A Data Storage Method for Kidney Disease Patient Status Based on Image Data Processing

Abstract

This invention relates to the field of kidney disease data storage technology and discloses a method for storing kidney disease patient status data based on image data processing. The method acquires a continuous time-series medical image dataset of kidney disease patients; processes the continuous time-series medical image dataset using an image feature extraction algorithm to generate a multi-dimensional feature vector set; iteratively groups the multi-dimensional feature vector set using a feature clustering algorithm to form a kidney disease status category set; in each iterative grouping stage, key feature sets are identified, and adjustment parameters for the feature clustering algorithm are calculated based on historical stored data sets; the grouping rules of the feature clustering algorithm are updated according to the adjustment parameters, and the updated grouping rules are synchronized to the kidney disease status data repository. This method, by dynamically adjusting the clustering grouping rules, makes the storage of kidney disease status data more closely reflect the characteristics of disease progression, improving the effectiveness and relevance of status data storage.

Description

Technical Field

[0001] This invention relates to the field of kidney disease data storage technology, specifically a method for storing the status of kidney disease patients based on image data processing. Background Technology

[0002] In the diagnosis and treatment of kidney disease, the effective storage of patient status data is a crucial foundation for tracking changes in the condition and developing treatment plans. Kidney disease often exhibits dynamic changes, and its status data includes not only physiological indicators but also a large amount of medical image data, such as continuous time-series images from ultrasound, CT, and MRI. These images contain key information about changes in kidney structure and function, and can intuitively reflect the progression or remission of the disease.

[0003] The storage of data on patients with kidney disease often relies on traditional file archiving or simple database classification methods. These methods typically use basic patient information or examination time as an index, storing medical images and related data in a piecemeal fashion, lacking in-depth mining and integration of the inherent characteristics of the image data. Continuous time-series medical images are characterized by strong temporal correlation, high feature dimensionality, and significant individual differences. The image characteristics of the same patient at different stages of the disease can vary significantly, and even patients with similar conditions may have different image presentations due to factors such as physical condition and etiology.

[0004] The limitations of traditional storage methods are becoming increasingly apparent: Because effective feature extraction from image data is not performed, the stored raw images are difficult to use directly for state category classification. Medical staff must spend a significant amount of time sifting through massive amounts of data to extract key information, greatly impacting data utilization efficiency. Existing clustering methods often use algorithms with fixed parameters, which cannot be dynamically adjusted based on patterns observed in historical stored data. When new image data is added, fixed grouping rules may lead to biases in state category classification; for example, classifying rapidly progressing patients into stable state categories, or failing to distinguish similar but fundamentally different pathological features, making it difficult for the stored state data to accurately reflect the patient's actual condition.

[0005] The lack of identification of key features is also a major shortcoming of traditional storage methods. Differences in kidney disease states are often reflected in specific image features, such as renal parenchymal echogenicity and changes in renal pelvis morphology. The changing trends of these features in continuous time series are the core basis for distinguishing state categories. However, traditional methods either ignore these features or rely solely on subjective human judgment, resulting in a lack of objective and unified standards for classifying stored state categories, further reducing the reliability and comparability of the data. This chaotic storage of states not only hinders long-term data management but also impedes subsequent work such as disease analysis and treatment efficacy evaluation. Summary of the Invention

[0006] The purpose of this invention is to provide a method for storing the status data of kidney disease patients based on image data processing, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for storing the status data of kidney disease patients based on image data processing, the method comprising:

[0008] Acquire a collection of continuous time-series medical image data of patients with kidney disease;

[0009] The continuous time-series medical image data set is processed using an image feature extraction algorithm to generate a multi-dimensional feature vector set.

[0010] A feature clustering algorithm is used to perform iterative grouping operations on the multi-dimensional feature vector set to form a set of kidney disease state categories;

[0011] In each iteration of the grouping operation, key feature sets are identified, and the adjustment parameters of the feature clustering algorithm are calculated based on the historical stored data set.

[0012] The grouping rules of the feature clustering algorithm are updated according to the adjusted parameters, and the updated grouping rules are synchronized to the kidney disease status data repository.

[0013] Preferably, the acquisition of the continuous time-series medical image data set of kidney disease patients includes:

[0014] Multiple frames of images of the kidney region of patients with kidney disease are acquired using medical imaging equipment to form an initial image sequence;

[0015] Image enhancement processing, including contrast adjustment and noise filtering, is performed on the initial image sequence to generate a preprocessed image set;

[0016] Extract timestamp information from the preprocessed image set, align image frames from different acquisition times, and generate a time-synchronized image sequence;

[0017] The time-synchronized image sequence is input as a continuous time-series medical image data set into subsequent processing steps.

[0018] Preferably, the applied image feature extraction algorithm processes the continuous time series medical image data set to generate a multi-dimensional feature vector set, including:

[0019] The continuous time-series medical image data set is divided into multiple image block regions;

[0020] A spatial domain feature extractor is used to analyze the texture features and edge distribution of each image patch region to generate a primary feature matrix.

