Image processing method and apparatus, device, and storage medium

Abstract

Embodiments of the present application provide an image processing method and device, equipment and a storage medium, belonging to the technical field of artificial intelligence. The method comprises: acquiring an original sample image and a sample image label; inputting the original sample image into a preset original correlation degree prediction model to obtain an original disease correlation degree; taking the original disease correlation degree and the sample image label as model parameters of a preset distribution model to obtain an original distribution model; optimizing the original distribution model through a preset target loss function to obtain a target distribution model; inputting the original disease correlation degree into the target distribution model to obtain an updated disease correlation degree; adjusting parameters of the original correlation degree prediction model according to the updated disease correlation degree to obtain a target correlation degree prediction model; acquiring a target image and inputting the target image into the target correlation degree prediction model to obtain a target disease correlation degree. The embodiments of the present application can improve the accuracy of disease correlation degree prediction.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an image processing method, apparatus, device, and storage medium. Background Technology

[0002] Currently, training disease association prediction models using deep learning networks requires preparing a large amount of sample data beforehand. However, during model training, different samples predict different probability distributions, and the predicted probability distributions for samples with the same disease category label should be different. However, current disease association prediction models predict the same probability distribution for samples with the same disease category label, which does not match the data characteristics of the sample data and thus affects the accuracy of disease association prediction. Summary of the Invention

[0003] The main objective of this application is to provide an image processing method, apparatus, device, and storage medium, which aims to improve the accuracy of disease correlation prediction.

[0004] To achieve the above objectives, a first aspect of this application provides an image processing method, the method comprising:

[0005] Obtain pre-labeled sample image data; wherein, the sample image data includes original sample images and sample image labels, the sample image labels being used to characterize the disease association of the original sample images;

[0006] The original sample images are input into a preset original correlation prediction model to predict the disease correlation and obtain the original disease correlation.

[0007] The original disease correlation degree and the sample image label are used as model parameters of the preset distribution model to obtain the original distribution model;

[0008] The original distribution model is optimized using a preset target loss function to obtain the target distribution model;

[0009] The original disease correlation score is input into the target distribution model for correction processing to obtain the updated disease correlation score;

[0010] The parameters of the original correlation prediction model are adjusted based on the updated disease correlation to obtain the target correlation prediction model;

[0011] Acquire the target image;

[0012] The target image is input into the target association prediction model to predict the disease association and obtain the target disease association.

[0013] In some embodiments, the step of using the original disease association degree and the sample image label as model parameters of a preset distribution model to obtain the original distribution model includes:

[0014] Based on the original disease correlation degree and the sample image label, disease measurement data is obtained by calculating disease measurement data.

[0015] The disease measurement data, the original disease correlation degree, and the sample image labels are used as model parameters for the preset distribution model to construct the original distribution model.

[0016] In some embodiments, optimizing the original distribution model using a preset target loss function to obtain a target distribution model includes:

[0017] The original distribution model is transformed using the target loss function to obtain the prediction loss model;

[0018] The target loss value is obtained by calculating the loss of the prediction loss model using the target loss function.

[0019] The original distribution model is adjusted according to the target loss value to obtain the target distribution model.

[0020] In some embodiments, adjusting the parameters of the original association prediction model based on the updated disease association to obtain the target association prediction model includes:

[0021] The loss is calculated on the updated disease correlation degree and the original disease correlation degree to obtain the correlation degree loss value;

[0022] The parameters of the original correlation prediction model are adjusted based on the correlation loss value to obtain the target correlation prediction model.

[0023] In some embodiments, the target association prediction model includes: a target feature extraction network, a target normalization layer, and a target unimodal loss function; the step of inputting the target image into the target association prediction model to predict the disease association and obtain the target disease association includes:

[0024] The target image is input into the target feature extraction network for feature extraction to obtain the target feature vector;

[0025] The target feature vector is subjected to feature correlation analysis through the target normalization layer to obtain the candidate disease correlation distribution information;

[0026] The candidate disease association distribution information is adjusted by applying the target unimodal loss function to obtain the target disease association distribution information.

[0027] Based on the distribution information of the target disease association, disease association is filtered to obtain the association degree of the target disease.

[0028] In some embodiments, the candidate disease association distribution information includes: the predicted association degree of each candidate disease category; the step of applying a single-peak adjustment to the candidate disease association distribution information using the target single-peak loss function to obtain the target disease association distribution information includes:

[0029] The predicted correlation degree is obtained by peak filtering through the target unimodal loss function;

[0030] The predicted correlations on both sides of the target correlation are subjected to unimodal adjustment to obtain the distribution information of the target disease correlation.

[0031] In some embodiments, the step of filtering disease associations based on the target disease association distribution information to obtain the target disease association includes:

[0032] Peak data is searched for based on the distribution information of the correlation between the target disease and the target disease to obtain the target peak value;

[0033] The predicted correlation is filtered based on the target peak value to obtain the target disease correlation.

[0034] To achieve the above objectives, a second aspect of this application provides an image processing apparatus, the apparatus comprising:

[0035] The data acquisition module is used to acquire pre-labeled sample image data; wherein, the sample image data includes original sample images and sample image labels, and the sample image labels are used to characterize the disease association of the original sample images;

[0036] The original prediction module is used to input the original sample image into a preset original correlation prediction model to predict the disease correlation and obtain the original disease correlation.

[0037] The model building module is used to take the original disease correlation degree and the sample image label as model parameters of the preset distribution model to obtain the original distribution model;

[0038] The model optimization module is used to optimize the original distribution model using a preset target loss function to obtain the target distribution model.

