Risk assessment method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to an artificial intelligence-based risk assessment method and device, computer equipment and a storage medium, and the method comprises the steps: carrying out the preprocessing of collected face image data of a user, and obtaining standardized face data; performing feature extraction on the standardized face data to obtain visual features and semantic features, and fusing the visual features and the semantic features to obtain target face feature representation; processing the target face feature representation based on a contour expert network, an expression expert network and a comprehensive expert network to obtain a first risk assessment result, a second risk assessment result and a third risk assessment result; and based on the gating network and the target face feature representation, performing comprehensive calculation on the first risk assessment result, the second risk assessment result and the third risk assessment result, and generating a Parkinson risk assessment result of the user. The Parkinson risk assessment method can be applied to a Parkinson risk assessment scene in the medical field, and the accuracy of Parkinson risk assessment is effectively improved through the Parkinson risk assessment method.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and can be applied to the medical field, and in particular to a risk assessment method and device based on artificial intelligence, a computer device and a storage medium. BACKGROUND

[0002] Parkinson's disease is a common neurodegenerative disease, and its incidence rate is increasing year by year with the acceleration of population aging, which seriously affects the quality of life of patients and brings heavy burden to families and society. Therefore, timely and accurate risk assessment of Parkinson's disease is of great clinical significance and social value in order to intervene and treat as early as possible.

[0003] Traditional Parkinson's disease risk assessment methods mainly rely on clinical symptom observation and professional instrument detection. Clinical symptom observation relies on the professional experience and subjective judgment of doctors, and there may be assessment differences between different doctors. In the early stage of the disease, the clinical symptoms of patients are often atypical and easily ignored or misdiagnosed, resulting in low efficiency and strong subjectivity of Parkinson's disease risk assessment. Although professional instrument detection can provide more objective data, the equipment is usually expensive and complex to operate, and requires professional personnel to operate and interpret the results, which is difficult to be widely applied in primary medical institutions, limiting its ability to conduct large-scale screening and early assessment.

[0004] In recent years, studies have shown that Parkinson's disease patients exhibit unique subtle changes in their faces, such as expression rigidity, reduced blink frequency, and eyelid tremor. These facial features provide a new approach and direction for using facial information to assess the risk of Parkinson's disease. By analyzing these subtle changes in facial images or videos, it is possible to achieve non-contact, convenient, and efficient risk assessment of Parkinson's disease. However, there are many challenges in using facial information to assess the risk of Parkinson's disease. Existing single models usually only focus on a certain type of facial feature or information in a certain local area, making it difficult to fully capture the subtle pathological features of the face. For example, some models may only focus on analyzing changes in facial expressions, while ignoring other important features such as blink frequency and eyelid tremor; or only extracting features from specific areas of the face, missing other areas that may contain critical information. This one-sided feature extraction method makes it difficult for the model to fully utilize facial information, which in turn reduces the accuracy of Parkinson's disease risk assessment and fails to meet the needs of clinical practice.

[0005] Therefore, it is of great practical significance to develop a method that can improve the accuracy of Parkinson's disease risk assessment. SUMMARY

[0006] The embodiment of the application aims to provide an artificial intelligence-based risk assessment method, device, computer equipment and storage medium, to solve the technical problem of low accuracy of existing Parkinson's disease risk assessment.

[0007] In a first aspect, an artificial intelligence-based risk assessment method is provided, comprising:

[0008] Collecting face image data of a user;

[0009] Preprocessing the face image data to obtain corresponding standardized face data;

[0010] Extracting visual features from the standardized face data and extracting semantic features from the standardized face data;

[0011] Fusing the visual features and the semantic features to obtain corresponding target face feature representation;

[0012] Processing the target face feature representation based on a preset contour expert network, an expression expert network and a comprehensive expert network to obtain a first risk assessment result, a second risk assessment result and a third risk assessment result;

[0013] Based on the preset gating network and the target face feature representation, the first risk assessment result, the second risk assessment result and the third risk assessment result are comprehensively calculated and processed to generate the Parkinson's risk assessment result of the user.

[0014] In a second aspect, an artificial intelligence-based risk assessment device is provided, comprising:

[0015] The acquisition module is configured to collect face image data of a user;

[0016] The preprocessing module is configured to preprocess the face image data to obtain corresponding standardized face data;

[0017] The extraction module is configured to extract visual features from the standardized face data and extract semantic features from the standardized face data;

[0018] The fusion module is configured to fuse the visual features and the semantic features to obtain corresponding target face feature representation;

[0019] The processing module is configured to process the target face feature representation based on a preset contour expert network, an expression expert network and a comprehensive expert network to obtain a first risk assessment result, a second risk assessment result and a third risk assessment result;

[0020] The calculation module is used to perform comprehensive calculation and processing on the first risk assessment result, the second risk assessment result, and the third risk assessment result based on a preset gating network and the target facial feature representation, to generate the user's Parkinson's risk assessment result.

[0021] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned artificial intelligence-based risk assessment method.

[0022] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned artificial intelligence-based risk assessment method.

