A high altitude pulmonary edema risk prediction method and system based on multi-modal data

Abstract

The present application relates to the technical field of data analysis, in particular to a highland pulmonary edema risk prediction method and system based on multi-modal data, the method or system provided by the present application mainly includes obtaining the first diagnosis image and the first monitoring data of the target current medical record data, the second diagnosis image and the second monitoring data of the historical medical record data; based on the comprehensive score model, the comprehensive score is output, and the judgment threshold is set, and the prediction result is output based on the comprehensive score and the judgment threshold. Through the above method, the influence of the historical medical record data on the current situation is combined, the misjudgment caused by the good current situation is avoided, and the final prediction result is improved for the comprehensive judgment of medical personnel.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and more specifically, to a method and system for predicting the risk of high-altitude pulmonary edema based on multimodal data. Background Technology

[0002] High-altitude pulmonary edema (HAPE) is a life-threatening non-cardiac pulmonary edema commonly seen in individuals rapidly ascending to altitudes above 2500 meters. Its pathophysiological basis is pulmonary hypertension induced by high-altitude hypoxia. Due to the rapid decrease in alveolar oxygen partial pressure, hypoxic pulmonary vasoconstriction (HPV) occurs unevenly, with some vessels excessively constricting. This forces blood flow to areas where blood flow is not constricted or is less constricted, causing a sudden increase in capillary wall pressure and permeability in these areas. Ultimately, protein-rich fluid seeps into the alveolar spaces, forming lung edema.

[0003] Current technologies generally analyze and predict risks based on current patient data, without incorporating historical medical records, which may lead to errors in risk prediction. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for predicting the risk of high-altitude pulmonary edema based on multimodal data, in order to solve the above-mentioned problems in the prior art.

[0005] This invention is achieved through the following technical solution:

[0006] Firstly, a method for predicting the risk of high-altitude pulmonary edema based on multimodal data includes:

[0007] Acquire the first diagnostic image and first monitoring data of the target's current medical record data, and the second diagnostic image and second monitoring data of the historical medical record data;

[0008] Obtain control diagnostic images of pulmonary edema from the database, and output a first evaluation score and a second evaluation score based on the comparison between the control diagnostic images and the first and second diagnostic images.

[0009] The third and fourth evaluation scores are output based on the first and second monitoring data and the normal thresholds of various types of data in the monitoring data.

[0010] The first calculated weight is output based on the first evaluation score and the second evaluation score, and the second calculated weight is output based on the third evaluation score and the fourth evaluation score.

[0011] A comprehensive scoring model is established using the first evaluation score, the second evaluation score, the third evaluation score, the fourth evaluation score, the first calculated weight, and the second calculated weight. A comprehensive score is output based on the comprehensive scoring model, and a judgment threshold is set. A prediction result is output based on the comprehensive score and the judgment threshold.

[0012] Preferably, the comparison between the control diagnostic image and the first and second diagnostic images includes identifying the diagnostic images and extracting pathological features:

[0013] An image recognition network model is established to obtain historical images of patients diagnosed with pulmonary edema. After processing the historical images, a dataset is generated, and the dataset is divided into a training set and a test set.

[0014] Set the training rounds and learning rate, train the image recognition network model using the training set based on the training rounds and learning rate, and output the trained image recognition network model;

[0015] The regions for identifying pathological features in the control diagnostic image, the first diagnostic image, and the second diagnostic image were labeled and extracted using an image recognition network model.

[0016] Based on the area of ​​the pathological feature region and the area of ​​the diagnostic image, the first pathological percentage, the second pathological percentage, and the third pathological percentage of the control diagnostic image, the first diagnostic image, and the second diagnostic image are output respectively.

[0017] Preferably, the establishment of the image recognition network model includes:

[0018] The input layer is used for preprocessing the current image data;

[0019] The intermediate layer is used to receive the results from the input layer and perform feature extraction and nonlinear transformation.

[0020] The output layer maps the high-level features extracted and processed by the intermediate layers to a specific output space.

[0021] The attention mechanism module is used to selectively enhance or suppress local features generated by intermediate layers, focusing the network's attention on intermediate layer features that are beneficial to image classification decisions.

[0022] The preferred attention mechanism module includes a main branch and an attention branch, wherein the attention branch is used to guide feature selection of the main branch;

[0023] The feature maps of the main branches and the attention masks of the attention branches are fused together:

[0024]

[0025] In the formula, To incorporate the attention-based feature maps, For attention masking, This is a feature diagram of the main branches.

