Intelligent self-inspection and data protection integrated method and system

Abstract

The invention discloses an intelligent self-checking and data protection integrated method and system. The method comprises the following steps: acquiring a to-be-analyzed image by adopting anonymization and minimization principles; inputting the to-be-analyzed image into a pre-trained multi-reference object and disease feature relation segmentation model for focus region segmentation and feature extraction; acquiring a historical health image or a standard health image of the user as a comparison sample image by adopting anonymization and minimization principles; inputting the to-be-analyzed image and the comparison sample image into a disease feature change estimation model, and calculating the feature change of the lesion region in combination with the segmentation feature vector to obtain a second feature vector; calculating a final disease feature evaluation result according to the first feature vector and the second feature vector; and outputting a final disease feature evaluation result and a corresponding focus area mask position. By implementing the method provided by the invention, the accuracy of disease detection can be improved, and the security and privacy of the user data in the whole processing process can be ensured.

Description

Technical Field

[0001] This invention relates to personal health management methods, and more specifically to an intelligent self-testing and data protection integrated method and system. Background Technology

[0002] With the rapid development of information and medical technologies, an increasing number of intelligent health management systems are being applied to the field of personal health management. These systems utilize image analysis technology to assist in the diagnosis of diseases, particularly in areas such as skin lesions and X-ray imaging. They analyze medical images using machine learning and deep learning algorithms to identify potential health problems.

[0003] However, existing self-testing systems often rely on a single reference or fixed feature set for disease assessment, which may lead to misdiagnosis or missed diagnosis. Furthermore, the lack of precision in calculating specific parameters for lesion areas affects the reliability of the final diagnostic results. While providing efficient and convenient services, effectively protecting users' personal health information security has become an urgent issue. Some current systems fail to adequately consider data encryption and anonymization measures, increasing the risk of user information exposure. Some models trained on specific types of samples may exhibit poor adaptability when facing individual differences, limiting their broad applicability and clinical value. Traditional feature extraction methods typically focus only on local or global features, neglecting the correlation between the two and the interaction between multi-level features, thus reducing diagnostic accuracy.

[0004] Therefore, it is necessary to design a new method to improve the accuracy of disease detection while ensuring the security and privacy of user data throughout the entire process. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent self-inspection and data protection integrated method and system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent self-inspection and data protection integrated method, comprising:

[0007] The image to be analyzed is obtained using the principles of anonymization and minimization.

[0008] The image to be analyzed is input into a pre-trained multi-reference object and disease feature relationship segmentation model for lesion region segmentation and feature extraction to obtain a preliminary feature set. The preliminary feature set includes a first feature vector and a segmentation feature vector, wherein the first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information.

[0009] The user's historical health images or standard health images are obtained as comparison sample images using the principles of anonymization and minimization;

[0010] The image to be analyzed and the comparison sample image are input into the disease feature change estimation model, and the feature change of the lesion region is calculated by combining the segmentation feature vector to obtain the second feature vector;

[0011] The final disease feature assessment result is calculated based on the first feature vector and the second feature vector;

[0012] Output the final disease feature assessment results and the corresponding lesion area mask location.

[0013] The further technical solution is as follows: the image to be analyzed is input into a pre-trained multi-reference object and disease feature relationship segmentation model for lesion region segmentation and feature extraction to obtain a preliminary feature set. The preliminary feature set includes a first feature vector and a segmentation feature vector, wherein the first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information, including:

[0014] After performing data augmentation on the image to be analyzed, it is input into the multi-reference object and disease feature relationship segmentation model;

[0015] A feature vector is generated based on the image to be analyzed by a multi-reference object and disease feature relationship segmentation model.

[0016] The feature vector is processed sequentially through a normalization layer, a self-attention mechanism, and a multilayer perceptron to obtain lesion segmentation information and related pathological feature information. The lesion segmentation information includes segmentation feature vectors, and the related pathological feature information includes the mask position of the lesion region and its corresponding pathological feature value.

[0017] Calculate the specific parameters of the abnormal region based on the proportional relationship between the pathological features and the actual physiological structure.

[0018] The specific parameters of the abnormal region, the segmentation feature vector, the mask position of the lesion region and its corresponding pathological feature value are combined to form a preliminary feature set.

[0019] The further technical solution is as follows: the segmentation model based on the relationship between multiple reference objects and disease features generates a feature vector according to the image to be analyzed, including:

[0020] The convolutional layer of the multi-reference object and disease feature relationship segmentation model performs convolution operations and downsampling, and then processes it through batch normalization and activation function to obtain the superficial lesion shape features and surrounding tissue features of the image to be analyzed.

[0021] The C2f layer is used to extract deep internal lesion structural features, lesion marker features, and surrounding tissue detail features from the image to be analyzed.

[0022] SPPF layers are used to perform pooling operations on shallow lesion shape features and surrounding tissue features, deep lesion internal structure features, lesion marker features, and surrounding tissue detail features to generate feature vectors.

[0023] The further technical solution is as follows: the calculation of specific parameters of the abnormal region based on the proportional relationship between the pathological feature information and the actual physiological structure includes:

[0024] Determine the mask pixel size information of the abnormal area based on the pathological feature information;

[0025] Obtain the actual physiological structure size ratio stored from the multi-reference object and disease feature relationship segmentation model;

[0026] The actual size of the abnormal region is calculated based on the pixel size information of the abnormal region mask, and compared with the proportion of the actual physiological structure size to obtain specific parameters.

[0027] The further technical solution is as follows: The image to be analyzed and the comparison sample image are input into the disease feature change estimation model, and the lesion region feature changes are calculated by combining the segmentation feature vector to obtain the second feature vector, including:

[0028] The image to be analyzed and the comparison sample image are input into the disease feature change estimation model for encoding, forming two sets of feature vectors;

[0029] Extract corresponding surrounding tissue features, detail features, lesion shape features, internal structural features, and biomarker features from the image to be analyzed and the comparison sample image;

[0030] Based on the two sets of feature vectors and the segmentation feature vector, the feature changes of the lesion region are calculated to obtain preliminary feature vectors;

[0031] The spatial size of the preliminary feature vector is adjusted using upsampling and downsampling techniques in the multiple feature coding layer, and a fusion operation is performed to integrate information from different scales to obtain a fused preliminary feature map.

[0032] Based on the fused preliminary feature map, the number of channels is adjusted to 1, and a hybrid structure is used for downsampling to obtain the adjusted environmental feature map, considering the surrounding tissue features and detailed features.

[0033] For the lesion shape features, internal structure features and marker features, based on the fused preliminary feature map, the number of channels is adjusted by convolution and upsampling is performed using the nearest neighbor interpolation method to obtain the adjusted lesion feature map;

[0034] The adjusted environmental feature map and the adjusted lesion feature map are convolved once, and then concatenated along the channel dimension to form the final fused feature map.

[0035] The difference between the fused feature vectors is calculated by combining the segmented feature vectors with the final fused feature map to determine the second feature vector.

[0036] The further technical solution is as follows: The step of calculating the difference in the fused feature vectors by combining the segmented feature vectors with the final fused feature map to determine the second feature vector includes:

[0037] Align the segmented feature vector with the final fused feature map;

[0038] The difference between the feature vectors after feature alignment is calculated using a mathematical distance metric to form a second feature vector.

