Information processing method and device for cerebral apoplexy classification

Abstract

The embodiment of the invention discloses an information processing method and device for cerebral apoplexy classification. The method comprises the steps that a data set is preprocessed; inputting the training set into a preset model, and training the model; wherein during training, the preset models comprise CNN-2, VGG-16 and ResNet-50, and hyper-parameter adjustment data of the models are as follows: the learning rate is equal to 0.0001, the maximum number of iterations is equal to 50, and the small batch is equal to 8, 16, 32, 64 and 128; during training, different optimizers are used to train different preset models. According to the method, a pre-trained CNN-2 model, a pre-trained VGG-16 model and a pre-trained ResNet-50 model are adopted, and fine hyper-parameter optimization is carried out, so that the accuracy and reliability of brain CT image classification are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an information processing method and device for stroke classification. Background Art

[0002] For experienced radiologists, quickly and accurately interpreting emergency brain CT images remains a time-consuming and tedious task. Traditional CAD systems may be limited to detecting specific diseases and require independent workstations, while deep learning can handle more complex situations. Existing deep learning models typically use unbalanced and limited amounts of data. Training traditional convolutional neural networks (CNNs) usually requires a large amount of data, and building a large enough dataset can be labor-intensive, time-consuming, or even impossible in some cases. Therefore, the results analyzed by current learning models are unsatisfactory. The symptoms of hemorrhagic stroke and ischemic stroke are similar, and misdiagnosis may lead to the use of the wrong treatment (such as anticoagulants), which can be fatal. It is difficult to understand the "black box" mechanism in the operation of deep learning models, especially when there is a lack of reference standards when adjusting hyperparameters.

[0003] On the basis of relevant technologies, how to improve the accuracy and efficiency of stroke classification is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the present invention is to provide an information processing method and device for stroke classification to solve the technical problems existing in the related art.

[0005] In order to achieve the above-mentioned purpose, according to a first aspect of the present invention, there is provided an information processing method for stroke classification, comprising preprocessing a data set; inputting the training set into a preset model to train the model; wherein, during training, the preset models include CNN-2, VGG-16, ResNet-50, and the hyperparameter adjustment data of these models are learning rate = 0.0001, maximum number of iterations = 50, mini-batch = 8, 16, 32, 64 and 128; during training, different preset models are trained using different optimizers.

[0006] Optionally, the CNN-2 includes 2 convolutional layers, each of which is connected to a pooling layer and a dense layer, the dense layer contains 3 hidden layers, and the dense layer uses a rectified linear unit activation function; the VGG-16 is used for transfer learning, which includes n convolutional layers, each of which is connected to a pooling layer, and the last pooling layer is connected to a dense layer, the convolutional layer is used for a feature extractor, and the dense layer is trained as a classifier; the dense layer uses a rectified linear unit activation function; ResNet-50 adopts a residual network architecture, including multiple convolutional layers, and dense layers are connected after all convolutional layers, wherein the connection between the convolutional layers is specified, and the dense layers connected after all convolutional layers do not use the rectified linear unit activation function.

[0007] Optionally, when training based on the small batch value, using different optimizers to train different preset models includes: using Adam optimizer, SGD optimizer, and RMSProp optimizer to train different preset models respectively, and obtaining the precision, sensitivity, accuracy and F1 score of different preset models under different optimizers.

[0008] Optionally, training is performed at different cycles to obtain the accuracy of different preset models at different cycles.

[0009] Optionally, different preset models are trained using different data scales during training to obtain training accuracy values ​​of different preset models under different data scales.

[0010] Optionally, a model for stroke classification is determined based on data obtained during a training process.

[0011] According to a second aspect of the present invention, there is provided an information processing device for stroke classification, comprising a training set acquisition unit for acquiring a preprocessed training set; a training unit configured to input the training set into a preset model to train the model; wherein, during training, the preset models include CNN-2, VGG-16, and ResNet-50, and the hyperparameter adjustment data of these models are learning rate = 0.0001, maximum number of iterations = 50, and mini-batch = 8, 16, 32, 64, and 128; during training, different preset models are trained using different optimizers.

[0012] According to a third aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method described in any one of the implementations of the first aspect.

[0013] According to a fourth aspect of the present invention, there is provided an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor executes the method described in any one implementation of the first aspect.