[0021] The initial feature matrix is ​​processed using the time series analysis module to calculate the feature change trend value;

[0022] The primary feature matrix and feature change trend values ​​are combined to generate a multi-dimensional feature vector set, which is then output to the feature clustering algorithm.

[0023] Preferably, the step of using a feature clustering algorithm to perform iterative grouping operations on the multi-dimensional feature vector set to form a set of kidney disease state categories includes:

[0024] Initialize the set of cluster centers for the feature clustering algorithm;

[0025] In each iteration, the similarity distance between the multi-dimensional feature vector and the cluster center is calculated, and the feature vector is assigned to the nearest cluster center.

[0026] Update the cluster center positions and generate new cluster groups based on the allocation results;

[0027] Repeat the iteration until the change in cluster center points is below a stable threshold, then output the set of kidney disease status categories and store it in the database.

[0028] Preferably, the step of identifying key feature sets and calculating adjustment parameters for the feature clustering algorithm based on historical stored data sets during each iteration grouping operation phase includes:

[0029] Extract frequently occurring feature vectors from the current iterative grouping operation phase as the key feature set;

[0030] Retrieve reference feature vectors for the same kidney disease state category from the historical stored dataset;

[0031] Calculate the difference measure between the key feature set and the reference feature vector;

[0032] Based on the difference metric and the cluster centroid stability factor, adjustment parameters are generated for the update step.

[0033] Preferably, the step of updating the grouping rules of the feature clustering algorithm according to the adjusted parameters and synchronizing the updated grouping rules to the kidney disease status data repository includes:

[0034] Analyze and adjust the cluster radius correction factor and weight allocation values ​​in the parameters;

[0035] Modify the distance calculation rules of the feature clustering algorithm to incorporate a clustering radius correction factor;

[0036] Adjust the weighting values ​​in the importance score of the feature vectors and reconstruct the grouping rules;

[0037] Write the restructured grouping rules into the kidney disease status data repository, overriding the old version of the rules.

[0038] Preferably, the method further includes:

[0039] Read the latest set of kidney disease status categories from the kidney disease status data repository;

[0040] Receive newly collected medical image data from patients with kidney disease;

[0041] The updated feature clustering algorithm is applied to process newly acquired medical image data of kidney disease patients to generate new status category labels;

[0042] Compare the new status category labels with the stored historical status categories to identify abnormal status data points;

[0043] If the number of abnormal data points exceeds a preset threshold, the data verification process is triggered.

[0044] Preferably, the updated feature clustering algorithm is used to process newly acquired medical image data of kidney disease patients to generate new status category labels, including:

[0045] The newly collected medical image data of kidney disease patients are input into the image feature extraction algorithm to generate new feature vectors;

[0046] Calculate the membership degree between the new feature vector and the cluster center using the updated grouping rules;

[0047] Assign new state category labels based on the maximum membership degree;

[0048] The new status category label is transmitted to the abnormal status detection module.

[0049] Preferably, the step of comparing the new state category label with the stored historical state categories to identify abnormal state data points includes:

[0050] Obtain the boundary values ​​of the distribution range of historical state categories;

[0051] Calculate the degree of deviation between the new state category label and the distribution range boundary value;

[0052] If the deviation exceeds the tolerance limit, it is marked as an abnormal data point;

[0053] Aggregate abnormal state data points into an abnormal dataset and output it to the repository update module.

[0054] Preferably, the method further includes:

[0055] Construct a data storage system for kidney disease status, including an image acquisition interface and a data processing engine;

[0056] The image acquisition interface receives a set of continuous time-series medical image data transmitted by the medical imaging equipment;

[0057] The data processing engine executes image feature extraction algorithms and feature clustering algorithms;

[0058] After each iteration of the grouping operation, the data processing engine writes the set of kidney disease status categories to the distributed storage nodes;

[0059] Distributed storage nodes adjust the data index structure based on the updated grouping rules.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] This image data processing-based method for storing the status data of kidney patients brings many positive impacts to the storage of status data of kidney patients through in-depth processing of continuous time-series medical image data.

[0062] Acquiring a continuous time-series medical image dataset can fully preserve the temporal characteristics of disease progression in patients with kidney disease. The progression of kidney disease is often a dynamic process, and there is an inherent correlation between medical images at different time points. This correlation is crucial for accurately reflecting the evolution of the disease. Traditional storage methods often simply arrange image data chronologically, making it difficult to reflect this temporal correlation. This method, however, integrates continuous sequences, allowing the stored underlying data to naturally carry information about the trajectory of disease development.

