A computer vision-based upper airway muscle group training monitoring method and system
Patent Information
- Application Number
- CN202410559273.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-08
- Publication Date
- 2026-10-23
AI Technical Summary
[0003]针对上述情况，为克服现有技术的缺陷，本发明提供一种基于计算机视觉的上气道肌群训练监测方法及系统，针对现有上气道肌群训练方法只有专业人士指导，效率低下，监测效果较差的问题，本发明采用计算机视觉的方法，对上气道图像进行采集、特征提取与特征分类预测，根据分类预测结果对上气道肌群的训练进行自动的监测，节省人工成本；针对现有的基于计算机视觉的医学图像分割技术在用于上气道肌群训练监测时，对上气道MRI图像分割效果较差，难以根据分割图像对上气道肌群的训练状况进行准确监测的问题，本发明采用基于横向注意力优化的交叉切片注意力方法，使用图像切片之间的特征传递与融合，捕捉MRI图像横切面之间的空间关系，实现对上气道MRI图像的精准分割，得到上气道肌群的图像，再对上气道肌群图像进行特征提取与分类预测，得到上气道肌群的训练状况，实现对上气道肌群训练状况的精准判断，提高监测效率
[0038]（1）针对现有上气道肌群训练方法只有专业人士指导，效率低下，监测效果较差的问题，本发明采用计算机视觉的方法，对上气道图像进行采集、特征提取与特征分类预测，根据分类预测结果对上气道肌群的训练进行自动的监测，节省人工成本；
Smart Images
Figure CN122966493A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical image recognition, specifically a method and system for monitoring upper airway muscle training based on computer vision. Background Technology
[0002] The upper airway muscles, including the hyoid, pharyngeal, and soft palate muscles, are crucial for functions such as breathing, chewing, swallowing, and speech production. Upper airway muscle training is a training method for the respiratory system aimed at strengthening and improving the function and coordination of the upper airway muscles. The main reason for upper airway muscle training is that abnormalities in upper airway muscle function can lead to breathing problems and related health issues. Upper airway muscle training can enhance the strength of upper airway muscles and promote upper airway patency. However, current upper airway muscle training methods rely on professional guidance. Professionals create customized training plans based on individual characteristics and needs. Manual monitoring is inefficient, and the actual monitoring results are greatly influenced by the instructor's skill level, making it difficult to develop the optimal training plan based on each individual's specific situation. Existing computer vision-based medical image segmentation technology has poor segmentation results for upper airway MRI images when used for upper airway muscle training monitoring, making it difficult to accurately monitor the training status of the upper airway muscles based on the segmented images. Summary of the Invention
[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a computer vision-based method and system for monitoring upper airway muscle training. Addressing the problems of existing upper airway muscle training methods requiring professional guidance, resulting in low efficiency and poor monitoring effectiveness, this invention employs computer vision to acquire, extract, and classify upper airway images. Based on the classification and prediction results, it automatically monitors the training of upper airway muscles, saving labor costs. Furthermore, addressing the issue that existing computer vision-based medical image segmentation techniques for upper airway muscle training monitoring often produce poor segmentation results on upper airway MRI images, making it difficult to accurately monitor the training status of upper airway muscles based on segmented images, this invention uses a cross-slice attention method based on lateral attention optimization. By utilizing feature transfer and fusion between image slices, it captures the spatial relationships between cross-sections of MRI images, achieving precise segmentation of upper airway MRI images to obtain images of the upper airway muscles. Then, feature extraction and classification prediction are performed on these images to determine the training status of the upper airway muscles, enabling accurate judgment of the upper airway muscle training status and improving monitoring efficiency.
[0004] The technical solution adopted in this invention is as follows: A method for monitoring upper airway muscle group training based on computer vision, specifically including the following steps:
[0005] Step S1: Image acquisition. Using an MRI scanner, images are acquired from the population that needs upper airway muscle training to obtain MRI images, and all MRI images are divided into training set and test set.
[0006] Step S2: Image slicing. All MRI images in the training set are sliced to obtain image slices. Each MRI image slice includes a central slice and two adjacent slices.