[0039] The further technical solution is as follows: the calculation of the final disease feature assessment result based on the first feature vector and the second feature vector includes:

[0040] Multiply the first feature vector and the second feature vector together to obtain the feature vector multiplication result;

[0041] The result of multiplying the feature vectors is sequentially passed through convolutional layers, deconvolutional layers, multiple convolutional layers, and activation functions in the decoder to recover the actual pathological feature value represented by each pixel;

[0042] By combining the mask position of the lesion area, the corresponding pathological feature values ​​are extracted from the recovered pathological feature map, and the final disease feature assessment result is determined.

[0043] This invention also provides an intelligent self-testing and data protection integrated system, including:

[0044] The first acquisition unit is used to acquire the image to be analyzed using the principles of anonymization and minimization.

[0045] The segmentation and extraction unit is used to input the image to be analyzed into a pre-trained multi-reference object and disease feature relationship segmentation model to perform lesion region segmentation and feature extraction to obtain a preliminary feature set. The preliminary feature set includes a first feature vector and a segmentation feature vector, wherein the first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information.

[0046] The second acquisition unit is used to acquire the user's historical health image or standard health image as a comparison sample image by adopting the principles of anonymization and minimization.

[0047] The change calculation unit is used to input the image to be analyzed and the comparison sample image into the disease feature change estimation model, and calculate the feature change of the lesion region in combination with the segmentation feature vector to obtain the second feature vector;

[0048] The final result calculation unit is used to calculate the final disease feature assessment result based on the first feature vector and the second feature vector;

[0049] The output unit is used to output the final disease feature assessment results and the corresponding lesion area mask location.

[0050] The present invention also 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 above-described method.

[0051] The present invention also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0052] The advantages of this invention compared to existing technologies are as follows: This invention ensures user privacy by employing anonymization and minimization principles, protecting user personal data during the acquisition of images to be analyzed and historical health images; it accurately segments and extracts features from lesion regions by inputting the images to be analyzed into a pre-trained multi-reference object and disease feature relationship segmentation model, obtaining a preliminary feature set containing lesion region mask locations and pathological feature values, thus improving detection accuracy; it uses historical health images as comparison samples and calculates feature changes in lesion regions through a disease feature change estimation model, further enhancing the ability to identify disease changes; it combines preliminary features and change features to comprehensively calculate the final disease feature evaluation result, improving the accuracy and reliability of diagnosis; and the entire process improves disease detection effectiveness through intelligent algorithms without disclosing sensitive user information.

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart illustrating the intelligent self-testing and data protection integrated method provided in this embodiment of the invention;

[0056] Figure 2 A schematic block diagram of an intelligent self-testing and data protection integrated system provided in an embodiment of the present invention;

[0057] Figure 3 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0060] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0061] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0062] Please see Figure 1 , Figure 1This is a schematic flowchart illustrating the intelligent self-inspection and data protection integrated method provided in this embodiment of the invention. This method is applied to a server that interacts with terminals. By combining intelligent self-inspection with data protection, it improves the accuracy of disease detection and the security of user data. First, the image to be analyzed is processed using anonymization and minimization principles to ensure user privacy is not compromised. Next, a multi-reference object and disease feature relationship segmentation model is used to segment lesion regions and extract features, generating high-quality feature vectors for disease assessment. Through data augmentation and deep learning models, pathological features and tissue details of the image are extracted, further improving the accuracy of lesion detection. By comparing historical healthy images or standard images with the segmentation features, changes in lesion regions are assessed to ensure accurate prediction of disease progression. Self-attention mechanisms and multilayer perceptrons are used to optimize feature processing, and refined analysis is performed in conjunction with environmental and lesion features, enhancing detection sensitivity. Finally, the difference in feature vectors is calculated to obtain accurate disease assessment results. Throughout the process, data protection technologies, such as image anonymization and minimization, are employed to ensure the security and privacy of user data during processing.

[0063] Figure 1 This is a flowchart illustrating the intelligent self-inspection and data protection integrated method provided in this embodiment of the invention. Figure 1 As shown, the method includes the following steps S110 to S160.

[0064] S110. Obtain the image to be analyzed using the principles of anonymization and minimization.

[0065] In this embodiment, before acquiring any health-related images, it is necessary to

[0066] Obtain explicit consent from users. This means clearly informing users why these images are needed, how they will be used, and who can access this information.

[0067] Remove or blur all personally identifiable information from images, such as names, dates of birth, and medical record numbers. For digital image files, ensure that personal information in the metadata is also removed. Employ strong encryption to ensure that even if an image is intercepted during transmission, its content cannot be easily deciphered. Simultaneously, use encryption techniques internally to protect statically stored data. Ensure that the direct link between the image and its original source is severed, for example, by creating a unique ID that cannot be traced back to the user, replacing the original identity information.

[0068] Strictly limit the amount of data collected to the minimum required to complete the specific task. For example, if only X-rays of a specific area are needed for diagnosis, additional CT scans or other unnecessary imaging data should not be requested. Only authorized professionals should have access to these images, and they should only access the portions necessary to complete the work. Furthermore, all access should be logged for auditing purposes. Images should be deleted immediately after analysis, or at least removed from the active database and converted to a form that is not easily accessible.

[0069] Strict anonymization measures significantly reduce the risk of sensitive personal information leakage, protecting patients' privacy rights. By reducing unnecessary data collection, the corresponding storage, management, and protection costs also decrease, while potentially reducing economic losses from data breaches.

[0070] S120. The image to be analyzed is input into a pre-trained multi-reference object and disease feature relationship segmentation model to perform lesion region segmentation and feature extraction to obtain a preliminary feature set. The preliminary feature set includes a first feature vector and a segmentation feature vector. The first feature vector includes the lesion region mask position and its corresponding pathological feature value. The segmentation feature vector includes image semantic information.

[0071] In this embodiment, the first feature vector refers to a comprehensive feature representation, which not only includes the location of the lesion (mask location), but also associates with the specific pathological feature values ​​at that location, such as the size of the mass and density distribution.

[0072] Segmentation feature vectors represent the understanding of the overall semantics of an image from the perspective of each pixel, i.e., which regions belong to lesions and which are normal tissues. This vectorized representation can help machine learning algorithms better interpret complex patterns in images.

[0073] The feature vector is a series of numerical features generated by the multi-reference and disease feature relationship segmentation model based on the input image. These features can reflect the location, shape, size and associated pathological attributes of lesions in the image.

[0074] The multi-reference-disease feature relationship segmentation model includes the following components:

[0075] The encoder is responsible for extracting low- to high-level spatial features from the original image. This stage may contain multiple convolutional layers, each followed by batch normalization and an activation function (such as ReLU). Additionally, downsampling operations may be performed to reduce the resolution.

[0076] Middle module:

[0077] C2f layer: used to extract deep features, especially the internal structural features of the lesion, lesion marker features, and details of the surrounding tissue.

[0078] SPPF layer: Performs spatial pyramid pooling, enabling the model to understand features at different scales and output fixed-length feature vectors.

[0079] Decoder: Designed to recover details and reconstruct high-resolution segmentation maps. The decoder may use skip connections to pass low-level features from the encoder, thus helping to preserve more boundary information.

[0080] Custom components, such as self-attention mechanisms and multilayer perceptrons, are used to enhance the quality of feature representations.

[0081] In one embodiment, step S120 described above may include steps S121 to S125.

[0082] S121. After performing data augmentation processing on the image to be analyzed, input it into the multi-reference object and disease feature relationship segmentation model;

[0083] S122. A feature vector is generated based on the image to be analyzed by a multi-reference object and disease feature relationship segmentation model.

