Machine vision-based dysphagia risk identification method and system for elderly patients

Abstract

The invention discloses a machine vision-based dysphagia risk identification method and system for an elderly patient, and belongs to the technical field of dysphagia, and the method comprises the following steps: S1, collecting real-time image data of the swallowing of the elderly patient; s2, processing the swallowing real-time image data of the elderly patient, and determining swallowing feature data of the elderly patient; s3, determining an optimal elderly patient dysphagia risk identification model; s4, analyzing and identifying the swallowing characteristic data of the elderly patient, and judging whether the elderly patient has the risk of dysphagia or not; s5, early warning feedback intervention measures are taken in time, and life safety and life quality of the elderly patients are guaranteed. The problem that effective risk identification and timely protection intervention cannot be carried out on dysphagia of old patients in the prior art is solved. According to the invention, effective risk identification and timely protection intervention can be carried out on dysphagia of the elderly patient, and life safety and life quality of the elderly patient can be effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of dysphagia, and in particular to a method and system for identifying dysphagia risks in elderly patients based on machine vision. Background Art

[0002] Dysphagia is a common clinical problem among elderly patients, manifested primarily by a weakened swallowing reflex and delayed swallowing movements. It is often caused by neurological disorders (such as stroke and Parkinson's disease) or aging. Dysphagia not only affects patients' quality of life but can also lead to serious complications such as aspiration pneumonia and malnutrition, and can even be life-threatening.

[0003] Chinese patent publication number CN114469100A discloses a dysphagia treatment workstation, comprising a main unit and multiple separate subunits. The main unit is connected to an electrical stimulation probe handle and a tongue muscle probe handle. The workstation provides comprehensive dysphagia clinical assessment, tongue pressure detection, and electrical stimulation therapy. Furthermore, tongue muscle resistance training can simultaneously train multiple tongue muscles, focusing on maximum average isometric force, target muscle strength, and endurance training. It can also reflect temporal changes in tongue pressure in real time, facilitating dynamic adjustments by therapists. However, this patent suffers from the following drawbacks:

[0004] Existing technologies cannot effectively identify the risks of dysphagia in elderly patients, resulting in the inability to provide timely protective intervention for elderly patients with dysphagia and the inability to effectively protect the life safety and quality of life of elderly patients. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for identifying the risk of swallowing disorders in elderly patients based on machine vision, which can effectively identify the risks of swallowing disorders in elderly patients, provide timely protective intervention for elderly patients with swallowing disorders, effectively protect the life safety and quality of life of elderly patients, and solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for identifying the risk of dysphagia in elderly patients based on machine vision includes the following steps:

[0008] S1: Based on machine vision technology, obtain swallowing facial image data Swa facial image data , swallowing neck image data Swa neck image data and swallowing chest image data Swa chest imagedata ,collecting real-time image data of swallowing of elderly patients based on machine vision;

[0009] S2: De-noising, image enhancement, and feature extraction are performed on the real-time swallowing image data of elderly patients based on machine vision to determine the swallowing feature data of elderly patients based on machine vision;

[0010] S3: Train a deep learning-based risk identification model for dysphagia in elderly patients, test and adjust the model, and determine the optimal risk identification model for dysphagia in elderly patients;

[0011] S4: Analyze and identify swallowing feature data of elderly patients based on machine vision to determine whether elderly patients have swallowing disorder risks and determine the swallowing disorder risk identification results for elderly patients;

[0012] S5: Based on the risk identification results of dysphagia in elderly patients, timely early warning and feedback intervention measures should be taken to ensure the life safety and quality of life of elderly patients.

[0013] Preferably, in S1, real-time swallowing image data of elderly patients based on machine vision is collected, and the following operations are performed:

[0014] Based on machine vision technology, image acquisition devices are deployed in swallowing monitoring scenarios for elderly patients;

[0015] Camera-based facial area analysis of elderly patients when swallowing facial Real-time monitoring and collection are performed to obtain swallowing facial image data of elderly patients Sw facial image data ;

[0016] Camera-based monitoring of the neck area of ​​elderly patients when swallowing neck Real-time monitoring and collection are performed to obtain swallowing neck image data of elderly patients. neck image data ;

[0017] Based on the camera, the chest area of ​​elderly patients is monitored when swallowing. Chestl Real-time monitoring and collection are performed to obtain chest image data of elderly patients swallowing Swa chest image data ;

[0018] Among them, based on the swallowing facial image data of elderly patients, Sw facial image data , swallowing neck image data of elderly patients Swa neckimage data and elderly patients swallowing chest image data Swa chest image data ,determine the real-time image data of swallowing in elderly patients based on machine vision.

[0019] Preferably, in S2, processing the real-time swallowing image data of elderly patients based on machine vision includes:

[0020] Based on wavelet transform, real-time swallowing image data of elderly patients based on machine vision are denoised;

[0021] Among them, the real-time swallowing image data of elderly patients based on machine vision was analyzed on a multi-scale basis based on Symlets wavelet, and the real-time swallowing image data of elderly patients based on machine vision was decomposed into sub-bands of different frequencies;

[0022] The low-frequency subband contains the main structure and smooth area of ​​the real-time swallowing image data of elderly patients based on machine vision, and the high-frequency subband contains the details and noise of the real-time swallowing image data of elderly patients based on machine vision. The coefficients in the high-frequency subband are processed by thresholding to remove the noise in the high-frequency subband while retaining the details of the real-time swallowing image data of elderly patients based on machine vision.

