A cancer data processing method
Through cancer data processing methods, texture feature extraction and multi-level index construction are used to solve the efficiency and accuracy problems in medical image data processing, efficient image data analysis and rapid retrieval are achieved, and real-time diagnostic decision-making is supported.
Patent Information
- Application Number
- CN202411067439.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-06
AI Technical Summary
There are huge challenges in medical imaging data processing, including large amount of data, long analysis time, easy fatigue, high misjudgment rate, and inconsistent subjectivity and data quality of diagnosis results, resulting in a lack of unified standards.
A cancer data processing method is proposed, by obtaining image data, performing texture feature extraction, Gaussian distribution feature calculation, hash calculation and information entropy calculation, constructing multi-level indexes, realizing content-based image search, and improving image data processing and analysis efficiency.
It realizes efficient processing and analysis of large-scale cancer image data, improves identification and classification accuracy, reduces errors, supports real-time analysis and decision-making, and improves the consistency and efficiency of data processing.
Smart Images

Figure CN118588246B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of content-based image retrieval, and in particular to a cancer data processing method. Background Art
[0002] In modern medical diagnosis, medical image data processing and analysis technology plays a vital role. With the rapid development of medical imaging equipment (such as CT, MRI, PET, ultrasound, etc.), it has become easier to obtain high-resolution and high-quality medical image data. However, the massive amount of medical image data has brought great challenges to medical image analysis. The amount of medical image data is huge, and manual analysis requires a lot of time and effort, and is prone to fatigue and misjudgment. Different doctors have different experiences and judgment standards, which leads to the subjectivity of diagnostic results and the lack of unified standards. Summary of the invention
[0003] In order to solve the above technical problems, the present invention proposes a cancer data processing method to solve at least one of the above technical problems.
[0004] The present application provides a cancer data processing method, which is applied to a medical imaging big data platform and includes the following steps:
[0005] Step S1: Acquire cancer imaging data;
[0006] Step S2: extracting texture features according to the cancer image data to obtain cancer image texture feature data;
[0007] Step S3: performing Gaussian distribution feature calculation based on cancer image texture feature data to obtain cancer image Gaussian distribution feature data;
[0008] Step S4: performing hash calculation and information entropy calculation according to the Gaussian distribution characteristic data of the cancer image to obtain cancer image hash data and cancer image information entropy data;
[0009] Step S5: construct a cancer image similarity index based on the cancer image hash data and the cancer image information entropy data to obtain cancer image similarity index data for cancer data processing.
[0010] The present invention realizes content-based image search through systematic feature extraction, statistical analysis, hash calculation and similarity index construction, thereby realizing efficient processing and analysis of large-scale cancer imaging data. Compared with conventional image search, the present invention builds a multi-level index based on content features and feature dimensionality reduction processing, realizes the uniqueness of image search, and provides the ability of fast search. The combination of multidimensional texture features and statistical features provides rich feature information and improves the recognition and classification accuracy of cancer image fast search. The efficient similarity index structure and hierarchical index optimization ensure the efficient storage and fast retrieval of large-scale data, and support real-time analysis and decision-making.
[0011] Preferably, step S1 specifically includes:
[0012] Collecting data through a pre-stored cancer imaging database to obtain primary cancer imaging data;
[0013] The primary cancer imaging data is standardized to obtain cancer imaging data.
[0014] The present invention uses the existing cancer imaging database for data collection, which can significantly shorten the data collection time and quickly obtain a large amount of imaging data. Standardization processing ensures the consistency and quality of the imaging data, making it more suitable for subsequent feature extraction and analysis. It improves the accuracy and reliability of the analysis results and reduces errors and deviations caused by data quality issues.
[0015] Preferably, step S2 specifically includes:
[0016] Extracting first texture features according to the cancer image data to obtain first cancer image texture feature data;
[0017] Performing a second texture feature extraction based on the cancer image data to obtain second cancer image texture feature data, wherein the first texture feature extraction and the second texture feature extraction are different texture feature extraction methods;
[0018] Texture feature fusion is performed according to the first cancer image texture feature data and the second cancer image texture feature data to obtain cancer image texture feature data.
[0019] The present invention adopts different texture feature extraction methods (such as GLCM and LBP) to capture multi-dimensional features in the image. The ability to capture the details and overall features of cancer images is improved, which is conducive to more comprehensive and accurate image analysis. Different feature extraction methods can extract complementary feature information. For example, GLCM is good at capturing the spatial relationship of grayscale, while LBP focuses on local texture patterns. Through feature complementarity, the richness and distinguishing ability of feature representation can be improved. Fusion of feature data from different sources can comprehensively utilize the advantages of various features and reduce the deviation caused by a single feature. The fused feature data has higher expressive power and stability, which improves the accuracy of model training and prediction. The data is standardized and normalized before fusion to ensure the consistency of scales between different features. Data deviation is reduced and the effect and reliability of feature fusion are improved. Comprehensive use of multiple feature extraction methods and fusion strategies can more finely analyze and identify cancer images and improve the accuracy of cancer image classification.
[0020] Preferably, the first texture feature extraction is specifically:
[0021] Performing grayscale processing on the cancer image data to obtain cancer image grayscale data;
[0022] Calculate the gray-level co-occurrence matrix of the gray-scale data of cancer images to obtain gray-level co-occurrence matrix data;
[0023] The texture features are extracted from the gray level co-occurrence matrix data to obtain the first cancer image texture feature data.
[0024] In the present invention, the brightness information of the image is retained through grayscale processing, making the subsequent grayscale co-occurrence matrix calculation more accurate. The interference of color on texture feature extraction is eliminated, and the accuracy and consistency of feature extraction are improved. The grayscale co-occurrence matrix can effectively capture the spatial relationship between grayscale values in the image and reflect the local texture characteristics of the image. It can describe the subtle changes in the image in detail, which is helpful for identifying and classifying cancer images. By calculating multiple features such as contrast, energy, entropy and homogeneity, multi-dimensional texture feature information is provided. The feature space is enriched and the model's ability to distinguish different image features is improved.