[0026] Preferably, the output of the output layer is:

[0027]

[0028]

[0029] In the formula, For feature maps with attention mechanisms, For activation function, For convolution operations, For the first It has been passed. Feature map of the next upsampling operation. As input to the attention branch, The output of the main branch, For splicing operations, For the number of network layers, For the first Layer weights, for The output of the previous layer.

[0030] Preferably, it further includes a penalized cross-entropy loss module, the penalized cross-entropy loss module comprising:

[0031]

[0032]

[0033] In the formula, To penalize the cross-entropy loss function, For the true label of the sample, This is the model's predicted value for this sample. This is a penalty factor.

[0034] Preferably, the step of comparing the control diagnostic image with the first diagnostic image and the second diagnostic image to output a first evaluation score and a second evaluation score includes:

[0035]

[0036]

[0037] In the formula, The first evaluation score, This is the second evaluation score. The proportion of the first pathological pathology This represents the proportion of the second pathological pathology. This represents the third pathological percentage.

[0038] Preferably, the third evaluation score and the fourth evaluation score are output based on the first monitoring data, the second monitoring data, and the normal thresholds of various types of data in the monitoring data;

[0039]

[0040]

[0041] In the formula, This is the third evaluation score. This is the fourth evaluation score. The total number of categories in the first monitoring data. The first monitoring data Class input data, The first monitoring data The maximum threshold value of the input data. The first monitoring data The minimum threshold value of the input data. The total number of categories in the first monitoring data. The second monitoring data Class input data, The second monitoring data The maximum threshold value of the input data. The second monitoring data The minimum threshold value of the input data.

[0042] Preferably, the step of outputting a first calculated weight based on a first evaluation score and a second evaluation score, and outputting a second calculated weight based on a third evaluation score and a fourth evaluation score, includes:

[0043] When the first evaluation score is greater than or equal to the second evaluation score When the third evaluation score is greater than or equal to the fourth evaluation score, ;

[0044] When the first evaluation score is less than the second evaluation score:

[0045]

[0046] When the third evaluation score is less than the fourth evaluation score:

[0047]

[0048] In the formula, As the first calculation weight, The second calculation weight;

[0049] The establishment of the comprehensive scoring model includes:

[0050]

[0051] In the formula, This is a comprehensive score.

[0052] Secondly, the present invention also provides a high-altitude pulmonary edema risk prediction system based on multimodal data, used to execute the above-mentioned high-altitude pulmonary edema risk prediction method based on multimodal data, comprising:

[0053] The data processing module is configured to acquire the first diagnostic image and first monitoring data of the current medical record data of the target, and the second diagnostic image and second monitoring data of the historical medical record data; acquire the control diagnostic image of pulmonary edema in the database, and output the first evaluation score and the second evaluation score based on the comparison between the control diagnostic image and the first and second diagnostic images; and output the third evaluation score and the fourth evaluation score based on the first monitoring data, the second monitoring data and the normal threshold of various types of data in the monitoring data.

[0054] The results output module is configured to output a first calculated weight based on a comparison of the first evaluation score and the second evaluation score, and output a second calculated weight based on the third evaluation score and the fourth evaluation score. A comprehensive scoring model is established based on the first evaluation score, the second evaluation score, the third evaluation score, the fourth evaluation score, the first calculated weight, and the second calculated weight. A comprehensive score is output based on the comprehensive scoring model, and a judgment threshold is set. A prediction result is output based on the comprehensive score and the judgment threshold.

[0055] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0056] The method or system provided by this invention mainly includes acquiring a first diagnostic image and first monitoring data of the target's current medical record data, and a second diagnostic image and second monitoring data of historical medical record data; outputting a comprehensive score based on a comprehensive scoring model, setting a judgment threshold, and outputting a prediction result based on the comprehensive score and the judgment threshold. This method incorporates the influence of historical medical record data on the current situation, avoiding misjudgments due to a favorable current situation, and improving the final prediction result for medical personnel to make a comprehensive judgment. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0059] Figure 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0061] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. The naming or numbering of steps in this application does not imply that the steps in the method flow must be executed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical objective, as long as the same or similar technical effect is achieved.

[0062] Furthermore, the connection, coupling, or communication in this application can be a direct connection, coupling, or communication between related objects, or an indirect connection, coupling, or communication through other devices. Moreover, the connection, coupling, or communication between objects can be electrical or other similar forms, and this application does not impose any limitations on these. Independently described modules or sub-modules may or may not be physically separated; they may be implemented in software or hardware, and some modules or sub-modules may be implemented in software, with the processor calling the software to implement the function of these modules or sub-modules, while other modules or sub-modules may be implemented in hardware, such as through hardware circuits. Furthermore, some or all of the modules can be selected to achieve the purpose of this application's solution according to actual needs.