[0084] In one embodiment, step S122 described above may include steps S1221 to S1223.

[0085] S1221. The convolutional layer of the multi-reference object and disease feature relationship segmentation model performs convolution operations and downsampling, and then performs batch normalization and activation function processing to obtain the superficial lesion shape features and surrounding tissue features of the image to be analyzed.

[0086] S1222. Use the C2f layer to extract the deep internal structural features of the lesion, the features of lesion markers, and the detailed features of the surrounding tissue in the image to be analyzed;

[0087] S1223. Use the SPPF layer to perform pooling operations on the shape features of superficial lesions and surrounding tissue features, the internal structural features of deep lesions, lesion marker features, and surrounding tissue detail features to generate feature vectors.

[0088] Convolutional layers extract shallow features from the image to be analyzed, such as the shape of lesions and the basic contours of surrounding tissues, and downsampling is used to reduce spatial dimensionality. Local texture and edge information are captured, laying the foundation for subsequent deep feature extraction.

[0089] A customized C2f layer was applied to further uncover deeper internal structural features of the lesion, lesion marker characteristics, and details of surrounding tissues. This enhanced the understanding of complex patterns in the lesion region and improved the richness and accuracy of feature representation.

[0090] Spatial pyramid pooling (SPP) or its variants (such as SPPF) are used to fuse features at different scales, generating a fixed-length feature vector. This ensures that the output feature representation remains consistent regardless of the input image size, facilitating downstream processing.

[0091] In this embodiment, the shape characteristics of the superficial lesion and the characteristics of the surrounding tissue are described.

[0092] Superficial lesion shape characteristics refer to the geometric morphological properties of the lesion area (such as tumors, inflammation, etc.), including but not limited to boundary contours, area size, length-to-width ratio, roundness, or irregularity. These characteristics typically reflect the general appearance and location of the lesion. Superficial surrounding tissue characteristics involve the appearance of healthy tissue immediately adjacent to the lesion area, such as texture patterns, density variations, and contrast. They can help distinguish the boundary between lesions and normal tissue and provide information about the extent of the lesion's influence.

[0093] Deeper lesion internal structural features delve deeper into the lesion's interior, describing its microscopic structure, such as cell arrangement, blood vessel distribution, and cyst formation. These are crucial for understanding the nature of the lesion and help determine its type and progression.

[0094] Disease markers are biomolecules or other indicators that specifically indicate the presence of a certain disease state. On imaging, they manifest as signals of specific intensity or abnormal focal points, such as certain types of calcifications or metabolically active areas. These features are often highly specific and are key evidence for diagnosing specific diseases.

[0095] In addition to the immediate adjacent tissue directly surrounding the lesion, detailed features of the surrounding tissue also encompass subtle changes in areas further away that may still be affected by the lesion. This includes localized edema, fibrosis, infiltration, and any indirect effects caused by the lesion. Understanding these details helps in a comprehensive assessment of the extent and severity of the lesion's impact.

[0096] In short, in medical image analysis, the aforementioned features focus on the macroscopic morphology, internal structure, unique biomarkers, and changes in the surrounding environment of lesions. Through comprehensive analysis of this multi-dimensional information, doctors can gain a more accurate and comprehensive understanding of the condition, thus providing a solid foundation for developing treatment plans. When using deep learning models such as multi-reference disease feature relationship segmentation models, the models are designed to automatically identify and utilize these different levels of features for efficient lesion detection and classification.

[0097] S123. The feature vector is processed sequentially through a normalization layer, a self-attention mechanism, and a multilayer perceptron to obtain lesion segmentation information and related pathological feature information. The lesion segmentation information includes segmentation feature vectors, and the related pathological feature information includes the lesion region mask position and its corresponding pathological feature value.

[0098] In this embodiment, the numerical range is adjusted sequentially through a normalization layer, key parts are highlighted using a self-attention mechanism, and then nonlinear transformation is performed through a multilayer perceptron (MLP) to finally obtain lesion segmentation information and pathological feature information. This optimizes feature quality and improves model prediction accuracy; the self-attention mechanism helps the model focus on the most important feature regions, while the MLP can learn complex mapping relationships.

[0099] S124. Calculate the specific parameters of the abnormal region based on the proportional relationship between the pathological feature information and the actual physiological structure.

[0100] In one embodiment, step S124 described above may include steps S1241 to S1243.

[0101] S1241. Determine the mask pixel size information of the abnormal area based on the pathological feature information;

[0102] S1242. Obtain the stored actual physiological structure size ratio from the multi-reference object and disease feature relationship segmentation model;

[0103] S1243. Calculate the actual size of the abnormal region based on the pixel size information of the abnormal region mask, and compare it with the actual physiological structure size ratio to obtain specific parameters.

[0104] In this embodiment, the pathological feature information includes a binary image generated from the boundaries of the identified lesion tissue, where the lesion is marked as the foreground (e.g., white), while the normal tissue background remains the background color (e.g., black). This binary image serves as a mask for the abnormal region, or can be considered a mask for the lesion region. Then, the number and distribution of all pixels belonging to the foreground in this mask are calculated to determine the size and shape of the abnormal region.

[0105] A quantitative method is used to represent the location, extent, and internal structural characteristics of the lesion area. This improves the accuracy of subsequent processing and makes the assessment of the lesion area more accurate and reliable.

[0106] In this embodiment, the multi-reference model for disease feature segmentation has learned the standard size proportions of different organ or tissue types through learning from a large number of standard samples. When a new case is input, the model can automatically match the closest standard template based on pathological feature information and extract the corresponding size proportion information. This ensures that the measurement results are comparable and consistent, and even data collected under different devices and conditions can be interpreted in a unified and standardized manner.

[0107] Combining the abnormal region mask pixel size information obtained in S1241 and the actual physiological structure size ratio provided in S1242, appropriate mathematical transformation formulas (such as linear scaling, nonlinear mapping, etc.) are applied to convert pixel-level measurements into real physical units (such as millimeters). Then, the converted dimensions are compared with a standard template, and the degree of change in gain, loss, or other forms is calculated as the specific parameters of the final output.

[0108] By precisely calculating the specific parameters of abnormal areas, human error is reduced, helping doctors to more accurately determine the severity of diseases. Standardized proportional relationships ensure that data from different sources can be compared on the same basis, which is beneficial for scientific research and clinical collaboration.

[0109] S125. Combine the specific parameters of the abnormal region, the segmentation feature vector, the mask position of the lesion region and its corresponding pathological feature value to form a preliminary feature set.

[0110] In this embodiment, the preliminary feature set includes three parts:

[0111] The location of the lesion area mask and its corresponding pathological feature value (i.e., the first feature vector);

[0112] Specific parameters of the abnormal area (such as size, shape, etc., as supplementary explanation of pathological characteristic values).

[0113] The segmentation feature vector includes image semantic information.

[0114] In intelligent disease self-examination systems, pathological feature values ​​are typically obtained from image data (such as CT and MRI scans) through algorithmic analysis. These features are used to identify and quantify specific attributes of lesion areas. In other words, case feature values ​​are among the key pieces of information extracted from lesion areas after processing by a segmentation model. They not only include location information (i.e., mask location) but also encompass various quantitative values ​​describing the characteristics of the lesion, assisting doctors in making more accurate diagnostic decisions. These pathological feature values ​​are a crucial component in constructing the first feature vector and play a vital role in subsequent analysis.