[0025] Preferably, the second cancer image texture feature data includes static local texture feature data and dynamic local texture feature data, and the second texture feature extraction is specifically as follows:
[0026] Performing convolution calculation according to the cancer image data to obtain cancer image convolution data;
[0027] Perform filtering processing on cancer image convolution data to obtain static local texture feature data;
[0028] Performing morphological processing on the cancer image convolution data to obtain cancer image morphological processing data;
[0029] Performing closed space segmentation on the cancer image morphological processing data to obtain cancer image closed space data;
[0030] Calculate the texture energy of cancer image closed space data to obtain dynamic local texture feature data;
[0031] The dynamic local texture feature data includes local texture gradient change feature data and local texture multi-frequency feature data. The texture energy calculation is specifically as follows:
[0032] Perform gradient calculation on cancer image closed space data to obtain closed space gradient data;
[0033] Perform gradient amplitude feature extraction and gradient direction feature extraction according to closed space gradient data to obtain gradient amplitude feature data and gradient direction feature data respectively;
[0034] Calculate the change rate according to the gradient amplitude characteristic data and the gradient direction characteristic data to obtain the gradient characteristic change rate data;
[0035] Generate a gradient sliding window according to the gradient feature change rate data to obtain gradient sliding window data;
[0036] Extract local gradient features from closed space gradient data according to gradient sliding window data to obtain local texture gradient change feature data;
[0037] Perform wavelet decomposition on the closed space data of cancer images to obtain closed space wavelet decomposition data;
[0038] Perform convolution calculation on the closed space wavelet decomposition data to obtain wavelet decomposition convolution data;
[0039] Perform sparse multi-head self-attention calculation based on wavelet decomposition convolution data to obtain wavelet decomposition sparse multi-head attention data;
[0040] Energy calculation is performed based on wavelet decomposition of sparse multi-head attention data to obtain local texture multi-frequency feature data.
[0041] The convolution calculation in the present invention can automatically extract local features in the image, especially edge and texture information. The automation and accuracy of feature extraction are improved, and manual intervention is reduced. Filtering processing can effectively extract static texture features of the image, such as direction, frequency, etc. Providing detailed static local texture information is helpful for accurately locating and analyzing cancer images, and providing relatively static texture feature dimension data for subsequent index construction. Morphological processing can highlight structural features in the image, such as boundaries and shapes. The structural information of the image is enhanced, making subsequent segmentation and analysis more accurate. Closed space segmentation can accurately extract closed areas in the image, which are usually cancer image areas of interest. Segmenting key areas provides precise targets for subsequent texture analysis. Texture energy calculation can quantify dynamic texture features of the image, such as rate of change and energy distribution. Capturing dynamic change information in the image enhances the richness and accuracy of feature representation. Combining static and dynamic texture features, comprehensively describing different dimensional information of cancer images, improving the expressive power and classification accuracy of features, and helping to more comprehensively identify and analyze the distinctiveness and uniqueness of cancer images, thereby achieving distinction and dimensionality reduction in content dimensions, and reducing the difficulty of search and storage.
[0042] In the present invention, the image data can be decomposed into sub-bands of different scales by wavelet decomposition method, so as to extract texture features at different scales and provide rich feature information. However, the existing technical means often adopt a single-scale or fixed-scale feature extraction method, which is difficult to fully capture the multi-scale information in the image data. The present invention provides rich feature information through multi-scale and multi-angle feature extraction and fusion, improves the feature description ability, realizes efficient feature extraction and energy calculation, and meets the real-time processing requirements.
[0043] Preferably, the closed space segmentation is specifically as follows:
[0044] Connectivity pixel extraction is performed on cancer image morphological processing data to obtain connectivity pixel data;
[0045] Mapping and segmenting the cancer image morphological processing data according to the connectivity pixel data to obtain connectivity pixel mapping segmentation data;
[0046] Integrate the data with pixel overlap in the connectivity pixel map segmentation data into connectivity pixel overlap data;
[0047] Calculate the overlap degree of the connectivity pixel coincidence data to obtain the overlap degree data;
[0048] When it is determined that the overlapping degree data is greater than the preset overlapping degree threshold data, the smallest data in the connectivity pixel coincidence data is determined as the cancer image closed space data;
[0049] When it is determined that the overlapping degree data is less than or equal to the preset overlapping degree threshold data, the largest data in the connectivity pixel coincidence data is determined as the cancer image closed space data.
[0050] The present invention can accurately extract connected pixel areas through connectivity analysis to ensure the integrity and accuracy of the segmented area. The accuracy of the segmentation results is improved, making the extracted closed space data more representative, which is helpful for subsequent analysis and processing. By identifying and integrating overlapping pixel data, the problem of data duplication that may be caused by overlapping areas is solved. Data redundancy is reduced, and the efficiency and accuracy of data processing are improved. The selection criteria of closed spaces are dynamically adjusted according to the overlapping degree data, making the segmentation results more flexible and adaptable. The robustness of the segmentation method is improved, and it can adapt to image data of different complexities.
[0051] Preferably, step S3 specifically includes:
[0052] Performing clustering calculation according to cancer image texture feature data to obtain cancer image texture clustering feature data;
[0053] Performing Gaussian distribution processing on cancer image texture clustering feature data to obtain cancer image texture Gaussian distribution data;
[0054] Statistical feature extraction is performed on the cancer image texture Gaussian distribution data to obtain the cancer image Gaussian distribution feature data.
[0055] The clustering algorithm in the present invention can group similar texture features, reduce the dimension of data, and improve the efficiency and accuracy of analysis. Through clustering, the patterns and structures in texture features can be better captured, and the accuracy of feature representation can be improved. Gaussian distribution processing can fit cluster features and provide more detailed feature descriptions. Through the Gaussian distribution model, the distribution of features can be more accurately characterized and the accuracy of data modeling can be improved. Statistical features (such as mean, variance, skewness, kurtosis) extracted from Gaussian distribution data can provide rich analysis information. Through clustering and Gaussian distribution processing, the dimension of data can be effectively reduced and the data structure can be simplified. The simplified data is easier to process and analyze, which improves the efficiency of data processing.
[0056] Preferably, step S4 is specifically:
[0057] Perform hash calculation according to the Gaussian distribution characteristic data of cancer images to obtain cancer image hash data;
[0058] Performing spatial mapping according to Gaussian distribution characteristic data of cancer images to obtain cancer image spatial mapping data;
[0059] Information entropy calculation is performed based on the cancer image spatial mapping data and the cancer image texture feature data to obtain cancer image information entropy data.
[0060] The hash calculation in the present invention can quickly generate a unique hash value, which is convenient for data indexing and retrieval. The efficiency of data processing and retrieval is improved, especially in the application of large-scale data sets. By mapping Gaussian distribution feature data to high-dimensional space through spatial mapping methods (such as t-SNE, PCA, UMAP), the intrinsic structure of the data can be better captured. The accuracy and expressiveness of feature representation are improved, which is helpful for subsequent analysis and classification. Information entropy, as an indicator of uncertainty, can quantify the complexity of spatial mapping data and texture feature data. A quantitative description of data complexity and uncertainty is provided, which helps to improve the classification and recognition capabilities of the model. Through multi-step processing of hash calculation, spatial mapping and information entropy calculation, data can be analyzed and represented from different angles, emphasizing uniqueness while increasing the difficulty of data storage.
