Medical image management method and device based on Internet of Things

Abstract

The invention discloses a medical image management method and device based on the Internet of Things, and relates to the technical field of image management, and the management method comprises the steps: deploying a medical image collection device, collecting data, and forming an original medical image data set; performing image preprocessing on the original medical image data set; carrying out multi-modal image fusion by utilizing a generative adversarial network (GAN); processing the fused image by using a deep learning noise reduction technology to obtain a first image; marking the first image, and uploading marked data and the image to a database to form a cloud medical data warehouse; on the basis of a cloud data warehouse, remote diagnosis is carried out, doctors or experts can remotely access medical image data and annotation information of patients, analysis and diagnosis are carried out by utilizing the medical image data and the annotation information, and timely medical services are provided for the patients.

Description

Technical Field

[0001] The present invention relates to the technical field of image management, and particularly to a medical image management method and device based on the Internet of Things. Background Art

[0002] Through the integration of technologies such as intelligent devices, sensors, and cloud computing, the Internet of Things technology realizes the real-time collection, transmission, and processing of data, and can be applied to the remote management of medical images. However, there are problems in the current medical image management system. Existing medical image acquisition devices often have a single function and cannot meet the requirements of multi-modal image acquisition. At the same time, the quality of the collected original image data is uneven, and complex preprocessing is required to be used for subsequent diagnosis and analysis, which increases the burden of manual operation and also affects the accuracy and efficiency of diagnosis. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art, and provide a medical image management method and device based on the Internet of Things. Compared with the prior art, the display system and the medical image management method based on the Internet of Things disclosed in the present invention can solve the above problems.

[0004] The present invention is realized through the following technical solutions: The present invention discloses a medical image management method based on the Internet of Things, and its structure includes:

[0005] Deploy medical image acquisition devices to collect data and form an original medical image data set;

[0006] Perform image preprocessing on the original medical image data set;

[0007] Use a generative adversarial network (GAN) for multi-modal image fusion;

[0008] Adopt deep learning denoising technology to process the fused image to obtain a first image;

[0009] Annotate the first image, and upload the annotated data and images to the database to form a cloud medical data warehouse;

[0010] Based on the cloud data warehouse, perform remote diagnosis.

[0011] Preferably, the steps of forming the original medical image data set include:

[0012] S101: Deploy multiple medical image acquisition devices and process the data through a dynamic sensor calibration algorithm;

[0013] S102: Synchronize the data from different devices in time to generate highly consistent synchronized image data;

[0014] S103: Apply the enhanced SVM anomaly detection algorithm to screen and eliminate outliers in the data;

[0015] S104: Integrate data from different modalities into the original medical image dataset.

[0016] Preferably, the steps of performing multi-modal image fusion using a generative adversarial network (GAN) include:

[0017] S301: Adopt a random sampling algorithm to select representative data as the GAN training set;

[0018] S302: Construct a deep convolutional generative adversarial network (DCGAN) to perform deep feature extraction and fusion on multi-modal data;

[0019] S303: Introduce the Wasserstein loss function optimization algorithm to stabilize the training process;

[0020] S304: Apply an adaptive image post-processing algorithm to generate a fused and enhanced image dataset.

[0021] Preferably, the steps of annotating the first image include:

[0022] S501: Augment the data to expand the training samples;

[0023] S502: Apply a deep CNN to extract image features and construct a feature space.

[0024] S503: Map the features to the label space through a fully connected network layer to achieve preliminary annotation.

[0025] S504: Use a conditional random field (CRF) to optimize pixel label assignment and generate accurately annotated medical image data.

[0026] Preferably, the steps of remotely diagnosing include

[0027] S601: Use a deep CNN to extract medical image features;

[0028] S602: Combine a recurrent neural network (RNN) to analyze the patient's historical data and obtain health status features;

[0029] S603: Fuse the medical image features and health status features, and use a deep belief network (DBN) to predict diagnostic suggestions;

[0030] S604: Combine the medical knowledge base and expert experience to correct and improve the diagnostic suggestions and generate a remote diagnosis report.

[0031] Preferably, the steps of generating a remote diagnosis report include:

[0032] Extract the patient's basic information and historical health records;

[0033] Apply decision tree algorithm for health data analysis to identify health patterns;

[0034] Compare with the health knowledge base to refine health suggestions;

[0035] Conduct manual verification and revision to generate a remote diagnosis report.