[0007] Step S3: Data preprocessing. Preprocess all image slices to obtain preprocessed image slices;
[0008] Step S4: Image segmentation. The preprocessed image slices are segmented using a cross-slice attention method based on lateral attention optimization to obtain upper airway muscle group images.
[0009] Step S5: Image feature extraction. Feature extraction is performed on the upper airway muscle group image to obtain the features of the upper airway muscle group image.
[0010] Step S6: Feature classification prediction. Using an activation function, the features of the upper airway muscle group image are classified and predicted to obtain the classification prediction results of the upper airway muscle group training.
[0011] Step S7: Model evaluation and optimization. The cross-slice attention method based on lateral attention optimization is evaluated and optimized using a test set.
[0012] Step S8: Training status monitoring. The MRI image is segmented using the optimized cross-slice attention method based on lateral attention optimization to obtain upper airway muscle group images. Feature extraction and classification are performed on the upper airway muscle group images, and the training status of the upper airway muscle group is obtained based on the classification results.
[0013] Furthermore, step S4 specifically includes the following steps:
[0014] Step S41: Use a convolutional neural network to extract features from a center slice and two adjacent slices of the preprocessed image slice, respectively, to obtain a primary center feature with dimensions (H, W, C) and two primary neighbor features with dimensions (H, W, C), where H represents height, W represents width, and C represents the number of channels.
[0015] Step S42: Multi-level feature extraction. Perform secondary and tertiary feature extraction on the primary center feature and primary neighbor feature to obtain one secondary center feature, one tertiary center feature, two secondary neighbor features, and two tertiary neighbor features. The dimensions of all secondary features are (H / 2, W / 2, 2C), and the dimensions of all tertiary features are (H / 4, W / 4, 4C).
[0016] Step S43: Lateral attention feature fusion and intra-slice attention feature fusion. Using one third-level center feature and two third-level neighbor features, lateral attention feature fusion and intra-slice attention feature fusion are performed to obtain two lateral fusion features and one intra-slice fusion feature.
[0017] Furthermore, step S43 specifically includes the following steps:
[0018] Step S431: Perform a 1*1 convolution on a third-level neighbor feature to obtain the convolutional neighbor feature, and perform two 1*1 convolutions on the third-level center feature to obtain the convolutional center feature a1 and the convolutional center feature a2.
[0019] Step S432: Multiply the convolutional neighbor features and the convolutional center feature a1 by matrix multiplication to obtain the primary lateral fusion features;
[0020] Step S433: Multiply the primary lateral fusion feature with the convolution center feature a2 by matrix and perform a 1*1 convolution to obtain the secondary lateral fusion feature;
[0021] Step S434: Merge the secondary lateral fusion feature with the tertiary central feature to obtain a lateral fusion feature;
[0022] Step S435: Repeat steps S431 to S434 to obtain another lateral fusion feature;
[0023] Step S436: Perform three 1*1 convolutions on a third-level central feature to obtain convolutional central feature a3, convolutional central feature a4, and convolutional central feature a5;
[0024] Step S437: Multiply the convolution center feature a3 and the convolution center feature a4 by matrix to obtain the primary intra-slice fusion feature;
[0025] Step S438: Multiply the primary intra-piece fusion feature with the convolution center feature a5 by matrix and perform a 1*1 convolution to obtain the secondary intra-piece fusion feature;
[0026] Step S439: Merge the secondary intra-piece fusion feature with the tertiary central feature to obtain an intra-piece fusion feature;
[0027] Step S44: Fuse one of the two lateral fusion features with the intra-piece fusion feature to obtain the hidden feature. Perform linear projection and downsampling on the hidden feature to obtain the dimensionality-reduced feature.
[0028] Step S45: Use a visual transformer to enhance the dimensionality-reduced features to obtain enhanced features, and then reshape the enhanced features to obtain reshaped features;
[0029] Step S46: Perform convolution-activation-upsampling-concatenation operations sequentially on the reshaped features, tertiary center features, secondary center features, and primary center features to obtain segmentation features;
[0030] Step S47: Restore the segmentation features to obtain an image of the upper airway muscle group.