[0014] The present embodiment is an information processing method and device for stroke classification, wherein the method includes preprocessing a data set; inputting the training set into a preset model to train the model; wherein, during training, the preset models include CNN-2, VGG-16, and ResNet-50, and the hyperparameter adjustment data of these models are learning rate = 0.0001, maximum number of iterations = 50, and small batches = 8, 16, 32, 64, and 128; during training, different preset models are trained using different optimizers. The use of pretrained CNN-2, VGG-16, and ResNet-50 models, and fine hyperparameter optimization, greatly improves the accuracy and reliability of brain CT image classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 is a flow chart of an information processing method for stroke classification according to an embodiment of the present invention;

[0017] Figure 2 is a schematic diagram of the CNN model structure of an embodiment of the present invention;

[0018] Figure 3 is a schematic diagram of VGG-16 model transfer learning in an embodiment of the present invention;

[0019] Figure 4 Schematic diagram of the ResNet-50 model architecture of an embodiment of the present invention;

[0020] Figure 5 is a schematic diagram of training performance data of an embodiment of the present invention;

[0021] Figure 6 is a schematic diagram of a confusion matrix according to an embodiment of the present invention;

[0022] Figure 7It is an overall logic diagram of the information processing method for stroke classification according to an embodiment of the present invention;

[0023] Figure 8 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present invention described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0027] According to an embodiment of the present invention, an information processing method for stroke classification is provided. Figure 1 As shown, it includes the following steps 101 to 103:

[0028] Step 101: Preprocess the data set.

[0029] As an optional implementation method of this embodiment, preprocessing the data set includes: acquiring normal brain images, images of hemorrhagic patients, and brain images of infarction patients and then labeling them; after cropping the labeled images, downsampling each image; transforming the images to obtain an image set, and dividing the image set into a training set, a verification data set, and a test data set, wherein the changes include horizontal flipping, rotation, translation, scaling, and shearing.

[0030] In this implementation, the dataset contains a large number of healthy people, hemorrhage patients and infarction patients, and these images are annotated. They are randomly divided into training dataset, validation dataset and test dataset at the case-based level.

[0031] After acquiring the dataset, the MicroDicom multi-stage image processing algorithm was used to crop the original image to a standard size of 224 × 224 × 3. MicroDicom is a free medical image viewer that is mainly used to view, edit and analyze medical images, especially images in the Digital Imaging and Communications in Medicine (DICOM) format.

[0032] After cropping, data normalization is performed to downsample the pixel values ​​of the input image to 224×224×3 by simply scaling each channel in the range of 0-1. The purpose is to change the range of pixel intensity values ​​and convert the input image to a more familiar or normal sensory pixel value range. It is faster and more stable for gradient descent.

[0033] After normalization, data augmentation is performed, and a series of transformation procedures are used while retaining the original labels, with the goal of doubling the number of images. Therefore, horizontal flipping, rotation, translation, scaling, and shearing are randomly applied to each image in the dataset, so that the data size ranges from 1 to 2 times the original data.

[0034] Transfer learning after data enhancement, deep learning is data-dependent, and the network needs a lot of information to be fully trained to recognize data patterns. Ideally, the training data and test data are assumed to have the same distribution and characteristics. However, building a large enough dataset can be tedious and time-consuming. In this case, transfer learning from one pre-trained field to another is beneficial and efficient.

[0035] Step 102: Input the training set into a preset model to train the model; wherein, during training, the preset models include CNN-2, VGG-16, and ResNet-50, and the hyperparameter adjustment data of these models are learning rate = 0.0001, maximum number of iterations = 50, and small batches = 8, 16, 32, 64, and 128; during training, different preset models are trained using different optimizers.

[0036] In this step, the hyperparameters of the models were tuned as follows: learning rate = 0.0001, maximum number of iterations = 50, mini-batches = 8, 16, 32, 64, and 128. Early stopping was applied if gradient explosion occurred. The dropout rates for all models were 0.3 and 0.5. For further comparison, ResNet-50 was applied without dropout.

[0037] Based on the verification accuracy of the CNN model trained with different mini-batch values, when the mini-batch size is 8, 16, 32, 64, and 128, the verification accuracy of CNN-2 is 0.9787, 0.9829, 0.9840, 0.9808, and 0.9872, respectively; VGG-16 is 0.8480, 0.8948, 0.9543, 0.9575, and 0.9479, respectively; ResNet-50 with dropout is 0.9606, 0.9659, 0.9734, 0.9681, and 0.9638, respectively; ResNet-50 without dropout is 0.9770, 0.9851, 0.9776, 0.9808, and 0.9872, respectively. Therefore, CNN-2, VGG-16, ResNet-50 with dropout, and ResNet-50 without dropout have the highest accuracy at mini-batch values ​​of 128, 64, 32, and 128, respectively, as shown in Table 1.