[0061] Preferably, step S5 is specifically:
[0062] Perform fixed data segmentation according to the cancer image hash data and the cancer image information entropy data to obtain hash fixed data segment data and information entropy fixed data segment data respectively;
[0063] Cancer image index is constructed according to hash fixed data segment data and information entropy fixed data segment data to obtain primary cancer image similarity index data;
[0064] Construct a cancer image index based on cancer image hash data and cancer image information entropy data to obtain secondary cancer image similarity index data;
[0065] The primary cancer image similarity index data and the secondary cancer image similarity index data are hierarchically indexed and stored to obtain cancer image similarity index data for cancer data processing operations.
[0066] In the present invention, the processing of hash data and information entropy data is standardized by fixed data segmentation. The consistency and efficiency of data processing are improved, and subsequent index construction and rapid retrieval are facilitated. By constructing primary and secondary indexes, a multi-level retrieval mechanism is provided. The primary index can quickly locate roughly similar image data, and the secondary index provides a more refined match, which improves the accuracy and efficiency of retrieval. The hierarchical index storage stores index data of different levels separately, which optimizes the data management and retrieval process. The organization of data storage and the speed of retrieval are improved, which is convenient for processing large-scale image data. Combining hash data and information entropy data for index construction provides multi-dimensional feature information. The flexibility and robustness of data processing are enhanced, and it can better adapt to data of different types and complexity. Through the multi-level and multi-dimensional indexing mechanism, the retrieval and analysis efficiency of image data is significantly improved. It is convenient to quickly find similar image data and support real-time analysis and diagnostic decisions. Through fixed data segment segmentation and multi-level indexing mechanism, data redundancy and repeated calculations are reduced. The efficiency and accuracy of data processing are improved, and computing resources and time are saved.
[0067] Preferably, the present application also provides a cancer data processing system for executing the cancer data processing method as described above, the cancer data processing system comprising:
[0068] A cancer imaging data acquisition module, used for acquiring cancer imaging data;
[0069] A cancer image texture feature extraction module is used to extract texture features based on cancer image data to obtain cancer image texture feature data;
[0070] A cancer image Gaussian distribution feature calculation module is used to calculate Gaussian distribution features based on cancer image texture feature data to obtain cancer image Gaussian distribution feature data;
[0071] A cancer image hash and information entropy calculation module, used to perform hash calculation and information entropy calculation based on the Gaussian distribution characteristic data of cancer images to obtain cancer image hash data and cancer image information entropy data;
[0072] The cancer image similarity index construction module is used to construct a cancer image similarity index based on cancer image hash data and cancer image information entropy data to obtain cancer image similarity index data for cancer data processing operations.
[0073] The beneficial effects of the present invention are: improving the consistency and efficiency of data processing and ensuring data quality. By extracting features through multiple methods, the richness and accuracy of feature expression are improved. Compared with the traditional texture feature extraction method, the accuracy of the extracted feature representation for search is improved by 33%. A detailed description of data distribution is provided, which enhances the expressive power of the model. The efficiency of data indexing and retrieval is improved, and the quantification of data complexity is provided. The data management and retrieval process is optimized, the efficiency and accuracy of large-scale data processing are improved, and the search efficiency is improved by more than 90% through hierarchical indexing. BRIEF DESCRIPTION OF THE DRAWINGS
[0074] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:
[0075] Figure 1 A flowchart showing the steps of a cancer data processing method according to an embodiment is shown;
[0076] Figure 2 A flowchart showing a method for extracting texture features of cancer images according to an embodiment of the present invention is shown;
[0077] Figure 3 A flowchart showing a method for calculating Gaussian distribution characteristics of cancer images according to an embodiment is shown;
[0078] Figure 4 A flowchart of a method for calculating hash and information entropy of cancer images according to an embodiment is shown;
[0079] Figure 5 A flowchart of the steps of a method for constructing a cancer image similarity index according to an embodiment is shown. DETAILED DESCRIPTION
[0080] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.
[0081] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.
[0082] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.
[0083] See also Figures 1 to 5 The present application provides a cancer data processing method, which is applied to a medical imaging big data platform and includes the following steps:
[0084] Step S1: Acquire cancer imaging data;
[0085] Specifically, connect to a pre-stored cancer imaging database (such as a PACS system). Retrieve relevant imaging data based on specific query conditions (such as patient ID, examination date, imaging type, etc.). Use denoising algorithms such as Gaussian filtering and median filtering to remove noise. Normalize image pixel values to a range of 0 to 1.
[0086] Step S2: extracting texture features according to the cancer image data to obtain cancer image texture feature data;
[0087] Specifically, the color image is converted into a grayscale image. The gray-level co-occurrence matrix with directions (0 degrees, 45 degrees, 90 degrees, 135 degrees) and a distance of 1 is calculated. Contrast, energy, entropy, and homogeneity features are extracted. The gray-level co-occurrence matrix feature data (GLCMFeature Data) is obtained.
[0088] Calculate the grayscale difference between each pixel and the surrounding pixels to generate LBP code. Calculate LBP histogram to obtain local binary pattern feature data (LBP Feature Data).
[0089] The GLCM features and LBP features are normalized and fused to obtain cancer image texture feature data.
[0090] Step S3: performing Gaussian distribution feature calculation based on cancer image texture feature data to obtain cancer image Gaussian distribution feature data;
[0091] Specifically, the K-means clustering algorithm is used to cluster the texture feature data. K cluster centers are initialized. The cluster centers are updated iteratively until convergence. Texture cluster feature data is obtained.
[0092] Calculate Gaussian distribution parameters (mean and variance) for each cluster (texture cluster feature data). Fit a Gaussian distribution model. Extract statistical features such as mean, variance, skewness, and kurtosis from Gaussian distribution data. Obtain Gaussian distribution feature data.
[0093] Step S4: performing hash calculation and information entropy calculation according to the Gaussian distribution characteristic data of the cancer image to obtain cancer image hash data and cancer image information entropy data;
[0094] Specifically, the SHA-256 algorithm is used to hash the Gaussian distribution feature data, generate a hash value for each feature vector, and obtain cancer image hash data.
[0095] The t-SNE algorithm is used to map Gaussian distribution feature data to low-dimensional space, retaining the local proximity relationship in high-dimensional space, and obtaining cancer image space mapping data.