[0039] The correction module is used to input the original disease correlation degree into the target distribution model for correction processing to obtain the updated disease correlation degree;

[0040] The parameter adjustment module is used to adjust the parameters of the original correlation prediction model according to the updated disease correlation to obtain the target correlation prediction model;

[0041] The image acquisition module is used to acquire the target image;

[0042] The target calculation module is used to input the target image into the target correlation prediction model to predict the disease correlation and obtain the target disease correlation.

[0043] To achieve the above objectives, a third aspect of this application provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the image processing method as described in the first aspect.

[0044] To achieve the above objectives, a fourth aspect of the present application provides a storage medium storing a computer program that, when executed by a processor, implements the image processing method as described in the first aspect.

[0045] The image processing method, apparatus, device, and storage medium proposed in this application first optimize the original distribution model to obtain a target distribution model, then obtain the updated disease correlation output by the target distribution model based on the original disease correlation, adjust the parameters of the original correlation prediction model according to the updated disease correlation to obtain the target correlation prediction model, and finally predict the disease correlation of the target image through the target correlation prediction model to obtain the target disease correlation, so as to use the target disease correlation as reference data for disease category prediction, thereby improving the accuracy of disease category prediction. Attached Figure Description

[0046] Figure 1 This is a flowchart of the image processing method provided in the embodiments of this application;

[0047] Figure 2 yes Figure 1 The flowchart of step S103 in the process;

[0048] Figure 3 yes Figure 1 The flowchart of step S104 in the process;

[0049] Figure 4 yes Figure 1 The flowchart of step S106 in the process;

[0050] Figure 5 yes Figure 1The flowchart of step S108 in the process;

[0051] Figure 6 yes Figure 5 The flowchart of step S502 in the document;

[0052] Figure 7 yes Figure 5 The flowchart of step S504 in the process;

[0053] Figure 8 This is a schematic diagram of the operation flow of the image processing method provided in the embodiments of this application;

[0054] Figure 9 This is a schematic diagram of the structure of the image processing apparatus provided in the embodiments of this application;

[0055] Figure 10 This is a schematic diagram of the hardware structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0058] 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 belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0059] First, let's analyze some of the terms used in this application:

[0060] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0061] Deep Learning (DL): Deep learning is a new research direction in the field of Machine Learning (ML). It was introduced into machine learning to bring it closer to its original goal—artificial intelligence. Deep learning learns the inherent patterns and hierarchical representations of sample data. The information gained during this learning process greatly helps in interpreting data such as text, images, and sound. Its ultimate goal is to enable machines to have analytical and learning capabilities like humans, and to be able to recognize data such as text, images, and sound.

[0062] The Softmax function is an activation function that normalizes a numerical vector into a probability distribution vector, where the sum of the probabilities is 1. The Softmax function normalizes the raw data from the previous layer, transforming it into a set of probabilities that can be used as a probability distribution for multi-class classification predictions. The Softmax function is typically used as the last layer in a neural network, receiving input values ​​from the previous layer and converting them into probabilities.

[0063] Ground-Truth: In machine learning, ground-truth refers to the precise set value of the training set for classification results in supervised learning. It is generally used for error estimation and performance evaluation. Ground-Truth is the reference standard, typically used for error quantification. Ground-Truth is also used in reinforcement learning mechanisms, which incorporate reward and punishment mechanisms into the learning process. For example, the closer the program's output is to the ground-truth, the greater the weight of the data that produced that result.

[0064] Label Distribution Learning (LDL): To better represent the distribution of a sample, training a model that learns the label distribution of the sample is called label distribution learning. The main difficulty of label distribution learning is that label distributions are extremely difficult to obtain. Most classification datasets do not have this condition. To obtain the true label distribution, theoretically, it is necessary to label a large number of samples to construct training samples and then learn from the label distribution of the training samples.

[0065] Normal distribution: Also known as the "normal distribution" or Gaussian distribution. The normal curve is bell-shaped, low at both ends and high in the middle, and symmetrical. Because of its bell-shaped curve, it is often called the bell curve. If a random variable X follows a normal distribution with expected value μ and variance σ², denoted as N(μ, σ²), its probability density function is given. The expected value μ of the normal distribution determines its position, and its standard deviation σ determines the amplitude of the distribution. The normal distribution when μ = 0 and σ = 1 is the standard normal distribution.

[0066] Likelihood function: The likelihood function is a function of the parameters in a statistical model, representing the likelihood of those parameters. "Likelihood" is similar in meaning to "probability" or "possibility," both referring to the likelihood of an event occurring. However, in statistics, there is a clear distinction between "likelihood" and "probability." Probability is used to predict the outcome of subsequent observations given some parameters, while likelihood is used to estimate parameters related to the properties of things given the outcomes of certain observations.

[0067] With the development of artificial intelligence, image processing methods are combined with machine learning and deep learning to predict the probability of a user having a disease and the probability of having the specific type of disease through the analysis and judgment of disease images. This data serves as a reference for doctors in disease diagnosis and analysis. However, intelligent analysis of disease correlation requires the construction of a disease correlation prediction model. This often involves learning and training the model based on single-label annotations of disease images at different stages of differentiation. However, disease differentiation occurs in different stages, and the labels are ordered, with inherent connections between labels at different differentiation stages. Therefore, using only single-label learning cannot fully utilize the labeled data of disease images. Furthermore, the adaptability of the disease correlation prediction model to samples is crucial. The model predicts disease correlations based on different sample data, and these correlations represent the correlation of each disease category. Therefore, the output of multiple disease correlations is represented by prediction curves. However, the prediction curves obtained from different sample data should differ. Even for sample data with the same image label, the prediction curves obtained by the disease correlation prediction model should be different, even if the sample category corresponding to the image label is the same. Therefore, the prediction curves obtained by traditional disease association prediction models do not conform to the actual situation of the sample data, thus reducing the accuracy of disease association prediction.