[0023] Preferably, in said S2, processing the real-time image data of swallowing of elderly patients based on machine vision further comprises:

[0024] Perform image enhancement on real-time swallowing image data of elderly patients based on machine vision;

[0025] Among them, based on histogram equalization, the grayscale distribution of the real-time swallowing image data of elderly patients based on machine vision is adjusted to enhance the contrast of the real-time swallowing image data of elderly patients based on machine vision;

[0026] Based on gamma correction, the brightness of real-time swallowing image data of elderly patients based on machine vision is adjusted through nonlinear transformation;

[0027] Based on the Sobel operator, the edge information of the real-time swallowing image data of elderly patients based on machine vision is highlighted through gradient calculation, and the edges of the real-time swallowing image data of elderly patients based on machine vision are enhanced;

[0028] Extract features from real-time swallowing image data of elderly patients based on machine vision;

[0029] Among them, the feature extraction method based on key point detection and image segmentation extracts facial muscle activity features, laryngeal movement trajectory features, swallowing frequency features, swallowing action duration and interval time features related to swallowing of elderly patients from real-time swallowing image data of elderly patients based on machine vision, and determines the swallowing feature data of elderly patients based on machine vision.

[0030] Preferably, according to the test results, the parameters and structure of the deep learning-based elderly patient dysphagia risk identification model are adjusted, and the deep learning-based elderly patient dysphagia risk identification model is continuously iteratively optimized to determine the optimal elderly patient dysphagia risk identification model, including:

[0031] Extracting a first test result related to a model parameter of a deep learning-based risk identification model for elderly patients with dysphagia from the test results, and extracting a second test result related to a model network structure of the deep learning-based risk identification model for elderly patients with dysphagia from the test results;

[0032] Matching the first test result with each parameter in the preset parameter optimization database one by one, thereby extracting the parameter optimization solution with the highest degree of matching with the first test result as the first parameter optimization solution;

[0033] Matching the second test result with each network structure in the preset test-structure-optimization database one by one, thereby extracting the network structure optimization solution with the highest degree of matching with the second test result as the corresponding first network structure optimization solution;

[0034] Optimizing and adjusting the parameters and structure of the deep learning-based elderly patient dysphagia risk identification model according to the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model;

[0035] Determine the current model convergence performance of the risk adjustment model and compare the current model convergence performance with the target model convergence performance of the deep learning-based dysphagia risk identification model for elderly patients;

[0036] If the convergence performance of the current model is greater than that of the target model, the risk adjustment model is judged to be superior to the deep learning-based risk identification model for dysphagia in elderly patients, and the risk adjustment model is used as the benchmark model for deep learning-based risk identification of dysphagia in elderly patients.

[0037] The deep learning-based benchmark model for identifying dysphagia risks in elderly patients was trained based on the training set, and the performance of the model training results of the benchmark model was tested again using the test set. The benchmark model was optimized and adjusted again based on the performance test results, and the optimal dysphagia risk identification model for elderly patients was determined based on the optimization and adjustment results.

[0038] Preferably, the parameters and structure of the deep learning-based elderly patient dysphagia risk identification model are optimized and adjusted according to the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model, including:

[0039] Inputting the first parameter optimization scheme, the first network structure optimization scheme, and the deep learning-based elderly patient dysphagia risk identification model into a preset virtual machine, and simulating and determining the model adaptability of the first parameter optimization scheme and the first network structure optimization scheme with the deep learning-based elderly patient dysphagia risk identification model;

[0040] If the model fitness of the first parameter optimization scheme and the first network structure optimization scheme with the deep learning-based elderly patient dysphagia risk identification model is higher than the preset minimum model fitness, then the deep learning-based elderly patient dysphagia risk identification model is optimized and adjusted based on the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model;

[0041] If the model fitness of the first parameter optimization scheme or the first network structure optimization scheme and the deep learning-based swallowing disorder risk identification model for elderly patients is not higher than the preset minimum model fitness, the first test result corresponding to the first parameter optimization scheme with a fitness not higher than the preset minimum model fitness is matched one by one with each parameter in the parameter optimization database to extract the second parameter optimization scheme with the second highest degree of matching with the first test result, or each network structure in the first network structure optimization scheme with a fitness not higher than the preset minimum model is matched one by one to extract the second network structure optimization scheme with the second highest degree of matching with the second test result, and simulation judgment is performed again to obtain the risk adjustment model.

[0042] Preferably, in S4, determining whether the elderly patient has a risk of dysphagia includes:

[0043] Deploy the optimal elderly patient dysphagia risk identification model in a real-world elderly patient dysphagia risk identification environment based on machine vision;

[0044] Input the swallowing feature data of elderly patients based on machine vision into the optimal elderly patients' swallowing disorder risk identification model. Based on the optimal elderly patients' swallowing disorder risk identification model, the swallowing feature data of elderly patients based on machine vision are analyzed and identified to determine whether the elderly patients have swallowing disorder risks and determine the elderly patients' swallowing disorder risk identification results.