[0096] Calculate the Shannon entropy of the mapping data and texture feature data, , is the cancer imaging information entropy data, is the probability function of cancer image space mapping data and cancer image texture feature data, Spatial mapping data for cancer imaging, is the texture feature data of cancer images. The information entropy data of cancer images is obtained.
[0097] Step S5: construct a cancer image similarity index based on the cancer image hash data and the cancer image information entropy data to obtain cancer image similarity index data for cancer data processing.
[0098] Specifically, a fixed length is determined, and the hash data and information entropy data are segmented, for example, the first five digits or the last five digits are intercepted. A primary index is constructed using the hash fixed data segment and the information entropy fixed data segment. The primary index is implemented using a data structure such as an inverted index or a B-tree. Primary cancer image similarity index data is obtained.
[0099] Directly use the original hash data and information entropy data to build a secondary index. Use data structures such as KD tree or LSH (local sensitive hashing) to implement the secondary index. Obtain secondary cancer image similarity index data.
[0100] The primary index and the secondary index are stored in different storage levels respectively, and cancer image similarity index data is obtained to perform cancer data processing operations.
[0101] The present invention realizes content-based image search through systematic feature extraction, statistical analysis, hash calculation and similarity index construction, thereby realizing efficient processing and analysis of large-scale cancer imaging data. Compared with conventional image search, the present invention builds a multi-level index based on content features and feature dimensionality reduction processing, realizes the uniqueness of image search, and provides the ability of fast search. The combination of multidimensional texture features and statistical features provides rich feature information and improves the recognition and classification accuracy of cancer image fast search. The efficient similarity index structure and hierarchical index optimization ensure the efficient storage and fast retrieval of large-scale data, and support real-time analysis and decision-making.
[0102] Preferably, step S1 specifically includes:
[0103] Collecting data through a pre-stored cancer imaging database to obtain primary cancer imaging data;
[0104] Specifically, a pre-stored cancer imaging database (such as a PACS system) is connected to retrieve relevant imaging data according to specific query conditions (such as patient ID, examination date, imaging type, etc.).
[0105] The primary cancer imaging data is standardized to obtain cancer imaging data.
[0106] Specifically, operations such as denoising, grayscale, and normalization are performed to ensure the consistency and quality of the image data. Algorithms such as Gaussian filtering and median filtering are used to remove image noise. Color images are converted to grayscale images, retaining only brightness information. The pixel values of grayscale images are normalized to the range of [0, 1]. Standardized cancer image data is obtained.
[0107] The present invention uses the existing cancer imaging database for data collection, which can significantly shorten the data collection time and quickly obtain a large amount of imaging data. Standardization processing ensures the consistency and quality of the imaging data, making it more suitable for subsequent feature extraction and analysis. It improves the accuracy and reliability of the analysis results and reduces errors and deviations caused by data quality issues.
[0108] Preferably, step S2 specifically includes:
[0109] Step S21: extracting first texture features according to the cancer image data to obtain first cancer image texture feature data;
[0110] Specifically, the spatial symbiosis relationship of the grayscale values in the image is calculated to extract the texture features. The grayscale symbiosis matrix of different directions (0 degrees, 45 degrees, 90 degrees, 135 degrees) and distances (such as 1) is calculated to obtain the first cancer image texture feature data (GLCM feature).
[0111] Step S22: performing a second texture feature extraction according to the cancer image data to obtain second cancer image texture feature data, wherein the first texture feature extraction and the second texture feature extraction are different texture feature extraction methods;
[0112] Specifically, the local binary pattern code of each pixel in the image is calculated to extract the texture feature. The LBP code of each pixel is calculated to generate an LBP image and perform statistical histograms to obtain the texture feature data (LBP feature) of the second cancer image.
[0113] Step S23: performing texture feature fusion according to the first cancer image texture feature data and the second cancer image texture feature data to obtain cancer image texture feature data.
[0114] Specifically, the GLCM features and LBP features are normalized and fused to generate comprehensive texture feature data.
[0115] The present invention adopts different texture feature extraction methods (such as GLCM and LBP) to capture multi-dimensional features in the image. The ability to capture the details and overall features of cancer images is improved, which is conducive to more comprehensive and accurate image analysis. Different feature extraction methods can extract complementary feature information. For example, GLCM is good at capturing the spatial relationship of grayscale, while LBP focuses on local texture patterns. Through feature complementarity, the richness and distinguishing ability of feature representation can be improved. Fusion of feature data from different sources can comprehensively utilize the advantages of various features and reduce the deviation caused by a single feature. The fused feature data has higher expressive power and stability, which improves the accuracy of model training and prediction. The data is standardized and normalized before fusion to ensure the consistency of scales between different features. Data deviation is reduced and the effect and reliability of feature fusion are improved. Comprehensive use of multiple feature extraction methods and fusion strategies can more finely analyze and identify cancer images and improve the accuracy of cancer image classification.
[0116] Preferably, the first texture feature extraction is specifically:
[0117] Performing grayscale processing on the cancer image data to obtain cancer image grayscale data;
[0118] Specifically, the color image is converted into a grayscale image, retaining only the brightness information and removing the color information.
[0119] Calculate the gray-level co-occurrence matrix of the gray-scale data of cancer images to obtain gray-level co-occurrence matrix data;
[0120] Specifically, the gray-level co-occurrence matrix of different directions and distances is calculated to count the co-occurrence relationship between gray values. The parameters (direction and distance) of the gray-level co-occurrence matrix are determined, and the gray-level co-occurrence matrix is calculated for the gray-scale data of the cancer image according to the parameters of the gray-level co-occurrence matrix to obtain the gray-level co-occurrence matrix data.
[0121] The texture features are extracted from the gray level co-occurrence matrix data to obtain the first cancer image texture feature data.
[0122] Specifically, texture features such as contrast, energy, entropy, and homogeneity are extracted from the gray level co-occurrence matrix. Texture features are calculated using the gray level co-occurrence matrix. Contrast, energy, entropy, and homogeneity features are extracted. The first cancer image texture feature data (First Cancer Image Texture Feature Data) is obtained, including contrast, energy, entropy, and homogeneity features.
[0123] In the present invention, the brightness information of the image is retained through grayscale processing, making the subsequent grayscale co-occurrence matrix calculation more accurate. The interference of color on texture feature extraction is eliminated, and the accuracy and consistency of feature extraction are improved. The grayscale co-occurrence matrix can effectively capture the spatial relationship between grayscale values in the image and reflect the local texture characteristics of the image. It can describe the subtle changes in the image in detail, which is helpful for identifying and classifying cancer images. By calculating multiple features such as contrast, energy, entropy and homogeneity, multi-dimensional texture feature information is provided. The feature space is enriched and the model's ability to distinguish different image features is improved.