[0063] Please refer to Figures 1-2 A first aspect, a method for predicting the risk of high-altitude pulmonary edema based on multimodal data, includes:

[0064] S101: Acquire the first diagnostic image and first monitoring data of the target's current medical record data, and the second diagnostic image and second monitoring data of the historical medical record data;

[0065] In this embodiment, both the first diagnostic image and the second diagnostic image are diagnostic images. The diagnostic image is a lung CT image in medical testing. Subsequent data analysis is performed on the CT image. Both the first monitoring data and the second monitoring data are monitoring data. The monitoring data includes oxygen partial pressure, carbon dioxide partial pressure, acid-base balance, blood routine, and biochemical indicators, including basic medical examination data such as white blood cells, electrolytes, and liver and kidney function.

[0066] S102: Obtain control diagnostic images of pulmonary edema from the database, and output a first evaluation score and a second evaluation score based on the comparison between the control diagnostic images and the first and second diagnostic images.

[0067] Among them, typical CT image data of the disease can be selected as the basis for comparison in the diagnostic comparison.

[0068] S103: Output the third and fourth evaluation scores based on the first and second monitoring data and the normal thresholds of various types of data in the monitoring data;

[0069] The normal threshold for each type of data can be obtained directly from existing data.

[0070] S104: Output the first calculated weight based on the first evaluation score and the second evaluation score, and output the second calculated weight based on the third evaluation score and the fourth evaluation score;

[0071] S105: Establish a comprehensive scoring model based on the first evaluation score, second evaluation score, third evaluation score, fourth evaluation score, first calculated weight, and second calculated weight. Output a comprehensive score based on the comprehensive scoring model and set a judgment threshold. Output the prediction result based on the comprehensive score and the judgment threshold.

[0072] The threshold setting can be based on target patients with different disease severity as data collection samples, and different comprehensive scores obtained through this scheme can be used as different thresholds. The comprehensive score is located within the range of the threshold, and the corresponding predicted disease severity result is output.

[0073] The method or system provided by this invention mainly includes acquiring a first diagnostic image and first monitoring data of the target's current medical record data, and a second diagnostic image and second monitoring data of historical medical record data; outputting a comprehensive score based on a comprehensive scoring model, setting a judgment threshold, and outputting a prediction result based on the comprehensive score and the judgment threshold. This method incorporates the influence of historical medical record data on the current situation, avoiding misjudgments due to a favorable current situation, and improving the final prediction result for medical personnel to make a comprehensive judgment.

[0074] In one exemplary embodiment of the present invention, the comparison of the control diagnostic image with the first diagnostic image and the second diagnostic image includes identifying the diagnostic image and extracting pathological features:

[0075] An image recognition network model is established to obtain historical images of patients diagnosed with pulmonary edema. After processing the historical images, a dataset is generated, and the dataset is divided into a training set and a test set.

[0076] Set the training rounds and learning rate, train the image recognition network model using the training set based on the training rounds and learning rate, and output the trained image recognition network model;

[0077] The regions for identifying pathological features in the control diagnostic image, the first diagnostic image, and the second diagnostic image were labeled and extracted using an image recognition network model.

[0078] Based on the area of ​​the pathological feature region and the area of ​​the diagnostic image, the first pathological percentage, the second pathological percentage, and the third pathological percentage of the control diagnostic image, the first diagnostic image, and the second diagnostic image are output respectively.

[0079] The pathological proportion can be interpreted as the ratio of the area of ​​the pathological feature region to the area of ​​the entire diagnostic image, that is, the ratio of the area of ​​the pathological feature region of the lung to the entire lung.

[0080] Specifically, this embodiment preferably uses the Att-ResNet network structure to establish the image recognition network model, including:

[0081] The input layer is used for preprocessing the current image data;

[0082] The intermediate layer is used to receive the results from the input layer and perform feature extraction and nonlinear transformation.

[0083] The output layer maps the high-level features extracted and processed by the intermediate layers to a specific output space.

[0084] The attention mechanism module is used to selectively enhance or suppress local features generated by intermediate layers, focusing the network's attention on intermediate layer features that are beneficial to image classification decisions.