[0068] Based on this, compared to traditional disease association prediction models that do not consider the adaptability of sample data when constructing, the embodiments of this application provide an image processing method, apparatus, device, and storage medium. The aim is to obtain the original disease association by inputting the original sample image into a preset original association prediction model to predict the disease association, then using the original disease association and sample image labels as model parameters of a preset distribution model to obtain the original distribution model, optimizing the original distribution model through a preset target loss function to obtain the target distribution model, and inputting the original disease association into the target distribution model for correction to obtain the updated disease association, adjusting the parameters of the original association prediction model according to the updated disease association to obtain the target association prediction model, and finally inputting the acquired target image into the target association prediction model to predict the disease association to obtain the target disease association. Therefore, by first constructing the original distribution model corresponding to the original disease correlation, optimizing the original distribution model to obtain the target distribution model, adjusting the original disease correlation through the target distribution model to obtain the updated disease correlation, and then adjusting the parameters of the original correlation prediction model based on the updated disease correlation to obtain the target correlation prediction model, the updated disease correlation is adapted to the original sample image, thus constructing a target correlation prediction model with higher prediction accuracy, thereby improving the accuracy of disease correlation prediction, and using the target disease correlation as reference data for disease diagnosis to assist in disease diagnosis.

[0069] The image processing methods, apparatus, devices, and storage media provided in this application are specifically described through the following embodiments. First, the image processing methods in the embodiments of this application are described.

[0070] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use 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.

[0071] 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.

[0072] The image processing method provided in this application relates to the field of artificial intelligence technology. The image processing method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the image processing method, but is not limited to the above forms.

[0073] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0074] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user image data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to a confirmation page. Only after obtaining the user's separate permission or consent is the necessary user-related data for the normal operation of the embodiments of this application obtained.

[0075] Figure 1 This is an optional flowchart of the image processing method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S108.

[0076] Step S101: Obtain pre-labeled sample image data; wherein, the sample image data includes original sample images and sample image labels, and the sample image labels are used to characterize the disease association of the original sample images;

[0077] Step S102: Input the original sample image into the preset original correlation prediction model to predict the disease correlation and obtain the original disease correlation.

[0078] Step S103: Use the original disease correlation degree and sample image labels as model parameters of the preset distribution model to obtain the original distribution model;

[0079] Step S104: Optimize the original distribution model using a preset target loss function to obtain the target distribution model;

[0080] Step S105: Input the original disease association degree into the target distribution model for correction processing to obtain the updated disease association degree;

[0081] Step S106: Adjust the parameters of the original correlation prediction model based on the updated disease correlation to obtain the target correlation prediction model;

[0082] Step S107: Obtain the target image;

[0083] Step S108: Input the target image into the target association prediction model to predict the disease association and obtain the target disease association.

[0084] Steps S101 to S108 of this embodiment involve acquiring pre-labeled original sample images and sample image labels, inputting the original sample images into a preset original correlation prediction model to predict the original disease correlation, and constructing an original distribution model using the original disease correlation and sample image labels as model parameters. To make the original disease correlation more suitable for the original sample images, the original distribution model is first optimized using a preset target loss function to obtain a target distribution model adapted to the original sample images. Then, the target distribution model is used to correct the original disease correlation to obtain an updated disease correlation, which is more consistent with the original sample images. Based on the updated disease correlation and the original disease correlation, the parameters of the original correlation prediction model are adjusted to obtain a target correlation prediction model. The target correlation prediction model outputs a target disease correlation that is more suitable for the target images, thus making the target correlation prediction model more accurate in predicting the target disease correlation. Therefore, by constructing an original distribution model associated with the original disease correlation, and then optimizing the original distribution model according to a preset target loss function to obtain a target distribution model, the original disease correlation is then corrected according to the target distribution model to obtain an updated disease correlation. The updated disease correlation is more consistent with the original sample image. Then, the parameters of the original correlation prediction model are adjusted according to the updated disease correlation and the original disease correlation to obtain a target correlation prediction model. The target disease correlation predicted by the target correlation prediction model is more consistent with the target image, thereby improving the accuracy of disease correlation prediction.

[0085] In step S101 of some embodiments, pre-annotated sample image data can be obtained from a sample database. Sample image data can also be obtained directly through other methods, and is not limited to these. For example, a system architecture can be built with multiple hospital platforms to collect patient data uploaded from multiple hospital platforms. From the patient data, disease progression data with the disease type "already infected" can be selected. Then, disease status information and disease diagnosis image data can be obtained from the disease progression data. Sample image labels are generated based on the patient status information, and the disease diagnosis image data is used as the original sample image. The sample image label is used to characterize the disease correlation of the original sample image, that is, to determine the correlation of the original sample image to which disease category it belongs. First, the corresponding disease category is determined based on the patient status information. Then, sample image labels are generated based on the correlation of the disease category. The correlation of the sample image label to the corresponding disease category is 100%. For example, if the disease category of the sample image label is lupus erythematosus, then the correlation of the sample image label to lupus erythematosus is determined to be 100%. The sample image label characterizes the disease correlation and can also characterize the severity of the same disease category. Each original sample image corresponds to one sample image label.