[0063] The application of image feature extraction algorithms transforms raw medical images into a set of multi-dimensional feature vectors, a process that achieves the structuring and quantification of image information. Medical images often contain rich visual information, such as grayscale distribution, texture features, and morphological structures. If this information is stored in the form of raw images, it is difficult to effectively utilize it in subsequent grouping operations. Multi-dimensional feature vectors can integrate this scattered information into a computable and comparable numerical form, enabling the feature differences between different images to be clearly quantified, providing a solid foundation for subsequent clustering and grouping.

[0064] The iterative grouping operation of the feature clustering algorithm, combined with the identification of key feature sets, enables the division of kidney disease state categories to better reflect the actual characteristics of the disease. Each iteration focuses on key features, avoiding interference from irrelevant features and ensuring that the core basis for category division always revolves around the essential differences in kidney disease states. Simultaneously, the iterative approach allows for continuous optimization of the grouping process. As data accumulates and key features are continuously identified, the boundaries of state categories become clearer, and the consistency within each category gradually improves, thus avoiding the rigidity of category division inherent in traditional fixed grouping methods.

[0065] By calculating and adjusting parameters based on historical stored data sets, and updating grouping rules accordingly, the feature clustering algorithm gains the ability to adapt to dynamic data changes. The status data of kidney disease patients constantly changes with disease progression, treatment interventions, and other factors, and the patterns contained in historical data also evolve accordingly. By using historical data as the basis for parameter adjustment, the grouping rules can remain synchronized with the data's changing trends, ensuring that the classification of status categories remains highly reasonable at different stages and under different data characteristics, avoiding grouping bias caused by rigid rules.

[0066] The design of synchronizing grouping rules to the renal disease status data repository enables real-time linkage between the storage system and the grouping logic. Data in the repository is no longer isolated fragments of information, but organized based on unified and dynamically updated rules. This allows data retrieval, access, and analysis to be performed based on consistent classification standards, improving the availability and consistency of stored data. Whether healthcare professionals are accessing data for specific status categories or performing statistical analysis on the overall data, this can be efficiently accomplished within the structured storage system, reducing the complexity of data processing. Attached Figure Description

[0067] Figure 1 This is a schematic diagram illustrating the working principle of the image data processing-based method for storing the status of kidney disease patients as described in this invention.

[0068] Figure 2 A flowchart for obtaining a continuous time-series medical image dataset of patients with kidney disease;

[0069] Figure 3 A flowchart illustrating the application of image feature extraction algorithms to process continuous time-series medical image datasets;

[0070] Figure 4 A flowchart illustrating the process of identifying key feature sets and calculating adjustment parameters during each iteration of the grouping operation;

[0071] Figure 5 This is a schematic diagram of medical image acquisition.

[0072] Figure 6 This is a comparison chart showing the effects of image enhancement processing. Detailed Implementation

[0073] 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 embodiments of the present invention, and not all embodiments. 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.

[0074] Please see Figure 1 This invention provides a method for storing the status data of kidney disease patients based on image data processing, the method comprising:

[0075] A collection of continuous medical images of the kidney region in patients with kidney disease, arranged in chronological order, is obtained through the interface of a medical imaging device.

[0076] The image dataset is processed using a pre-built image feature extraction algorithm. The algorithm performs image segmentation, spatial domain feature extraction, and time series dynamic analysis, and then fuses these to generate a multi-dimensional feature vector set that characterizes the structural state and temporal changes of the kidney.

[0077] A predefined feature clustering algorithm is used to iteratively group the multi-dimensional feature vector set. During each iteration, the system identifies the key feature element combination (key feature set) that plays a decisive role in the grouping and queries the benchmark information of the same type of state stored in the historical data set. Based on the comparison results of the difference between the key feature set and the historical benchmark, and combined with the convergence stability state of the current cluster center, a set of adjustment parameter values ​​for fine-tuning the clustering algorithm is calculated.

[0078] The grouping decision logic inside the feature clustering algorithm is updated in real time using the adjusted parameter values, including but not limited to the parameters in the distance calculation formula, feature dimension weights, or cluster boundary thresholds. The updated grouping rules are then written completely and in real time into the rule storage area of ​​the system's core kidney disease status data repository, replacing the old rules.

[0079] Through multiple iterations and updates until convergence, a stable set of grouping results representing different stages of kidney disease development is formed, namely a set of kidney disease status categories. This set of categories and its corresponding latest grouping rules are persistently stored in the repository.

[0080] Example 1: See Figure 2 The acquisition of continuous time-series medical image data of the kidney status of patients with kidney disease begins with the image acquisition operation of medical imaging equipment. Medical imaging equipment includes computed tomography (CT) scanners, magnetic resonance imaging (MRI) scanners, or ultrasound scanners. These devices perform image acquisition tasks on the kidney regions of the target patients according to preset time periods and examination specifications. The time period is dynamically adjusted based on clinical monitoring needs, commonly set to once daily, multiple times weekly, or once every two weeks. After the device's acquisition action is triggered, it generates one or more raw scan images corresponding to a single examination of the kidney region. The raw images generated from multiple consecutive examinations are naturally arranged in chronological order of acquisition to form an initial image sequence. This sequence is the raw data directly output by the device and usually contains inherent technical limitations.