[0115] Various numerical values ​​that quantitatively describe the characteristics of a lesion can include:

[0116] Morphological characteristics: Describe the shape, size, and margin clarity of the lesion. For example, the diameter, area, and volume of the tumor.

[0117] Density or signal intensity: For different imaging modalities, such as Hounsfield units (HU) in CT images and signal intensity in MRI images, they reflect the density or contrast characteristics of tissue or lesions.

[0118] Texture features: These represent the variations in grayscale distribution patterns in an image, such as uniformity and roughness, which helps distinguish between normal and abnormal tissues.

[0119] Functional parameters: If functional imaging techniques (such as PET, fMRI) are used, indicators related to metabolic activity, blood perfusion, etc. can also be obtained.

[0120] Statistical measurement: Statistical information based on pixel level, such as mean, standard deviation, etc., or the results output by more complex statistical models.

[0121] Other biomarkers: These also include implicit features learned by AI models from a large number of cases, which are difficult to observe directly but are helpful for diagnosis.

[0122] S130. Obtain the user's historical health image or standard health image as a comparison sample image by adopting the principles of anonymization and minimization.

[0123] In this embodiment, while adhering to strict data privacy regulations, the system collects users' past health records or image data referencing standard health to the most necessary extent. This information is anonymized to ensure that personally identifiable information is untraceable, thereby protecting user privacy.

[0124] This provides a reliable benchmark for subsequent analysis while ensuring user data security. It also enhances the system's credibility and user experience.

[0125] This step is the same as step S110, and will not be repeated here.

[0126] S140. Input the image to be analyzed and the comparison sample image into the disease feature change estimation model, and calculate the feature change of the lesion area in combination with the segmentation feature vector to obtain the second feature vector.

[0127] In this embodiment, the second feature vector is a multi-dimensional numerical representation generated by comparing and analyzing the image to be analyzed (the current patient's lesion image) with the comparison sample image (images under historical healthy conditions or other standard healthy conditions), and extracting and quantifying the changes in lesion region features using a deep learning model. It reflects the degree of change of the lesion relative to the normal state.

[0128] In this embodiment, the disease feature change estimation model is a deep learning model specifically designed for medical image analysis. It aims to estimate changes in lesion region features by comparing the image to be analyzed (e.g., a lesion image of a current patient) with a comparison sample image (e.g., an image from a historical health state or other standard health state). The specific structure and training method of this model typically include the following aspects:

[0129] The model structure includes:

[0130] Encoder: Employs a convolutional neural network (CNN) or its variants (such as ResNet, DenseNet, etc.) to extract multi-layer feature representations of the input image.

[0131] The encoder converts the raw image into low-dimensional feature vectors that capture important information in the image, such as shape, texture, and location.

[0132] Feature extraction module: Further refines the encoding process, extracting more specific features for specific tasks (such as lesion detection and segmentation). Includes multi-scale feature extraction layers to accommodate lesions of different sizes and shapes.

[0133] Fusion Layer: Used to merge feature vectors from the image to be analyzed and the comparison sample images.

[0134] Fusion techniques can range from simple connection operations to more complex mechanisms, such as attention mechanisms, which allow the model to focus on the most relevant features.

[0135] Feature adjustment and difference calculation module: contains upsampling and downsampling layers, used to adjust spatial size and perform feature fusion.

[0136] Mathematical distance metrics (such as Euclidean distance, cosine similarity, etc.) are used to calculate the difference between two sets of features, forming the final feature vector representation.

[0137] Decoder (optional): If a segmentation map or other form of output is required, a decoder section may be included to reconstruct a high-resolution image or feature map from low-dimensional feature vectors.

[0138] Design special loss functions based on task requirements, such as combining classification loss, regression loss, and custom feature matching loss, to guide the model in learning effective representations of lesion feature changes.

[0139] The training process includes: collecting and labeling a large dataset of medical images, including images of patients' lesions at different time points and their corresponding healthy or standard healthy states. The dataset is designed to include sufficiently diverse cases to enhance the model's generalization ability.

[0140] Standardize the image data, such as resizing and normalizing pixel values, to ensure input consistency; in some cases, data augmentation may also be required to expand the training set.

[0141] Initialize the model weights using either random initialization or transfer learning. If you choose transfer learning, you can start with a model that has already been trained on a large-scale general image dataset and then fine-tune the model according to the characteristics of medical images.

[0142] Using the dataset constructed above, and according to the predetermined hyperparameter settings such as learning rate and batch size, the loss function is minimized through the backpropagation algorithm, and the model parameters are iteratively updated until satisfactory performance metrics or convergence conditions are achieved.

[0143] Evaluate the model's performance on an independent validation set, and adjust the model architecture or hyperparameters based on the results; finally, test the model's practical application effectiveness on a completely unseen test set.

[0144] Based on the test results, additional post-processing steps may be required for the model's predictions, such as morphological operations to remove noise, or further optimization of the model to improve efficiency and accuracy.

[0145] In summary, designing and training a disease feature change estimation model is a complex process involving expertise and techniques from multiple fields, including but not limited to deep learning, medical image processing, and biostatistics. A successful model should not only accurately identify changes in lesion features but also possess good interpretability and clinical applicability.

[0146] In one embodiment, step S140 described above may include steps S141 to S148.

[0147] S141. Input the image to be analyzed and the comparison sample image into the disease feature change estimation model for encoding to form two sets of feature vectors.

[0148] In this embodiment, the image to be analyzed (the patient's lesion image) and the comparison sample image (images of historical health status or other standard health status) are input into a model. The input images are automatically encoded to generate two different but structurally similar feature vectors. Through encoding, the system can compare the two images within the same feature space, thereby ensuring the consistency and accuracy of subsequent processing.

[0149] Specifically, these feature vectors are representative information extracted from images, reflecting different aspects of the image, such as shape, texture, and color. Through encoding, the model can transform complex image data into more concise feature vectors for subsequent analysis and comparison.

[0150] S142. Extract the corresponding surrounding tissue features, detail features, lesion shape features, internal structural features, and marker features from the image to be analyzed and the comparison sample image.

[0151] In this embodiment, for each image, the system extracts a series of features, including surrounding tissue features, detail features, lesion shape features, internal structural features, and landmark features. These features can be obtained through multi-layer convolution operations and other computer vision techniques.

[0152] A rich feature set enables the model to capture lesion-specific information more accurately, improving the reliability of diagnosis.

[0153] Specifically, the characteristics of surrounding tissues refer to the morphology, structure, and other features of the normal tissues surrounding the lesion.

[0154] Lesion shape characteristics: Describes the shape, size, and boundaries of the lesion area.

[0155] Internal structural features: the internal texture, density, and tissue structure of the lesion area.

[0156] Biomarker characteristics: refers to specific biomarkers that may be present in a lesion, such as characteristic signals of a tumor.

[0157] These features are extracted to comprehensively describe changes in the lesion and its surrounding environment, so that the model can more accurately estimate the nature of the lesion.

[0158] S143. Based on the two sets of feature vectors and the segmentation feature vector, calculate the feature changes of the lesion region to obtain a preliminary feature vector.

[0159] In this embodiment, firstly, it is ensured that the feature vectors from different sources (i.e., the image to be analyzed and the comparison sample image) are spatially corresponding. This step may involve image registration or feature alignment techniques to ensure that features at the same anatomical location can be accurately matched. For non-rigid deformation cases, elastic registration methods can be used.