[0045] Among them, the identification result of dysphagia risk in elderly patients is that the elderly patients have dysphagia risk or the elderly patients do not have dysphagia risk.

[0046] Preferably, in S5, taking early warning feedback intervention measures includes:

[0047] When elderly patients are at risk of swallowing disorders, an audible and visual warning message will be automatically issued to remind nursing staff to take protective intervention measures, including adjusting diet, improving the eating environment and conducting swallowing training for elderly patients. Based on the actual swallowing situation of the elderly patients, the protective intervention measures will be flexibly adjusted in conjunction with professional medical staff to ensure the life safety and quality of life of the elderly patients.

[0048] According to another aspect of the present invention, a system for identifying the risk of dysphagia in elderly patients based on machine vision is provided, which is used to implement the above-mentioned method for identifying the risk of dysphagia in elderly patients based on machine vision, comprising:

[0049] Data acquisition module for elderly patients' facial area during swallowing facial , Neck areaAre neck and chest areaAre Chestl Collect and determine real-time swallowing image data of elderly patients based on machine vision;

[0050] A data processing module is used to process the real-time image data of swallowing of elderly patients collected based on machine vision, and determine the swallowing feature data of elderly patients based on machine vision;

[0051] The risk identification module is used to analyze and identify swallowing feature data of elderly patients based on machine vision using deep learning technology, determine whether elderly patients have swallowing disorder risks, and determine the swallowing disorder risk identification results for elderly patients;

[0052] The early warning intervention module is used to take early warning feedback intervention measures in a timely manner to provide protective intervention for elderly patients based on the risk identification results of dysphagia in elderly patients.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention obtains swallowing facial image data Swa of elderly patients based on machine vision facial image data , swallowing neck image data Swa neck image data and swallowing chest image data Swa chest image data, determine the real-time swallowing image data of elderly patients based on machine vision, and process the real-time swallowing image data of elderly patients based on machine vision, determine the swallowing feature data of elderly patients based on machine vision, and analyze and identify the swallowing feature data of elderly patients based on machine vision based on the optimal elderly patients' swallowing disorder risk identification model, judge whether the elderly patients have the risk of swallowing disorder, and determine the risk identification results of elderly patients' swallowing disorder. According to the risk identification results of elderly patients' swallowing disorder, timely early warning feedback intervention measures are taken to ensure the life safety and quality of life of elderly patients. It can effectively identify the risk of swallowing disorder in elderly patients, and can carry out timely protective intervention for elderly patients with swallowing disorder, which can effectively ensure the life safety and quality of life of elderly patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the method for identifying dysphagia risk in elderly patients according to the present invention;

[0056] Figure 2 This is a module diagram of the swallowing disorder risk identification system for elderly patients of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0058] In order to solve the existing problem of inability to effectively identify the risk of dysphagia in elderly patients, resulting in the inability to provide timely protective intervention for elderly patients with dysphagia and the inability to effectively protect the life safety and quality of life of elderly patients, please refer to Figure 1-Figure 2 , this embodiment provides the following technical solutions:

[0059] A method for identifying the risk of dysphagia in elderly patients based on machine vision includes the following steps:

[0060] S1: Based on machine vision technology, obtain swallowing facial image data Swa facial image data , swallowing neck image data Swa neck image data and swallowing chest image data Swa chest image data ,collecting real-time image data of swallowing of elderly patients based on machine vision;

[0061] In this embodiment, real-time swallowing image data of elderly patients is collected based on machine vision, and the following operations are performed:

[0062] Based on machine vision technology, image acquisition devices are deployed in swallowing monitoring scenarios for elderly patients;

[0063] Camera-based facial area analysis of elderly patients when swallowing facial Real-time monitoring and collection are performed to obtain swallowing facial image data of elderly patients Sw facial image data ;

[0064] Camera-based monitoring of the neck area of ​​elderly patients when swallowing neck Real-time monitoring and collection are performed to obtain swallowing neck image data of elderly patients. neck image data ;

[0065] Based on the camera, the chest area of ​​elderly patients is monitored when swallowing. Chestl Real-time monitoring and collection are performed to obtain chest image data of elderly patients swallowing Swa chest image data ;

[0066] Among them, based on the swallowing facial image data of elderly patients, Sw facial image data , Swallowing neck image data of elderly patients neck image data and elderly patients swallowing chest image data Swa chest image data ,determine the real-time image data of swallowing in elderly patients based on machine vision.

[0067] S2: De-noising, image enhancement, and feature extraction are performed on the real-time swallowing image data of elderly patients based on machine vision to determine the swallowing feature data of elderly patients based on machine vision;

[0068] In this embodiment, the real-time swallowing image data of elderly patients based on machine vision is processed, including:

[0069] Based on wavelet transform, real-time swallowing image data of elderly patients based on machine vision are denoised;

[0070] Among them, the real-time swallowing image data of elderly patients based on machine vision was analyzed on a multi-scale basis based on Symlets wavelet, and the real-time swallowing image data of elderly patients based on machine vision was decomposed into sub-bands of different frequencies;

[0071] The low-frequency subband contains the main structure and smooth area of ​​the real-time swallowing image data of elderly patients based on machine vision, and the high-frequency subband contains the details and noise of the real-time swallowing image data of elderly patients based on machine vision. The coefficients in the high-frequency subband are processed by thresholding to remove the noise in the high-frequency subband while retaining the details of the real-time swallowing image data of elderly patients based on machine vision.