[0023] In the aforementioned scheme implemented by the AI-based risk assessment method, device, computer equipment, and storage medium, the user's facial image data is first collected; then, the facial image data is preprocessed to obtain corresponding standardized facial data; subsequently, visual features are extracted from the standardized facial data to obtain visual features, and semantic features are extracted from the standardized facial data to obtain semantic features; subsequently, the visual features and semantic features are fused to obtain a corresponding target facial feature representation; further, the target facial feature representation is processed based on a preset contour expert network, expression expert network, and comprehensive expert network to obtain corresponding first risk assessment results, second risk assessment results, and third risk assessment results; finally, based on a preset gating network and the target facial feature representation, the first risk assessment results, second risk assessment results, and third risk assessment results are comprehensively calculated to generate the user's Parkinson's disease risk assessment result. Based on the above processing flow, this application preprocesses the collected user facial image data to obtain standardized facial data, then extracts features from the standardized facial data to obtain visual and semantic features, and fuses the visual and semantic features to obtain a target facial feature representation. Furthermore, it uses a combination of multiple expert networks, including contour expert networks, expression expert networks, and comprehensive expert networks, as well as a gating network, to perform Parkinson's disease risk assessment on the target facial feature representation. This fully leverages the learning advantages of different networks for facial features, accurately capturing subtle facial changes related to Parkinson's disease, achieving high-precision Parkinson's risk assessment, improving the accuracy of Parkinson's risk assessment, and ensuring the accuracy of the obtained Parkinson's risk assessment results. Attached Figure Description

[0024] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0026] Figure 2 This is a flowchart of an embodiment of the AI-based risk assessment method according to this application;

[0027] Figure 3 This is a schematic diagram of a structure of an embodiment of the AI-based risk assessment device according to this application;

[0028] Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0030] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0032] like Figure 1As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0033] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0034] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0035] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0036] It should be noted that the AI-based risk assessment method provided in this application is generally executed by a server / terminal device, and correspondingly, the AI-based risk assessment device is generally installed in the server / terminal device.

[0037] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0038] Continue to refer to Figure 2The flowchart illustrates an embodiment of the AI-based risk assessment method according to this application. The order of steps in the flowchart can be changed, and some steps can be omitted, depending on different needs. The AI-based risk assessment method provided in this application can be applied to any scenario requiring Parkinson's risk assessment, and therefore can be applied to products in these scenarios. The AI-based risk assessment method includes the following steps:

[0039] Step S201: Collect the user's facial image data.

[0040] In this embodiment, the AI-based risk assessment method operates on electronic devices (e.g., Figure 1 The server / terminal device shown can acquire facial image data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wide wireless band) connections, and other currently known or future wireless connection methods. The implementing entity of this application is a risk assessment system, which can be simply referred to as the system. This application can be applied to Parkinson's disease risk assessment scenarios in the medical field. Specifically, facial image data of users can be collected and processed based on a standardized acquisition process. First, a uniform indoor environment is selected, ensuring it is quiet and clean, minimizing external interference. For example, the indoor temperature should be suitable to avoid unnatural expressions and movements caused by cold or heat. Then, a high-definition camera with a resolution of no less than 1920×1080 is used to ensure clear capture of subtle facial features. The camera is pre-calibrated to ensure good image quality and accurate color reproduction. Furthermore, the user is required to stand approximately 1.5 meters away from the camera, maintaining an upright posture and a relaxed head. Multiple angle photos, including frontal and 45-degree side profile shots, were taken to comprehensively acquire information from different areas of the face. Simultaneously, short videos (approximately 10-15 seconds) were recorded, showcasing natural expressions, blinking, and smiling, allowing observation of dynamic changes in the patient's face under different facial expressions. Furthermore, while collecting facial image data, detailed information such as the user's age and gender was recorded and uploaded to the system. This information is valuable for subsequent analysis and modeling, as patients of different ages and genders may exhibit differences in facial features.

[0041] Standardized facial data collection and preprocessing is a fundamental step in the entire Parkinson's disease risk assessment process. A standardized collection procedure ensures the consistency and reliability of the collected data, reducing variability caused by factors such as environment, equipment, and patient condition.

[0042] Step S202: Preprocess the face image data to obtain the corresponding standardized face data.

[0043] In this embodiment, the specific implementation process of preprocessing the facial image data to obtain the corresponding standardized facial data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0044] Step S203: Visual features are extracted from the standardized face data to obtain visual features, and semantic features are extracted from the standardized face data to obtain semantic features.

[0045] In this embodiment, the semantic feature extraction process includes: utilizing pre-trained large visual language models, such as CLIP and Vi LBERT. These models are trained on large-scale image and text data and possess powerful semantic understanding capabilities. Then, the standardized facial images are input into the pre-trained model, which can understand the semantic information in the images, capture subtle semantic connections between facial expressions and Parkinson's disease, and output corresponding semantic features. For example, the pre-trained model can identify semantic expressions corresponding to pathological features such as facial stiffness and abnormal eye muscles. Through an attention mechanism, the model can focus on key facial regions related to the diagnosis of Parkinson's disease, generating deep semantic feature sequences. The specific implementation process of extracting visual features from the standardized facial data will be further described in detail in subsequent embodiments of this application and will not be elaborated upon here. Furthermore, standardized data acquisition and preprocessing, as well as multi-level feature extraction methods, enable the model to effectively cope with interference factors such as individual differences, lighting changes, and facial expression changes, enhancing the model's robustness and generalization ability in different scenarios.

[0046] Step S204: The visual features and the semantic features are fused to obtain the corresponding target face feature representation.

[0047] In this embodiment, the specific implementation process of fusing the visual features and the semantic features to obtain the corresponding target face feature representation will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0048] Step S205: Based on the preset contour expert network, expression expert network and comprehensive expert network, the target face feature representation is processed to obtain the corresponding first risk assessment result, second risk assessment result and third risk assessment result.