[0031] The present invention provides an upper airway muscle group training monitoring system based on computer vision, comprising a central control module, a data acquisition module, a data processing module, a cross-slice attention module, and a feature extraction and classification module;
[0032] The central control module directly controls the data acquisition module, data processing module, cross-slicing attention module, and feature extraction and classification module, sets parameters and supervises the operation of all modules, and stores all data generated by all modules.
[0033] The data acquisition module uses an MRI scanner to acquire images of people who need upper airway muscle training, obtain MRI images, and transmit all MRI images to the data processing module.
[0034] The data processing module divides all MRI images into training and testing sets, slices all MRI images, and transmits all image slices to the cross-slice attention module after preprocessing.
[0035] The cross-slice attention module uses a cross-slice attention method based on lateral attention optimization to segment the preprocessed image slices to obtain upper airway muscle group images, and then transmits the upper airway muscle group images to the feature extraction and classification module.
[0036] The feature extraction and classification module extracts and classifies features from the upper airway muscle group images, and obtains the training status of the upper airway muscle group based on the classification results.
[0037] The beneficial results achieved by the present invention using the above solution are as follows:
[0038] (1) In view of the problem that existing upper airway muscle training methods are only guided by professionals, which are inefficient and have poor monitoring results, this invention uses computer vision to collect, extract features and predict features of upper airway images, and automatically monitor the training of upper airway muscle groups based on the classification prediction results, thus saving labor costs.
[0039] (2) In view of the problem that the existing computer vision-based medical image segmentation technology has poor segmentation effect on upper airway MRI images when used for upper airway muscle group training monitoring, and it is difficult to accurately monitor the training status of upper airway muscle groups based on segmented images, this invention adopts a cross-slice attention method based on lateral attention optimization. It uses feature transfer and fusion between image slices to capture the spatial relationship between cross sections of MRI images, realizes accurate segmentation of upper airway MRI images, obtains images of upper airway muscle groups, and then performs feature extraction and classification prediction on the upper airway muscle group images to obtain the training status of upper airway muscle groups, realizes accurate judgment of the training status of upper airway muscle groups, and improves monitoring efficiency. Attached Figure Description
[0040] Figure 1 A flowchart illustrating a computer vision-based upper airway muscle group training monitoring method provided by the present invention;
[0041] Figure 2 A schematic diagram of the cross-slice attention method based on lateral attention optimization provided by the present invention;
[0042] Figure 3 This is a schematic diagram of a computer vision-based upper airway muscle training monitoring system provided by the present invention.
[0043] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0044] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0045] Example 1, see Figure 1 The present invention provides a computer vision-based method for monitoring upper airway muscle training, which specifically includes the following steps:
[0046] Step S1: Image acquisition. Using an MRI scanner, images are acquired from the population that needs upper airway muscle training to obtain MRI images, and all MRI images are divided into training set and test set.
[0047] Step S2: Image slicing. All MRI images in the training set are sliced to obtain image slices. Each MRI image slice includes a central slice and two adjacent slices.
[0048] Step S3: Data preprocessing. Preprocess all image slices to obtain preprocessed image slices;
[0049] Step S4: Image segmentation. The preprocessed image slices are segmented using a cross-slice attention method based on lateral attention optimization to obtain upper airway muscle group images.
[0050] Step S5: Image feature extraction. Feature extraction is performed on the upper airway muscle group image to obtain the features of the upper airway muscle group image.
[0051] Step S6: Feature classification prediction. Using an activation function, the features of the upper airway muscle group image are classified and predicted to obtain the classification prediction results of the upper airway muscle group training.
[0052] Step S7: Model evaluation and optimization. The cross-slice attention method based on lateral attention optimization is evaluated and optimized using a test set.
[0053] Step S8: Training status monitoring. The MRI image is segmented using the optimized cross-slice attention method based on lateral attention optimization to obtain upper airway muscle group images. Feature extraction and classification are performed on the upper airway muscle group images, and the training status of the upper airway muscle group is obtained based on the classification results.