[0072] In this embodiment, the processing of real-time swallowing image data of elderly patients based on machine vision also includes:

[0073] Perform image enhancement on real-time swallowing image data of elderly patients based on machine vision;

[0074] Among them, based on histogram equalization, the grayscale distribution of the real-time swallowing image data of elderly patients based on machine vision is adjusted to enhance the contrast of the real-time swallowing image data of elderly patients based on machine vision;

[0075] Based on gamma correction, the brightness of real-time swallowing image data of elderly patients based on machine vision is adjusted through nonlinear transformation;

[0076] Based on the Sobel operator, the edge information of the real-time swallowing image data of elderly patients based on machine vision is highlighted through gradient calculation, and the edges of the real-time swallowing image data of elderly patients based on machine vision are enhanced;

[0077] Extract features from real-time swallowing image data of elderly patients based on machine vision;

[0078] Among them, the feature extraction method based on key point detection and image segmentation extracts facial muscle activity features, laryngeal movement trajectory features, swallowing frequency features, swallowing action duration and interval time features related to swallowing of elderly patients from real-time swallowing image data of elderly patients based on machine vision, and determines the swallowing feature data of elderly patients based on machine vision.

[0079] It should be noted that the completion of swallowing action is judged by tracking the movement of Adam's apple; abnormal swallowing behavior is identified by analyzing the movement pattern of facial muscles; and the normality of swallowing function is evaluated by counting the number and time of swallowing per unit time.

[0080] S3: Train a deep learning-based risk identification model for dysphagia in elderly patients, test and adjust the model, and determine the optimal risk identification model for dysphagia in elderly patients;

[0081] In this embodiment, determining the optimal risk identification model for dysphagia in elderly patients includes:

[0082] Extracting a first test result related to a model parameter of a deep learning-based risk identification model for elderly patients with dysphagia from the test results, and extracting a second test result related to a model network structure of the deep learning-based risk identification model for elderly patients with dysphagia from the test results;

[0083] Matching the first test result with each parameter in the preset parameter optimization database one by one, thereby extracting the parameter optimization solution with the highest degree of matching with the first test result as the first parameter optimization solution;

[0084] Matching the second test result with each network structure in the preset test-structure-optimization database one by one, thereby extracting the network structure optimization solution with the highest degree of matching with the second test result as the corresponding first network structure optimization solution;

[0085] Optimizing and adjusting the parameters and structure of the deep learning-based elderly patient dysphagia risk identification model according to the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model;

[0086] Determine the current model convergence performance of the risk adjustment model and compare the current model convergence performance with the target model convergence performance of the deep learning-based dysphagia risk identification model for elderly patients;

[0087] If the convergence performance of the current model is greater than that of the target model, the risk adjustment model is judged to be superior to the deep learning-based risk identification model for dysphagia in elderly patients, and the risk adjustment model is used as the benchmark model for deep learning-based risk identification of dysphagia in elderly patients.

[0088] The deep learning-based benchmark model for identifying dysphagia risks in elderly patients was trained based on the training set, and the performance of the model training results of the benchmark model was tested again using the test set. The benchmark model was optimized and adjusted again based on the performance test results, and the optimal dysphagia risk identification model for elderly patients was determined based on the optimization and adjustment results.

[0089] In this embodiment, the parameters and structure of the deep learning-based elderly patient dysphagia risk identification model are optimized and adjusted according to the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model, including:

[0090] The first parameter optimization scheme, the first network structure optimization scheme, and the deep learning-based elderly patient dysphagia risk identification model are input into a preset virtual machine, and the model adaptability of the first parameter optimization scheme and the first network structure optimization scheme with the deep learning-based elderly patient dysphagia risk identification model is simulated and judged:

[0091]

[0092] Among them, T is the model fitness of the first parameter optimization scheme and the first network structure optimization scheme with the elderly patient swallowing disorder risk identification model based on deep learning, α is the weight coefficient of parameter matching, β is the weight coefficient of structure matching, γ is the weight coefficient of virtual machine performance, h i is the performance value of the i-th performance indicator in the virtual machine, μ i is the performance conversion coefficient of the i-th performance indicator, t j is the network structure of the jth structure type in the first network structure optimization scheme, f j is the model network structure of the jth structural type corresponding to the deep learning-based elderly patient dysphagia risk identification model, A is the parameter similarity between each parameter value corresponding to the first parameter optimization scheme and the parameter value of the deep learning-based elderly patient dysphagia risk identification model corresponding to the first test result, n is the number of performance indicators in the virtual machine, and m is the number of structural types of the model network structure corresponding to the current model;