[0049] In this embodiment, the aforementioned contour expert network is an expert network constructed using a CNN (Convolutional Neural Network) architecture. CNNs have powerful spatial feature extraction capabilities and can effectively learn subtle changes in facial contours. Through a combination of multiple convolutional and pooling layers, the network can automatically extract contour features at different levels, from simple edges and lines to complex shapes and structures. The aforementioned expression expert network is an expert network constructed based on the Transformer architecture. The Transformer architecture has advantages in processing sequential data and capturing long-distance dependencies, and can uncover complex correlation features between facial expressions and Parkinson's disease. It can model the dynamic changes in facial expressions and analyze the interrelationships between various facial regions under different expression states. Furthermore, by combining the advantages of contour expert networks and expression expert networks, a comprehensive expert network is constructed. This comprehensive expert network can handle complex combinations of facial features, comprehensively considering multiple aspects of information such as facial contours and expressions, thereby improving the diagnostic ability for Parkinson's disease. Each expert network independently receives multi-dimensional facial feature representations as input and learns and analyzes them in its own feature domain.

[0050] The construction process of the aforementioned contour expert network will be described in further detail in subsequent specific embodiments of this application, and will not be elaborated upon here. Furthermore, the construction process of the expression expert network and the comprehensive expert network can refer to the aforementioned construction process of the contour expert network.

[0051] Furthermore, multi-dimensional target facial feature representations are input into the contour expert network, expression expert network, and comprehensive expert network, respectively. The contour expert network focuses on learning subtle changes in facial contours, the expression expert network mines features related to facial expressions and Parkinson's disease, and the comprehensive expert network combines the advantages of the former two to handle complex combinations of facial features. After targeted training, each expert network (contour expert network, expression expert network, and comprehensive expert network) analyzes and processes the user's target facial feature representation and outputs corresponding first risk assessment results, second risk assessment results, and third risk assessment results.

[0052] Step S206: Based on the preset gating network and the target face feature representation, the first risk assessment result, the second risk assessment result, and the third risk assessment result are comprehensively calculated and processed to generate the user's Parkinson's risk assessment result.

[0053] In this embodiment, the specific implementation process of generating the user's Parkinson's risk assessment result by comprehensively calculating and processing the first risk assessment result, the second risk assessment result, and the third risk assessment result based on the preset gating network and the target facial feature representation will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0054] This application first collects the user's facial image data; then preprocesses the facial image data to obtain corresponding standardized facial data; subsequently, it extracts visual features from the standardized facial data to obtain visual features, and extracts semantic features from the standardized facial data to obtain semantic features; subsequently, it fuses the visual features and semantic features to obtain a corresponding target facial feature representation; further, it processes the target facial feature representation based on a preset contour expert network, expression expert network, and comprehensive expert network to obtain corresponding first risk assessment results, second risk assessment results, and third risk assessment results; finally, based on a preset gating network and the target facial feature representation, it performs comprehensive calculation processing on the first risk assessment results, second risk assessment results, and third risk assessment results to generate the user's Parkinson's disease risk assessment result. Based on the above processing flow, this application preprocesses the collected user facial image data to obtain standardized facial data, then extracts features from the standardized facial data to obtain visual and semantic features, and fuses the visual and semantic features to obtain a target facial feature representation. Furthermore, it uses a combination of multiple expert networks, including contour expert networks, expression expert networks, and comprehensive expert networks, as well as a gating network, to perform Parkinson's disease risk assessment on the target facial feature representation. This fully leverages the learning advantages of different networks for facial features, accurately capturing subtle facial changes related to Parkinson's disease, achieving high-precision Parkinson's risk assessment, improving the accuracy of Parkinson's risk assessment, and ensuring the accuracy of the obtained Parkinson's risk assessment results.

[0055] In some alternative implementations, step S204 includes the following steps:

[0056] The visual features and the semantic features are fused together to obtain the corresponding first feature.

[0057] In this embodiment, the visual features and semantic features described above can be concatenated and fused to form a longer feature vector, namely the first feature. This allows features of different types and levels to be integrated, giving full play to their respective advantages.

[0058] Invoke the preset multilayer perceptron.

[0059] In this embodiment, the multilayer perceptron is specifically an MLP model.

[0060] The first feature is transformed and dimensionality reduced based on the multilayer perceptron to obtain the corresponding second feature.

[0061] In this embodiment, the concatenated first feature can be transformed and dimensionality reduced using a multilayer perceptron (MLP). The MLP model can remove redundant information from the features through multilayer nonlinear transformations, extracting more representative and discriminative features. After processing, a multi-dimensional facial feature representation, i.e., the aforementioned second feature, is obtained, which can be used to provide rich and effective input data for subsequent expert networks.

[0062] The second feature is used as the target face feature representation.

[0063] In this embodiment, visual feature extraction captures basic facial shape and texture information, which is intuitive and easy to understand. Deep semantic feature mining, on the other hand, leverages the powerful semantic understanding capabilities of pre-trained models to discover potential semantic associations in facial expressions related to Parkinson's disease. By fusing and optimizing these two different types of features, their advantages can be integrated to obtain a more comprehensive and accurate representation of facial features, thus providing a more reliable basis for subsequent Parkinson's risk assessment.