[0160] Segmentation feature vectors contain semantic information about the image, which guides how to correctly match and compare two feature vectors. For example, when performing feature alignment, the region of interest can be defined by the lesion mask information in the segmentation feature vectors, thereby improving matching accuracy.

[0161] For each pixel or small region, calculate the difference between the two feature vectors. Common methods include, but are not limited to:

[0162] Euclidean distance is applicable to the absolute differences between quantifiable numerical features.

[0163] Cosine similarity measures the angular difference between two vectors, making it suitable for assessing directional consistency rather than magnitude difference.

[0164] Mutual information takes into account the statistical dependency between two variables, which helps to discover more complex association patterns.

[0165] The choice of metric depends on the specific application scenario and the nature of the extracted features. Furthermore, multiple metrics can be combined to comprehensively evaluate feature differences.

[0166] The local feature differences calculated above are integrated into a global feature change representation. This step can be performed in the following ways:

[0167] Weighted summation: Summing the results by assigning different weights to the differences in each locality based on their importance.

[0168] Maximum / Minimum Value Selection: Select the most significant change as a representative of the overall degree of change.

[0169] Convolution operation: The convolution kernel is used to smooth the local difference map, so as to obtain a more stable and reliable representation of feature changes.

[0170] Self-attention mechanism: allows the model to automatically learn which parts have more important differences and adjust their contributions to the final representation accordingly.

[0171] After the above processing, a preliminary feature vector is obtained, a new feature vector that can reflect the changes in the features of the lesion area. This feature vector not only contains the spatial structure information of the original image, but also incorporates dynamic changes in the time dimension, providing a solid foundation for subsequent more in-depth disease feature assessment.

[0172] S144. The spatial size of the preliminary feature vector is adjusted using upsampling and downsampling techniques in the multiple feature coding layer, and a fusion operation is performed to integrate information at different scales to obtain a fused preliminary feature map.

[0173] In this embodiment, upsampling and downsampling techniques in the multiple feature encoding layer are used to adjust the spatial size of the initial feature vector. Upsampling increases the spatial resolution of the feature map, while downsampling reduces the resolution. This size adjustment allows the model to capture information at different scales, helping it extract more useful features and merge visual information from different scales.

[0174] S145. Based on the fused preliminary feature map, the number of channels is adjusted to 1, and a hybrid structure is used for downsampling to obtain the adjusted environmental feature map, considering the surrounding tissue features and detailed features.

[0175] In this embodiment, the extracted surrounding tissue features and detail features are downsampled based on the fused preliminary feature map to obtain an adjusted environmental feature map. During this process, the number of channels is adjusted to 1, indicating that only single-channel feature information is available.

[0176] By downsampling, the model reduces redundant information and focuses on important surrounding tissue features to improve the contrast between lesion areas and healthy areas.

[0177] S146. For the lesion shape features, internal structure features and marker features, based on the fused preliminary feature map, the number of channels is adjusted by convolution, and upsampling is performed using the nearest neighbor interpolation method to obtain the adjusted lesion feature map.

[0178] In this embodiment, for lesion shape features, internal structural features, and marker features, based on the fused preliminary feature map, the number of channels is adjusted by convolution, and then upsampling is performed using the nearest neighbor interpolation method to obtain the adjusted lesion feature map.

[0179] Upsampling improves the resolution of the feature map of the lesion area, enabling more accurate capture of lesion details and enhancing the ability to identify the lesion area.

[0180] S147. Perform a convolution operation on the adjusted environmental feature map and the adjusted lesion feature map, and concatenate them in the channel dimension to form the final fused feature map.

[0181] In this embodiment, the adjusted environmental feature map and lesion feature map are convolved once, and then concatenated along the channel dimension to form the final fused feature map. This fused feature map combines multi-dimensional information from the lesion and surrounding tissues.

[0182] By using fusion operations, the model can simultaneously process the features of the lesion and surrounding tissues, thereby improving the accuracy and detail of the overall analysis.

[0183] S148. Calculate the difference in the fused feature vectors by combining the segmented feature vectors with the final fused feature map, so as to determine the second feature vector.

[0184] In this embodiment, in the final step, the feature vector differences between the segmented feature vectors and the final fused feature map are calculated to form a second feature vector. This step uses a mathematical distance metric to quantify the differences between the feature vectors.

[0185] By calculating the differences in feature vectors, the system can clearly distinguish between diseased and normal regions, thereby generating a high-dimensional feature vector for disease analysis and judgment. This feature vector provides data support for subsequent disease prediction, diagnosis, and treatment decisions.

[0186] In one embodiment, step S148 described above may include steps S1481 to S1482.

[0187] S1481. Align the segmented feature vector with the final fused feature map.

[0188] In this embodiment, the purpose of S1481 is to ensure that the features from the segmented feature vector and the final fused feature map are consistent in spatial location so as to make an effective comparison.

[0189] If two feature representations have different spatial resolutions, appropriate transformations (such as bilinear interpolation, nearest neighbor interpolation, etc.) need to be applied to make them have the same size.

[0190] Use registration algorithms (such as rigid body transformation, affine transformation, or non-rigid transformation) to correct for possible rotation, scaling, and translation errors, ensuring that the anatomical structures of the two sets of features correspond consistently.

[0191] If the feature vector and feature map have different numbers of channels, they can be matched by increasing or decreasing the number of channels. This typically involves feature selection or aggregation strategies to retain the most relevant feature information.

[0192] The aligned and consistent feature representations are ready for use in the next step of distance metric calculation.

[0193] S1482. Calculate the difference between the feature vectors after feature alignment using a mathematical distance metric to form a second feature vector.

[0194] In this embodiment, this step is to quantify the similarity and differences between the feature vectors after feature alignment in order to capture the changes in lesion region features over time or other factors.

[0195] You can choose an appropriate mathematical distance metric based on the task requirements. Commonly used metrics include Euclidean distance, cosine similarity, and Manhattan distance. Each metric has its own characteristics and is suitable for different application scenarios.

[0196] For each pixel or local region, the distance between the two feature representations is calculated to obtain a new feature map that reflects the feature differences.

[0197] To eliminate the influence of different feature scales, the calculated distance values ​​can be normalized to make the results more stable and easier to interpret.

[0198] All calculated distance values ​​are integrated into a new feature vector that can compactly represent the differences between the original features and serve as the basis for subsequent analysis or classification.

[0199] The second feature vector reflects the changes in lesion features relative to the comparison sample and can be used for further diagnostic analysis or as one of the targets for model training.

[0200] The two steps described above can effectively capture the changing patterns of lesion area characteristics, which is of great significance for early detection, progression monitoring, and treatment efficacy evaluation of the disease.

[0201] S150. Calculate the final disease feature assessment result based on the first feature vector and the second feature vector.

[0202] In this embodiment, the disease characteristic assessment result refers to

[0203] In one embodiment, step S150 described above may include steps S151 to S153.

[0204] S151. Multiply the first feature vector and the second feature vector to obtain the feature vector multiplication result.

[0205] In this embodiment, the result of multiplying the eigenvectors refers to the vector obtained by multiplying the first eigenvector and the second eigenvector.

[0206] Element-wise multiplication is performed on the two feature vectors. This process can be viewed as a weighted combination of feature strengths in each dimension, thereby enhancing features that are significant in both vectors and suppressing irrelevant or conflicting information. The result integrates information from the original and modified features, providing a foundation for subsequent decoding.