[0093] If the model fitness of the first parameter optimization scheme and the first network structure optimization scheme with the deep learning-based elderly patient dysphagia risk identification model is higher than the preset minimum model fitness, then the deep learning-based elderly patient dysphagia risk identification model is optimized and adjusted based on the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model;

[0094] If the model fitness of the first parameter optimization scheme or the first network structure optimization scheme and the deep learning-based swallowing disorder risk identification model for elderly patients is not higher than the preset minimum model fitness, the first test result corresponding to the first parameter optimization scheme with a fitness not higher than the preset minimum model fitness is matched one by one with each parameter in the parameter optimization database to extract the second parameter optimization scheme with the second highest degree of matching with the first test result, or each network structure in the first network structure optimization scheme with a fitness not higher than the preset minimum model is matched one by one to extract the second network structure optimization scheme with the second highest degree of matching with the second test result, and simulation judgment is performed again to obtain the risk adjustment model.

[0095] In this embodiment, the deep learning-based swallowing disorder risk identification model for elderly patients is a model constructed using deep learning technology, which is used to identify and predict whether elderly patients have the risk of swallowing disorders by processing and analyzing the swallowing-related data of elderly patients.

[0096] In this embodiment, model parameters refer to the numerical values ​​that need to be trained and adjusted in deep learning models, such as weights and biases. Model parameters need to be optimized through the training process so that the model can better complete the task.

[0097] In this embodiment, the model network structure refers to the architecture of the deep learning model, including the model's hierarchical structure and the connections between layers. Different network structures process data differently, which affects model performance. For example, optimizing the model network structure includes increasing network depth, adjusting convolution kernel size, introducing regularization techniques, and using pre-trained models.

[0098] In this embodiment, the first test result refers to the part related to the model parameters extracted from the test result, and the second test result refers to the part related to the model network structure extracted from the test result.

[0099] In this embodiment, the preset parameter optimization database is a database containing various parameter optimization schemes, which is used to compare with the first test results to find the most matching parameter optimization scheme, wherein the preset parameter optimization database contains all possible parameter values ​​of the model parameters and the parameter optimization scheme corresponding to each parameter value, wherein the parameter optimization scheme corresponding to each parameter value is not unique.

[0100] In this embodiment, the preset test-structure-optimization database is a database containing network structure optimization solutions, which is used to compare with the second test results to find the most matching network structure optimization solution, wherein the preset test-structure-optimization database includes all test results related to the network structure, the model network structure corresponding to each test result, and the network structure optimization solution for each network structure, wherein the network structure optimization solution corresponding to each network structure is not unique.

[0101] In this embodiment, model fitness refers to the compatibility or adaptability between the first parameter optimization scheme or the first network structure optimization scheme and the model of the elderly patient swallowing disorder risk identification model based on deep learning. A high-fit optimization scheme means that it can continue to work better with the elderly patient swallowing disorder risk identification model based on deep learning.

[0102] In this embodiment, the preset minimum model fitness is an adaptation threshold used to determine whether the model fitness is high enough to perform model optimization adjustment, wherein the value range of the preset minimum model fitness is (0.5, 0.95).

[0103] In this embodiment, the first risk adjustment model refers to a model that has been optimized and adjusted using the first parameter optimization solution and the first network structure optimization solution.

[0104] In this embodiment, the model convergence performance refers to the ability of the model to gradually converge to the optimal state during the training process.

[0105] In this embodiment, the baseline model refers to the first risk adjustment model after optimization and adjustment through the first parameter optimization scheme and the first network structure optimization scheme, or the first risk adjustment model after optimization and adjustment through the second parameter optimization scheme and the second network structure optimization scheme, wherein the model convergence performance of the first risk adjustment model is better than the model convergence performance of the elderly patient swallowing disorder risk identification model based on deep learning, and therefore is used as the starting model for model optimization and adjustment and model training.

[0106] In this embodiment, the performance test result refers to the result obtained after evaluating the model using the test set.

[0107] In this embodiment, the optimal risk identification model for dysphagia in elderly patients refers to the optimal risk identification model for dysphagia in elderly patients in terms of accuracy, recall rate, F1 Score and model convergence performance after multiple optimization adjustments.

[0108] The working principle of the above technical solution is as follows: First, the parts related to the model parameters and network structure are separated from the test results, and matched with the preset optimization database respectively to find the best optimization solution. Then, these optimization solutions are combined with the elderly patient swallowing disorder risk identification model based on deep learning, and the model fitness is simulated and judged in the virtual machine. If the model fitness meets the standard, the elderly patient swallowing disorder risk identification model based on deep learning is optimized and adjusted to obtain the first risk adjustment model. If the model fitness does not meet the standard, the suboptimal solution is selected and simulated and judged again until the matching model parameter optimization solution and model network structure optimization solution are obtained, thereby comprehensively determining the first risk adjustment model. The convergence performance of the first risk adjustment model is then evaluated. If it is better than the elderly patient swallowing disorder identification model based on deep learning, the first risk adjustment model is used as the baseline model for further training and optimization adjustment. Finally, based on the performance test results of the training set and the test set, the optimal elderly patient swallowing disorder risk identification model is determined.