[0036] Preferably, the content of the remote diagnosis report can be customized, and the specific content of customization includes the selection of report templates and the editing of relevant items in the report.

[0037] Preferably, the steps of remote diagnosis further include:

[0038] Establish a patient health record system, integrate the patient's historical medical data and current medical imaging data, and provide comprehensive information for diagnosis;

[0039] Apply artificial intelligence-assisted diagnosis algorithm, combine medical knowledge base and clinical experience to provide preliminary diagnosis suggestions;

[0040] Multiple doctors review the preliminary diagnosis suggestions through a remote consultation mechanism.

[0041] Preferably, the steps of fusing medical imaging features and health status features include:

[0042] Use the attention mechanism to dynamically adjust the importance of different features and highlight key information;

[0043] Adopt a multi-modal fusion framework to fuse medical imaging features and health status features at multiple levels;

[0044] Apply graph convolutional network (GCN) to process the graph structure information of the patient's health data.

[0045] The present invention also discloses a medical imaging management device based on the Internet of Things. The management device includes:

[0046] A medical imaging acquisition module configured to deploy multiple medical imaging acquisition devices, process data through a dynamic sensor calibration algorithm, and output an original medical imaging data set;

[0047] An image processing module connected to the medical imaging acquisition module, configured to perform image preprocessing on the original medical imaging data set and conduct multi-modal image fusion to obtain a first image;

[0048] A data annotation and upload module connected to the image processing module, configured to annotate the first image and upload the annotated data and images to a database to form a cloud medical data warehouse;

[0049] The remote diagnosis module is connected to the data annotation and upload module.

[0050] The present invention discloses a medical image management method and device based on the Internet of Things. Compared with the prior art:

[0051] A medical image management method based on the Internet of Things in the present invention has the following key steps in its process: Deploy medical image acquisition devices (step 100), set up devices with Internet of Things functions such as X-ray machines and CT scanners inside medical institutions to collect two-dimensional, three-dimensional or time-series image data of patients, and store them in digital format as the original medical image dataset, providing a reliable basis for subsequent image processing; Image preprocessing (step 200), perform geometric correction, gray-scale adjustment, noise removal, cropping, etc. on the original dataset to improve the image quality and ensure the clarity and accuracy of the images; Use a generative adversarial network (GAN) for multi-modal image fusion (step 300). Since different modalities of medical images have their own characteristics and are complementary, through the adversarial training of the generator and discriminator of the GAN, fuse images of different modalities such as CT and MRI to generate a fused image that contains both structural information and functional information, providing a more comprehensive diagnostic basis for doctors; Deep learning noise reduction technology processing (step 400), train a deep neural network to learn the mapping relationship between noise and clean images, perform noise reduction on the fused image, and output a high-quality first image to assist doctors in accurately judging the condition; Annotate the first image (step 500), including information such as the lesion location and size, and upload the annotation data and the image to the cloud database to form a cloud medical data warehouse. The cloud database uses distributed storage technology to ensure the security and reliability of the data and supports remote access and sharing. Perform remote diagnosis based on the cloud data warehouse (step 600), doctors or experts can remotely access the medical image data and annotation information of patients, and use professional software for analysis and diagnosis to provide timely medical services for patients. Brief Description of the Drawings

[0052] Figure 1 It is a schematic diagram of the process of the image management method in an embodiment;

[0053] Figure 2 It is a schematic diagram of the process of forming the original medical image dataset in an embodiment;

[0054] Figure 3 It is a schematic diagram of the step process of multi-modal image fusion in an embodiment;

[0055] Figure 4 It is a schematic diagram of the process of the step of annotating the first image in an embodiment;

[0056] Figure 5 It is a schematic diagram of the step process of remotely diagnosing in an embodiment;

[0057] Figure 6 It is a schematic structural diagram of an image management device in an embodiment. Specific implementation manners

[0058] The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of various exemplary embodiments of the present disclosure as defined by the claims and their equivalents. This description includes various details to assist in the understanding, but these details will be considered merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications can be made to the various embodiments described herein without departing from the scope and spirit of the present disclosure. In addition, descriptions of well-known functions and structures may be omitted for clarity and conciseness.