[0038]

[0039] As an optional implementation method of this embodiment, the CNN-2 includes 2 convolutional layers, each of which is connected to a pooling layer and a dense layer, the dense layer contains 3 hidden layers, and the dense layer uses a rectified linear unit activation function; the VGG-16 is used for transfer learning, which includes n convolutional layers, each of which is connected to a pooling layer, and the last pooling layer is connected to a dense layer, the convolutional layer is used for a feature extractor, and the dense layer is trained as a classifier; the dense layer uses a rectified linear unit activation function; ResNet-50 adopts a residual network architecture, including multiple convolutional layers, and dense layers are connected after all convolutional layers, wherein the connection between the convolutional layers is specified, and the dense layer connected after all convolutional layers does not use the rectified linear unit activation function.

[0040] In the above implementation, the included CNN models are pre-trained CNN-2, VGG-16, and ResNet-50 from ImageNet.

[0041] like Figure 2As shown in the figure, the CNN-2 model consists of 2 convolutional layers, each followed by a pooling layer and a fully connected layer. The fully connected layer is also called a dense layer. The dense layer of the CNN-2 model contains 3 hidden layers, a total of 50,176 input units and 4 output units. The dense layers of VGG-16 and CNN-2 use the rectified linear unit (ReLU) activation function, while the dense layers of ResNet-50 do not use it.

[0042] like Figure 3 As shown in the figure, a schematic diagram of transfer learning using VGG-16 is shown. The convolutional layer can be used as a feature extractor, and the fully connected layer can be trained as a classifier. Compared with CNN-2, VGG-16 has a simpler architecture, which can be composed of 13 convolutional layers and 3 fully connected layers, using ReLU activation function and L2 regularization technology.

[0043] like Figure 4 As shown in the figure, the ResNet-50 model contains 50 parameter layers and adopts a residual network architecture. The dashed short-circuit connection is used to increase the dimension. ResNet-50 adopts an innovative residual connection design to solve the gradient vanishing problem of deep learning models, allowing the model to be trained to a deeper number of layers, thereby significantly improving performance. The architecture of ResNet-50 consists of 50 convolutional layers and 3 fully connected layers, using the ReLU activation function and batch normalization technology.

[0044] As an optional implementation method of this embodiment, based on the small batch value training, different preset models are trained using different optimizers, including: using Adam optimizer, SGD optimizer, and RMSProp optimizer to train different preset models respectively, to obtain the precision, sensitivity, accuracy and F1 score of different preset models under different optimizers.

[0045] In this optional implementation, the method adds Adam (Adaptive Moment Estimation), Stochastic Gradient Descent (SGD), and Root Mean Square Propagation (RMSProp) respectively. Adam (Adaptive Moment Estimation) is an optimization algorithm that combines momentum and adaptive learning rate characteristics, suitable for weight updates in deep learning. Stochastic Gradient Descent (SGD) is one of the most basic optimization algorithms, which updates model parameters by calculating the gradient of each training sample. Root Mean Square Propagation (RMSProp) is an optimization algorithm for adaptive learning rate, which adjusts the learning rate by exponentially weighted moving average of the squares of recent gradients.

[0046] As shown in Table 2, the performance of CNN models trained with different optimization methods shows that the Adam optimizer is more optimized and can provide the most effective results. Therefore, the Adam optimizer can be used to train all models.

[0047] Table 2

[0048]

[0049] As an optional implementation method of this embodiment, training is performed at different periods to obtain the accuracy of different preset models at different periods.

[0050] In this optional implementation, the training performance data includes training loss, validation loss, and validation accuracy of the selected model at different cycles, as shown in Table 3.

[0051] Table 3

[0052]

[0053] Figure 5 The training performance data shown in the figure shows the training performance data of the selected VGG-16, CNN-2, and ResNet-50 models. The horizontal axis represents the number of training iterations, and the vertical axis represents the accuracy of the model. It can be observed that the ResNet-50 model is trained faster and achieves higher accuracy with fewer iterations.