[0109] The beneficial effects of the above technical solution are: by analyzing the test results related to the model parameters in the test results to determine the parameter optimization scheme, and at the same time analyzing the test results related to the model network structure in the test results to determine the network structure optimization scheme, thereby performing a model adaptability test with the elderly patient swallowing disorder risk identification model based on deep learning, and then adjusting the model, which can make the model optimization and adjustment of the elderly patient swallowing disorder risk identification model based on deep learning more accurate. At the same time, the convergence performance of the model optimization and adjustment results is judged, so as to obtain a more accurate optimal elderly patient swallowing disorder risk identification model, thereby improving the swallowing disorder risk identification performance.

[0110] S4: Analyze and identify swallowing feature data of elderly patients based on machine vision to determine whether elderly patients have swallowing disorder risks and determine the swallowing disorder risk identification results for elderly patients;

[0111] In this embodiment, determining whether an elderly patient has a risk of dysphagia includes:

[0112] Deploy the optimal elderly patient dysphagia risk identification model in a real-world elderly patient dysphagia risk identification environment based on machine vision;

[0113] Input the swallowing feature data of elderly patients based on machine vision into the optimal elderly patients' swallowing disorder risk identification model. Based on the optimal elderly patients' swallowing disorder risk identification model, the swallowing feature data of elderly patients based on machine vision are analyzed and identified to determine whether the elderly patients have swallowing disorder risks and determine the elderly patients' swallowing disorder risk identification results.

[0114] Among them, the identification result of dysphagia risk in elderly patients is that the elderly patients have dysphagia risk or the elderly patients do not have dysphagia risk.

[0115] S5: Based on the risk identification results of dysphagia in elderly patients, timely early warning and feedback intervention measures should be taken to ensure the life safety and quality of life of elderly patients.

[0116] In this embodiment, early warning feedback intervention measures are taken, including:

[0117] When elderly patients are at risk of swallowing disorders, an audible and visual warning message will be automatically issued to remind nursing staff to take protective intervention measures, including adjusting diet, improving the eating environment and conducting swallowing training for elderly patients. Based on the actual swallowing situation of the elderly patients, the protective intervention measures will be flexibly adjusted in conjunction with professional medical staff to ensure the life safety and quality of life of the elderly patients.

[0118] It should be noted that non-contact monitoring based on machine vision technology does not require the wearing of any equipment, reducing discomfort to patients and is particularly suitable for elderly patients; it can monitor and analyze swallowing behavior in real time and detect abnormalities in a timely manner. Based on deep learning technology, it can automatically identify the risk of swallowing disorders and reduce manual intervention. It can be customized and expanded according to different elderly patient groups and needs; it can be used in nursing homes, hospitals and home care scenarios. When used in nursing homes, it is used to monitor the swallowing function of elderly patients and prevent complications caused by swallowing disorders; when used in hospitals, it is used for swallowing function assessment and monitoring in rehabilitation or geriatric departments; when used in home care, it is used for daily swallowing function monitoring of elderly patients at home to help family members detect abnormalities in a timely manner.

[0119] To better demonstrate the principle of dysphagia risk identification for elderly patients based on machine vision, this embodiment provides a dysphagia risk identification system for elderly patients based on machine vision, which is used to implement the above-mentioned dysphagia risk identification method for elderly patients based on machine vision, including:

[0120] Data acquisition module for elderly patients' facial area during swallowing facial , Neck areaAre neck and chest areaAre Chestl Collect and determine real-time swallowing image data of elderly patients based on machine vision;

[0121] A data processing module is used to process the real-time image data of swallowing of elderly patients collected based on machine vision, and determine the swallowing feature data of elderly patients based on machine vision;

[0122] The risk identification module is used to analyze and identify swallowing feature data of elderly patients based on machine vision using deep learning technology, determine whether elderly patients have swallowing disorder risks, and determine the swallowing disorder risk identification results for elderly patients;

[0123] The early warning intervention module is used to take early warning feedback intervention measures in a timely manner to provide protective intervention for elderly patients based on the risk identification results of dysphagia in elderly patients.

[0124] In summary, the swallowing facial image data Swa of elderly patients is obtained based on machine vision. facial image data , swallowing neck image data Swa neck image data and swallowing chest image data Swa chest image data, determine the real-time swallowing image data of elderly patients based on machine vision, and process the real-time swallowing image data of elderly patients based on machine vision, determine the swallowing feature data of elderly patients based on machine vision, and analyze and identify the swallowing feature data of elderly patients based on machine vision based on the optimal elderly patients' swallowing disorder risk identification model, judge whether the elderly patients have the risk of swallowing disorder, and determine the risk identification results of elderly patients' swallowing disorder. According to the risk identification results of elderly patients' swallowing disorder, timely early warning feedback intervention measures are taken to ensure the life safety and quality of life of elderly patients. It can effectively identify the risk of swallowing disorder in elderly patients, and can carry out timely protective intervention for elderly patients with swallowing disorder, which can effectively ensure the life safety and quality of life of elderly patients.