[0054] By performing the above operations, this invention addresses the problems of existing upper airway muscle training methods, which require professional guidance, are inefficient, and have poor monitoring results. It uses computer vision to collect, extract, and classify upper airway images, and automatically monitors the training of upper airway muscles based on the classification and prediction results, thus saving labor costs.
[0055] Example 2, this example is based on the above example, step S3 specifically includes the following steps:
[0056] Step S31: Adjust the image size to a uniform size, adjusting it to 256*256 pixels;
[0057] Step S32: Use min-max to normalize all images to the range [0, 1] to obtain normalized images;
[0058] Step S33: Perform data augmentation processing on the normalized image, including random rotation and translation, and contrast enhancement.
[0059] Example 3, see Figure 2 This embodiment is based on the above embodiment. Step S4 specifically includes the following steps:
[0060] Step S41: Use a convolutional neural network to extract features from a center slice and two adjacent slices of the preprocessed image slice, respectively, to obtain a primary center feature with dimensions (H, W, C) and two primary neighbor features with dimensions (H, W, C), where H represents height, W represents width, and C represents the number of channels.
[0061] Step S42: Multi-level feature extraction. Perform secondary and tertiary feature extraction on the primary center feature and primary neighbor feature to obtain one secondary center feature, one tertiary center feature, two secondary neighbor features, and two tertiary neighbor features. The dimensions of all secondary features are (H / 2, W / 2, 2C), and the dimensions of all tertiary features are (H / 4, W / 4, 4C).
[0062] Step S43: Lateral attention feature fusion and intra-film attention feature fusion. Using one third-level center feature and two third-level neighbor features, lateral attention feature fusion and intra-film attention feature fusion are performed to obtain two lateral fused features and one intra-film fused feature. Specifically, this includes the following steps:
[0063] Step S431: Perform a 1*1 convolution on a third-level neighbor feature to obtain the convolutional neighbor feature, and perform two 1*1 convolutions on the third-level center feature to obtain the convolutional center feature a1 and the convolutional center feature a2.
[0064] Step S432: Multiply the convolutional neighbor features and the convolutional center feature a1 by matrix multiplication to obtain the primary lateral fusion features;
[0065] Step S433: Multiply the primary lateral fusion feature with the convolution center feature a2 by matrix and perform a 1*1 convolution to obtain the secondary lateral fusion feature;
[0066] Step S434: Merge the secondary lateral fusion feature with the tertiary central feature to obtain a lateral fusion feature;
[0067] Step S435: Repeat steps S431 to S434 to obtain another lateral fusion feature;
[0068] Step S436: Perform three 1*1 convolutions on a third-level central feature to obtain convolutional central feature a3, convolutional central feature a4, and convolutional central feature a5;
[0069] Step S437: Multiply the convolution center feature a3 and the convolution center feature a4 by matrix to obtain the primary intra-slice fusion feature;
[0070] Step S438: Multiply the primary intra-piece fusion feature with the convolution center feature a5 by matrix and perform a 1*1 convolution to obtain the secondary intra-piece fusion feature;
[0071] Step S439: Merge the secondary intra-piece fusion feature with the tertiary central feature to obtain an intra-piece fusion feature;
[0072] Step S44: Fuse one of the two lateral fusion features with the intra-piece fusion feature to obtain the hidden feature. Perform linear projection and downsampling on the hidden feature to obtain the dimensionality-reduced feature.
[0073] Step S45: Use a visual transformer to enhance the dimensionality-reduced features to obtain enhanced features, and then reshape the enhanced features to obtain reshaped features;
[0074] Step S46: Perform convolution-activation-upsampling-concatenation operations sequentially on the reshaped features, tertiary center features, secondary center features, and primary center features to obtain segmentation features;
[0075] Step S47: Restore the segmentation features to obtain an image of the upper airway muscle group.
[0076] Example 4: Based on the above examples, this example sets the parameters as follows when training the model using the training set:
[0077] During training, the ADAM optimizer was used with a learning rate of 0.001 and a weight decay of 0.00001. A pre-trained 12-layer visual transformer was selected. The segmentation performance was evaluated using the 95th percentile of the dice coefficients and the Hausdorff distance.