[0059] It will be understood that when an element or layer is referred to as being "on," "connected to," or "coupled to" another element or layer, it can be directly on, directly connected or coupled to, that other element or layer, or there may also be one or more intervening elements or layers. When an element is referred to as being "directly on," "directly connected to," or "directly coupled to" another element or layer, there are no intervening elements or layers.

[0060] It will be understood that although terms such as first, second, third, etc. may be used herein to describe various elements, components, regions, layers, and / or sections, these elements, components, regions, layers, and / or sections should not be limited by these terms. These terms are used to distinguish one element, component, region, layer, or section from another element, component, region, layer, or section. Thus, without departing from the teachings of the exemplary embodiments, the first element, first component, first display region, first layer, or first section discussed below may be referred to as the second element, second component, second display region, second layer, or second section. In the drawings, for the sake of clarity, the dimensions of various elements, layers, etc. may be exaggerated.

[0061] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0062] During the medical diagnosis process, especially in scenarios such as dermatology or cosmetology that require detailed observation and analysis of the patient's skin condition, doctors use professional image acquisition devices to capture high-definition images of the patient's face for the diagnosis and analysis of skin problems.

[0063] Please refer to Figure 1 , a medical image management method based on the Internet of Things, the steps of which include:

[0064] Step 100: Deploy a medical image acquisition device, collect data, and form an original medical image data set;

[0065] Specifically, the establishment of the original dataset provides a reliable data foundation for subsequent image processing, analysis, and diagnosis. Various medical imaging acquisition devices can be deployed within medical institutions (such as hospitals, clinics, etc.), such as X-ray machines, CT scanners, MRI (Magnetic Resonance Imaging) devices, ultrasound devices, etc. The selected relevant medical imaging acquisition devices have Internet of Things (IoT) capabilities, that is, they can exchange data with a central system or other devices through a network; when a patient undergoes a medical imaging examination, the device will collect the patient's image data according to a preset program. The collected data may include two-dimensional images, three-dimensional volume data, time series data, etc., specifically depending on the type of the device and the purpose of the examination. The present invention does not make specific limitations. The collected data is saved as an original medical image dataset. The relevant dataset can be stored in a digital format, such as DICOM (Digital Imaging and Communications in Medicine) format, or in other formats. The original dataset contains all the information of the patient's images.

[0066] Step 200: Perform image preprocessing on the original medical image dataset;

[0067] The quality of the image is improved through image preprocessing. The present invention does not limit the specific preprocessing process. It can perform geometric correction on the original image to eliminate image distortion caused by reasons such as inaccurate device calibration and patient movement, perform gray-scale correction to adjust the brightness and contrast of the image to make the image clearer, or apply filtering techniques to remove noise in the image, such as Gaussian noise, salt-and-pepper noise, etc., or crop irrelevant parts in the image according to needs and only retain the region of interest.

[0068] Step 300: Perform multi-modal image fusion using a Generative Adversarial Network (GAN);

[0069] In clinical diagnosis, it is necessary to comprehensively consider medical imaging information of multiple modalities, such as the structural information of CT and the functional information of MRI. However, images of different modalities often have differences and complementarities. The present application solves related problems through a Generative Adversarial Network (GAN).

[0070] GAN consists of two networks, a generator and a discriminator. The generator is responsible for generating fake images, and the discriminator is responsible for distinguishing real images from fake images. Through continuous adversarial training, the generator can generate increasingly realistic fake images, and the discriminative ability of the discriminator will gradually improve. In the multi-modal image fusion of the present invention, medical images of different modalities are used as inputs, and GAN is used for fusion. The fused images combine the advantages of different modalities, providing more comprehensive information and helping doctors understand the patient's condition more accurately. Specifically, images of different modalities such as CT and MRI are input into the GAN model. The generator will attempt to generate a fake image that combines the information of CT and MRI, and the discriminator will strive to distinguish this fake image from the real CT or MRI images. Through continuous training and optimization, the generator can gradually learn how to generate fused images that have both the structural information of CT and the functional information of MRI. The generated fused images can provide a more comprehensive and accurate basis for diagnosis.