[0086] For example, if disease association represents the severity of cancer, then the sample image labels represent the severity of cancer. If the sample image labels for different differentiation stages of cancer are 1, 2, 3, 4…, C, and the sample image labels are ordered, 1 represents normal tissue, 2 represents benign lesions, …, C represents the most severe malignant lesions. For the i-th original sample image, its sample image label is y. i That is, the Ground-Truth of the i-th original sample image is y. i , then y i If ∈{1, 2, ..., C}, then the severity of cancer in the original sample image follows a normal distribution, i.e., d ~ N(μ, σ). 2 If the probability that the i-th original sample image belongs to the j-th class of sample image labels can be expressed as formula (1):

[0087]

[0088] In the formula, d i , j Let be the probability that the i-th original sample image belongs to the j-th class of sample images, where π, σ, and μ are real parameters, and σ > 0. The probabilities of the same original sample image belonging to each class of sample images are summed to 1, that is... j =1. Therefore, by obtaining the original sample images and sample image labels representing the disease categories of the original sample images, a pre-defined original association prediction model is trained based on the original sample images and sample image labels.

[0089] In step S102 of some embodiments, the original sample image is input into a preset original correlation prediction model to predict the disease correlation, that is, the predicted correlation of each disease category is calculated based on the original sample image. Therefore, the original disease correlation corresponding to the original sample image is determined based on the predicted correlation of each disease category, making the original disease correlation prediction operation simple.

[0090] Specifically, the original correlation prediction model includes: an original feature extraction network, an original normalization layer, and an original unimodal loss function. The original feature extraction network extracts features from the original sample images to obtain original feature vectors. The original normalization layer performs correlation analysis on the original feature vectors, that is, it analyzes the correlation between each original feature and a preset feature, where the preset feature is the image feature corresponding to each disease category. Then, it calculates the correlation between each original feature and the preset feature to obtain the initial disease correlation distribution information. The original unimodal loss function adjusts the initial disease correlation distribution information to have only one peak, thus reducing the impact of other peaks on disease correlation prediction. Finally, the original disease correlation is predicted using the original disease correlation distribution information, which includes the predicted correlation for each disease category. The highest predicted correlation is obtained to obtain the original disease correlation, and the original disease correlation is used as reference data for disease type prediction to analyze the disease type, thereby assisting in disease type prediction and improving the accuracy of disease type prediction.

[0091] Please see Figure 2 In some embodiments, step S103 may include, but is not limited to, steps S201 to S202:

[0092] Step S201: Calculate disease metrics based on the original disease correlation degree and sample image labels to obtain disease metric data;

[0093] Step S202: The disease measurement data, the original disease correlation degree, and the sample image labels are used as model parameters of the preset distribution model to construct the model and obtain the original distribution model.

[0094] In step S201 of some embodiments, the sample image label represents the disease category of the original sample image. Disease measurement data is obtained by calculating the disease measurement based on the original disease correlation and the sample image label. Specifically, the variance between the original disease correlation and the sample image label is calculated to obtain the disease measurement data. The prediction accuracy of the original correlation prediction model can be determined through the disease measurement data. Wherein, v iLet represent the disease measurement data of the i-th original sample image, and the calculation formula for the disease measurement data is shown in formula (2):

[0095]

[0096] In the formula, P i,j C is the predicted correlation between the label of the i-th original sample image and the label of the j-th sample image, where C is a constant. Let be the original disease association degree of the i-th original sample image, and let the original disease association degree be . j is the label of the sample image. Therefore, the disease measurement data can be easily calculated using formula (2), and the original disease correlation can be corrected by adjusting the disease measurement data.

[0097] In step S202 of some embodiments, a model is constructed by using disease measurement data, original disease correlation, and sample image labels as model parameters to obtain an original distribution model, which is then used as a reference model for parameter adjustment of the original correlation prediction model.

[0098] Specifically, disease measurement data, original disease correlation, and sample image labels are used as model parameters of the preset distribution model, and the likelihood function is used to characterize the model parameters, representing the likelihood of the model parameters. The likelihood function is shown in formula (3):

[0099]

[0100] In the formula, P i is the correlation distribution function of the i-th original sample image, which is also the predicted correlation of the original disease correlation in this application. N is the total number of original sample images, y i is the sample image label of the i-th original sample image. Here, the likelihood function is a function of the model parameters, given the original disease association P. i When the likelihood function with respect to parameter θ is equal to the updated disease association degree calculated after the given parameter θ, the likelihood function is adjusted according to the output probability, thereby adjusting the model parameters to update the original distribution model.

[0101] Steps S201 and S202, as illustrated in this embodiment, involve calculating disease metrics based on the original disease correlation and sample image labels. Specifically, this involves calculating the variance of the original disease correlation to obtain disease metric data. The disease metric data, the original disease correlation, and the sample image labels are then used as model parameters to construct the original distribution model. The model parameters of the original distribution model are characterized by a likelihood function, allowing the calculation of the optimal model parameters for training. This results in a more accurate correction of the original disease correlation by the original distribution model.

[0102] Please see Figure 3 In some embodiments, step S104 may include, but is not limited to, steps S301 to S303:

[0103] Step S301: Transform the original distribution model using the target loss function to obtain the prediction loss model;

[0104] Step S302: Calculate the loss of the prediction loss model using the target loss function to obtain the target loss value;

[0105] Step S303: Adjust the parameters of the original distribution model according to the target loss value to obtain the target distribution model.

[0106] In step S301 of some embodiments, the target loss function is used to optimize the original distribution model, that is, to adjust the parameters of the original distribution model so that the trained original distribution model corrects the original disease association degree to better match the original sample images, thereby adjusting the original association degree prediction model. Therefore, the target loss function is used to transform the original distribution model to obtain the prediction loss model, that is, the target loss function is used to extract the model parameters of the original distribution model, and the likelihood function is used to characterize the model parameters. Since the model parameters are disease measurement data, original disease association degree and sample image labels, the logarithm of the likelihood function is used to calculate the prediction loss model, and the prediction loss model represents the model loss value of the original distribution model.