[0064] This application obtains a first feature by concatenating and fusing the visual and semantic features; then, it invokes a preset multilayer perceptron; subsequently, it transforms and reduces the dimensionality of the first feature based on the multilayer perceptron to obtain a second feature; finally, it uses the second feature as the target face feature representation. Based on the above processing flow, this application obtains a first feature by concatenating and fusing visual and semantic features, and then transforms and reduces the dimensionality of the first feature using a multilayer perceptron, thereby achieving efficient and accurate fusion processing of visual and semantic features, ensuring the accuracy and comprehensiveness of the generated target face feature representation.

[0065] In some optional implementations of this embodiment, before step S205, the electronic device may further perform the following steps:

[0066] Obtain first facial data from Parkinson's disease patients and second facial data from healthy users.

[0067] In this embodiment, first facial data of Parkinson's disease patients and second facial data of healthy users can be collected within a specified time period. The selection of the specified time period is not specifically limited and can be determined according to actual business needs; for example, it can be set to the past two years.

[0068] The first face data and the second face data are labeled to obtain the corresponding sample data.

[0069] In this embodiment, by labeling the first and second facial data, it is determined whether each data belongs to a Parkinson's disease patient or a healthy person, and the sample data is ensured to have sufficient diversity and representativeness, covering patients of different ages, genders, and disease severity.

[0070] Invoke the preset initial contour expert network.

[0071] In this embodiment, the initial contour expert network can specifically be a convolutional neural network (CNN). The initial network architecture corresponding to the expression expert network adopts the Transformer architecture.

[0072] Determine the target loss function and target optimizer corresponding to the initial contour expert network.

[0073] In this embodiment, the cross-entropy loss function (i.e., the target loss function) is used as the optimization objective. The cross-entropy loss function measures the difference between the model's prediction and the true label. By minimizing the cross-entropy loss, the model's prediction can be made closer to the reality. Furthermore, the AdamW optimizer (i.e., the target optimizer) is used for parameter updates. The AdamW optimizer combines the adaptive learning rate advantage of the Adam optimizer with the weight decay regularization method, enabling fast convergence during training and avoiding overfitting.

[0074] Based on the target loss function, the target optimizer, and the preset learning rate decay strategy, the initial contour expert network is trained using the sample data until a specified network that meets the construction requirements is obtained.

[0075] In this embodiment, the initial contour expert network is trained using sample data by utilizing the aforementioned objective loss function, objective optimizer, and learning rate decay strategy, with the learning rate gradually decreasing during training. As training progresses, the model gradually approaches the optimal solution. A smaller learning rate allows for more precise parameter adjustment, improving the model's stability and generalization ability, ultimately resulting in a contour expert network that meets the construction requirements. Specifically, each expert network undergoes targeted training, learning from a large amount of data to achieve optimal performance in its respective feature domain, thereby improving the ability to recognize facial features of Parkinson's patients.

[0076] The specified network is used as the contour expert network.

[0077] This application obtains first facial data of a Parkinson's disease patient and second facial data of a healthy user; then, it annotates the first and second facial data to obtain corresponding sample data; next, it calls a preset initial contour expert network; and determines the target loss function and target optimizer corresponding to the initial contour expert network; subsequently, based on the target loss function, the target optimizer, and a preset learning rate decay strategy, it trains the initial contour expert network using the sample data until a specified network that meets the construction requirements is obtained; finally, the specified network is used as the contour expert network. Based on the above processing flow, this application obtains sample data by annotating the first facial data of a Parkinson's disease patient and the second facial data of a healthy user, and then trains the initial contour expert network using the sample data based on the combined use of the target loss function, target optimizer, and learning rate decay strategy. This ensures that the model converges quickly during training, avoids overfitting, and enables the generated expert network to achieve optimal performance in its area of ​​expertise, thereby providing an accurate basis for subsequent Parkinson's risk assessment.

[0078] In some alternative implementations, step S206 includes the following steps:

[0079] The target facial features are represented, and the first risk assessment result, the second risk assessment result, and the third risk assessment result are input into the gating network.

[0080] In this embodiment, the gating network design includes: Input determination: The input to the gating network is the output of each expert network and the original multi-dimensional facial feature representation. The output of each expert network reflects its analysis results of the input facial image in its respective feature domain, while the original multi-dimensional facial feature representation contains comprehensive facial information. Weight calculation: The weight coefficients of each expert network in the decision-making process are calculated through fully connected layers and a Softmax function. The fully connected layers can linearly combine and non-linearly transform the input features, while the Softmax function converts the output into a probability distribution, ensuring that the sum of all weight coefficients is 1. The gating network can dynamically adjust the contribution of each expert network according to the feature characteristics of the input facial image. For example, when facial contour features are abnormally obvious, the weight of the contour expert network is increased; when expression features are highly correlated with Parkinson's disease, the weight of the expression expert network is increased.

[0081] The weight coefficients of each expert network in the decision-making process are calculated based on the fully connected layers and the Softmax function in the gated network; wherein, the expert network includes the contour expert network, the expression expert network, and the comprehensive expert network.

[0082] In this embodiment, based on the use of fully connected layers and the Softmax function in the above-mentioned gated network, the input features are linearly combined and nonlinearly transformed by the fully connected layers, and the Softmax function converts the output into a probability distribution, so that the sum of all weight coefficients is 1, so as to obtain the weight coefficient of each expert network in the decision-making process.

[0083] The first risk assessment result, the second risk assessment result, and the third risk assessment result are weighted and summed based on the weighting coefficients to obtain the corresponding specified risk assessment result.

