous module into a vector. Since this vector is very long, the classification result is output through two fully connected layers.

[0076] In the embodiments of the present specification, the input module, the first module, the second module, and the output module together constitute the MCA classification model. The MCA classification model is a model obtained by pre-training through a neural network based on brain image data and its corresponding clinical features. It should be noted that the neural network can adopt the model structure, the number of layers of the model, and the number of convolutional kernels shown in Embodiments 1 and 2 of the present specification.

[0077] It should be particularly noted that in the appendix of this specification Figure 5 and 6 in, taking the first convolutional module in Figure 5 as an example, in 64*64*32*128, 64*64*32 represents the size or dimension of each corresponding feature map, 128 represents that the corresponding number of channels is 128, and 128 also represents that the number of convolutional kernels is 128.

[0078] The above describes specific embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the specific order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0079] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of the device, electronic device, and non-volatile computer storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiments.

[0080] The device, electronic device, and non-volatile computer storage medium provided by the embodiments of this specification correspond to the method. Therefore, the device, electronic device, and non-volatile computer storage medium also have beneficial technical effects similar to those of the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding device, electronic device, and non-volatile computer storage medium will not be elaborated here.

[0081] In the 1990s, it was obvious to distinguish whether an improvement in a technology was an improvement in hardware (e.g., improvement in circuit structures such as diodes, transistors, switches, etc.) or an improvement in software (improvement in method flows). However, with the development of technology, many improvements in method flows today can be regarded as direct improvements in hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structures by programming the improved method flows into the hardware circuits. Therefore, it cannot be said that an improvement in a method flow cannot be implemented with hardware entity modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program by themselves to "integrate" a digital system on a single PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called Hardware Description Language (HDL). And there is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones currently are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that as long as the method flow is slightly logically programmed with the above-mentioned several hardware description languages and programmed into the integrated circuit, it is easy to obtain the hardware circuit that implements the logic method flow.

[0082] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to implement the same function by logically programming the method steps so that the controller takes the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0083] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0084] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0085] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, the embodiments of this specification can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0086] This specification is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0087] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0089] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0090] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

[0091] A computer-readable medium includes permanent and non-permanent, removable and non-removable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, 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, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0092] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0093] This specification can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The specification can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0094] Each embodiment in this specification is described in a progressive manner, and the same or similar parts among the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, they are described relatively simply, and the relevant parts can be referred to the description of the method embodiments.

[0095] The above are only examples of this specification and are not intended to limit this application. For those skilled in the art, various modifications and changes can be made to this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the scope of the claims of this application.

Claims

1. A neural network model for typing, characterized in that: The model includes: Input module, first module, second module and output module; the first module includes a first convolution module and a second convolution module, the features extracted by the first convolution module are input into the second module, and after being processed by the second module, they are input into the second convolution module, the output module is composed of a global pooling layer and a fully connected layer, or a one-dimensional processing layer and a fully connected layer, and the first layer of the output module is a global pooling layer or a one-dimensional processing layer; in, The input module is configured to receive image data to be processed, wherein the image data is at least CTA or MRA image data; The first module is used to extract a first feature of the image data to be processed, where the first feature is a local feature of the image data to be processed; The second module is used to extract a second feature of the image data to be processed, where the second feature is a global feature of the image data to be processed; The output module is used to perform typing based on the first feature of the image data to be processed and the second feature of the image data to be processed to obtain a typing result of the image data to be processed, The model also includes a preprocessing module, which is used to preprocess the image data to be processed. in: The preprocessing includes: removing the skull from the image data to be processed, performing normalization processing, and obtaining preprocessed image data to extract the first feature of the image data to be processed and the second feature of the image data to be processed; The global pooling layer or one-dimensional processing layer of the output module is used to flatten the tensors of the first feature of the image data to be processed and the second feature of the image data to be processed or reconstruct the dimensions of the tensors, and flatten the multi-dimensional vector into a one-dimensional vector.

2. An application of a neural network model for classification, characterized in that: The neural network model described in claim 1 can be used for the classification of imaging data, and the imaging data is preferably CTA imaging data or MRA imaging data, and the classification is to classify the imaging data into middle cerebral artery types.

Images