[0085] Attention mechanisms provide a channel for enhancing local information. By using attention mechanisms as feature selectors, local features generated by intermediate layers of the network can be selectively enhanced or suppressed. This allows the network's attention to be focused on intermediate layer features that are beneficial to the classification decision of the image, while suppressing the interference of irrelevant features. This can improve the expressive power of image features to a certain extent.

[0086] Secondly, the attention mechanism module includes a main branch and an attention branch, wherein the attention branch is used to guide the feature selection of the main branch;

[0087] The attention mechanism consists of a main branch and an attention branch. Their main functions are as follows: the main branch is part of the main network and performs normal convolution operations; the attention branch uses the features of the main branch to generate an attention mask through the encoder-decoder structure, which serves as the control gate for the neurons of the main branch to guide the feature selection of the main branch.

[0088] The feature maps of the main branches and the attention masks of the attention branches are fused together:

[0089]

[0090] In the formula, To incorporate the attention-based feature maps, For attention masking, This is a feature diagram of the main branches.

[0091] This residual attention mechanism transforms the attention mechanism from a feature selector into a feature enhancer, selectively enhancing local features in the backbone branches and mitigating the loss of effective information during forward propagation due to erroneous activation of shallow attention. While the residual structure effectively improves attention fusion, it is not suitable for medical image recognition in terms of attention generation. The residual attention mechanism is primarily designed for natural image recognition; the encoder-decoder structure used for attention generation is not specifically designed for medical image recognition.

[0092] Its main drawbacks are:

[0093] (1) Attention is generated by an encoder-decoder structure based on upsampling and downsampling. Although information can be compressed in the encoder structure, the resolution of the feature map is still low when it is restored to its original size in the decoder structure.

[0094] (2) The number of upsampling and downsampling of the encoder and decoder is fixed, which makes the traditional attention structure less effective in compressing and restoring information of multi-scale targets in medical images.

[0095] To address the aforementioned issues, this paper improves the encoder-decoder structure in the attention branch of the residual attention structure, making it more suitable for medical image recognition. Specifically, the improvements made to address the above shortcomings are as follows:

[0096] (1) To address the issue of low resolution when reconstructing feature maps using traditional structures, this paper adds skip connections between corresponding layers of the attention branch encoder-decoder structure. By using skip connections, shallower convolutional layer features can be fused in the decoder structure. These features contain richer low-level information, which will result in more refined attention masks.

[0097] (2) To address the problem that traditional structures struggle to extract multi-scale target features, the improved structure in this paper utilizes a shared encoder structure and adds a decoder structure to achieve multiple encoders and decoders with different upsampling and downsampling times, thereby enhancing feature extraction of targets at different scales in medical images. When two decoders exist, skip connections and a multi-codec structure are added.

[0098] Specifically, the output of the output layer is:

[0099]

[0100]

[0101] In the formula, For feature maps with attention mechanisms, For activation function, For convolution operations, For the first It has been passed. Feature map of the next upsampling operation. As input to the attention branch, The output of the main branch, For splicing operations, For the number of network layers, For the first Layer weights, for The output of the previous layer.

[0102] After pixel-by-pixel activation, feature selection is achieved within the current feature map. Furthermore, due to the addition factor, the generated attention does not directly discard or retain the original features, but rather selectively enhances or suppresses them. This can, to some extent, mitigate the impact of accumulated errors in early feature selection on decision-making as the network deepens.

[0103] This paper proposes a penalized cross-entropy loss method to correct prediction results. The penalized cross-entropy loss module includes:

[0104]

[0105]

[0106] In the formula, To penalize the cross-entropy loss function, For the true label of the sample, This is the model's predicted value for this sample. This is a penalty factor.

[0107] In one exemplary embodiment of the present invention, the step of comparing the control diagnostic image with the first diagnostic image and the second diagnostic image to output a first evaluation score and a second evaluation score includes:

[0108]

[0109]

[0110] In the formula, The first evaluation score, This is the second evaluation score. The proportion of the first pathological pathology This represents the proportion of the second pathological pathology. This represents the third pathological percentage.

[0111] In one exemplary embodiment of the present invention, a third evaluation score and a fourth evaluation score are output based on the first monitoring data, the second monitoring data, and the normal thresholds of various types of data in the monitoring data;

[0112]

[0113]

[0114] In the formula, This is the third evaluation score. This is the fourth evaluation score. The total number of categories in the first monitoring data. The first monitoring data Class input data, The first monitoring data The maximum threshold value of the input data. The first monitoring data The minimum threshold value of the input data. The total number of categories in the first monitoring data. The second monitoring data Class input data, The second monitoring data The maximum threshold value of the input data. The second monitoring data The minimum threshold value of the input data.