[0054] also Figure 6The confusion matrices of all models with test data are given, and the confusion matrices of the selected VGG-16, CNN-2, and ResNet-50 models are shown. CVA (Cerebrovascular Accident) means infarction, Hx (Hemorrhage) means hemorrhage, NP (Normal) means normal, and Oth (Others) means others. The number in each cell indicates the number of times that the category is misclassified as another category.

[0055] As an optional implementation method of this embodiment, different preset models are trained using different data scales during training to obtain training accuracy values ​​of different preset models under different data scales.

[0056] In this optional implementation, the dataset sizes are divided into four categories: <1000, 1000-5000, 5000-9000 and >10,000. This method compares the performance of different dataset sizes in different models under the best environment for each model.

[0057] Comparison of different data sizes in training: The training data sizes provided are <1000, 1000-5000, 5000-9000, and >10000. The validation accuracies of CNN-2 are 0.3312, 0.6111, 0.8655, and 0.9872, respectively; VGG-16 is 0.5250, 0.7075, 0.8884, and 0.9575; ResNet-50 is 0.5750, 0.7266, 0.9567, and 0.9872, as shown in Table 4. In general, the accuracy of all included models increases proportionally with the data size, and the accuracy is highest when the data size is >10000. However, for data sizes between 5000-9000 and >10000, the accuracy values ​​of ResNet-50 are very close, which shows that even for data sizes >5000, a high accuracy (0.9567) can be achieved using ResNet-50.

[0058] Table 4

[0059]

[0060] As an optional implementation of this embodiment, a model for stroke classification is determined based on the data obtained during the training process.

[0061] Through the above methods, we can determine a very good model and training parameters. The overall complete process reference Figure 7 The results determined are the third part. First, the CNN model is trained based on different mini-batch values ​​to obtain different verification accuracy, training performance (training performance data), comparison results of different optimization models, and comparison results of different data in training.

[0062] This embodiment achieves the following technical effects:

[0063] Significantly improve classification accuracy and reliability: The use of pre-trained CNN-2, VGG-16 and ResNet-50 models, and after fine hyperparameter optimization, has greatly improved the accuracy and reliability of brain CT image classification. Among them, the overall accuracy of CNN-2 and ResNet-50 models is as high as 0.9872, far exceeding traditional methods, which can effectively reduce the risk of misdiagnosis and missed diagnosis, and provide patients with more accurate diagnosis results.

[0064] Saving computing resources and energy consumption: Thanks to the optimized model, the image classification task can be completed in a shorter time, which significantly reduces the demand for computing resources and energy consumption. Especially when processing large-scale medical data, the advantages of this method are more obvious, which can effectively save the operating costs of medical institutions and reduce the impact on the environment.

[0065] Make full use of data resources: Data normalization and data enhancement techniques are used to effectively expand the size of the data set, and transfer learning methods are used to obtain high-precision classification results even on smaller data sets. This method not only improves data utilization, but also reduces dependence on large-scale data sets, which has important practical significance.

[0066] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0067] According to an embodiment of the present invention, an information processing device for stroke classification is also provided, including a training set acquisition unit for acquiring a preprocessed training set; a training unit is configured to input the training set into a preset model to train the model; wherein, during training, the preset models include CNN-2, VGG-16, ResNet-50, and the hyperparameter adjustment data of these models are learning rate = 0.0001, maximum number of iterations = 50, small batch = 8, 16, 32, 64 and 128; during training, different optimizers are used to train different preset models.

[0068] As an optional implementation method of this embodiment, the CNN-2 includes 2 convolutional layers, each of which is connected to a pooling layer and a dense layer, the dense layer contains 3 hidden layers, and the dense layer uses a rectified linear unit activation function; the VGG-16 is used for transfer learning, which includes n convolutional layers, each of which is connected to a pooling layer, and the last pooling layer is connected to a dense layer, the convolutional layer is used for a feature extractor, and the dense layer is trained as a classifier; the dense layer uses a rectified linear unit activation function; ResNet-50 adopts a residual network architecture, including multiple convolutional layers, and dense layers are connected after all convolutional layers, wherein the connection between the convolutional layers is specified, and the dense layer connected after all convolutional layers does not use the rectified linear unit activation function.

[0069] As an optional implementation method of this embodiment, based on the small batch value training, different preset models are trained using different optimizers, including: using Adam optimizer, SGD optimizer, and RMSProp optimizer to train different preset models respectively, to obtain the precision, sensitivity, accuracy and F1 score of different preset models under different optimizers.