[0107] In step S302 of some embodiments, the loss of the prediction loss model is calculated using a target loss function, that is, the target loss value is obtained by taking the derivative of the prediction loss model using the target loss function. The target loss value characterizes the trend of the model loss value and is used to characterize whether the training level of the original distribution model is sufficient. During training, the model loss value of the original distribution model decreases, but the variance increases. Therefore, if the model loss value is high but the variance is low, the original distribution model is considered underfitting; otherwise, it is considered overfitting. Therefore, the target loss value is used to determine whether the training level of the original distribution model meets the requirements, and the model parameters are adjusted based on the disease metric data to optimize the original distribution model.

[0108] In step S303 of some embodiments, since the target loss value characterizes whether the training level of the original distribution model is sufficient, the parameters of the original distribution model are adjusted according to the target loss value to adjust the disease measurement data in the model parameters in order to construct a better target distribution model. Therefore, the original disease correlation is corrected by the target distribution model to better match the original sample image.

[0109] Specifically, if the target loss value is positive, the disease measurement data in the original distribution model is reduced; if the target loss value is negative, the disease measurement data in the original distribution model is increased; when the target loss value is 0, the parameter adjustment of the original distribution model is stopped to obtain the target distribution model.

[0110] In steps S301 to S303 of the embodiments of this application, the original distribution model is transformed using a target loss function, i.e., the model parameters of the original distribution model are extracted. The model parameters are characterized by a likelihood function, and the logarithm of the likelihood function is then calculated to obtain a prediction loss model. The target loss value is obtained by taking the derivative of the prediction loss model using the target loss function. The target loss value indicates whether the training level of the original distribution model is sufficient. Then, the parameters of the original distribution model are adjusted according to the target loss value, i.e., the disease measurement data in the original distribution model is adjusted to obtain a target distribution model with higher correction accuracy. Thus, the target distribution model can correct the original disease correlation to better match the original sample image.

[0111] In step S105 of some embodiments, a target distribution model is obtained by optimizing the original distribution model. This target distribution model can correct the original disease association degree to make it more consistent with the original sample image. The original association degree prediction model outputs original disease association degree distribution information, which includes the predicted association degree for each disease category. This original disease association degree distribution information is input into the original distribution model, which is a pre-trained model. Therefore, for original sample images with obvious features, the predicted association degree corresponding to the original disease association degree is increased, while the predicted association degree corresponding to other disease categories is decreased, making the original disease association degree more obvious. If the features of the original sample image are not obvious, the predicted association degree corresponding to the original disease association degree in the original disease association degree distribution information is decreased, while the predicted association degree corresponding to the original disease association degree is increased, but the predicted association degree corresponding to the original disease association degree remains the highest. Therefore, by decreasing the probability value corresponding to the original disease association degree, the original disease association degree distribution information is corrected to obtain updated disease association degree distribution information, which then better matches the original sample image. Therefore, the updated disease correlation output by the target distribution model based on the original disease correlation is more consistent with the original sample image, and the updated disease correlation distribution information is also more consistent with the original sample image.

[0112] Please see Figure 4 In some embodiments, step S106 may include, but is not limited to, steps S401 to S402:

[0113] Step S401: Calculate the loss between the updated disease correlation degree and the original disease correlation degree to obtain the correlation degree loss value;

[0114] Step S402: Adjust the parameters of the original correlation prediction model based on the correlation loss value to obtain the target correlation prediction model.

[0115] In step S401 of some embodiments, since the updated disease correlation is more adapted to the original sample image, a correlation loss value is obtained by calculating the loss between the updated disease correlation and the original disease correlation, so as to determine the error rate of the original correlation prediction model through the correlation loss value. Specifically, the correlation loss value is obtained by calculating the difference between the updated disease correlation and the original disease correlation.

[0116] In step S402 of some embodiments, the parameters of the original correlation prediction model are adjusted according to the correlation loss value, that is, the parameters of the original correlation prediction model are adjusted so that the correlation loss value approaches zero, so as to obtain the target correlation prediction model.

[0117] In steps S401 to S402 of the embodiments of this application, the difference between the updated disease correlation and the original disease correlation is calculated to obtain a correlation loss value. The model parameters of the original correlation prediction model are then adjusted based on the correlation loss value, so that the correlation loss value approaches zero, thereby obtaining the target correlation prediction model. Therefore, by calculating the correlation loss value and then adjusting the original correlation prediction model based on the correlation loss value, a target correlation prediction model with higher prediction accuracy is trained.

[0118] In step S107 of some embodiments, a target image is acquired, and the target image is the disease diagnosis image data of the patient collected in real time from the hospital platform. The disease diagnosis image data can be obtained by real-time shooting or by scanning the human body with a medical scanning device. There are no specific restrictions on the acquisition method of the disease diagnosis image data.

[0119] Please see Figure 5 In some embodiments, the target correlation prediction model includes: a target feature extraction network, a target normalization layer, and a target unimodal loss function; step S108 may include, but is not limited to, steps S501 to S504:

[0120] Step S501: Input the target image into the target feature extraction network for feature extraction to obtain the target feature vector;

[0121] Step S502: Perform feature correlation analysis on the target feature vector through the target normalization layer to obtain the distribution information of candidate disease correlation degree;

[0122] Step S503: The candidate disease association degree distribution information is adjusted by a single peak using the target single-peak loss function to obtain the target disease association degree distribution information;

[0123] Step S504: Filter disease categories based on the distribution information of the correlation between the target disease and the target disease category.