[0071] Step 400: Process the fused image using deep learning denoising technology to obtain the first image;

[0072] Specifically, the above-mentioned GAN can generate images that fuse multiple modalities of information, but some noise or artifacts may be introduced during the fusion process. These noise or artifacts will affect the quality and readability of the images. Therefore, we need to perform denoising processing on the fused images. The deep learning denoising technology is trained using a large number of pairs of noisy images and corresponding clean images to learn the mapping relationship between the noise and the clean images. Through training, the deep neural network can remove the noise and retain the details of the images. When processing the fused images, the fused images can be input into the trained deep learning denoising model, and the model performs denoising processing on the images and outputs the first image. The first image has higher quality and readability, and doctors can more accurately judge the patient's condition and formulate treatment plans by observing the first image.

[0073] Step 500: Annotate the first image, and upload the annotated data and the image to the database to form a cloud medical data warehouse;

[0074] Specifically, doctors or radiologists mark and annotate lesions, organs, tissues, etc. in the images. The annotated data can include information such as the location, size, shape, and type of the lesions, and can be used for medical research and teaching. After the generation of the first image is completed, the annotated data and the first image are uploaded to the cloud database together. The cloud database is a platform for centralized storage and management of medical image data. In one embodiment, the cloud database adopts distributed storage and backup technology to ensure the security and reliability of the data. At the same time, the cloud database also supports remote access and sharing of data, enabling doctors or experts to access the required medical image data anytime and anywhere.

[0075] Step 600: Conduct remote diagnosis based on the cloud data warehouse.

[0076] Specifically, doctors or experts can remotely access the patient's medical imaging data and annotation information via the Internet, analyze and diagnose the imaging data using professional medical imaging diagnosis software, thereby obtaining the patient's condition and treatment suggestions. Remote diagnosis enables doctors or experts to provide medical services to patients anytime and anywhere. Regions with remote locations or scarce medical resources can obtain professional doctors' diagnoses and treatment suggestions through remote diagnosis without having to visit the hospital in person.

[0077] Please refer to Figure 2 , in one embodiment, the steps of forming the original medical imaging data set include:

[0078] S101: Deploy multiple medical imaging acquisition devices and process the data through a dynamic sensor calibration algorithm;

[0079] Specifically, the multiple deployed medical imaging acquisition devices include but are not limited to X-ray machines, CT (Computed Tomography) scanners, MRI (Magnetic Resonance Imaging) machines, ultrasound devices, etc. Different devices capture different types of tissue structures and pathological changes. During the data acquisition process, using a dynamic sensor calibration algorithm can monitor and adjust the parameters of the sensors in real time, such as sensitivity, gain, offset, etc., to compensate for performance drift caused by environmental temperature changes, equipment aging, or long-term use.

[0080] S102: Synchronize the data from different devices in time to generate highly consistent synchronized imaging data;

[0081] Specifically, a high-precision time synchronization protocol, such as the Network Time Protocol (NTP) or the Precision Time Protocol (PTP), can be used to ensure that the system clocks of all devices in the network are highly consistent. For medical imaging data, time synchronization ensures the continuity and consistency of the imaging sequence in time. For example, in cardiac dynamic imaging, images of different modalities (such as ultrasound and MRI) are precisely synchronized to accurately capture the cardiac motion cycle.

[0082] S103: Apply an enhanced SVM anomaly detection algorithm to screen and eliminate outliers in the data; the enhanced SVM anomaly detection algorithm constructs a hyperplane in the feature space to distinguish normal data from abnormal data.

[0083] S104: Integrate the data from different modalities into the original medical imaging data set.

[0084] Specifically, integrating data from different modalities can form a more comprehensive and three-dimensional medical image dataset. The integration process takes into account the spatial alignment, resolution matching, and information complementarity between different modality data. Spatial alignment usually relies on image registration, which aligns images of different modalities into the same coordinate system to ensure spatial consistency. Resolution matching upsamples low-resolution images or downsamples high-resolution images to make the resolutions of different modality images consistent. Information complementarity fully considers the characteristics of different modality data during analysis, such as combining the bone structure of CT with the soft tissue information of MRI to more accurately evaluate the scope and nature of lesions.

[0085] Please refer to Figure 3 , in one embodiment, the steps of using a generative adversarial network (GAN) for multi-modal image fusion include:

[0086] S301: Adopt a random sampling algorithm to select representative data as the GAN training set;

[0087] S302: Construct a deep convolutional generative adversarial network (DCGAN) to perform deep feature extraction and fusion on multi-modal data;

[0088] S303: Introduce the Wasserstein loss function optimization algorithm to stabilize the training process;

[0089] S304: Apply an adaptive image post-processing algorithm to generate a fused and enhanced image dataset.