[0084] In this embodiment, the outputs of each expert network (i.e., the first risk assessment result, the second risk assessment result, and the third risk assessment result) are weighted and summed according to the weight coefficients calculated by the gating network to obtain a comprehensive assessment result, namely the specified risk assessment result mentioned above. This comprehensive assessment result takes into account the analysis results of different expert networks in their respective fields, and can more comprehensively and accurately reflect the patient's risk of developing Parkinson's disease.

[0085] The specified risk assessment result is numerically compared with the preset risk threshold to obtain the corresponding comparison result.

[0086] In this embodiment, the selection of the aforementioned risk threshold is not specifically limited and can be set according to actual business needs, for example, it can be set to 0.6. The comparison result may include: a specified risk assessment result higher than the risk threshold, or a specified risk assessment result lower than the risk threshold.

[0087] The Parkinson's risk assessment result for the user is generated based on the comparison results.

[0088] In this embodiment, when the specified risk assessment result is higher than the risk threshold, it is determined to be a suspected case of Parkinson's disease, and a Parkinson's risk assessment result indicating a risk of Parkinson's disease is generated; when the specified risk assessment result is lower than the risk threshold, it is determined to be healthy, and a Parkinson's risk assessment result indicating no risk of Parkinson's disease is generated. Simultaneously, the assessment probabilities and gating weights of each expert network can also be output. The assessment probabilities of each expert network reflect its contribution to the diagnostic result in its respective feature domain, while the gating weights reflect the dynamic adjustment of the importance of each expert network by the gating network. This information provides interpretability for the Parkinson's risk assessment results, facilitating doctors' understanding of the risk assessment logic of each expert network and the importance of each feature. This allows them to understand the model's decision-making process, increases doctors' trust in the Parkinson's risk assessment results, and helps promote the practical application of this technology in clinical analysis.

[0089] This application represents the target facial features, and inputs the first risk assessment result, the second risk assessment result, and the third risk assessment result into a gating network. Then, based on the fully connected layers and the Softmax function in the gating network, the weight coefficients of each expert network in the decision-making process are calculated. The expert networks include the contour expert network, the expression expert network, and the comprehensive expert network. Subsequently, the first risk assessment result, the second risk assessment result, and the third risk assessment result are weighted and summed based on the weight coefficients to obtain the corresponding specified risk assessment result. The specified risk assessment result is then numerically compared with a preset risk threshold to obtain a comparison result. Finally, the user's Parkinson's disease risk assessment result is generated based on the comparison result. Based on the above processing flow, this application, through the use of a gating network, enables the system to dynamically adjust the contribution of each expert network according to the feature characteristics of the input facial image, fully leveraging the advantages of different expert networks. Furthermore, by weighting and summing the output results of each expert network to obtain a comprehensive assessment result, and setting a risk threshold for judgment to generate the final Parkinson's disease risk assessment result, the accuracy and reliability of Parkinson's disease risk assessment can be effectively improved.

[0090] In some alternative implementations, step S202 includes the following steps:

[0091] The face image data is subjected to face detection and cropping processing to obtain the corresponding first face data.

[0092] In this embodiment, the face detection algorithm in the OpenCV library is used to scan the acquired face image data, such as photos and video frames, to accurately locate the face region. Then, based on the location results, the face is cropped from the original image to obtain a standard face image, which is used as the first face data mentioned above. This removes irrelevant information such as background and reduces interference in subsequent processing.

[0093] The first face data is processed using a preset image enhancement algorithm to obtain the corresponding second face data.

[0094] In this embodiment, a deep learning-based image enhancement algorithm, such as CycleGAN, is employed. This algorithm can automatically learn the distribution characteristics of an image, perform denoising to remove noise caused by factors such as the shooting device and environment; perform brightness equalization to make the brightness distribution of the image more uniform and avoid overly bright or dark areas affecting feature extraction; and perform contrast enhancement to highlight details and texture information in the image, thereby improving image quality.

[0095] The second face data is normalized to obtain the corresponding third face data.

[0096] In this embodiment, the enhanced images are uniformly adjusted to a fixed size, such as 224×224 pixels. This facilitates subsequent model processing, as most deep learning models have certain requirements on the size of the input image. Simultaneously, mapping the pixel values ​​of the image to the [0,1] interval ensures that the pixel values ​​have a uniform numerical range, which is beneficial for model training and convergence. After these processes, a standardized face dataset, i.e., standardized face data, is formed.

[0097] The third-person facial data is used as the standardized facial data.

[0098] In this embodiment, the data preprocessing operation further optimizes the acquired face image data by removing noise, adjusting brightness and contrast, and unifying size and pixel value range, so that subsequent feature extraction and model training can be based on high-quality data, thereby improving the accuracy and stability of the entire risk assessment system.

[0099] This application obtains first face data by performing face detection and cropping on the face image data; then, it performs image enhancement processing on the first face data based on a preset image enhancement algorithm to obtain second face data; subsequently, it performs normalization processing on the second face data to obtain third face data; and finally, it uses the third face data as the standardized face data. This application achieves efficient and accurate preprocessing of face image data by performing face detection and cropping, image enhancement, and normalization processing, which helps improve the quality and standardization of the generated standardized face data.

[0100] In some optional implementations of this embodiment, step S203, which involves extracting visual features from the standardized face data, includes the following steps:

[0101] The standardized face data is processed using a preset image processing algorithm to extract facial contour features, thereby obtaining the corresponding facial contour features.

[0102] In this embodiment, computer vision methods are used to calculate the aspect ratio of the face using image processing algorithms, thus determining the approximate shape of the patient's face, such as round, square, or oblong. Simultaneously, the relative positions of facial features are measured, such as the distance between the eyes and the relative position of the nose and mouth. These features reflect the overall structure of the facial contour, i.e., the aforementioned facial contour features.