[0081] Raw scan images often suffer from insufficient contrast. Imaging equipment may produce images with uneven brightness under different operating conditions or environments. Voluntary or involuntary movements of the patient during acquisition can lead to image blurring artifacts. Internal electronic noise in the equipment manifests as randomly distributed bright and dark spots in the image. Electromagnetic interference from the surrounding environment or cross-use of equipment by other patients can introduce specific patterns of image artifacts. These defects directly affect the visual recognition of renal anatomy and lesion areas in the images and must be improved through image enhancement processing.

[0082] Image enhancement processing is performed frame-by-frame on the initial image sequence, comprising two interconnected but complementary processing units. The first unit performs contrast adjustment. This unit calculates the global distribution of image grayscale. For images with a narrow overall grayscale range, linear or nonlinear transformations are used to stretch the grayscale interval, widening the difference between dark and bright areas. For images with uneven grayscale distribution, a local window-based histogram equalization algorithm is applied. This algorithm divides the image into several small blocks, independently calculates the grayscale histogram within each small window, and remaps the grayscale values, making the grayscale distribution more uniform in each small region, enhancing local detail contrast, and using a constraint function to prevent noise amplification caused by excessive enhancement of local areas. The second unit performs noise filtering. Noise is prevalent and diverse in form in medical images. A nonlocal mean filtering algorithm is used to suppress random point noise. This algorithm searches for structurally similar regions in the image and replaces the current pixel value with the weighted average of these regions, smoothing noise while preserving edge integrity. For images containing stair-step or patchy artifacts, anisotropic diffusion filtering is applied. This filter calculates the gradient intensity of the image, performing more smoothing in low-texture areas with low gradient intensity and less smoothing or suppressing smoothing in edge areas with high gradient intensity. The images processed in these two steps have a clearer visual structure, more defined contours, and a cleaner background, forming a preprocessed image set.

[0083] While the preprocessed image set has been optimized for quality, the temporal dimension is not yet standardized. Different inspection devices record timestamps in different locations and formats. The timestamp extraction module parses the metadata information fields of the preprocessed image set. It reads the acquisition time recording field from proprietary data tags defined by the device manufacturer or international standard tags. It identifies and converts the binary time encoding from the complex file header structure into a readable date and time format. The extracted timestamps are accurate to the millisecond or second level, depending on the device's recording precision. This process ensures that each image is bound to the exact time it was generated.

[0084] Image sets from multiple consecutive examinations require alignment along the timeline. Different examinations may experience time drift between planned acquisition points due to equipment resource constraints. Patients frequently fail to arrive at the scheduled time. The timestamp alignment module reorganizes the image order based on extracted precise time points. For continuous monitoring scenarios, images are arranged into a linear time sequence strictly according to their timestamps. For non-continuous examination scenarios, interpolation techniques are used to handle missing time points. Virtual timestamps are generated between two known valid examination time points. The kidney state corresponding to the virtual time points is estimated based on a motion model, and interpolated images are generated using adjacent actual images. When the same patient is examined on different devices, spatial differences exist between the coordinate systems of the different devices. A location matching algorithm based on reference point calibration is used to address this issue. Imageable positioning markers are set on the patient's body surface. During acquisition, the spatial coordinates of these markers in images from different devices are identified, and spatial transformation parameters between coordinate systems are calculated. Affine or elastic transformations are applied to spatially resample all images, ensuring all examination images are in the same anatomical standard spatial coordinate system. The spatially aligned image set retains the original timestamp information, forming a strictly spatiotemporally aligned continuous time-series medical image data set.

[0085] This dataset is organized using a specific data structure. Each time point corresponds to a set of image slices or multimodal image fusion results. The dataset is stored using a tree-indexed structure, with top-level nodes indexed by patient unique identifiers, second-level nodes indexed by time, and bottom-level nodes storing single-frame image data and its metadata. This dataset serves as the foundational input data source for subsequent image feature extraction algorithms. Its completeness, spatiotemporal consistency, and quality stability directly determine the effectiveness of subsequent analysis processes. The entire acquisition and preprocessing process is automated, reducing manual intervention and making data preparation efficient and reliable. A stable communication protocol is established between the data processing logic module and the medical imaging equipment interface, supporting data access capabilities from multiple vendors. The operation log records the execution status and abnormal alarm information for each step, providing a trajectory basis for operation auditing.

[0086] See Figure 5 This image presents a comprehensive view of the original medical scan image, the enhanced image, the processing differences, and the time-series grayscale changes. The top left corner shows the original kidney scan image, containing typical noise and artifact features; the top right corner shows the enhanced image after contrast adjustment and noise filtering, making the kidney structure more clearly visible; the heatmap in the bottom left corner shows the distribution of differences before and after image processing, with red areas indicating areas of significant change; the curve in the bottom right corner shows the changes in the average grayscale value of the images acquired at different time points. The original image (circled markers) shows larger fluctuations, while the enhanced image (square markers) shows more stable changes, indicating that the image enhancement processing effectively reduced the impact of differences in acquisition conditions.