[0124] Preferably, the second cancer image texture feature data includes static local texture feature data and dynamic local texture feature data, and the second texture feature extraction is specifically as follows:
[0125] Performing convolution calculation according to the cancer image data to obtain cancer image convolution data;
[0126] Specifically, choose an appropriate convolution kernel (such as or ) (such as the maximum convolution layer, minimum convolution layer, average convolution layer and Gaussian convolution layer, the offset is 1 or 2, and the zero padding operation is performed.) Perform convolution operations on image data.
[0127] Perform filtering processing on cancer image convolution data to obtain static local texture feature data;
[0128] Specifically, specific filters (such as Gabor filters and Sobel filters) are applied to process the convolution data to extract static local texture features.
[0129] Performing morphological processing on the cancer image convolution data to obtain cancer image morphological processing data;
[0130] Specifically, morphological operations (such as dilation, erosion, opening, and closing) are used to process the convolution data to emphasize structural features. Morphological operations (such as dilation and erosion) are selected. Morphological processing is performed on the convolution data.
[0131] Performing closed space segmentation on the cancer image morphological processing data to obtain cancer image closed space data;
[0132] Specifically, the morphologically processed data is segmented using a connectivity analysis method to extract closed space regions. Connectivity analysis is performed to identify closed spaces. The closed space regions are segmented to obtain closed space data of cancer images.
[0133] Texture energy calculation is performed on closed-space data of cancer images to obtain dynamic local texture feature data.
[0134] Specifically, a specific method (such as the Laws texture energy method) is used to calculate the texture energy of the closed space data and extract the dynamic local texture features. The texture energy is calculated using the Laws texture energy method. Energy features are extracted to obtain dynamic local texture feature data.
[0135] The dynamic local texture feature data includes local texture gradient change feature data and local texture multi-frequency feature data. The texture energy calculation is specifically as follows:
[0136] Perform gradient calculation on cancer image closed space data to obtain closed space gradient data;
[0137] Specifically, the Sobel operator or Prewitt operator is used to calculate the gradient of the image, calculate the horizontal gradient and the vertical gradient, and combine the gradient data to obtain the closed space gradient data.
[0138] Perform gradient amplitude feature extraction and gradient direction feature extraction according to closed space gradient data to obtain gradient amplitude feature data and gradient direction feature data respectively;
[0139] Specifically, the gradient magnitude and gradient direction formulas are used for feature extraction. Gradient magnitude feature: , is the closed space gradient data, is the horizontal gradient, is the vertical gradient. Gradient direction formula: .
[0140] Calculate the change rate according to the gradient amplitude characteristic data and the gradient direction characteristic data to obtain the gradient characteristic change rate data;
[0141] Specifically, the closed space data of the cancer image is divided by preset division window data to obtain closed space division data; gradient calculation is performed based on the closed space division data to obtain closed space division gradient data; gradient amplitude feature extraction and gradient direction feature extraction are performed based on the closed space division gradient data to obtain closed space division gradient amplitude feature data and closed space division gradient direction feature data; the change rate of gradient amplitude and direction is calculated based on the closed space division gradient amplitude feature data and the closed space division gradient direction feature data to obtain gradient feature change rate data. , is the gradient characteristic change rate data, For the Gradient amplitude feature data and gradient direction feature data, For the Gradient amplitude feature data and gradient direction feature data.
[0142] Specifically, the rate of change is:
[0143] ;
[0144] ;
[0145] ;
[0146] is the gradient characteristic change rate data, For Coordinate increment data in direction, For Coordinate increment data in direction, is the pixel horizontal coordinate data, For The gradient amplitude feature data at is the pixel ordinate data, For The gradient amplitude feature data at for The gradient direction change rate data in the direction, for The gradient direction change rate data in the direction, for The influence of the gradient direction change rate data on the adjustment data, is the gradient direction feature data, for The direction of the gradient direction change rate data affects the adjustment data.
[0147] Generate a gradient sliding window according to the gradient feature change rate data to obtain gradient sliding window data;
[0148] Specifically, the sliding window technique is used to smooth the gradient change rate. , is the gradient sliding window data, is the quantitative data of gradient characteristic change rate, is the order term data of gradient characteristic change rate, is the initial data of gradient characteristic change rate, For the Gradient feature change rate data.
[0149] Specifically, matching is performed based on the gradient feature change rate data and the preset sliding window change rate matching list data to obtain the gradient sliding window data (such as the sliding window size, the sliding window offset, and whether to fill the edge with 0).
[0150] Extract local gradient features from closed space gradient data according to gradient sliding window data to obtain local texture gradient change feature data;
[0151] Specifically, local feature extraction is performed on the smoothed gradient change data to extract local gradient features and form local texture gradient change feature data.
[0152] Specifically, local gradient features are extracted from closed space gradient data according to the gradient sliding window data to form local texture gradient change feature data. Local features are extracted from closed space gradient data using the matched sliding window data. Local features are extracted from closed space gradient data using the sliding window to generate local texture gradient change feature data.
[0153] Perform wavelet decomposition on the closed space data of cancer images to obtain closed space wavelet decomposition data;
[0154] Specifically, the image is decomposed using wavelet transform. A wavelet basis (such as Haar wavelet or Daubechies wavelet) is selected. Wavelet decomposition is performed to obtain multi-scale wavelet coefficients.
[0155] Perform convolution calculation on the closed space wavelet decomposition data to obtain wavelet decomposition convolution data;
[0156] Specifically, the wavelet coefficients are processed using a convolution operation. A convolution kernel (such as a maximum kernel, a minimum kernel, an average kernel, an extreme kernel, and a Gaussian kernel) is selected. The convolution operation is performed to obtain the wavelet decomposition data after convolution.
[0157] Perform sparse multi-head self-attention calculation based on wavelet decomposition convolution data to obtain wavelet decomposition sparse multi-head attention data;
[0158] Specifically, the convolution data is processed using a sparse multi-head self-attention mechanism. Query, Key, and Value vectors are calculated. Attention scores are calculated and mask operations are performed. Attention weights are calculated and weighted summed.