[0207] S152. The result of multiplying the feature vectors is sequentially passed through the convolutional layer, deconvolutional layer, multiple convolutional layers and activation function in the decoder to recover the actual pathological feature value represented by each pixel.

[0208] In this embodiment, convolution operations are applied to extract local feature patterns, which helps capture complex relationships in the multiplication result. At this stage, the multiplied feature vectors are fed into convolutional layers in the decoder, which extract low-level features of the image (such as edges, textures, etc.).

[0209] Deconvolutional layers (also known as transposed convolutions) perform upsampling, progressively restoring the spatial resolution of the image. Deconvolutional layers, or unconvolutional layers, are primarily used to restore low-resolution feature maps to a higher resolution, approaching the size of the original image. The purpose of this step is to recover the spatial distribution of the lesion region, enabling the model to locate specific lesion areas.

[0210] Multiple convolutional layers further refine the feature representation, ensuring the final output accurately maps to the original image space. During decoding, multiple convolutional layers further process the feature maps, allowing the pathological features of each pixel to be gradually recovered. These convolutional layers and activation functions (such as ReLU and Sigmoid) help extract and optimize high-level features of lesion regions. Activation functions help the network perform non-linear mappings, enhancing its ability to express complex patterns.

[0211] The activation function uses a non-linear activation function (such as ReLU, Sigmoid, or Tanh) to introduce non-linearity into the model, enabling the network to learn more complex mapping relationships. Finally, after processing through multiple layers of convolution and activation functions, a restored feature map is obtained, where each pixel represents the actual pathological feature value at that location, such as the concentration, morphological features, and pathological type of the lesion area.

[0212] Output the actual pathological feature value represented by each pixel, that is, the high-dimensional feature map obtained after the above processing, which should better reflect the specific pathological features of the lesion area.

[0213] S153. Combine the mask position of the lesion area to extract the corresponding pathological feature values ​​from the restored pathological feature map, and determine the final disease feature assessment result.

[0214] In this embodiment, meaningful evaluation indicators are extracted from the recovered pathological feature map based on the location information of the lesion area.

[0215] Lesion area mask location: Specifies the specific location of the lesion. It is usually a binary mask in which the lesion area is marked as 1 and the background area is marked as 0.

[0216] Using a lesion area mask as a guide, only feature values ​​belonging to the lesion portion are selected from the pathological feature map. This allows focus on the characteristics of the lesion itself, avoiding the influence of surrounding normal tissue. Combining the extracted pathological feature values ​​with other clinical information (such as medical history and laboratory test results), one or more quantitative indicators are calculated, such as lesion volume, density changes, and edge clarity scores. These indicators collectively constitute the final disease feature assessment result.

[0217] The final disease characteristic assessment result can be output in the form of specific numerical values, classification labels, or visual reports.

[0218] Specifically, in medical image analysis, lesion regions are typically identified using a mask. A mask is a binary image that identifies the area where the lesion is located in the image. In this step, the model uses this mask to extract features from the lesion region, ensuring that the calculation is focused solely on the lesion area and unaffected by surrounding normal areas. Combining the mask information, the model extracts pathological feature values ​​within the corresponding region. This process combines the lesion region with the pathological feature map generated by the decoder to obtain specific feature values ​​within the lesion region, such as the lesion's area, shape, and boundary features. Based on the extracted pathological features, combined with medical knowledge and diagnostic criteria, the model calculates the final disease feature assessment result. These results can be quantitative (e.g., lesion size, volume) or qualitative (e.g., lesion type, whether it is malignant), and can provide clinical decision support. The final assessment result may be classified into different disease stages (e.g., early, middle, late) or output a disease type based on specific diagnostic criteria.

[0219] S160, output the final disease feature assessment results and the corresponding lesion area mask location.

[0220] The final disease characteristic assessment results and the corresponding lesion area mask location are output to the terminal.

[0221] In this embodiment, the above-mentioned method adopts the principles of anonymization and minimization to avoid exposing users' sensitive data and ensure the protection of personal privacy during disease diagnosis. By segmenting lesion regions using a multi-reference object and disease feature relationship model, the features of the lesion region can be more accurately located and extracted, improving the accuracy of disease diagnosis. Simultaneously, the mask position of the lesion region, pathological feature values, and image semantic information are extracted, providing multi-dimensional diagnostic evidence and helping to improve the comprehensiveness and depth of diagnosis. Data augmentation processing of the image to be analyzed helps improve the model's adaptability to different scenes and image qualities, reducing overfitting and improving the universality of disease diagnosis. Through multi-layer feature extraction modules such as convolutional layers, C2f layers, and SPPF layers, different levels of image features can be extracted, thereby enhancing the ability to discriminate lesion shape, structure, and marker features. By calculating the mask size information and physiological structure ratio of abnormal regions, the actual size of abnormal regions can be estimated more accurately, avoiding misdiagnosis or missed diagnosis. Comparing the size of the lesion region with the actual physiological structure ensures that the diagnostic results are consistent with the patient's actual physical condition, improving the scientific nature of the diagnosis. By comparing feature changes between the image to be analyzed and the comparison sample image, the changes in lesions can be estimated in real time, providing dynamic information on disease progression and helping doctors make more accurate judgments. Combining multiple features such as surrounding tissue, lesion shape, and internal structure allows for a more comprehensive capture of disease changes, improving the accuracy of disease prediction. Through feature vector alignment and difference calculation, the differences between the image to be analyzed and the comparison sample image can be evaluated more precisely, helping to better identify changes in lesion areas. The fusion of feature information at different scales, through upsampling and downsampling techniques, enhances the processing capability for complex image information, providing more detailed analysis of disease changes. By combining the first and second feature vectors and performing a final evaluation, various feature information can be comprehensively considered, providing a comprehensive disease assessment result. Through multi-layer convolution, deconvolution, and activation function processing in the decoder, the actual pathological feature values ​​of each pixel are recovered, generating high-quality pathological feature maps and ensuring the accuracy and reliability of the final disease feature assessment result.

[0222] The method in this embodiment employs a multi-reference system and multi-feature fusion approach, combining lesion region segmentation, feature extraction, and change estimation models to achieve comprehensive capture and accurate analysis of features from superficial to deep layers, thereby improving the accuracy of disease feature assessment. Through detailed calculations of specific parameters in abnormal regions, it provides doctors with more reliable data support, helping to make more accurate diagnostic decisions.

[0223] The system simplifies the user experience, allowing users to obtain detailed disease assessment reports by simply uploading relevant medical images, including lesion location, size, and pathological characteristics. Furthermore, the system can send diagnostic results to a health management platform, facilitating long-term tracking and management of individual health status and improving the overall quality and efficiency of healthcare services.

[0224] End-to-end data encryption is implemented throughout the self-testing process to ensure that all transmitted and stored data is protected with high strength; the principles of anonymization and minimization are applied to collect only the minimum amount of data required to complete the disease diagnosis, and the data is deleted or anonymized immediately after analysis; strict access control policies restrict access to user data and ensure the security of user information.

[0225] The aforementioned intelligent self-inspection and data protection integrated method ensures user privacy and security by employing anonymization and minimization principles to protect user personal data when acquiring images to be analyzed and historical health images. By inputting the images to be analyzed into a pre-trained multi-reference object and disease feature relationship segmentation model, it accurately segments lesion regions and extracts features, obtaining a preliminary feature set containing lesion region mask locations and pathological feature values, thus improving detection accuracy. Using historical health images as comparison samples, it calculates feature changes in lesion regions through a disease feature change estimation model, further enhancing the ability to identify disease changes. Combining preliminary and change features, it comprehensively calculates the final disease feature evaluation result, improving the accuracy and reliability of diagnosis. The entire process enhances disease detection effectiveness through intelligent algorithms without disclosing sensitive user information.