[0087] See Figure 6 This image compares the original and enhanced images at four consecutive time points. The top row shows the original scan images, revealing significant noise interference, insufficient contrast, and artifacts. The bottom row shows the results after histogram equalization and median filtering, resulting in a significant improvement in image quality. Specifically, the kidney edges are clearer, the internal tissue structure details are richer, and noise interference is greatly reduced. Particularly noteworthy is the increased prominence of lesion areas (highlighted spots in the images) in the enhanced images, which is crucial for subsequent feature extraction and disease analysis.

[0088] Example 2: See Figure 3 The algorithm processes continuous time-series medical image datasets as input image feature extraction data. It first processes a single-time-point medical image, which typically contains complex anatomical structures and potential lesion information. To effectively capture local details and reduce overall computational load, the image is uniformly divided into multiple rectangular image blocks. The size of the image blocks is dynamically configured based on the average size of the kidney in the image and the target region of interest, employing either overlapping or non-overlapping block strategies. This block division allows the analysis to focus on the structural characteristics of local kidney regions.

[0089] Each resulting image block is processed by a spatial domain feature extractor. Spatial domain analysis focuses on the spatial arrangement and grayscale variation patterns of pixels within the block. Texture feature analysis calculates the relationships between the grayscale values ​​of pixels within the block. By constructing a grayscale co-occurrence matrix, the joint probability distribution of paired pixel grayscale values ​​under specific spatial relationships is statistically analyzed. Various metrics are then quantified and extracted from this matrix, one core indicator being the texture contrast of the image block region, calculated as follows:

[0090]

[0091] In this formula, the symbol... and These represent the row and column indices in the gray-level co-occurrence matrix, respectively, i.e., different gray levels. This indicates that, under the condition of satisfying specific spatial relationships within an image patch, a pixel's gray level is... And the gray level of another related pixel is The joint probability value is calculated. The texture contrast value reflects the intensity of local gray-level changes within the image patch; a larger value indicates a coarser texture and more pronounced edges. Furthermore, the energy value calculates the sum of squares of the gray-level distribution uniformity, the entropy value measures the randomness or information content of the texture pattern, and the correlation value quantifies the degree of linear dependence of gray levels. These statistical features collectively constitute a numerical description of the texture regularity. Simultaneously, a set of Gabor filters is applied to the image patch. Gabor filters simulate the receptive field of biological vision, possessing specific spatial frequency and directional selectivity. The image patch is convolved with Gabor filter kernels of different center frequencies and directions to obtain the corresponding complex response images. The energy or amplitude mean of each filtered response image is calculated to form a multi-scale, multi-directional texture spectral feature vector. Another set of analyses focuses on the edge information of the image patch region. The edge detection operator scans the pixel positions with significant gray-level changes within the image patch, connecting these points to form an edge contour map. The spatial distribution density of all edge points in the contour map is analyzed, i.e., the number of edge points per unit area. The angular distribution of all edge points is statistically analyzed to generate an orientation histogram, reflecting the main orientation characteristics of the boundary. The total length distribution and average length of the extracted edge contour segments are measured. Combining the texture and edge feature values, a set of primary descriptors for the image patch region is formed. At a given time, the primary descriptors of all image patch regions are arranged according to their position coordinates in the original image, forming a two-dimensional matrix structure. This matrix is ​​the primary feature matrix corresponding to that time. The rows of this matrix represent different image patch regions, and the columns represent different types of primary feature values ​​extracted within each patch.

[0092] The module combines time series analysis to process primary feature matrix sequences across multiple consecutive time points. It focuses on the behavioral pattern changes of each specific image patch region along the time axis. For each image patch region and its corresponding set of feature values ​​(such as texture contrast), it tracks their numerical trajectory at consecutive acquisition time points. Moving averages are used to analyze the short-term fluctuations and long-term trends of the feature values. The slope of the feature time series is calculated to describe the overall direction and intensity of feature value increases and decreases. The difference in feature values ​​between adjacent time points is examined, analyzing periodic or abrupt changes in these differences. The autocorrelation coefficient or cross-correlation coefficient of the feature sequence is calculated to discover temporal dependencies between feature values ​​or synchronous relationships between feature changes in different regions. Similarity measures are applied to compare whether the feature evolution patterns of different image patch regions converge or diverge. Based on the above analysis, a series of quantitative indicators are generated for the image patch region, such as the short-term deviation of a feature value from its long-term value, the estimated slope of the feature value's upward or downward trend, the average value of changes at adjacent time points, the coefficient of variation of the feature sequence, and a trend stability score. These indicators are collectively referred to as the feature change trend value of the image patch region. These trend values ​​characterize the dynamic behavior of the structural properties of the corresponding spatial locations as they evolve over time.