[0159] Perform linear transformation on the wavelet decomposition convolution data to obtain query, key, and value vectors. Divide the query, key, and value vectors into multiple heads, and calculate attention independently for each head. Sparsely process the score matrix of each attention head, retaining only the important attention scores to reduce computational complexity. Use the Top-k sparse method to retain only the top attention scores. The highest attention scores are selected and the rest are set to zero. The dot product of the query and the key is calculated to get the attention score matrix. A sparse mask is applied to set the non-important attention scores to negative infinity to ignore them during Softmax normalization. Softmax normalize the sparsely processed attention score matrix to get the attention weights. The value vector is weighted summed using the attention weights to get the output of each attention head. The outputs of all attention heads are concatenated to get the output of the sparse multi-head attention.
[0160] Energy calculation is performed based on wavelet decomposition of sparse multi-head attention data to obtain local texture multi-frequency feature data.
[0161] Specifically, the energy of the attention data is calculated (such as summation, weighted summation, or activation calculation and weighted summation) to extract multi-frequency features.
[0162] The convolution calculation in the present invention can automatically extract local features in the image, especially edge and texture information. The automation and accuracy of feature extraction are improved, and manual intervention is reduced. Filtering processing can effectively extract static texture features of the image, such as direction, frequency, etc. Providing detailed static local texture information is helpful for accurately locating and analyzing cancer images, and providing relatively static texture feature dimension data for subsequent index construction. Morphological processing can highlight structural features in the image, such as boundaries and shapes. The structural information of the image is enhanced, making subsequent segmentation and analysis more accurate. Closed space segmentation can accurately extract closed areas in the image, which are usually cancer image areas of interest. Segmenting key areas provides precise targets for subsequent texture analysis. Texture energy calculation can quantify dynamic texture features of the image, such as rate of change and energy distribution. Capturing dynamic change information in the image enhances the richness and accuracy of feature representation. Combining static and dynamic texture features, comprehensively describing different dimensional information of cancer images, improving the expressive power and classification accuracy of features, and helping to more comprehensively identify and analyze the distinctiveness and uniqueness of cancer images, thereby achieving distinction and dimensionality reduction in content dimensions, and reducing the difficulty of search and storage.
[0163] In the present invention, the image data can be decomposed into sub-bands of different scales by wavelet decomposition method, so as to extract texture features at different scales and provide rich feature information. However, the existing technical means often adopt a single-scale or fixed-scale feature extraction method, which is difficult to fully capture the multi-scale information in the image data. The present invention provides rich feature information through multi-scale and multi-angle feature extraction and fusion, improves the feature description ability, realizes efficient feature extraction and energy calculation, and meets the real-time processing requirements.
[0164] Preferably, the closed space segmentation is specifically as follows:
[0165] Connectivity pixel extraction is performed on cancer image morphological processing data to obtain connectivity pixel data;
[0166] Specifically, a connected component labeling algorithm (such as 4-connected or 8-connected) is used to extract connected pixel regions. A connectivity analysis algorithm is used to analyze the morphologically processed data, identify connected regions, and obtain connected pixel data.
[0167] Mapping and segmenting the cancer image morphological processing data according to the connectivity pixel data to obtain connectivity pixel mapping segmentation data;
[0168] Specifically, the morphologically processed data is segmented using the connectivity pixel data as a mask, and the corresponding segmented area is extracted using the connectivity pixel data as a mask to obtain connectivity pixel map segmentation data.
[0169] Integrate the data with pixel overlap in the connectivity pixel map segmentation data into connectivity pixel overlap data;
[0170] Specifically, overlapping regions in the mapped segmented data are identified and integrated. Coincident regions are detected and integrated.
[0171] Calculate the overlap degree of the connectivity pixel coincidence data to obtain the overlap degree data;
[0172] Specifically, the degree of overlap of the overlapping regions is calculated, and the overlapping area or the number of pixels is quantified.
[0173] When it is determined that the overlapping degree data is greater than the preset overlapping degree threshold data, the smallest data in the connectivity pixel coincidence data is determined as the cancer image closed space data;
[0174] Specifically, the overlap degree data is compared with a preset threshold value. If the overlap degree data is greater than the preset threshold value, the minimum connectivity pixel data in the overlap area is selected as the closed space data.
[0175] When it is determined that the overlapping degree data is less than or equal to the preset overlapping degree threshold data, the largest data in the connectivity pixel coincidence data is determined as the cancer image closed space data.
[0176] Specifically, the overlap degree data is compared with a preset threshold value. If the overlap degree data is less than or equal to the preset threshold value, the maximum connectivity pixel data in the overlap area is selected as the closed space data.
[0177] The present invention can accurately extract connected pixel areas through connectivity analysis to ensure the integrity and accuracy of the segmented area. The accuracy of the segmentation results is improved, making the extracted closed space data more representative, which is helpful for subsequent analysis and processing. By identifying and integrating overlapping pixel data, the problem of data duplication that may be caused by overlapping areas is solved. Data redundancy is reduced, and the efficiency and accuracy of data processing are improved. The selection criteria of closed spaces are dynamically adjusted according to the overlapping degree data, making the segmentation results more flexible and adaptable. The robustness of the segmentation method is improved, and it can adapt to image data of different complexities.
[0178] Preferably, step S3 specifically includes:
[0179] Step S31: performing clustering calculation according to the cancer image texture feature data to obtain cancer image texture clustering feature data;
[0180] Specifically, use K-means, hierarchical clustering or DBSCAN algorithms to cluster the texture feature data. Select a clustering algorithm (such as K-means). Set clustering parameters (such as the number of clusters k). Execute the clustering algorithm to cluster the texture feature data to obtain cancer image texture clustering feature data.
[0181] Step S32: performing Gaussian distribution processing on the cancer image texture clustering feature data to obtain cancer image texture Gaussian distribution data;
[0182] Specifically, the Gaussian distribution parameters (mean and variance) are calculated for each cluster, and the Gaussian distribution model is fitted. The mean and variance are calculated for the data points of each cluster. The Gaussian distribution model is fitted.
[0183] Step S33: extracting statistical features from the cancer image texture Gaussian distribution data to obtain cancer image Gaussian distribution feature data.
[0184] Specifically, statistical features such as mean, variance, skewness, kurtosis, etc. are extracted from Gaussian distribution data. Statistical features are extracted from Gaussian distribution models. The mean, variance, skewness, and kurtosis of each Gaussian distribution are calculated.