[0124] In step S501 of some embodiments, the target image is input into a target feature extraction network to extract image features, and then the image features are vectorized to obtain target feature vectors. Therefore, the image features of the target image are represented by target feature vectors. Target images of different disease categories are also represented by target feature vectors to increase the correlation between different disease categories.

[0125] In step S502 of some embodiments, the target feature extraction network outputs a target feature vector to the target normalization layer. The target normalization layer performs feature correlation analysis on the target feature vector to calculate the correlation degree between each target feature and a preset feature in each target feature vector, i.e., the similarity between the target feature vector and the preset feature vector, to obtain candidate disease correlation distribution information. This candidate disease correlation distribution information is a correlation distribution and includes the predicted correlation degree for each disease category. Therefore, by inputting the target feature vector through the target normalization layer and outputting candidate disease classification distribution information, the disease prediction operation is simplified.

[0126] Specifically, the target normalization layer is the Softmax function, which transforms the target feature vector into mutually selected disease category distribution information, making the disease category prediction operation simple.

[0127] In step S503 of some embodiments, since the candidate disease association distribution information output by the target normalization layer has a maximum predicted association value for a certain disease category, and also a maximum predicted association value for other disease categories, but the lesion image has only one most likely disease category, and the probability of other disease categories is relatively small, the candidate disease association distribution information will have a multi-peak situation. Therefore, the candidate disease association distribution information is adjusted to a single peak by using a target single-peak function, that is, the peak with the highest predicted association value in the candidate disease association distribution information is obtained, and other peaks are corrected to obtain the target disease association distribution information, so that the association prediction of disease categories is more accurate.

[0128] In step S504 of some embodiments, after the target disease correlation distribution information is output by the target unimodal function, the disease correlation screening based on the target disease correlation distribution information is more accurate, so as to obtain a more accurate target disease category, thereby improving the accuracy of disease category prediction.

[0129] In steps S501 to S504 of the embodiments of this application, image features are extracted from the target image through a target feature extraction network. Then, the image features are vectorized to obtain target feature vectors. The target normalization layer performs feature correlation analysis on the target feature vectors to obtain candidate disease correlation distribution information. Then, the candidate disease correlation distribution information is adjusted to a single peak through a target unimodal function to obtain target disease correlation distribution information with only a single peak. Finally, the target disease correlation distribution information is filtered for disease correlation to obtain the target disease correlation degree, so as to obtain the target disease correlation degree with the highest value. The target disease correlation degree is the highest correlation degree of a certain disease category. The target disease correlation degree is used as reference data for disease category analysis to improve the accuracy of disease category prediction.

[0130] Please see Figure 6 In some embodiments, step S502 includes, but is not limited to, steps S601 to S602:

[0131] Step S601: The predicted correlation degree is obtained by peak filtering through the target unimodal loss function;

[0132] Step S602: Perform single-peak adjustment on the predicted correlations on both sides of the target correlation to obtain the distribution information of the target disease correlation.

[0133] In step S601 of some embodiments, since the candidate disease correlation distribution information includes the predicted correlation of each candidate disease category, the predicted correlation is filtered by the target unimodal loss function to obtain the highest predicted correlation as the target correlation after filtering out the peak according to the distribution of each predicted correlation.

[0134] Specifically, the partial derivative of the predicted correlation is obtained by taking the partial derivative of the target unimodal loss function, and the partial derivatives are sorted. If any partial derivative is positive and the next partial derivative is negative, it indicates that the peak value has been selected, and the predicted correlation corresponding to this partial derivative value is determined as the target correlation. The target unimodal loss function is shown in formula (4):

[0135]

[0136] In the formula, N is the total number of target images, and q i , j Let t be the predicted probability that the i-th target image belongs to the j-th disease category. i Let represent the target image label of the i-th target image. The target correlation can be obtained by filtering out the peak values ​​of the predicted correlation using a target unimodal function.

[0137] In step S603 of some embodiments, if the target correlation is selected, the predicted correlation of the candidate disease correlation distribution information is adjusted to a single peak based on the target correlation. That is, the predicted correlation to the left of the target correlation is adjusted to a monotonically increasing trend, while the predicted correlation to the right of the target correlation is adjusted to a monotonically decreasing trend, so as to obtain the target disease correlation distribution information with only a single peak distribution, so as to select the target disease category more accurately based on the target disease correlation distribution information.

[0138] In steps S601 to S603 as shown in the embodiments of this application, the predicted correlation is obtained by peak filtering through the target unimodal loss function, and the predicted correlation on both sides of the target correlation is adjusted to obtain the target disease correlation distribution information with only one peak, so as to make the prediction of the target disease category more accurate.

[0139] Please see Figure 7 In some embodiments, step S504 may include, but is not limited to, steps S701 to S702:

[0140] Step S701: Find the peak data of the target disease correlation distribution information to obtain the target peak value;

[0141] Step S702: Filter the predicted correlation based on the target peak value to obtain the target disease correlation.

[0142] In step S701 of some embodiments, the target disease correlation distribution information is a unimodal distribution of predicted correlation, so the predicted correlation corresponding to the peak value is directly obtained to obtain the target peak value, making it easy to obtain the target peak value without the influence of other peak values.