[0093] The fusion process integrates spatial and temporal features. The primary feature matrix output by the spatial feature extractor contains a static structural description (texture characteristics, edge distribution) of each image patch region at a single time point. The set of feature change trend values ​​output by the time series analysis module reflects the dynamic evolution of the corresponding image patch region's features over a period of time (trend direction, change intensity, fluctuation pattern). The fusion operation combines the static spatial feature vectors from the same image patch region with their dynamic temporal trend values ​​to form a higher-dimensional descriptive vector. The fusion strategy uses feature vector concatenation, directly merging a set of values ​​reflecting spatial structural features with associated values ​​reflecting temporal change features to form a long vector. Fusion is performed synchronously for all image patch regions at each time step. The fusion process outputs a complete set of vectors, i.e., a multi-dimensional feature vector set. Each feature vector in this set uniquely corresponds to a specific image patch region at a certain time point, and the vector dimension comprehensively expresses the spatial structural attributes of the region and its change characteristics over a continuous period of time. This set is then fed into a subsequent feature clustering algorithm for group analysis.

[0094] Feature clustering algorithms take a set of multi-dimensional feature vectors as input and aim to group and classify them based on the similarity of the vectors in the feature space. In the initial stage, the positions of the initial set of cluster centroids are set or automatically calculated. The number of cluster centroids is based on a pre-set expected number of categories based on knowledge of the physiological and pathological state of kidney disease. The core of the algorithm is the iterative optimization process. In each iteration, the similarity distance between each multi-dimensional feature vector in the set and all current cluster centroids is calculated one by one. The distance function can use Euclidean distance to measure the absolute numerical difference between vectors, Manhattan distance to calculate the sum of the absolute values ​​of dimensional differences, or cosine similarity to measure the consistency of vector direction. Based on the calculated distance values, each feature vector is classified into the cluster represented by the cluster centroid with the smallest distance. After all feature vectors have been classified and assigned, the positions of the cluster centroids are updated. The geometric centroid positions of all feature vectors contained in each newly formed cluster are recalculated, usually taking their dimensional mean as the new cluster centroid coordinates. The steps of assigning vectors to the nearest center and recalculating the centroid positions are repeated. The algorithm continuously monitors the convergence of the iteration process, tracking the movement of all cluster center points between iterations. It calculates the average displacement distance of all cluster center points and sets a small positive threshold as a stability standard. After an iteration, if the calculated average displacement distance of the cluster center points is less than this stability threshold, the clustering process is considered to have converged, and the grouping results tend to stabilize. Upon convergence, the final grouping results are output as a set of kidney disease state categories. Each category in this set consists of a group of data points with highly similar feature vectors, and each category is assigned a unique identifier. Category identifiers, cluster center point coordinates, and boundary rules are persistently stored in a database as a basis for identifying and classifying future new data points. The iterative logic and distance function rules of the feature clustering algorithm can be adjusted and updated in each iteration based on the output of other modules to optimize the grouping effect.

[0095] Example 3: See Figure 4In each iteration of the feature clustering algorithm's grouping operation, the key feature identification and parameter adjustment calculation process is executed synchronously. This process is embedded within the main loop framework of the clustering iteration, dynamically optimizing its rules without interrupting the grouping process. After an operation of assigning vectors to the nearest cluster centroid and recalculating the centroid position is completed, the system analyzes the temporary grouping structure formed in this iteration. Within each temporarily formed cluster, the distribution of feature dimension values ​​of all member feature vectors is statistically analyzed. Feature components that appear with significantly higher frequencies than other dimensions within the cluster, or subsets of dimensions with narrow and stable numerical distribution ranges, are identified. Simultaneously, feature vectors near the cluster boundaries are analyzed to determine which small changes in dimension values ​​lead to a change in their cluster affiliation. Combining frequency statistics and boundary influence analysis, combinations of feature elements that dominate the current grouping results are selected; these elements constitute the key feature set for this iteration stage. This key feature set represents the core information distinguishing different state categories under the current data distribution state.

[0096] The system then accesses the kidney disease status data repository. Based on the provisional kidney disease status category labels generated in the current iteration (e.g., provisional category numbers identified by cluster centroids), a query is performed in the historical record area of ​​the repository. The query target is located in a previously stored set of historical feature vectors that have the same or highly similar status category labels. A representative reference feature vector is extracted from this historical set. The selection of the representative vector can be based on the centroid vectors of the historical data or on typical pattern vectors that frequently appear in the historical distribution. This reference feature vector is considered the baseline pattern for that status category under historical experience.