[0185] The clustering algorithm in the present invention can group similar texture features, reduce the dimension of data, and improve the efficiency and accuracy of analysis. Through clustering, the patterns and structures in texture features can be better captured, and the accuracy of feature representation can be improved. Gaussian distribution processing can fit cluster features and provide more detailed feature descriptions. Through the Gaussian distribution model, the distribution of features can be more accurately characterized and the accuracy of data modeling can be improved. Statistical features (such as mean, variance, skewness, kurtosis) extracted from Gaussian distribution data can provide rich analysis information. Through clustering and Gaussian distribution processing, the dimension of data can be effectively reduced and the data structure can be simplified. The simplified data is easier to process and analyze, which improves the efficiency of data processing.
[0186] Preferably, step S4 is specifically:
[0187] Step S41: performing hash calculation according to the cancer image Gaussian distribution characteristic data to obtain cancer image hash data;
[0188] Specifically, a hash function (such as SHA-256) is used to encode the Gaussian distribution feature data to generate a hash value. The Gaussian distribution feature data is converted into a string or byte format. A hash function is used to generate a hash value.
[0189] Step S42: performing spatial mapping according to the cancer image Gaussian distribution feature data to obtain cancer image spatial mapping data;
[0190] Specifically, use dimensionality reduction or embedding algorithms (such as t-SNE, PCA, UMAP) to map Gaussian distribution feature data into a low-dimensional space for visualization and subsequent analysis. Select a spatial mapping algorithm (such as t-SNE). Set algorithm parameters (such as learning rate and number of iterations for t-SNE). Execute the spatial mapping algorithm to map Gaussian distribution feature data into a low-dimensional space.
[0191] Step S43: performing information entropy calculation based on the cancer image space mapping data and the cancer image texture feature data to obtain cancer image information entropy data.
[0192] Specifically, the information entropy calculation method (such as Shannon entropy) is used to calculate the entropy value of the spatial mapping data and texture feature data to quantify the complexity and uncertainty of the data. The spatial mapping data and texture feature data are combined to calculate the joint probability distribution. The Shannon entropy formula is used to calculate the information entropy.
[0193] The hash calculation in the present invention can quickly generate a unique hash value, which is convenient for data indexing and retrieval. The efficiency of data processing and retrieval is improved, especially in the application of large-scale data sets. By mapping Gaussian distribution feature data to high-dimensional space through spatial mapping methods (such as t-SNE, PCA, UMAP), the intrinsic structure of the data can be better captured. The accuracy and expressiveness of feature representation are improved, which is helpful for subsequent analysis and classification. Information entropy, as an indicator of uncertainty, can quantify the complexity of spatial mapping data and texture feature data. A quantitative description of data complexity and uncertainty is provided, which helps to improve the classification and recognition capabilities of the model. Through multi-step processing of hash calculation, spatial mapping and information entropy calculation, data can be analyzed and represented from different angles, emphasizing uniqueness while increasing the difficulty of data storage.
[0194] Preferably, step S5 is specifically:
[0195] Step S51: performing fixed data segment segmentation according to the cancer image hash data and the cancer image information entropy data to obtain hash fixed data segment data and information entropy fixed data segment data respectively;
[0196] Specifically, the hash data and information entropy data are divided into fixed-length data segments to ensure data consistency and standardization. The fixed length of each data segment is determined (such as 128 bits or 256 bits). The hash data and information entropy data are sliced to generate fixed-length data segments.
[0197] Step S52: constructing a cancer image index according to the hash fixed data segment data and the information entropy fixed data segment data to obtain primary cancer image similarity index data;
[0198] Specifically, a primary index is constructed based on a fixed data segment, and an inverted index or other index structure is used to achieve fast retrieval. A primary index is constructed using a hash fixed data segment and an information entropy fixed data segment. An index is implemented using an inverted index or a B-tree or other data structure.
[0199] Step S53: constructing a cancer image index according to the cancer image hash data and the cancer image information entropy data to obtain secondary cancer image similarity index data;
[0200] Specifically, a more refined secondary index is constructed based on the original hash data and information entropy data, and efficient retrieval is achieved using structures such as KD tree or LSH (local sensitive hashing). The secondary index is constructed using the original hash data and information entropy data. The index is implemented using data structures such as KD tree or LSH.
[0201] Step S54: hierarchically index and store the primary cancer image similarity index data and the secondary cancer image similarity index data to obtain cancer image similarity index data for cancer data processing.
[0202] Specifically, the primary and secondary index data are stored in a hierarchical manner to optimize the data management and retrieval process. The primary index data is stored in the first-level storage layer. The secondary index data is stored in the second-level storage layer. When performing a cancer imaging data search operation, the first-level storage layer is searched first, and when the conditions are met, the corresponding second-level storage layer is searched. Compared with conventional search operations, the present invention can achieve more than a thousand times of efficiency improvement, and the accuracy rate is improved by at least 30%.
[0203] In the present invention, the processing of hash data and information entropy data is standardized by fixed data segmentation. The consistency and efficiency of data processing are improved, and subsequent index construction and rapid retrieval are facilitated. By constructing primary and secondary indexes, a multi-level retrieval mechanism is provided. The primary index can quickly locate roughly similar image data, and the secondary index provides a more refined match, which improves the accuracy and efficiency of retrieval. The hierarchical index storage stores index data of different levels separately, which optimizes the data management and retrieval process. The organization of data storage and the speed of retrieval are improved, which is convenient for processing large-scale image data. Combining hash data and information entropy data for index construction provides multi-dimensional feature information. The flexibility and robustness of data processing are enhanced, and it can better adapt to data of different types and complexity. Through the multi-level and multi-dimensional indexing mechanism, the retrieval and analysis efficiency of image data is significantly improved. It is convenient to quickly find similar image data and support real-time analysis and diagnostic decisions. Through fixed data segment segmentation and multi-level indexing mechanism, data redundancy and repeated calculations are reduced. The efficiency and accuracy of data processing are improved, and computing resources and time are saved.
[0204] Preferably, the present application also provides a cancer data processing system for executing the cancer data processing method as described above, the cancer data processing system comprising:
[0205] A cancer imaging data acquisition module, used for acquiring cancer imaging data;
[0206] A cancer image texture feature extraction module is used to extract texture features based on cancer image data to obtain cancer image texture feature data;
[0207] A cancer image Gaussian distribution feature calculation module is used to calculate Gaussian distribution features based on cancer image texture feature data to obtain cancer image Gaussian distribution feature data;
[0208] A cancer image hash and information entropy calculation module, used to perform hash calculation and information entropy calculation based on the Gaussian distribution characteristic data of cancer images to obtain cancer image hash data and cancer image information entropy data;
[0209] The cancer image similarity index construction module is used to construct a cancer image similarity index based on cancer image hash data and cancer image information entropy data to obtain cancer image similarity index data for cancer data processing operations.