[0103] Texture features are extracted from the standardized face data based on the target operator to obtain the corresponding texture features.

[0104] In this embodiment, the target operator can specifically be the LBP (Local Binary Pattern) operator, which can extract local texture patterns from the standardized face data based on the selected target operator. The LBP operator generates a binary code by comparing the gray values ​​of a pixel with its surrounding neighboring pixels. This code can describe the texture features of a local region. By applying the LBP operator to the entire facial image, a texture feature map of the face can be obtained.

[0105] Geometric features are extracted from the standardized face data to obtain the corresponding geometric features.

[0106] In this embodiment, geometric features such as the interocular distance and nasolabial angle of the standardized facial data can be measured. The interocular distance reflects the horizontal dimensions of the face, while the nasolabial angle reflects the relative position and shape of the nose and lips. These geometric features are important for determining whether there are abnormal changes in the face.

[0107] The facial contour features, texture features, and geometric features are quantized and normalized to obtain the corresponding visual features.

[0108] In this embodiment, the extracted basic visual features are quantized and normalized to form a basic visual feature vector, which is then used as the aforementioned visual features.

[0109] This application extracts facial contour features from standardized face data using a preset image processing algorithm to obtain corresponding facial contour features; then, it extracts texture features from the standardized face data using a target operator to obtain corresponding texture features; subsequently, it extracts geometric features from the standardized face data to obtain corresponding geometric features; finally, it quantizes and normalizes the facial contour features, texture features, and geometric features to obtain corresponding visual features. Based on the above processing flow, this application achieves efficient and accurate visual feature extraction from standardized face data by extracting facial contour features, texture features, and geometric features, thereby improving the quality and standardization of the generated visual features.

[0110] In some optional implementations of this embodiment, after step S205, the electronic device may further perform the following steps:

[0111] Collect feedback data on the specified Parkinson's risk assessment results corresponding to multiple risk assessment cases.

[0112] In this embodiment, the aforementioned risk assessment cases may include correctly assessed cases, incorrectly assessed cases, and missed diagnosis cases. A comprehensive clinical feedback mechanism can be established through cooperation with medical institutions. The feedback process and methods should be clearly defined to ensure that doctors can conveniently and promptly provide feedback on the model evaluation results to the R&D team. Furthermore, doctors' evaluation information on the model evaluation results should be collected, including detailed analysis of correctly assessed cases, incorrectly assessed cases, and missed diagnosis cases. Information such as facial data, feature extraction results, expert network output, and gating weight distribution for each risk assessment case should be recorded. This detailed information can help the R&D team gain a deeper understanding of the model's (expert network's) performance under different conditions and identify problems and shortcomings in the model.

[0113] Based on the feedback data, the corresponding target optimization strategy is determined.

[0114] In this embodiment, the collected feedback data is used to optimize the model by combining online learning and offline training. Online learning allows for real-time updates of some parameters during model operation, quickly adapting to new data and situations; offline training, on the other hand, allows for comprehensive retraining of the entire model, improving its performance and stability.

[0115] Invoke the preset expert network.

[0116] In this embodiment, the expert network includes the contour expert network, the facial expression expert network, and the comprehensive expert network.

[0117] The expert network is iteratively optimized based on the stated objective optimization strategy.

[0118] In this embodiment, the expert network can be iteratively optimized based on the selected target optimization strategy. Specifically, this may include: in-depth analysis of the causes of erroneous evaluation cases and missed diagnoses. For example, if it is found that the model is prone to errors when processing certain specific facial features, the parameters or structure of the expert network can be adjusted in a targeted manner, and the weight calculation logic of the gating network can be optimized. Through continuous adjustment and optimization, the model can better recognize the facial features of Parkinson's patients, improving diagnostic accuracy. In addition, the entire model is retrained regularly using new case data. Over time, more case data will accumulate, which can reflect the characteristics and disease changes of different patient groups. Through regular retraining, the model can continuously learn and adapt to new data, improving the diagnostic accuracy and reliability of the model in different patient groups and complex scenarios, making it better suited to clinical diagnostic needs.

[0119] Feedback-driven continuous model optimization is a crucial step in ensuring the long-term effectiveness and accuracy of the Parkinson's disease risk assessment system. Clinical feedback collection provides real and detailed data support for model optimization, helping the research team understand the model's performance and existing problems in practical applications. By combining online learning with offline training, and by specifically adjusting model parameters and structure, problems can be quickly and effectively addressed, improving model performance. Regular retraining with new case data allows the model to continuously adapt to changing clinical diagnostic needs, maintaining its advanced nature and reliability, and providing continuous and accurate support for Parkinson's disease risk assessment.

[0120] This application collects feedback data from specified Parkinson's disease risk assessment results corresponding to multiple risk assessment cases; then determines a corresponding target optimization strategy based on the feedback data; subsequently, it invokes a preset expert network, which includes a contour expert network, an expression expert network, and a comprehensive expert network; and then iteratively optimizes the expert network based on the target optimization strategy. Based on the above processing flow, this application collects feedback data from specified Parkinson's disease risk assessment results corresponding to multiple risk assessment cases, determines a target optimization strategy based on the feedback data, and then iteratively optimizes the expert network based on the use of the target optimization strategy. This allows for targeted adjustment of model parameters and structure, and can quickly and effectively solve problems existing in the expert network, thus improving its performance.