[0226] Figure 2 This is a schematic block diagram of an intelligent self-testing and data protection integrated system 300 provided in an embodiment of the present invention. Figure 2 As shown, corresponding to the above-described intelligent self-testing and data protection integrated method, the present invention also provides an intelligent self-testing and data protection integrated system 300. This intelligent self-testing and data protection integrated system 300 includes a unit for executing the above-described intelligent self-testing and data protection integrated method, and the system can be configured in a server. Specifically, please refer to... Figure 2 The intelligent self-inspection and data protection integrated system 300 includes a first acquisition unit 301, a segmentation and extraction unit 302, a second acquisition unit 303, a change calculation unit 304, a final result calculation unit 305, and an output unit 306.

[0227] The first acquisition unit 301 is used to acquire the image to be analyzed using anonymization and minimization principles; the segmentation and extraction unit 302 is used to input the image to be analyzed into a pre-trained multi-reference object and disease feature relationship segmentation model to perform lesion region segmentation and feature extraction to obtain a preliminary feature set, the preliminary feature set including a first feature vector and a segmentation feature vector, wherein the first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information; the second acquisition unit 303 is used to acquire the user's historical healthy image or standard healthy image as a comparison sample image using anonymization and minimization principles; the change calculation unit 304 is used to input the image to be analyzed and the comparison sample image into a disease feature change estimation model, and calculate the lesion region feature change in combination with the segmentation feature vector to obtain a second feature vector; the final result calculation unit 305 is used to calculate the final disease feature evaluation result based on the first feature vector and the second feature vector; the output unit 306 is used to output the final disease feature evaluation result and the corresponding lesion region mask position.

[0228] In one embodiment, the segmentation and extraction unit 302 is used for:

[0229] After data augmentation processing, the image to be analyzed is input into a multi-reference object and disease feature relationship segmentation model. The multi-reference object and disease feature relationship segmentation model generates feature vectors based on the image to be analyzed. The feature vectors are then processed sequentially through a normalization layer, a self-attention mechanism, and a multilayer perceptron to obtain lesion segmentation information and related pathological feature information. The lesion segmentation information includes segmentation feature vectors, and the related pathological feature information includes the lesion region mask position and its corresponding pathological feature value. The specific parameters of the abnormal region are calculated based on the proportional relationship between the pathological feature information and the actual physiological structure. The specific parameters of the abnormal region, the segmentation feature vectors, the lesion region mask position, and its corresponding pathological feature value are combined to form a preliminary feature set.

[0230] In one embodiment, the segmentation and extraction unit 302 is further configured to:

[0231] The convolutional layer of the multi-reference-disease feature relationship segmentation model performs convolution operations and downsampling, and then processes the data through batch normalization and activation functions to obtain the superficial lesion shape features and surrounding tissue features of the image to be analyzed. The C2f layer is used to extract the deep internal structural features of the lesion, lesion marker features, and surrounding tissue detail features of the image to be analyzed. The SPPF layer is used to perform pooling operations on the superficial lesion shape features and surrounding tissue features, the deep internal structural features of the lesion, lesion marker features, and surrounding tissue detail features to generate feature vectors.

[0232] In one embodiment, the segmentation and extraction unit 302 is further configured to:

[0233] The abnormal region mask pixel size information is determined based on the pathological feature information; the actual physiological structure size ratio is obtained from the multi-reference object and disease feature relationship segmentation model; the actual size of the abnormal region is calculated based on the abnormal region mask pixel size information, and compared with the actual physiological structure size ratio to obtain specific parameters.

[0234] In one embodiment, the change calculation unit 304 is used for:

[0235] The image to be analyzed and the comparison sample image are input into a disease feature change estimation model for encoding, forming two sets of feature vectors. Corresponding surrounding tissue features, detail features, lesion shape features, internal structure features, and biomarker features are extracted from the image to be analyzed and the comparison sample image. Based on the two sets of feature vectors and the segmentation feature vector, feature changes in the lesion region are calculated to obtain preliminary feature vectors. The spatial dimensions of the preliminary feature vectors are adjusted using upsampling and downsampling techniques in a multiple feature encoding layer, and a fusion operation is performed to integrate information at different scales, resulting in a fused preliminary feature map. The surrounding tissue features and detail features are then analyzed. Based on the fused preliminary feature map, the number of channels is adjusted to 1, and downsampling is performed using a hybrid structure to obtain an adjusted environmental feature map. For the lesion shape features, internal structure features, and marker features, based on the fused preliminary feature map, the number of channels is adjusted by convolution, and upsampling is performed using nearest neighbor interpolation to obtain an adjusted lesion feature map. The adjusted environmental feature map and the adjusted lesion feature map are convolved once, and concatenated along the channel dimension to form the final fused feature map. The difference between the fused feature vectors is calculated using the segmented feature vectors combined with the final fused feature map to determine the second feature vector.

[0236] In one embodiment, the change calculation unit 304 is further configured to:

[0237] The segmented feature vectors are aligned with the final fused feature map; the difference between the aligned feature vectors is calculated using a mathematical distance metric to form a second feature vector.

[0238] In one embodiment, the final result calculation unit 305 is used for:

[0239] The first feature vector and the second feature vector are multiplied together to obtain the feature vector multiplication result. The feature vector multiplication result is then passed sequentially through a convolutional layer, a deconvolutional layer, multiple convolutional layers, and an activation function in the decoder to recover the actual pathological feature value represented by each pixel. The corresponding pathological feature value is extracted from the recovered pathological feature map in combination with the lesion area mask position, and the final disease feature evaluation result is determined.

[0240] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned intelligent self-inspection and data protection integrated system 300 and its various units can be referred to the corresponding descriptions in the foregoing method embodiments. For the sake of convenience and brevity, these details will not be repeated here.

[0241] The aforementioned intelligent self-inspection and data protection integrated system 300 can be implemented as a computer program, which can, for example... Figure 3 It runs on the computer device shown.

[0242] Please see Figure 3 , Figure 3 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0243] See Figure 3 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0244] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform an intelligent self-test and data protection integrated method.

[0245] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0246] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can perform an intelligent self-test and data protection integrated method.

[0247] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0248] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps:

[0249] The process involves: acquiring an image to be analyzed using anonymization and minimization principles; inputting the image to be analyzed into a pre-trained multi-reference object and disease feature relationship segmentation model for lesion region segmentation and feature extraction to obtain a preliminary feature set, which includes a first feature vector and a segmentation feature vector. The first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information. The process also involves acquiring the user's historical healthy images or standard healthy images as comparison sample images using anonymization and minimization principles; inputting the image to be analyzed and the comparison sample images into a disease feature change estimation model, and calculating the lesion region feature change in conjunction with the segmentation feature vector to obtain a second feature vector; calculating the final disease feature evaluation result based on the first feature vector and the second feature vector; and outputting the final disease feature evaluation result and the corresponding lesion region mask position.