[0143] In step S702 of some embodiments, candidate disease categories are filtered by the target peak value to obtain the predicted correlation degree corresponding to the target peak value to determine the target disease correlation degree. Since the target disease correlation degree distribution information is a single-peak type distribution information, the peak value is directly obtained from the target disease correlation degree distribution information to obtain the target peak value, and then the predicted correlation degree corresponding to the target peak value is used as the target disease correlation degree to reduce the influence of filtering disease correlation degrees corresponding to other peak values. Furthermore, the target disease correlation degree is selected as reference data for disease category prediction to assist in disease diagnosis, thereby improving the accuracy of disease prediction.

[0144] In steps S701 to S702 of the embodiments of this application, since the target disease correlation distribution information of the single-peak type is obtained by correcting it, the target peak can be obtained by searching the peak of the target disease correlation distribution information. Then, the predicted correlation corresponding to the target peak is obtained to obtain the target disease correlation. The target disease correlation is used as reference data for disease category prediction, making disease category prediction both simple and accurate.

[0145] Please refer to Figure 8 , Figure 8 This is a schematic diagram of the operation flow of the image processing method according to an embodiment of this application, where p iTo predict the correlation, where j represents the disease category, this embodiment acquires original sample images and inputs them into an original feature extraction network to extract original feature vectors. A target normalization layer calculates the correlation of these original feature vectors to obtain initial disease correlation distribution information. The initial disease correlation distribution information is then adjusted using the original unimodal loss function to obtain the original disease correlation distribution information. Based on this original disease correlation distribution information, disease correlation is predicted to obtain the original disease correlation. Then, disease measurement data is calculated based on the original disease correlation and sample image labels. The original distribution model is constructed by using the disease measurement data, original disease correlation, and sample image labels as model parameters for a preset distribution model. The model parameters of the original distribution model are characterized by a likelihood function. A logarithmic calculation is then performed on the likelihood function to obtain the prediction loss model. The target loss value is calculated by differentiating the prediction loss model using the target loss function. The parameters of the original distribution model are then adjusted based on the target loss value, i.e., the model parameters of the disease measurement data in the original distribution model are adjusted to obtain a target distribution model with higher accuracy. The target distribution model corrects the original disease association to obtain an updated disease association. The updated disease association distribution information better matches the original sample image. The difference between the updated and original disease associations is then calculated to obtain the association loss value. The original association prediction model is adjusted based on this loss value to train a target association prediction model with higher prediction accuracy. The target association prediction model includes: a target feature extraction network, a target normalization layer, and a target unimodal loss function. The target feature extraction network extracts features from the target image to obtain image features, which are then vectorized to obtain target feature vectors. The target normalization layer performs feature correlation analysis on the target feature vectors to obtain candidate disease association distribution information. The target unimodal function adjusts the candidate disease association distribution information to obtain the target disease association distribution information. Finally, the peak value is obtained from the target disease association distribution information to obtain the target peak value, and the predicted association value corresponding to the target peak value is obtained to obtain the target disease association. Therefore, by first adjusting the model parameters of the original correlation prediction model to obtain the target correlation prediction model, and then using this target correlation prediction model to predict the disease correlation of the target image, and by including a target unimodal loss function, the disease correlation prediction becomes more accurate and more closely resembles the actual situation of the target image. Thus, by using the target correlation prediction model to predict the disease correlation of the target image, the target disease correlation is obtained. This target disease correlation is then used as reference data for disease category prediction to assist in disease prediction, thereby improving the accuracy of disease prediction.

[0146] Please see Figure 9This application also provides an image processing apparatus that can implement the above-described image processing method. The apparatus includes:

[0147] The data acquisition module 901 is used to acquire pre-labeled sample image data; wherein, the sample image data includes original sample images and sample image labels, and the sample image labels are used to characterize the disease association of the original sample images;

[0148] The original prediction module 902 is used to input the original sample image into the preset original correlation prediction model to predict the disease correlation and obtain the original disease correlation.

[0149] The model building module 903 is used to take the original disease correlation degree and sample image labels as model parameters of the preset distribution model to obtain the original distribution model;

[0150] The model optimization module 904 is used to optimize the original distribution model using a preset target loss function to obtain the target distribution model.

[0151] The correction module 905 is used to input the original disease association degree into the target distribution model for correction processing to obtain the updated disease association degree.

[0152] The parameter adjustment module 906 is used to adjust the parameters of the original correlation prediction model based on the updated disease correlation to obtain the target correlation prediction model.

[0153] Image acquisition module 907 is used to acquire target images;

[0154] The target prediction module 908 is used to input the target image into the target correlation prediction model to predict the disease correlation and obtain the target disease correlation.

[0155] The specific implementation of this image processing device is basically the same as the specific embodiment of the image processing method described above, and will not be repeated here.

[0156] This application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described image processing method. This computer device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0157] Please see Figure 10 , Figure 10 The hardware structure of a computer device according to another embodiment is illustrated. The computer device includes:

[0158] The processor 1001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0159] The memory 1002 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001 using the image processing method of the embodiments of this application.

[0160] Input / output interface 1003 is used to implement information input and output;

[0161] The communication interface 1004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0162] Bus 1005 transmits information between various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003, and communication interface 1004);

[0163] The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device via bus 1005.

[0164] This application embodiment also provides a storage medium, which is a computer-readable storage medium, storing a computer program that, when executed by a processor, implements the above-described image processing method.

[0165] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0166] The image processing method, apparatus, device, and storage medium provided in this application first construct an original distribution model, then optimize the original distribution model according to a preset target loss function to obtain a target distribution model. The target distribution model can correct the original disease correlation to approximate the original sample image. By adjusting the original disease correlation using the target distribution model to adapt to the original sample image, an updated disease correlation is obtained. Then, the parameters of the original correlation prediction model are adjusted using the updated disease correlation to obtain a target correlation prediction model that more accurately reflects the actual image situation. Therefore, predicting the target disease category using the target correlation prediction model yields a more accurate and consistent prediction of the target disease category.