[0121] In some alternative implementations, the user information obtained is subject to user consent and complies with relevant laws and policies.

[0122] Furthermore, any software tools or components not belonging to our company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0123] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0124] It should be emphasized that, to further ensure the privacy and security of the above Parkinson's risk assessment results, the above Parkinson's risk assessment results can also be stored in a blockchain node.

[0125] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0126] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0127] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0128] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0129] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0130] Further referenceFigure 3 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of an artificial intelligence-based risk assessment device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0131] like Figure 3 As shown, the AI-based risk assessment device 300 described in this embodiment includes: a data acquisition module 301, a preprocessing module 302, an extraction module 303, a fusion module 304, a processing module 305, and a calculation module 306. Wherein:

[0132] Acquisition module 301 is used to acquire user facial image data;

[0133] Preprocessing module 302 is used to preprocess the face image data to obtain corresponding standardized face data;

[0134] The extraction module 303 is used to extract visual features from the standardized face data to obtain visual features, and to extract semantic features from the standardized face data to obtain semantic features.

[0135] The fusion module 304 is used to fuse the visual features and the semantic features to obtain the corresponding target face feature representation;

[0136] The processing module 305 is used to process the target face feature representation based on a preset contour expert network, expression expert network and comprehensive expert network respectively to obtain the corresponding first risk assessment result, second risk assessment result and third risk assessment result;

[0137] The calculation module 306 is used to perform comprehensive calculation and processing on the first risk assessment result, the second risk assessment result and the third risk assessment result based on the preset gating network and the target face feature representation, to generate the user's Parkinson's risk assessment result.

[0138] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the artificial intelligence-based risk assessment method in the aforementioned implementation method, and will not be repeated here.

[0139] In some optional implementations of this embodiment, the fusion module 304 includes:

[0140] The fusion submodule is used to perform concatenated fusion processing on the visual features and the semantic features to obtain the corresponding first feature;

[0141] Calling submodules, used by submodules to call preset multilayer perceptrons;

[0142] The first processing submodule is used to transform and reduce the dimensionality of the first feature based on the multilayer perceptron to obtain the corresponding second feature;

[0143] The first determining submodule is used to use the second feature as the target face feature representation.

[0144] In some optional implementations of this embodiment, the AI-based risk assessment device further includes:

[0145] The acquisition module is used to acquire the first facial data of Parkinson's disease patients and the second facial data of healthy users;

[0146] The annotation module is used to annotate the first face data and the second face data to obtain corresponding sample data;

[0147] The first calling module is used to call the preset initial contour expert network;

[0148] The first determining module is used to determine the target loss function and target optimizer corresponding to the initial contour expert network;

[0149] The training module is used to train the initial contour expert network using the sample data based on the target loss function, the target optimizer, and the preset learning rate decay strategy, until a specified network that meets the construction requirements is obtained.

[0150] The second determining module is used to designate the specified network as the contour expert network.

[0151] In some optional implementations of this embodiment, the calculation module 306 includes:

[0152] The input submodule is used to input the target facial feature representation, the first risk assessment result, the second risk assessment result, and the third risk assessment result into the gating network.

[0153] The computational submodule is used to calculate the weight coefficients of each expert network in the decision-making process based on the fully connected layers and the Softmax function in the gated network; wherein the expert networks include the contour expert network, the expression expert network, and the comprehensive expert network;

[0154] The second processing submodule is used to perform weighted summation on the first risk assessment result, the second risk assessment result and the third risk assessment result based on the weighting coefficient to obtain the corresponding specified risk assessment result.

[0155] The comparison submodule is used to compare the specified risk assessment result with a preset risk threshold to obtain the corresponding comparison result.

[0156] A generation submodule is used to generate the user's Parkinson's risk assessment result based on the comparison result.

[0157] In some optional implementations of this embodiment, the preprocessing module 302 includes:

[0158] The third processing submodule is used to perform face detection and cropping processing on the face image data to obtain the corresponding first face data;

[0159] The fourth processing submodule is used to perform image enhancement processing on the first face data based on a preset image enhancement algorithm to obtain the corresponding second face data;

[0160] The fifth processing submodule is used to normalize the second face data to obtain the corresponding third face data;

[0161] The second determining submodule is used to use the third face data as the standardized face data.

[0162] In some optional implementations of this embodiment, the extraction module 303 includes:

[0163] The first extraction submodule is used to extract facial contour features from the standardized face data based on a preset image processing algorithm to obtain the corresponding facial contour features.

[0164] The second extraction submodule is used to extract texture features from the standardized face data based on the target operator to obtain the corresponding texture features.

[0165] The third extraction submodule is used to extract geometric features from the standardized face data to obtain the corresponding geometric features;

[0166] The sixth processing submodule is used to quantize and normalize the facial contour features, texture features, and geometric features to obtain the corresponding visual features.

[0167] In some optional implementations of this embodiment, the AI-based risk assessment device further includes:

[0168] The collection module is used to collect feedback data on the specified Parkinson's risk assessment results corresponding to multiple risk assessment cases;

[0169] The third determining module is used to determine the corresponding target optimization strategy based on the feedback data;

[0170] The second invocation module is used to invoke a preset expert network; wherein the expert network includes the contour expert network, the expression expert network, and the comprehensive expert network;

[0171] An optimization module is used to perform iterative optimization processing on the expert network based on the target optimization strategy.