[0125] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0126] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying dysphagia risk in elderly patients based on machine vision, characterized in that: The steps include: S1: Based on machine vision technology, obtain swallowing facial image data Swa facialimagedata , swallowing neck image data Swa neckimagedata and swallowing chest image data Swa chestimagedata ,collecting real-time image data of swallowing of elderly patients based on machine vision; S2: De-noising, image enhancement, and feature extraction are performed on the real-time swallowing image data of elderly patients based on machine vision to determine the swallowing feature data of elderly patients based on machine vision; S3: Train a deep learning-based risk identification model for dysphagia in elderly patients, test and adjust the model, and determine the optimal risk identification model for dysphagia in elderly patients; S4: Analyze and identify swallowing feature data of elderly patients based on machine vision to determine whether elderly patients have swallowing disorder risks and determine the swallowing disorder risk identification results for elderly patients; S5: Based on the risk identification results of dysphagia in elderly patients, timely early warning and feedback intervention measures should be taken to ensure the life safety and quality of life of elderly patients.

2. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 1, characterized in that: In S1, real-time swallowing image data of elderly patients based on machine vision is collected, and the following operations are performed: Based on machine vision technology, image acquisition devices are deployed in swallowing monitoring scenarios for elderly patients; Camera-based facial area analysis of elderly patients when swallowing facial Real-time monitoring and collection are performed to obtain swallowing facial image data of elderly patients Sw facialimagedata ; Camera-based monitoring of the neck area of ​​elderly patients when swallowing neck Real-time monitoring and collection are performed to obtain swallowing neck image data of elderly patients. neckimagedata ; Based on the camera, the chest area of ​​elderly patients is monitored when swallowing. Chestl Real-time monitoring and collection are performed to obtain chest image data of elderly patients swallowing Swa chestimagedata ; Among them, based on the swallowing facial image data of elderly patients, Sw facialimagedata , swallowing neck image data of elderly patients Swa neckimagedata and elderly patients swallowing chest image data Swa chestimagedata ,determine the real-time image data of swallowing in elderly patients based on machine vision.

3. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 2, characterized in that: In S2, the real-time swallowing image data of elderly patients based on machine vision is processed, including: Based on wavelet transform, real-time swallowing image data of elderly patients based on machine vision are denoised; Among them, the real-time swallowing image data of elderly patients based on machine vision was analyzed on a multi-scale basis based on Symlets wavelet, and the real-time swallowing image data of elderly patients based on machine vision was decomposed into sub-bands of different frequencies; The low-frequency subband contains the main structure and smooth area of ​​the real-time swallowing image data of elderly patients based on machine vision, and the high-frequency subband contains the details and noise of the real-time swallowing image data of elderly patients based on machine vision. The coefficients in the high-frequency subband are processed by thresholding to remove the noise in the high-frequency subband while retaining the details of the real-time swallowing image data of elderly patients based on machine vision.

4. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 3, characterized in that: In the above S2, the real-time swallowing image data of elderly patients based on machine vision is processed, which also includes: Perform image enhancement on real-time swallowing image data of elderly patients based on machine vision; Among them, based on histogram equalization, the grayscale distribution of the real-time swallowing image data of elderly patients based on machine vision is adjusted to enhance the contrast of the real-time swallowing image data of elderly patients based on machine vision; Based on gamma correction, the brightness of real-time swallowing image data of elderly patients based on machine vision is adjusted through nonlinear transformation; Based on the Sobel operator, the edge information of the real-time swallowing image data of elderly patients based on machine vision is highlighted through gradient calculation, and the edges of the real-time swallowing image data of elderly patients based on machine vision are enhanced; Extract features from real-time swallowing image data of elderly patients based on machine vision; Among them, the feature extraction method based on key point detection and image segmentation extracts facial muscle activity features, laryngeal movement trajectory features, swallowing frequency features, swallowing action duration and interval time features related to swallowing of elderly patients from real-time swallowing image data of elderly patients based on machine vision, and determines the swallowing feature data of elderly patients based on machine vision.

5. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 4, characterized in that: In S3, the optimal risk identification model for dysphagia in elderly patients is determined by performing the following operations: According to the risk identification needs of elderly patients with swallowing disorders, the swallowing history image data of elderly patients based on machine vision is collected, and the collected swallowing history image data of elderly patients based on machine vision is divided to determine the training set and test set; Based on deep learning technology, a training set is used to train the deep learning model, allowing the deep learning model to autonomously learn the risk identification process and rules of dysphagia in elderly patients, determine whether elderly patients have dysphagia risks, and determine the deep learning-based dysphagia risk identification model for elderly patients; The performance of the deep learning-based risk identification model for dysphagia in elderly patients was tested based on the test set. The accuracy, recall rate, and F1 score were used to evaluate whether the deep learning-based risk identification model for dysphagia in elderly patients could achieve the expected results. According to the test results, the parameters and structure of the deep learning-based dysphagia risk identification model for elderly patients were adjusted, and the deep learning-based dysphagia risk identification model for elderly patients was continuously iteratively optimized to determine the optimal dysphagia risk identification model for elderly patients.

6. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 5, characterized in that: Based on the test results, the parameters and structure of the deep learning-based dysphagia risk identification model for elderly patients were adjusted. The deep learning-based dysphagia risk identification model for elderly patients was continuously iteratively optimized to determine the optimal dysphagia risk identification model for elderly patients, including: Extracting a first test result related to a model parameter of a deep learning-based risk identification model for elderly patients with dysphagia from the test results, and extracting a second test result related to a model network structure of the deep learning-based risk identification model for elderly patients with dysphagia from the test results; Matching the first test result with each parameter in the preset parameter optimization database one by one, thereby extracting the parameter optimization solution with the highest degree of matching with the first test result as the first parameter optimization solution; Matching the second test result with each network structure in the preset test-structure-optimization database one by one, thereby extracting the network structure optimization solution with the highest degree of matching with the second test result as the corresponding first network structure optimization solution; Optimizing and adjusting the parameters and structure of the deep learning-based elderly patient dysphagia risk identification model according to the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model; Determine the current model convergence performance of the risk adjustment model and compare the current model convergence performance with the target model convergence performance of the deep learning-based dysphagia risk identification model for elderly patients; If the convergence performance of the current model is greater than that of the target model, the risk adjustment model is judged to be superior to the deep learning-based risk identification model for dysphagia in elderly patients, and the risk adjustment model is used as the benchmark model for deep learning-based risk identification of dysphagia in elderly patients. The deep learning-based benchmark model for identifying dysphagia risks in elderly patients was trained based on the training set, and the performance of the model training results of the benchmark model was tested again using the test set. The benchmark model was optimized and adjusted again based on the performance test results, and the optimal dysphagia risk identification model for elderly patients was determined based on the optimization and adjustment results.

7. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 6, characterized in that: According to the first parameter optimization scheme and the first network structure optimization scheme, the parameters and structure of the deep learning-based elderly patient dysphagia risk identification model are optimized and adjusted to obtain a risk adjustment model, including: Inputting the first parameter optimization scheme, the first network structure optimization scheme, and the deep learning-based elderly patient dysphagia risk identification model into a preset virtual machine, and simulating and determining the model adaptability of the first parameter optimization scheme and the first network structure optimization scheme with the deep learning-based elderly patient dysphagia risk identification model; If the model fitness of the first parameter optimization scheme and the first network structure optimization scheme with the deep learning-based elderly patient dysphagia risk identification model is higher than the preset minimum model fitness, then the deep learning-based elderly patient dysphagia risk identification model is optimized and adjusted based on the first parameter optimization scheme and the first network structure optimization scheme to obtain a risk adjustment model; If the model fitness of the first parameter optimization scheme or the first network structure optimization scheme and the deep learning-based swallowing disorder risk identification model for elderly patients is not higher than the preset minimum model fitness, the first test result corresponding to the first parameter optimization scheme with a fitness not higher than the preset minimum model fitness is matched one by one with each parameter in the parameter optimization database to extract the second parameter optimization scheme with the second highest degree of matching with the first test result, or each network structure in the first network structure optimization scheme with a fitness not higher than the preset minimum model is matched one by one to extract the second network structure optimization scheme with the second highest degree of matching with the second test result, and simulation judgment is performed again to obtain the risk adjustment model.

8. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 5, characterized in that: In S4, determining whether the elderly patient has a risk of dysphagia includes: Deploy the optimal elderly patient dysphagia risk identification model in a real-world elderly patient dysphagia risk identification environment based on machine vision; Input the swallowing feature data of elderly patients based on machine vision into the optimal elderly patients' swallowing disorder risk identification model. Based on the optimal elderly patients' swallowing disorder risk identification model, the swallowing feature data of elderly patients based on machine vision are analyzed and identified to determine whether the elderly patients have swallowing disorder risks and determine the elderly patients' swallowing disorder risk identification results. Among them, the identification result of dysphagia risk in elderly patients is that the elderly patients have dysphagia risk or the elderly patients do not have dysphagia risk.

9. The method for identifying dysphagia risk in elderly patients based on machine vision according to claim 7, characterized in that: In S5, early warning feedback intervention measures are taken, including: When elderly patients are at risk of swallowing disorders, an audible and visual warning message will be automatically issued to remind nursing staff to take protective intervention measures, including adjusting diet, improving the eating environment and conducting swallowing training for elderly patients. Based on the actual swallowing situation of the elderly patients, the protective intervention measures will be flexibly adjusted in conjunction with professional medical staff to ensure the life safety and quality of life of the elderly patients.

10. A system for identifying dysphagia risk in elderly patients based on machine vision, used to implement the method for identifying dysphagia risk in elderly patients based on machine vision as claimed in claim 8, characterized in that: include: Data acquisition module for elderly patients' facial area during swallowing facial , Neck areaAre neck and chest areaAre Chestl Collect and determine real-time swallowing image data of elderly patients based on machine vision; A data processing module is used to process the real-time image data of swallowing of elderly patients collected based on machine vision, and determine the swallowing feature data of elderly patients based on machine vision; The risk identification module is used to analyze and identify swallowing feature data of elderly patients based on machine vision using deep learning technology, determine whether elderly patients have swallowing disorder risks, and determine the swallowing disorder risk identification results for elderly patients; The early warning intervention module is used to take early warning feedback intervention measures in a timely manner to provide protective intervention for elderly patients based on the risk identification results of dysphagia in elderly patients.

Images