[0115] In one exemplary embodiment of the present invention, the step of outputting a first calculated weight based on a first evaluation score and a second evaluation score, and outputting a second calculated weight based on a third evaluation score and a fourth evaluation score, includes:

[0116] When the first evaluation score is greater than or equal to the second evaluation score When the third evaluation score is greater than or equal to the fourth evaluation score, ;

[0117] When the first evaluation score is less than the second evaluation score:

[0118]

[0119] When the third evaluation score is less than the fourth evaluation score:

[0120]

[0121] In the formula, As the first calculation weight, This is the second weight to be calculated.

[0122] Using the method described in this embodiment, the calculation weights are set with different values ​​according to different situations to better reflect the real-world situation. When the first evaluation score is greater than or equal to the second evaluation score or the third evaluation score is greater than or equal to the fourth evaluation score, in order to avoid historical data reducing the severity of the current data, the calculation weights set erase the calculation of historical data and directly use the current data as the judgment standard. When the current data is not as severe as the historical data, the calculation weight of the historical data is increased to focus on the impact of the current situation of the medical history, try to avoid missing other diagnoses, and provide medical staff with more reliable reference data.

[0123] Specifically, the establishment of the comprehensive scoring model includes:

[0124]

[0125] In the formula, This is a comprehensive score.

[0126] Secondly, the present invention also provides a high-altitude pulmonary edema risk prediction system based on multimodal data, used to execute the above-mentioned high-altitude pulmonary edema risk prediction method based on multimodal data, comprising:

[0127] The data processing module is configured to acquire the first diagnostic image and first monitoring data of the current medical record data of the target, and the second diagnostic image and second monitoring data of the historical medical record data; acquire the control diagnostic image of pulmonary edema in the database, and output the first evaluation score and the second evaluation score based on the comparison between the control diagnostic image and the first and second diagnostic images; and output the third evaluation score and the fourth evaluation score based on the first monitoring data, the second monitoring data and the normal threshold of various types of data in the monitoring data.

[0128] The results output module is configured to output a first calculated weight based on a comparison of the first evaluation score and the second evaluation score, and output a second calculated weight based on the third evaluation score and the fourth evaluation score. A comprehensive scoring model is established based on the first evaluation score, the second evaluation score, the third evaluation score, the fourth evaluation score, the first calculated weight, and the second calculated weight. A comprehensive score is output based on the comprehensive scoring model, and a judgment threshold is set. A prediction result is output based on the comprehensive score and the judgment threshold.

[0129] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0130] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. This computer software product, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0131] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the risk of high-altitude pulmonary edema based on multimodal data, characterized in that, include: Acquire the first diagnostic image and first monitoring data of the target's current medical record data, and the second diagnostic image and second monitoring data of the historical medical record data; Obtain control diagnostic images of pulmonary edema from the database, and output a first evaluation score and a second evaluation score based on the comparison between the control diagnostic images and the first and second diagnostic images. The third and fourth evaluation scores are output based on the first and second monitoring data and the normal thresholds of various types of data in the monitoring data. The first calculated weight is output based on the first evaluation score and the second evaluation score, and the second calculated weight is output based on the third evaluation score and the fourth evaluation score. A comprehensive scoring model is established based on the first evaluation score, the second evaluation score, the third evaluation score, the fourth evaluation score, the first calculated weight, and the second calculated weight. A comprehensive score is output based on the comprehensive scoring model, and a judgment threshold is set. A prediction result is output based on the comprehensive score and the judgment threshold. Based on the area of ​​the pathological feature region and the area of ​​the diagnostic image, the first pathological percentage, the second pathological percentage, and the third pathological percentage of the control diagnostic image, the first diagnostic image, and the second diagnostic image are output respectively. Based on the comparison between the control diagnostic image and the first and second diagnostic images, the first evaluation score and the second evaluation score are output, including: In the formula, The first evaluation score, This is the second evaluation score. The proportion of the first pathological pathology This represents the proportion of the second pathological pathology. This represents the proportion of the third pathological category; The third and fourth evaluation scores are output based on the first and second monitoring data and the normal thresholds of various types of data in the monitoring data. In the formula, This is the third evaluation score. This is the fourth evaluation score. The total number of categories in the first monitoring data. The first monitoring data Class input data, The first monitoring data The maximum threshold value of the input data. The first monitoring data The minimum threshold value of the input data. The total number of categories in the first monitoring data. The second monitoring data Class input data, The second monitoring data The maximum threshold value of the input data. The second monitoring data The minimum threshold value of the input data.