[0078] Example 5, see Figure 3 The present invention provides an upper airway muscle group training monitoring system based on computer vision, comprising a central control module, a data acquisition module, a data processing module, a cross-slice attention module, and a feature extraction and classification module;
[0079] The central control module directly controls the data acquisition module, data processing module, cross-slicing attention module, and feature extraction and classification module, sets parameters and supervises the operation of all modules, and stores all data generated by all modules.
[0080] The data acquisition module uses an MRI scanner to acquire images of people who need upper airway muscle training, obtain MRI images, and transmit all MRI images to the data processing module.
[0081] The data processing module divides all MRI images into training and testing sets, slices all MRI images, and transmits all image slices to the cross-slice attention module after preprocessing.
[0082] The cross-slice attention module uses a cross-slice attention method based on lateral attention optimization to segment the preprocessed image slices to obtain upper airway muscle group images, and then transmits the upper airway muscle group images to the feature extraction and classification module.
[0083] The feature extraction and classification module extracts and classifies features from the upper airway muscle group images, and obtains the training status of the upper airway muscle group based on the classification results.
[0084] By performing the above operations, this invention addresses the problem that existing computer vision-based medical image segmentation techniques, when used for upper airway muscle group training monitoring, have poor segmentation results for upper airway MRI images, making it difficult to accurately monitor the training status of upper airway muscles based on segmented images. This invention employs a cross-slice attention method based on lateral attention optimization, using feature transfer and fusion between image slices to capture the spatial relationships between cross-sections of MRI images. This achieves precise segmentation of upper airway MRI images, obtaining images of the upper airway muscles. Furthermore, feature extraction and classification prediction are performed on these upper airway muscle group images to determine the training status of the upper airway muscles, enabling accurate judgment of the upper airway muscle group training status and improving monitoring efficiency.
[0085] Example 6: This example is based on the above examples, and the details of MRI image acquisition are as follows:
[0086] Using a magnetic resonance imaging (MRI) scanner, images were acquired from 68 individuals who required upper airway muscle training. The MRI images were then sliced, with each slice measuring 320*320 pixels and having an in-plane resolution of 0.50mm-0.75mm. Fifty samples were used as the training set, and 18 samples were used as the test set.
[0087] Example 7: This example is based on the above examples, and the code example for the cross-slicing method is as follows: class CrossSliceAttention(Layer): def__init__(self,num_slices,**kwargs): def build(self, input_shape): def call(self, inputs): queries=self.query_transform(inputs) keys=self.key_transform(inputs) values=self.value_transform(inputs) queries=tf.split(queries,self.num_slices,axis=-1) keys=tf.split(keys,self.num_slices,axis=-1) values=tf.split(values,self.num_slices,axis=-1) attended_slices=[ for i in range(self.num_slices): attended_slice=self.attention([queries[i],keys[i],values[i]]) attended_slices.append(attended_slice) attended=tf.concat(attended_slices,axis=-1) Return attended.
[0088] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
[0090] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A computer vision-based method for monitoring upper airway muscle training, characterized in that: Includes the following steps: Step S1: Image acquisition. Using an MRI scanner, images are acquired from the population that needs upper airway muscle training to obtain MRI images, and all MRI images are divided into training set and test set. Step S2: Image slicing. All MRI images in the training set are sliced to obtain image slices. Each MRI image slice includes a central slice and two adjacent slices. Step S3: Data preprocessing. Preprocess all image slices to obtain preprocessed image slices; Step S4: Image segmentation. The preprocessed image slices are segmented using a cross-slice attention method based on lateral attention optimization to obtain upper airway muscle group images. Step S5: Image feature extraction. Feature extraction is performed on the upper airway muscle group image to obtain the features of the upper airway muscle group image. Step S6: Feature classification prediction. Using an activation function, the features of the upper airway muscle group image are classified and predicted to obtain the classification prediction results of the upper airway muscle group training. Step S7: Model evaluation and optimization. The cross-slice attention method based on lateral attention optimization is evaluated and optimized using a test set. Step S8: Training status monitoring. The MRI image is segmented using the optimized cross-slice attention method based on lateral attention optimization to obtain upper airway muscle group images. Feature extraction and classification are performed on the upper airway muscle group images, and the training status of the upper airway muscle group is obtained based on the classification results.