[0090] In the above embodiments, in step S301, a random sampling algorithm is used to select representative data as the GAN training set. The random sampling algorithm can ensure that samples are evenly selected from the original multi-modal image database, avoiding data bias. During the selection process, considering the characteristics of different modal images, such as resolution, contrast, noise level, etc., to ensure that the training set can fully represent the characteristics of the entire database. In step S302, a deep convolutional generative adversarial network (DCGAN) is constructed to perform deep feature extraction and fusion on multi-modal data. DCGAN is a variant of GAN that combines the structure of a deep convolutional neural network (CNN) and can more effectively extract the deep features of images. In DCGAN, the generator is responsible for converting random noise into realistic images, while the discriminator is responsible for distinguishing between the generated images and the real images. In the task of multi-modal image fusion, the generator can receive images from different modalities as inputs and extract their respective features through deep convolutional layers. These features are fused at an intermediate layer of the generator to generate a fused image containing multi-modal information. The discriminator will continuously learn and optimize to more accurately distinguish between the fused images and the real images, thereby driving the generator to continuously generate higher-quality fused images. The training process of GAN often has unstable problems, such as mode collapse, vanishing gradients, etc. To solve these problems, in step S303, the Wasserstein loss function optimization algorithm is introduced. The Wasserstein loss function is a loss function based on the earth mover's distance, which can more stably measure the difference between the generated images and the real images. In step S304, an adaptive image post-processing algorithm is applied to generate a fused enhanced image dataset.

[0091] Please refer to Figure 4 , in one embodiment, the steps of annotating the first image include:

[0092] S501: Augment the data to expand the training samples. In one embodiment, various means such as rotation, scaling, flipping, cropping, and adding noise are used to transform the original image data, thereby generating new and diverse training samples;

[0093] S502: Apply a deep CNN to extract image features and construct a feature space.

[0094] S503: Map the features to the label space through a fully connected network layer to achieve preliminary annotation. The fully connected layer can integrate the features extracted by the previous layer and output the preliminary label corresponding to each pixel or region according to the requirements of the learning task.

[0095] S504: Optimize pixel label assignment using Conditional Random Field (CRF) to generate accurately labeled medical image data. Take the preliminary labeling result as the input of CRF, and optimize the preliminary labeling by considering the spatial adjacency relationship between pixels and the compatibility between labels. CRF can capture local and global context information in the image, correct possible errors in the preliminary labeling, and thus generate more accurate and consistent labeling results.

[0096] Please refer to Figure 5 , in one embodiment, the steps of remote diagnosis include

[0097] S601: Extract medical image features using a deep CNN;

[0098] Specifically, as an image processing tool, the deep CNN automatically learns and extracts high-level features in the image. In the embodiment, a labeled medical image dataset is formed, and the relevant dataset covers various common and rare diseases to ensure the generalization ability of the model. Process the dataset through a deep CNN architecture, which includes multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layer is responsible for extracting local features of the image, such as edges, textures, etc.; the pooling layer is used to reduce the dimension of the feature map and reduce the computational amount; the fully connected layer integrates these features to form an overall description of the image. Through training on a large-scale dataset, the deep CNN can gradually learn the typical image features of various diseases.

[0099] S602: Analyze the patient's historical data in combination with a Recurrent Neural Network (RNN) to obtain health status features;

[0100] Specifically, evaluate the patient's current health status through the patient's historical data, including medical records, physiological indicators, medication history, etc. The relevant data has the characteristics of time series, and there is an association between data at different time points. The RNN is a neural network model for processing sequence data. Organize the patient's historical data into a time series form, with each time step corresponding to a data point or a data segment. Through an RNN architecture, such as a Long Short-Term Memory network (LSTM) or a Gated Recurrent Unit (GRU), capture the time-dependent relationship in the data. Through training, the RNN learns the law of the patient's health status changing over time, and thus extracts the features reflecting the patient's current health status.