[0070] As an optional implementation method of this embodiment, training is performed at different periods to obtain the accuracy of different preset models at different periods.

[0071] As an optional implementation method of this embodiment, different preset models are trained using different data scales during training to obtain training accuracy values ​​of different preset models under different data scales.

[0072] As an optional implementation of this embodiment, a model for stroke classification is determined based on the data obtained during the training process.

[0073] As an optional implementation method of this embodiment, preprocessing the data set includes: acquiring normal brain images, images of hemorrhagic patients, and brain images of infarction patients and then labeling them; after cropping the labeled images, downsampling each image; transforming the images to obtain an image set, and dividing the image set into a training set, a verification data set, and a test data set, wherein the changes include horizontal flipping, rotation, translation, scaling, and shearing.

[0074] According to an embodiment of the present invention, the present invention also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can implement the method described in any of the above embodiments when executing.

[0075] According to an embodiment of the present invention, the present invention further provides a readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to implement the method described in any of the above embodiments when executed.

[0076] According to an embodiment of the present invention, the present invention further provides a computer program product, which can implement the method described in any of the above embodiments when executed by a processor.

[0077] Figure 8 A schematic block diagram of an example electronic device 300 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices.

[0078] like Figure 8 As shown, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 can also be stored. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0079] A number of components in the electronic device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the electronic device 300 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0080] The computing unit 301 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 301 performs the various methods and processes described above, such as the object matching method. For example, in some embodiments, the object matching method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the method described above may be performed.

[0081] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0082] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0083] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

Claims

1. An information processing method for stroke classification, characterized in that: include Preprocess the dataset; Inputting the training set into a preset model to train the model; Among them, during training, the preset models include CNN-2, VGG-16, and ResNet-50, and the hyperparameter adjustment data of these models are learning rate = 0.0001, maximum number of iterations = 50, and small batch = 8, 16, 32, 64, and 128; during training, different optimizers are used to train different preset models.

2. The information processing method for stroke classification according to claim 1, characterized in that: The CNN-2 includes 2 convolutional layers, each of which is connected to a pooling layer and a dense layer, the dense layer includes 3 hidden layers, and the dense layer uses a rectified linear unit activation function; The VGG-16 is used for transfer learning, which includes n convolutional layers, each convolutional layer is connected to a pooling layer, and the last pooling layer is connected to a dense layer, the convolutional layer is used as a feature extractor, and the dense layer is trained as a classifier; the dense layer uses a rectified linear unit activation function; ResNet-50 adopts a residual network architecture, including multiple convolutional layers, and dense layers are connected after all convolutional layers. The connection between convolutional layers is specified, and the dense layers connected after all convolutional layers do not use the rectified linear unit activation function.

3. The information processing method for brain group classification according to claim 2, characterized in that: When training based on the mini-batch value, using different optimizers to train different preset models includes: The Adam optimizer, SGD optimizer, and RMSProp optimizer were used to train different preset models respectively, and the precision, sensitivity, accuracy, and F1 scores of different preset models under different optimizers were obtained.

4. The information processing method for stroke classification according to claim 3, characterized in that: Training is performed at different cycles to obtain the accuracy of different preset models at different cycles.

5. The information processing method for stroke classification according to claim 4, characterized in that: During training, different data scales are used to train different preset models to obtain training accuracy values ​​of different preset models under different data scales.

6. The information processing method for stroke classification according to claim 5, characterized in that: A model for stroke classification is determined based on the data obtained during the training process.

7. The information processing method for stroke classification according to claim 1, characterized in that: The preprocessing of the dataset includes: Obtain normal brain images, images of hemorrhagic patients, and brain images of infarct patients and then annotate them; After cropping the annotated images, each image is downsampled; The image is transformed to obtain an image set, and the image set is divided into a training set, a validation data set, and a test data set, wherein the changes include horizontal flipping, rotation, translation, scaling, and shearing.

8. An information processing device for stroke classification, characterized in that: include: A training set acquisition unit, used for acquiring a preprocessed training set; A training unit, configured to input the training set into a preset model to train the model; Among them, during training, the preset models include CNN-2, VGG-16, and ResNet-50, and the hyperparameter adjustment data of these models are learning rate = 0.0001, maximum number of iterations = 50, and small batch = 8, 16, 32, 64, and 128; during training, different optimizers are used to train different preset models.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor executes the method described in any one of claims 1 to 7.

Images