[0097] The identified key feature set and the retrieved reference feature vectors enter the difference calculation stage. This stage focuses on a specific dimensional subset contained in the key feature set. Let the key feature set be represented as a vector in the current iteration, denoted as . ,in It is the number of key feature dimensions. Representing the The values ​​of each key dimension (which may be the mean of that dimension within the current cluster or other statistics). Let the projection of the corresponding reference feature vector retrieved from historical data onto the same key dimension be... Calculate the difference measure between the two. One calculation method uses the mean absolute deviation:

[0098]

[0099] In this formula, the symbol This represents the total number of feature dimensions contained in the key feature set. (Symbol) The key feature set represents the first Numerical representations of key dimensions (e.g., the mean of that dimension across all vectors in the current cluster). Symbols The historical reference feature vector is in the th Values ​​on the same key dimension. Symbols Indicates the calculation of the first In each dimension and The absolute value of the difference. (Symbol) This indicates that from the first dimension to the second dimension... The summation operation is performed on the absolute differences across all dimensions of the dimension. (Symbol) This means dividing the sum by the total number of dimensions. The mean absolute deviation value is obtained. This value It quantifies the average deviation of the current key feature pattern from the historical baseline pattern on key dimensions. The larger the value, the greater the difference between the current grouping pattern and historical experience.

[0100] Difference measure The calculation results need to be comprehensively evaluated in conjunction with the current convergence state of the clustering algorithm. The system monitors the movement distance of the cluster center points relative to the previous iteration in this iteration. The average displacement distance of all cluster center points is calculated. Define a stability factor. This factor and Negative correlation. For example, It can be defined as ,in It is a positive scaling factor. When the displacement of the center point is large ( (Large), indicating that the clustering is still undergoing drastic adjustments, stability factor The value is small; when the displacement is very small ( (Approaching zero) indicates that the clustering tends to stabilize. The value is close to 1. Adjust the parameters. The generation comprehensively considers the difference measure and stability factor One generation method is a linear combination: ,in It is the preset weighting coefficient (0 < <1). This formula assigns different weights to the difference measure and the centroid instability. Adjust the parameters. It is a scalar value whose magnitude indicates the urgency and direction in which the current grouping rules need to be adjusted. High A higher value usually indicates a greater need for rule adjustments. The parameters are adjusted based on the specific algorithm requirements. It may be directly mapped to specific operational parameters, such as cluster radius correction factors. (For example , (It is the initial radius), or used to calculate the feature dimension weight adjustment vector. (For example, increasing the weight ratio of key dimensions).

[0101] Based on the calculated adjustment parameters, the feature clustering algorithm's grouping rules are updated. First, the specific content of the adjustment parameters is parsed. If the parameters include a cluster radius correction factor... If so, modify the rules related to the distance threshold in the algorithm. For density-based clustering algorithms, the neighborhood radius parameter is updated by adding or subtracting the original value. For center-based algorithms, this factor may be incorporated into the distance calculation formula, for example, by modifying the Euclidean distance to... Alternatively, it can be used to dynamically adjust the distance scale. If the parameters include feature dimension weight assignment values... ( If the total number of feature dimensions is 0, then the distance calculation function is reconstructed. A weighted distance metric is used, such as weighted Euclidean distance: distance Weight The first one was directly enlarged or reduced. The contribution of each feature dimension to distance calculation. Higher weight dimensions have a stronger impact on grouping decisions, while lower weight dimensions have a weaker impact. Weight allocation is typically based on key feature identification results and historical difference analysis to strengthen the discriminative power of key dimensions. The reconstructed grouping rules include an updated distance calculation function, an updated neighborhood radius threshold, or an updated weight vector.

[0102] The reconstructed grouping rules take effect immediately and are used for feature vector allocation calculations in the next iteration. Simultaneously, the updated rules are written to the rule storage area of ​​the nephropathy status data repository. The synchronization operation includes serializing the data structure of the new rules, establishing an association identifier with the current algorithm version and iteration number, executing a database write transaction, and overwriting previously stored older versions of the rules in the repository. This overwrite operation ensures that the latest and most optimized grouping rule definitions are always retained in the nephropathy status data repository. Updating the repository provides a basis for processing new data points and allows the grouping logic of the entire system to continuously evolve with the accumulation of data processing experience. Rule update logs are recorded, including update time, iteration number, adjusted parameter values, and rule change summaries, for system status tracking and audit analysis. The entire identification, calculation, update, and synchronization process is executed automatically in each iteration, constituting a dynamic optimization mechanism for the clustering grouping process.

[0103] Example 4: This example describes how the system continuously processes newly acquired medical image data of kidney disease patients and identifies potential abnormal states based on historical clustering rules. Assume there is a chronic kidney disease patient, Mr. Zhang, who has been monitored by the system for a long period. His historical medical image data has been processed and formed into a set of kidney disease state categories and their rules stored in a kidney disease state data repository.