[0101] S603: Fuse medical image features and health status features, and use a Deep Belief Network (DBN) to predict diagnostic suggestions;

[0102] Specifically, medical image features and health status features are concatenated or weighted and fused to form a comprehensive feature vector. The feature vector is input into the DBN, and through the learning of multiple layers of latent variables, a probability distribution prediction of the disease type is obtained. Each layer of the DBN is regarded as a feature extractor, which maps the input data to a higher-level feature space. Through training, the DBN can gradually adjust its parameters to make the prediction of the disease type more accurate. Based on the output of the DBN, preliminary diagnostic suggestions can be given.

[0103] S604: Combine the medical knowledge base and expert experience to correct and improve the diagnostic suggestions and generate a remote diagnostic report.

[0104] Specifically, medical experts can be invited to review and evaluate the preliminary diagnostic suggestions. Based on their professional knowledge and clinical experience, experts can correct and improve the unreasonable or uncertain parts in the suggestions. At the same time, experts can also provide additional diagnostic opinions or suggestions to enrich the content of the remote diagnostic report.

[0105] In one embodiment, the steps of generating a remote diagnostic report include:

[0106] Extract the patient's basic information and historical health records. Specifically, information such as age, gender, past medical history, family genetic history, allergy history, and recent physiological indicators can be collected.

[0107] Apply the decision tree algorithm to analyze the health data, identify the health pattern, and construct a series of decision nodes based on the patient's health data, such as blood pressure, blood sugar level, cholesterol content, etc. Each node represents a possible health status or disease risk. By observing these nodes, the patient's current health status and potential health risks or disease trends can be identified.

[0108] For easy understanding, an example of the decision tree construction and analysis process is given:

[0109] Select age as the root node. Age is one of the important factors affecting health status.

[0110] Suppose this patient belongs to the middle-aged age group (for example, 45 - 60 years old).

[0111] The first-layer decision node (blood pressure) makes a decision on the patient's blood pressure data.

[0112] If the patient's blood pressure is within the normal range (for example, 120 / 80 mmHg), then we continue to analyze other indicators.

[0113] If the blood pressure is high (for example, 140 / 90 mmHg or higher), it is marked as a hypertension risk, and further monitoring or treatment may be recommended.

[0114] The second - layer decision node (blood glucose level), for patients with normal blood pressure, analyze the blood glucose level;

[0115] If the patient's blood glucose level is within the normal range (e.g., fasting blood glucose < 6.1 mmol / L), continue the analysis.

[0116] If the blood glucose is on the high side (e.g., fasting blood glucose 6.1 - 7.0 mmol / L), mark it as a pre - diabetes risk and recommend regular re - examination.

[0117] If the blood glucose reaches the diabetes range (e.g., fasting blood glucose ≥ 7.0 mmol / L), directly mark it as diabetes and recommend seeking medical treatment.

[0118] The third - layer decision node (cholesterol content), for patients with normal blood glucose levels, analyze the cholesterol content;

[0119] If the patient's cholesterol content is within the normal range (e.g., total cholesterol < 5.2 mmol / L), comprehensively evaluate the health status as good and recommend maintaining a healthy lifestyle;

[0120] If the cholesterol is on the high side (e.g., total cholesterol 5.2 - 6.2 mmol / L), mark it as a cardiovascular disease risk and recommend improving diet, increasing exercise, etc.;

[0121] If the cholesterol is very high (e.g., total cholesterol ≥ 6.2 mmol / L), directly mark it as a high cardiovascular disease risk and recommend seeking medical treatment for further evaluation and treatment.

[0122] In a scenario, the patient is 50 years old, with a blood pressure of 130 / 85 mmHg (normal), a blood glucose level of 5.5 mmol / L (normal), and a cholesterol content of 5.8 mmol / L (on the high side). At the root node, the patient is classified as middle - aged. At the first - layer decision node, the patient's blood pressure is normal, so continue the analysis. At the second - layer decision node, the patient's blood glucose level is also normal, so continue the analysis. At the third - layer decision node, the patient's cholesterol content is on the high side, so it is marked as a cardiovascular disease risk and recommended to improve diet, increase exercise, etc.

[0123] Compare with the health knowledge base to refine the health advice. The health knowledge base is a database containing a large amount of medical knowledge, clinical guidelines, and best practices. Through comparison, the health advice for the patient can be further refined to make it more in line with medical standards and the actual situation of the patient.