2. The method for monitoring upper airway muscle group training based on computer vision according to claim 1, characterized in that: Step S4 specifically includes the following steps: Step S41: Use a convolutional neural network to extract features from a center slice and two adjacent slices of the preprocessed image slice, respectively, to obtain a primary center feature and two primary neighbor features; Step S42: Multi-level feature extraction. Perform secondary and tertiary feature extraction on the primary center feature and primary neighbor feature to obtain one secondary center feature, one tertiary center feature, two secondary neighbor features, and two tertiary neighbor features. Step S43: Lateral attention feature fusion and intra-slice attention feature fusion. Using one third-level center feature and two third-level neighbor features, lateral attention feature fusion and intra-slice attention feature fusion are performed to obtain two lateral fusion features and one intra-slice fusion feature. Step S44: Fuse one of the two lateral fusion features with the intra-piece fusion feature to obtain the hidden feature. Perform linear projection and downsampling on the hidden feature to obtain the dimensionality-reduced feature. Step S45: Use a visual transformer to enhance the dimensionality-reduced features to obtain enhanced features, and then reshape the enhanced features to obtain reshaped features; Step S46: Perform convolution-activation-upsampling-concatenation operations sequentially on the reshaped features, tertiary center features, secondary center features, and primary center features to obtain segmentation features; Step S47: Restore the segmentation features to obtain an image of the upper airway muscle group.
3. The method for monitoring upper airway muscle group training based on computer vision according to claim 2, characterized in that: Step S43 specifically includes the following steps: Step S431: Perform a 1*1 convolution on a third-level neighbor feature to obtain the convolutional neighbor feature, and perform two 1*1 convolutions on the third-level center feature to obtain the convolutional center feature a1 and the convolutional center feature a2. Step S432: Multiply the convolutional neighbor features and the convolutional center feature a1 by matrix multiplication to obtain the primary lateral fusion features; Step S433: Multiply the primary lateral fusion feature with the convolution center feature a2 by matrix and perform a 1*1 convolution to obtain the secondary lateral fusion feature; Step S434: Merge the secondary lateral fusion feature with the tertiary central feature to obtain a lateral fusion feature; Step S435: Repeat steps S431 to S434 to obtain another lateral fusion feature; Step S436: Perform three 1*1 convolutions on a third-level central feature to obtain convolutional central feature a3, convolutional central feature a4, and convolutional central feature a5; Step S437: Multiply the convolution center feature a3 and the convolution center feature a4 by matrix to obtain the primary intra-slice fusion feature; Step S438: Multiply the primary intra-piece fusion feature with the convolution center feature a5 by matrix and perform a 1*1 convolution to obtain the secondary intra-piece fusion feature; Step S439: Merge the secondary intra-film fusion feature with the tertiary central feature to obtain an intra-film fusion feature.
4. A computer vision-based upper airway muscle group training monitoring system, used to implement the computer vision-based upper airway muscle group training monitoring method according to any one of claims 1 to 3, characterized in that: A computer vision-based upper airway muscle training monitoring system includes a central control module, a data acquisition module, a data processing module, a cross-slice attention module, and a feature extraction and classification module. The central control module directly controls the data acquisition module, data processing module, cross-slicing attention module, and feature extraction and classification module, sets parameters and supervises the operation of all modules, and stores all data generated by all modules. The data acquisition module uses an MRI scanner to acquire images of people who need upper airway muscle training, obtain MRI images, and transmit all MRI images to the data processing module. The data processing module divides all MRI images into training and testing sets, slices all MRI images, and transmits all image slices to the cross-slice attention module after preprocessing. The cross-slice attention module uses a cross-slice attention method based on lateral attention optimization to segment the preprocessed image slices to obtain upper airway muscle group images, and then transmits the upper airway muscle group images to the feature extraction and classification module. The feature extraction and classification module extracts and classifies features from the upper airway muscle group images, and obtains the training status of the upper airway muscle group based on the classification results.