[0250] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0251] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0252] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform the following steps:

[0253] The process involves: acquiring an image to be analyzed using anonymization and minimization principles; inputting the image to be analyzed into a pre-trained multi-reference object and disease feature relationship segmentation model for lesion region segmentation and feature extraction to obtain a preliminary feature set, which includes a first feature vector and a segmentation feature vector. The first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information. The process also involves acquiring the user's historical healthy images or standard healthy images as comparison sample images using anonymization and minimization principles; inputting the image to be analyzed and the comparison sample images into a disease feature change estimation model, and calculating the lesion region feature change in conjunction with the segmentation feature vector to obtain a second feature vector; calculating the final disease feature evaluation result based on the first feature vector and the second feature vector; and outputting the final disease feature evaluation result and the corresponding lesion region mask position.

[0254] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0255] It should be noted that the functions or steps that the storage medium or computer device can achieve are described in the relevant descriptions in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0256] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0257] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each unit is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0258] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the system of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0259] 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 storage medium. Based on this understanding, the technical solution of the present invention, 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 several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0260] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A comprehensive method for intelligent self-inspection and data protection, characterized in that: include: The image to be analyzed is obtained using the principles of anonymization and minimization. The image to be analyzed is input into a pre-trained multi-reference object and disease feature relationship segmentation model for lesion region segmentation and feature extraction to obtain a preliminary feature set. The preliminary feature set includes a first feature vector and a segmentation feature vector, wherein the first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information. The user's historical health images or standard health images are obtained as comparison sample images using the principles of anonymization and minimization; The image to be analyzed and the comparison sample image are input into the disease feature change estimation model, and the feature change of the lesion region is calculated by combining the segmentation feature vector to obtain the second feature vector; The final disease feature assessment result is calculated based on the first feature vector and the second feature vector; Output the final disease feature assessment results and the corresponding lesion area mask location.

2. The intelligent self-inspection and data protection integrated method according to claim 1, characterized in that, The image to be analyzed is input into a pre-trained multi-reference object and disease feature relationship segmentation model for lesion region segmentation and feature extraction to obtain a preliminary feature set. The preliminary feature set includes a first feature vector and a segmentation feature vector. The first feature vector includes the lesion region mask position and its corresponding pathological feature value. The segmentation feature vector includes image semantic information, including: After performing data augmentation on the image to be analyzed, it is input into the multi-reference object and disease feature relationship segmentation model; A feature vector is generated based on the image to be analyzed by a multi-reference object and disease feature relationship segmentation model. The feature vector is processed sequentially through a normalization layer, a self-attention mechanism, and a multilayer perceptron to obtain lesion segmentation information and related pathological feature information. The lesion segmentation information includes segmentation feature vectors, and the related pathological feature information includes the mask position of the lesion region and its corresponding pathological feature value. Calculate the specific parameters of the abnormal region based on the proportional relationship between the pathological features and the actual physiological structure. The specific parameters of the abnormal region, the segmentation feature vector, the mask position of the lesion region and its corresponding pathological feature value are combined to form a preliminary feature set.

3. The intelligent self-inspection and data protection integrated method according to claim 2, characterized in that, The segmentation model based on the relationship between multiple reference objects and disease features generates feature vectors from the image to be analyzed, including: The convolutional layer of the multi-reference object and disease feature relationship segmentation model performs convolution operations and downsampling, and then processes it through batch normalization and activation function to obtain the superficial lesion shape features and surrounding tissue features of the image to be analyzed. The C2f layer is used to extract deep internal lesion structural features, lesion marker features, and surrounding tissue detail features from the image to be analyzed. SPPF layers are used to perform pooling operations on shallow lesion shape features and surrounding tissue features, deep lesion internal structure features, lesion marker features, and surrounding tissue detail features to generate feature vectors.

4. The intelligent self-inspection and data protection integrated method according to claim 3, characterized in that, The specific parameters for calculating the abnormal region based on the proportional relationship between the pathological feature information and the actual physiological structure include: Determine the mask pixel size information of the abnormal area based on the pathological feature information; Obtain the actual physiological structure size ratio stored from the multi-reference object and disease feature relationship segmentation model; The actual size of the abnormal region is calculated based on the pixel size information of the abnormal region mask, and compared with the proportion of the actual physiological structure size to obtain specific parameters.

5. The intelligent self-inspection and data protection integrated method according to claim 4, characterized in that, The step of inputting the image to be analyzed and the comparison sample image into the disease feature change estimation model, and calculating the feature change of the lesion region in combination with the segmentation feature vector to obtain the second feature vector includes: The image to be analyzed and the comparison sample image are input into the disease feature change estimation model for encoding, forming two sets of feature vectors; Extract corresponding surrounding tissue features, detail features, lesion shape features, internal structural features, and biomarker features from the image to be analyzed and the comparison sample image; Based on the two sets of feature vectors and the segmentation feature vector, the feature changes of the lesion region are calculated to obtain preliminary feature vectors; The spatial size of the preliminary feature vector is adjusted using upsampling and downsampling techniques in the multiple feature coding layer, and a fusion operation is performed to integrate information from different scales to obtain a fused preliminary feature map. Based on the fused preliminary feature map, the number of channels is adjusted to 1, and a hybrid structure is used for downsampling to obtain the adjusted environmental feature map, considering the surrounding tissue features and detailed features. For the lesion shape features, internal structure features and marker features, based on the fused preliminary feature map, the number of channels is adjusted by convolution and upsampling is performed using the nearest neighbor interpolation method to obtain the adjusted lesion feature map; The adjusted environmental feature map and the adjusted lesion feature map are convolved once, and then concatenated along the channel dimension to form the final fused feature map. The difference between the fused feature vectors is calculated by combining the segmented feature vectors with the final fused feature map to determine the second feature vector.

6. The intelligent self-inspection and data protection integrated method according to claim 5, characterized in that, The step of calculating the difference in the fused feature vectors by combining the segmented feature vectors with the final fused feature map to determine the second feature vector includes: Align the segmented feature vector with the final fused feature map; The difference between the feature vectors after feature alignment is calculated using a mathematical distance metric to form a second feature vector.

7. The intelligent self-inspection and data protection integrated method according to claim 6, characterized in that, The step of calculating the final disease feature assessment result based on the first feature vector and the second feature vector includes: Multiply the first feature vector and the second feature vector together to obtain the feature vector multiplication result; The result of multiplying the feature vectors is sequentially passed through convolutional layers, deconvolutional layers, multiple convolutional layers, and activation functions in the decoder to recover the actual pathological feature value represented by each pixel; By combining the mask position of the lesion area, the corresponding pathological feature values ​​are extracted from the recovered pathological feature map, and the final disease feature assessment result is determined.

8. An intelligent self-inspection and data protection integrated system, characterized in that: include: The first acquisition unit is used to acquire the image to be analyzed using the principles of anonymization and minimization. The segmentation and extraction unit is used to input the image to be analyzed into a pre-trained multi-reference object and disease feature relationship segmentation model to perform lesion region segmentation and feature extraction to obtain a preliminary feature set. The preliminary feature set includes a first feature vector and a segmentation feature vector, wherein the first feature vector includes the lesion region mask position and its corresponding pathological feature value; the segmentation feature vector includes image semantic information. The second acquisition unit is used to acquire the user's historical health image or standard health image as a comparison sample image by adopting the principles of anonymization and minimization. The change calculation unit is used to input the image to be analyzed and the comparison sample image into the disease feature change estimation model, and calculate the feature change of the lesion region in combination with the segmentation feature vector to obtain the second feature vector; The final result calculation unit is used to calculate the final disease feature assessment result based on the first feature vector and the second feature vector; The output unit is used to output the final disease feature assessment results and the corresponding lesion area mask location.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images