[0210] The beneficial effects of the present invention are: improving the consistency and efficiency of data processing and ensuring data quality. By extracting features through multiple methods, the richness and accuracy of feature expression are improved. Compared with the traditional texture feature extraction method, the extracted feature representation is searched with an accuracy rate increased by 33%. A detailed description of data distribution is provided, enhancing the expressive power of the model. The efficiency of data indexing and retrieval is improved, and the quantification of data complexity is provided. The data management and retrieval process is optimized, the efficiency and accuracy of large-scale data processing are improved, and the search efficiency is improved by more than 90% through hierarchical indexing.
[0211] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.
[0212] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.
Claims
1. A cancer data processing method, characterized in that: Applied to the medical imaging big data platform, it includes the following steps: Step S1: Acquire cancer imaging data; Step S2 comprises: Extracting first texture features according to the cancer image data to obtain first cancer image texture feature data; Extracting second texture features based on the cancer image data to obtain second cancer image texture feature data; Performing texture feature fusion according to the first cancer image texture feature data and the second cancer image texture feature data to obtain cancer image texture feature data; Step S3 includes: Performing clustering calculation according to cancer image texture feature data to obtain cancer image texture clustering feature data; Performing Gaussian distribution processing on cancer image texture clustering feature data to obtain cancer image texture Gaussian distribution data; Performing statistical feature extraction on cancer image texture Gaussian distribution data to obtain cancer image Gaussian distribution feature data; Step S4: performing hash calculation and information entropy calculation according to the Gaussian distribution characteristic data of the cancer image to obtain cancer image hash data and cancer image information entropy data; Step S5: constructing a cancer image similarity index according to the cancer image hash data and the cancer image information entropy data to obtain cancer image similarity index data for cancer data processing; The second cancer image texture feature data includes static local texture feature data and dynamic local texture feature data, and the second texture feature extraction is specifically as follows: Performing convolution calculation according to the cancer image data to obtain cancer image convolution data; Perform filtering processing on cancer image convolution data to obtain static local texture feature data; Performing morphological processing on the cancer image convolution data to obtain cancer image morphological processing data; Performing closed space segmentation on the cancer image morphological processing data to obtain cancer image closed space data; Calculate the texture energy of cancer image closed space data to obtain dynamic local texture feature data; The dynamic local texture feature data includes local texture gradient change feature data and local texture multi-frequency feature data. The texture energy calculation is specifically as follows: Perform gradient calculation on cancer image closed space data to obtain closed space gradient data; Perform gradient amplitude feature extraction and gradient direction feature extraction according to closed space gradient data to obtain gradient amplitude feature data and gradient direction feature data respectively; Calculate the change rate according to the gradient amplitude characteristic data and the gradient direction characteristic data to obtain the gradient characteristic change rate data; Generate a gradient sliding window according to the gradient feature change rate data to obtain gradient sliding window data; Extract local gradient features from closed space gradient data according to gradient sliding window data to obtain local texture gradient change feature data; Perform wavelet decomposition on the closed space data of cancer images to obtain closed space wavelet decomposition data; Perform convolution calculation on the closed space wavelet decomposition data to obtain wavelet decomposition convolution data; Perform sparse multi-head self-attention calculation based on wavelet decomposition convolution data to obtain wavelet decomposition sparse multi-head attention data; Energy calculation is performed based on wavelet decomposition of sparse multi-head attention data to obtain local texture multi-frequency feature data; The first texture feature extraction is specifically as follows: Performing grayscale processing on the cancer image data to obtain cancer image grayscale data; Calculate the gray-level co-occurrence matrix of the gray-scale data of cancer images to obtain gray-level co-occurrence matrix data; Extracting texture features from the gray-level co-occurrence matrix data to obtain first cancer image texture feature data; The closed space segmentation is as follows: Connectivity pixel extraction is performed on cancer image morphological processing data to obtain connectivity pixel data; Mapping and segmenting the cancer image morphological processing data according to the connectivity pixel data to obtain connectivity pixel mapping segmentation data; Integrate the data with pixel overlap in the connectivity pixel map segmentation data into connectivity pixel overlap data; Calculate the overlap degree of the connectivity pixel coincidence data to obtain the overlap degree data; When it is determined that the overlapping degree data is greater than the preset overlapping degree threshold data, the smallest data in the connectivity pixel coincidence data is determined as the cancer image closed space data; When it is determined that the overlapping degree data is less than or equal to the preset overlapping degree threshold data, the largest data in the connectivity pixel coincidence data is determined as the cancer image closed space data.
2. The method according to claim 1, characterized in that Step S1 is specifically as follows: Collecting data through a pre-stored cancer imaging database to obtain primary cancer imaging data; The primary cancer imaging data is standardized to obtain cancer imaging data.
3. The method according to claim 1, characterized in that Step S4 is specifically as follows: Perform hash calculation according to the Gaussian distribution characteristic data of cancer images to obtain cancer image hash data; Performing spatial mapping according to Gaussian distribution characteristic data of cancer images to obtain cancer image spatial mapping data; Information entropy calculation is performed based on the cancer image spatial mapping data and the cancer image texture feature data to obtain cancer image information entropy data.
4. The method according to claim 1, characterized in that: Step S5 is specifically as follows: Perform fixed data segmentation according to the cancer image hash data and the cancer image information entropy data to obtain hash fixed data segment data and information entropy fixed data segment data respectively; Cancer image index is constructed according to hash fixed data segment data and information entropy fixed data segment data to obtain primary cancer image similarity index data; Construct a cancer image index based on cancer image hash data and cancer image information entropy data to obtain secondary cancer image similarity index data; The primary cancer image similarity index data and the secondary cancer image similarity index data are hierarchically indexed and stored to obtain cancer image similarity index data for cancer data processing operations.
5. A cancer data processing system, characterized in that: For executing the cancer data processing method according to claim 1, the cancer data processing system comprises: A cancer imaging data acquisition module, used for acquiring cancer imaging data; A cancer image texture feature extraction module is used to extract texture features based on cancer image data to obtain cancer image texture feature data; A cancer image Gaussian distribution feature calculation module is used to calculate Gaussian distribution features based on cancer image texture feature data to obtain cancer image Gaussian distribution feature data; A cancer image hash and information entropy calculation module, used to perform hash calculation and information entropy calculation based on the Gaussian distribution characteristic data of cancer images to obtain cancer image hash data and cancer image information entropy data; The cancer image similarity index construction module is used to construct a cancer image similarity index based on cancer image hash data and cancer image information entropy data to obtain cancer image similarity index data for cancer data processing operations.