[0172] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0173] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0174] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0175] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions based on artificial intelligence risk assessment methods. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0176] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, such as executing computer-readable instructions for the artificial intelligence-based risk assessment method.

[0177] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0178] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the artificial intelligence-based risk assessment method described above.

[0179] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0180] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. An artificial intelligence-based risk assessment method, characterized by, The method comprises the following steps: Collecting face image data of a user; Preprocessing the face image data to obtain corresponding standardized face data; Extracting visual features from the standardized face data and semantic features from the standardized face data; Fusing the visual features and the semantic features to obtain corresponding target face feature representation; Processing the target face feature representation based on a preset contour expert network, an expression expert network, and a comprehensive expert network to obtain a first risk assessment result, a second risk assessment result, and a third risk assessment result; Based on the target face feature representation and a preset gating network, the first risk assessment result, the second risk assessment result, and the third risk assessment result are comprehensively calculated and processed to generate a Parkinson's risk assessment result of the user. 2.The artificial intelligence-based risk assessment method according to claim 1, characterized in that, The step of fusing the visual features and the semantic features to obtain the corresponding target face feature representation specifically comprises: Serially fusing the visual features and the semantic features to obtain a first feature; Calling a preset multilayer perceptron; Based on the multilayer perceptron, the first feature is transformed and dimensionally reduced to obtain a second feature; The second feature is used as the target face feature representation. 3.The artificial intelligence-based risk assessment method according to claim 1, characterized in that, Before the step of processing the target face feature representation based on the preset contour expert network, expression expert network, and comprehensive expert network to obtain the corresponding first risk assessment result, second risk assessment result, and third risk assessment result, it further comprises: Obtaining first face data of Parkinson's disease patients and second face data of healthy users; Labeling the first face data and the second face data to obtain corresponding sample data; Calling a preset initial contour expert network; Determining a target loss function and a target optimizer corresponding to the initial contour expert network; Based on the target loss function, the target optimizer, and a preset learning rate decay strategy, the initial contour expert network is trained using the sample data until a specified network meeting the construction requirements is obtained; The specified network is used as the contour expert network. 4.The artificial intelligence-based risk assessment method of claim 1, wherein, The step of generating the Parkinson's risk assessment result of the user based on the target face feature representation and the preset gating network, the first risk assessment result, the second risk assessment result, and the third risk assessment result specifically comprises: Inputting the target face feature representation, the first risk assessment result, the second risk assessment result, and the third risk assessment result into the gating network; Based on the fully connected layer and the Softmax function in the gating network, the weight coefficients of each expert network in the decision-making process are calculated; wherein the expert network includes the contour expert network, the expression expert network, and the comprehensive expert network; perform weighted sum processing on the first risk assessment result, the second risk assessment result and the third risk assessment result based on the weight coefficients to obtain a corresponding specified risk assessment result; perform numerical comparison on the specified risk assessment result and a preset risk threshold to obtain a corresponding comparison result; generate the Parkinson's risk assessment result of the user based on the comparison result. 5.The artificial intelligence-based risk assessment method according to claim 1, wherein, The step of preprocessing the face image data to obtain corresponding standardized face data specifically includes: performing face detection and cropping processing on the face image data to obtain corresponding first face data; performing image enhancement processing on the first face data based on a preset image enhancement algorithm to obtain corresponding second face data; performing normalization processing on the second face data to obtain corresponding third face data; taking the third face data as the standardized face data. 6.The artificial intelligence-based risk assessment method according to claim 1, wherein, The step of extracting visual features from the standardized face data specifically includes: extracting facial contour features from the standardized face data based on a preset image processing algorithm to obtain corresponding facial contour features; extracting texture features from the standardized face data based on a target operator to obtain corresponding texture features; extracting geometric features from the standardized face data to obtain corresponding geometric features; quantizing and normalizing the facial contour features, the texture features and the geometric features to obtain corresponding visual features. 7.The artificial intelligence-based risk assessment method according to claim 1, wherein, After the step of processing the target face feature representation based on the preset contour expert network, expression expert network and comprehensive expert network to obtain the first risk assessment result, second risk assessment result and third risk assessment result, the method further includes: collecting feedback data of specified Parkinson's risk assessment results corresponding to a plurality of risk assessment cases; determining a corresponding target optimization strategy based on the feedback data; calling a preset expert network; wherein the expert network includes the contour expert network, the expression expert network and the comprehensive expert network; performing iterative optimization processing on the expert network based on the target optimization strategy.

8. An artificial intelligence-based risk assessment device, characterized by, The method includes: a collection module configured to collect face image data of a user; a preprocessing module configured to preprocess the face image data to obtain corresponding standardized face data; an extraction module configured to extract visual features from the standardized face data and extract semantic features from the standardized face data; a fusion module configured to perform fusion processing on the visual features and the semantic features to obtain a corresponding target face feature representation; a processing module configured to process the target face feature representation based on a preset contour expert network, expression expert network and comprehensive expert network to obtain a first risk assessment result, a second risk assessment result and a third risk assessment result; The computing module is configured to perform comprehensive calculation and processing on the first risk assessment result, the second risk assessment result and the third risk assessment result based on the preset gating network and the target facial feature representation, and generate a Parkinson's risk assessment result of the user.

9. A computer device, comprising: The computer readable storage medium has stored thereon computer readable instructions which, when executed by a processor, implement the steps of the artificial intelligence-based risk assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium has stored thereon computer readable instructions which, when executed by a processor, implement the steps of the artificial intelligence-based risk assessment method according to any one of claims 1 to 7.

Images