[0124] Conduct manual verification and revision to generate a remote diagnosis report.

[0125] In one embodiment, the content of the remote diagnosis report can be customized. The specific content of customization includes the selection of report templates and the editing of relevant items in the report.

[0126] Specifically, in terms of the selection of report templates, a variety of preset report templates are provided. These templates are designed according to different medical scenarios, disease types, or health assessment needs. For example, for patients with cardiovascular diseases, a template including items such as electrocardiogram analysis, blood pressure monitoring results, and blood lipid levels can be selected; for patients with diabetes, a template including items such as blood glucose monitoring, glycated hemoglobin, and insulin usage can be selected. Doctors or medical institutions can select appropriate report templates according to the specific conditions and needs of patients to ensure the pertinence and practicality of the report content.

[0127] In terms of the editing of relevant items in the report, flexible editing functions are provided. Doctors or medical staff can fill in or modify the data in the report according to the actual examination results and health conditions of patients. At the same time, the system also supports adding custom items to meet some special or personalized assessment needs. For example, for certain specific diseases or symptoms, doctors can add relevant examination items or assessment indicators to more comprehensively reflect the health status of patients.

[0128] In one embodiment, the steps of remote diagnosis further include:

[0129] Establish a patient health record system, integrate the patient's historical medical data and current medical imaging data, and provide comprehensive information for diagnosis;

[0130] Apply artificial intelligence-assisted diagnosis algorithms, combine medical knowledge bases and clinical experience, and provide preliminary diagnosis suggestions;

[0131] Multiple doctors review the preliminary diagnosis suggestions through a remote consultation mechanism.

[0132] In one embodiment, the steps of fusing medical imaging features and health status features include:

[0133] Use the attention mechanism to dynamically adjust the importance of different features and highlight key information;

[0134] Adopt a multi-modal fusion framework to fuse medical imaging features and health status features at multiple levels;

[0135] Apply a graph convolutional network (GCN) to process the graph structure information of the patient's health data.

[0136] Please refer to Figure 6 , the present invention also discloses an Internet of Things-based medical imaging management device. The management device includes:

[0137] A medical image acquisition module, configured to deploy multiple medical image acquisition devices, process data through a dynamic sensor calibration algorithm, and output an original medical image dataset;

[0138] An image processing module, connected to the medical image acquisition module, configured to perform image preprocessing on the original medical image dataset, perform multi-modal image fusion, and obtain a first image;

[0139] A data annotation and uploading module, connected to the image processing module, configured to annotate the first image, and upload the annotated data and image to a database to form a cloud medical data warehouse;

[0140] A remote diagnosis module, connected to the data annotation and uploading module.

[0141] In summary, the present invention discloses a medical image management method based on the Internet of Things, and its process includes the following key steps: deploying medical image acquisition devices (step 100), setting devices with Internet of Things functions such as X-ray machines and CT scanners inside medical institutions to collect two-dimensional, three-dimensional or time-series image data of patients, and storing it in digital format as an original medical image dataset to provide a reliable basis for subsequent image processing; image preprocessing (step 200), performing geometric correction, gray level adjustment, noise removal, cropping, etc. on the original dataset to improve image quality and ensure the clarity and accuracy of the image; using a generative adversarial network (GAN) for multi-modal image fusion (step 300). Since different modalities of medical images have their own characteristics and are complementary, through the adversarial training of the generator and discriminator of GAN, images of different modalities such as CT and MRI are fused to generate a fused image that contains both structural information and functional information, providing a more comprehensive diagnostic basis for doctors; deep learning denoising technology processing (step 400), by training a deep neural network to learn the mapping relationship between noise and clean images, denoise the fused image, and output a high-quality first image to help doctors accurately judge the condition; annotating the first image (step 500), including information such as the location and size of the lesion, and uploading the annotated data and image to the cloud database to form a cloud medical data warehouse. The cloud database uses a distributed storage technology to ensure data security and reliability, and supports remote access and sharing. Remote diagnosis is performed based on the cloud data warehouse (step 600). Doctors or experts can remotely access the medical image data and annotation information of patients, and use professional software for analysis and diagnosis to provide timely medical services for patients.

[0142] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

[0143] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the said element.