2. The method for predicting the risk of high-altitude pulmonary edema based on multimodal data according to claim 1, characterized in that, The comparison of the diagnostic image with the first and second diagnostic images includes identifying the diagnostic images and extracting pathological features. An image recognition network model is established to obtain historical images of patients diagnosed with pulmonary edema. After processing the historical images, a dataset is generated, and the dataset is divided into a training set and a test set. Set the training rounds and learning rate, train the image recognition network model using the training set based on the training rounds and learning rate, and output the trained image recognition network model; The regions for identifying pathological features in the control diagnostic image, the first diagnostic image, and the second diagnostic image were labeled and extracted using an image recognition network model. Based on the area of ​​the pathological feature region and the area of ​​the diagnostic image, the first pathological percentage, the second pathological percentage, and the third pathological percentage of the control diagnostic image, the first diagnostic image, and the second diagnostic image are output respectively.

3. The method for predicting the risk of high-altitude pulmonary edema based on multimodal data according to claim 2, characterized in that, The establishment of the image recognition network model includes: The input layer is used for preprocessing the current image data; The intermediate layer is used to receive the results from the input layer and perform feature extraction and nonlinear transformation. The output layer maps the high-level features extracted and processed by the intermediate layers to a specific output space. The attention mechanism module is used to selectively enhance or suppress local features generated by intermediate layers, focusing the network's attention on intermediate layer features that are beneficial to image classification decisions.

4. The method for predicting the risk of high-altitude pulmonary edema based on multimodal data according to claim 3, characterized in that, The attention mechanism module includes a main branch and an attention branch, wherein the attention branch is used to guide the feature selection of the main branch; The feature maps of the main branches and the attention masks of the attention branches are fused together: In the formula, To incorporate the attention-based feature maps, For attention masking, This is a feature diagram of the main branches.

5. The method for predicting the risk of high-altitude pulmonary edema based on multimodal data according to claim 4, characterized in that, The output of the output layer is: In the formula, For feature maps with attention mechanisms, For activation function, For convolution operations, For the first It has been passed. Feature map of the next upsampling operation. As input to the attention branch, The output of the main branch, For splicing operations, For the number of network layers, For the first Layer weights, for The output of the previous layer.

6. The method for predicting the risk of high-altitude pulmonary edema based on multimodal data according to claim 5, characterized in that, It also includes a penalty cross-entropy loss module, which includes: In the formula, To penalize the cross-entropy loss function, For the true label of the sample, This is the model's predicted value for this sample. This is a penalty factor.

7. The method for predicting the risk of high-altitude pulmonary edema based on multimodal data according to claim 6, characterized in that, The step of outputting a first calculated weight based on a first evaluation score and a second evaluation score, and outputting a second calculated weight based on a third evaluation score and a fourth evaluation score, includes: When the first evaluation score is greater than or equal to the second evaluation score When the third evaluation score is greater than or equal to the fourth evaluation score, ; When the first evaluation score is less than the second evaluation score: When the third evaluation score is less than the fourth evaluation score: In the formula, As the first calculation weight, The second calculation weight; The establishment of the comprehensive scoring model includes: In the formula, This is a comprehensive score.

8. A high-altitude pulmonary edema risk prediction system based on multimodal data, characterized in that, A method for predicting the risk of high-altitude pulmonary edema based on multimodal data as described in any one of claims 1-7 includes: The data processing module is configured to acquire the first diagnostic image and first monitoring data of the current medical record data of the target, and the second diagnostic image and second monitoring data of the historical medical record data; acquire the control diagnostic image of pulmonary edema in the database, and output the first evaluation score and the second evaluation score based on the comparison between the control diagnostic image and the first and second diagnostic images; and output the third evaluation score and the fourth evaluation score based on the first monitoring data, the second monitoring data and the normal threshold of various types of data in the monitoring data. The results output module is configured to output a first calculated weight based on a comparison of the first evaluation score and the second evaluation score, and output a second calculated weight based on the third evaluation score and the fourth evaluation score. A comprehensive scoring model is established based on the first evaluation score, the second evaluation score, the third evaluation score, the fourth evaluation score, the first calculated weight, and the second calculated weight. A comprehensive score is output based on the comprehensive scoring model, and a judgment threshold is set. A prediction result is output based on the comprehensive score and the judgment threshold.

Images