[0167] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0168] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0169] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0170] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0171] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0172] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0173] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0174] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0175] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0176] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0177] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. An image processing method, characterized by, The method comprises: acquiring pre-labeled sample image data; wherein the sample image data comprises an original sample image and a sample image label, and the sample image label is used to represent the disease correlation degree of the original sample image; inputting the original sample image into a preset original correlation degree prediction model to predict the disease correlation degree, and obtaining an original disease correlation degree; using the original disease correlation degree and the sample image label as model parameters of a preset distribution model to obtain an original distribution model; optimizing the original distribution model through a preset target loss function to obtain a target distribution model; inputting the original disease correlation degree into the target distribution model for correction processing to obtain an updated disease correlation degree; adjusting the parameters of the original correlation degree prediction model according to the updated disease correlation degree to obtain a target correlation degree prediction model; acquiring a target image; inputting the target image into the target correlation degree prediction model to predict the disease correlation degree, and obtaining a target disease correlation degree; the step of using the original disease correlation degree and the sample image label as model parameters of a preset distribution model to obtain an original distribution model comprises: performing disease measurement calculation according to the original disease correlation degree and the sample image label to obtain disease measurement data; using the disease measurement data, the original disease correlation degree and the sample image label as model parameters of the preset distribution model to perform model construction, and obtaining the original distribution model; the target correlation degree prediction model comprises a target feature extraction network, a target normalization layer and a target unimodal loss function; the step of inputting the target image into the target correlation degree prediction model to predict the disease correlation degree, and obtaining a target disease correlation degree comprises: inputting the target image into the target feature extraction network to extract features, and obtaining a target feature vector; performing feature correlation analysis on the target feature vector through the target normalization layer to obtain candidate disease correlation degree distribution information; performing unimodal adjustment on the candidate disease correlation degree distribution information through the target unimodal loss function to obtain target disease correlation degree distribution information; performing disease correlation degree screening according to the target disease correlation degree distribution information to obtain the target disease correlation degree.

2. The method of claim 1, wherein, the step of optimizing the original distribution model through a preset target loss function to obtain a target distribution model comprises: performing model conversion on the original distribution model through the target loss function to obtain a prediction loss model; performing loss calculation on the prediction loss model through the target loss function to obtain a target loss value; adjusting the parameters of the original distribution model according to the target loss value to obtain the target distribution model.

3. The method of claim 1, wherein, the step of adjusting the parameters of the original correlation degree prediction model according to the updated disease correlation degree to obtain a target correlation degree prediction model comprises: performing loss calculation on the updated disease correlation degree and the original disease correlation degree to obtain a correlation degree loss value; adjusting the parameters of the original correlation degree prediction model according to the correlation degree loss value to obtain the target correlation degree prediction model.

4. The method of claim 3, wherein, The candidate disease association degree distribution information includes: a predicted association degree of each candidate disease category; the unimodal adjustment of the candidate disease association degree distribution information by the target unimodal loss function includes: Peak value screening processing of the predicted association degree by the target unimodal loss function to obtain a target association degree; Unimodal adjustment processing of the predicted association degrees on both sides of the target association degree to obtain the target disease association degree distribution information.

5. The method of claim 4, wherein, The disease association degree screening according to the target disease association degree distribution information to obtain the target disease association degree includes: Peak value data searching of the target disease association degree distribution information to obtain a target peak value; Screening processing of the predicted association degree according to the target peak value to obtain the target disease association degree.

6. An image processing apparatus characterized by comprising: The device includes: A data acquisition module configured to acquire pre-labeled sample image data; wherein the sample image data includes an original sample image and a sample image label, and the sample image label is used to represent a disease association degree of the original sample image; An original prediction module configured to input the original sample image into a preset original association degree prediction model to predict a disease association degree, and obtain an original disease association degree; A model construction module configured to use the original disease association degree and the sample image label as model parameters of a preset distribution model, and obtain an original distribution model; A model optimization module configured to use a preset target loss function to optimize the original distribution model, and obtain a target distribution model; A correction module configured to input the original disease association degree into the target distribution model for correction processing, and obtain an updated disease association degree; A parameter adjustment module configured to use the updated disease association degree to adjust parameters of the original association degree prediction model, and obtain a target association degree prediction model; An image acquisition module configured to acquire a target image; A target calculation module configured to input the target image into the target association degree prediction model to predict a disease association degree, and obtain a target disease association degree; The use of the original disease association degree and the sample image label as model parameters of a preset distribution model to obtain an original distribution model includes: Disease measurement calculation according to the original disease association degree and the sample image label to obtain disease measurement data; Model construction using the disease measurement data, the original disease association degree and the sample image label as model parameters of the preset distribution model to obtain the original distribution model; The target association degree prediction model includes a target feature extraction network, a target normalization layer and a target unimodal loss function; the input of the target image into the target association degree prediction model to predict a disease association degree includes: Feature extraction of the target image into the target feature extraction network to obtain a target feature vector; Feature association analysis of the target feature vector by the target normalization layer to obtain candidate disease association degree distribution information; The candidate disease association degree distribution information is unimodal adjusted by the target unimodal loss function to obtain target disease association degree distribution information. Disease association degrees are screened according to the target disease association degree distribution information to obtain the target disease association degrees.

7. A computer device, characterized by The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the image processing method of any one of claims 1 to 5 when executing the computer program.

8. A computer readable storage medium, the storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the image processing method of any one of claims 1 to 5.

Images