Claims

1. A medical image management method based on the Internet of Things, characterized in that the steps include: Deploy medical image acquisition equipment, collect data, and form original medical image data sets; Perform image preprocessing on the original medical image dataset; Multimodal image fusion using generative adversarial networks (GANs); The fused image is processed using deep learning noise reduction technology to obtain the first image; annotating the first image, and uploading the annotated data and the image to a database to form a cloud-based medical data warehouse; Remote diagnosis based on cloud data warehouse.

2. The method for managing medical images based on the Internet of Things according to claim 1, characterized in that: The steps to form the original medical image dataset include: S101: deploy multiple medical image acquisition devices and process data through a dynamic sensor calibration algorithm; S102: Time synchronization of data from different devices to generate highly consistent synchronized image data; S103: Applying the enhanced SVM anomaly detection algorithm to screen and remove outliers in the data; S104: Integrate data from different modalities into an original medical imaging dataset.

3. The method for managing medical images based on the Internet of Things as claimed in claim 1, characterized in that: The steps of using generative adversarial networks (GAN) for multimodal image fusion include: S301: Use a random sampling algorithm to select representative data as a GAN training set; S302: Construct a deep convolutional generative adversarial network (DCGAN) to perform deep feature extraction and fusion on multimodal data; S303: Introduce the Wasserstein loss function optimization algorithm to stabilize the training process; S304: Apply an adaptive image post-processing algorithm to generate a fused enhanced image data set.

4. The method for managing medical images based on the Internet of Things as claimed in claim 3, characterized in that: The steps of labeling the first image include: S501: Enhance the data and expand the training samples; S502: Apply deep CNN to extract image features and construct feature space; S503: Mapping features to label space through a fully connected network layer to achieve preliminary labeling; S504: Optimize pixel label assignment using conditional random fields (CRF) to generate accurately labeled medical imaging data.

5. The method for managing medical images based on the Internet of Things as claimed in claim 4, characterized in that: The steps to perform a remote diagnosis include: S601: Extracting medical image features using deep CNN; S602: Analyze the patient's historical data in combination with a recurrent neural network (RNN) to obtain health status characteristics; S603: Fusion of medical image features and health status features, using deep belief networks (DBNs) to predict diagnostic recommendations; S604: Combine the medical knowledge base and expert experience to revise and improve the diagnostic suggestions and generate a remote diagnosis report.

6. The method for managing medical images based on the Internet of Things as claimed in claim 4, characterized in that: The steps to generate a remote diagnostic report include: Extract basic patient information and historical health records; Apply decision tree algorithms to analyze health data and identify health patterns; Compare the health knowledge base and refine health recommendations; Perform manual verification and revision, and generate remote diagnostic reports.

7. The method for managing medical images based on the Internet of Things as claimed in claim 4, characterized in that: The content of the remote diagnosis report can be customized, and the specific content of the customization includes the selection of a report template and the editing of related items in the report.

8. The method for managing medical images based on the Internet of Things as claimed in claim 7, characterized in that: The remote diagnosis step further includes: Establish a patient health record system to integrate patients’ historical medical data and current medical imaging data to provide comprehensive information for diagnosis; Apply artificial intelligence-assisted diagnosis algorithms, combined with medical knowledge base and clinical experience, to provide preliminary diagnostic suggestions; Multiple doctors reviewed the preliminary diagnosis recommendations through a remote consultation mechanism.

9. The method for managing medical images based on the Internet of Things as claimed in claim 4, characterized in that: The steps of fusing medical image features and health status features include: Use the attention mechanism to dynamically adjust the importance of different features and highlight key information; A multimodal fusion framework is used to fuse medical image features and health status features at multiple levels; Graph convolutional networks (GCNs) are used to process the graph structure information of patient health data.

10. A medical image management device based on the Internet of Things, characterized in that: Used to execute the image management method according to any one of claims 1 to 9, the management device comprises: a medical image acquisition module configured to deploy a plurality of medical image acquisition devices, process data through a dynamic sensor calibration algorithm, and output a raw medical image data set; An image processing module, connected to the medical image acquisition module, configured to perform image preprocessing and multimodal image fusion on the original medical image data set to obtain a first image; a data annotation and uploading module, connected to the image processing module, configured to annotate the first image and upload the annotated data and the image to a database to form a cloud medical data warehouse; A remote diagnosis module is connected to the data labeling and uploading module.

Images