[0104] Example Scenario: Mr. Zhang has undergone regular renal MRI examinations over the past six months. Based on historical data clustering, the system defines three stable state categories: State_A represents the stable compensatory stage of renal function, characterized by uniform renal cortical texture, normal cortical thickness, and a gradual trend of change over time; State_B represents early mild damage, characterized by mild localized texture coarsening, slight cortical thinning, and a slow, progressive thinning trend; State_C represents the moderately progressive stage, characterized by significant and diffuse texture coarsening, a marked decrease in cortical thickness, and a continuously accelerating thinning trend. Mr. Zhang's three most recent examinations were classified as State_B, and the historical distribution boundary values ​​of his relevant feature vectors have been calculated and stored. These boundary values ​​include: a minimum permissible cortical thickness of 4.2 mm (the minimum observed value in this state over the past three months), an upper limit for the thinning trend slope of -0.15 mm / month (the maximum observed rate of change in the past), and a historical maximum Euclidean distance of 2.8 units from the feature vector of this category to the cluster center. The system sets tolerance limits for deviations, such as a cortical thickness deviation limit of 1.0 mm, a trend slope deviation limit of 0.1 mm / month, and a distance deviation limit of 3.5 units.

[0105] New Data Input and Processing: On the seventh follow-up examination day of Mr. Zhang, the system received his latest renal MRI scan data through the medical imaging equipment interface. The newly acquired medical image data (set as time point T7) was immediately input into the updated processing pipeline. The image feature extraction algorithm worked first: preprocessing (contrast enhancement and noise reduction) the T7 single-frame MRI image, segmenting it into the same image block grid structure as the previous processed images. The spatial domain feature extractor analyzed each block: calculating the gray-level co-occurrence matrix of the cortical region to obtain texture contrast and energy values; quantifying the sharpness of the cortical-medullic boundary through edge detection; and specifically measuring the cortical thickness values ​​at several key locations. The time series analysis module was launched: this module called the cortical thickness sequence of the same spatial block region in Mr. Zhang's three previous examinations (T4, T5, and T6 were all marked as State_B). The time series from T4 to T7 was analyzed, the instantaneous slope of the cortical thickness change was calculated (the thickness difference between the two most recent time points divided by the time interval), and the difference between this slope and its 12-week moving average was compared. The fusion operation produces a multidimensional feature vector, such as a vector containing: [average cortical thickness value of image patch 1, texture contrast of image patch 1, edge sharpness of image patch 1, instantaneous change slope of cortical thickness of image patch 1, magnitude of thickness deviation of image patch 1 from moving average, ..., corresponding feature value of image patch N], which together form a high-dimensional vector V7.

[0106] Feature Classification and Label Assignment: The multidimensional feature vector V7 is fed into the feature clustering algorithm module, which applies the latest grouping rules. The algorithm loads the current version of the grouping rules and the coordinates of the cluster centroids (corresponding to the centroids C_A, C_B, and C_C of State_A, State_B, and State_C) from the kidney disease state data repository. Based on the distance calculation method defined in the rules (e.g., considering the weighted Euclidean distance of each feature dimension), the system calculates the distances between V7 and C_A, C_B, and C_C. Assume the calculation results are: Dist(V7,C_A)=8.5, Dist(V7,C_B)=3.2, Dist(V7,C_C)=4.8. Based on the "minimum distance principle," the algorithm assigns V7 to the nearest State_B category and generates a new state category label, State_B, which is assigned to the T7 data point. This label, along with the complete feature vector V7 and the calculated distance values, is transmitted to the abnormal state detection module.

[0107] Abnormal State Identification Process: Upon receiving a new data point (V7, T7, label State_B), the abnormal state detection module immediately retrieves reference distribution data for that state category from the historical data repository based on its assigned category label State_B. This includes the aforementioned preset boundary values: minimum cortical thickness boundary = 4.2 mm, upper limit of thinning trend slope boundary = -0.15 mm / month, historical maximum distance boundary = 2.8 units; and tolerance limits: thickness deviation limit = 1.0 mm, slope deviation limit = 0.1 mm / month, distance deviation limit = 3.5 units. The module calculates three key deviation indicators for V7:

[0108] Core feature value deviation: Extract the feature component value representing the average cortical thickness from V7, for example, 3.8mm. Calculate its absolute deviation from the historical minimum thickness boundary of State_B (4.2mm): |3.8-4.2|=0.4mm. Then compare whether this deviation exceeds the thickness deviation limit of 1.0mm (i.e., determine 0.4mm<1.0mm).

[0109] Time trend deviation: Extract the feature component value representing the slope of instantaneous change in cortical thickness from V7, for example, -0.3 mm / month. Calculate its absolute deviation from the upper limit boundary of the historical slope in State_B (-0.15 mm / month): |-0.3-(-0.15)|=0.15 mm / month. Compare whether this deviation exceeds the slope deviation limit of 0.1 mm / month (i.e., determine if 0.15 mm / month > 0.1 mm / month).

[0110] Overall spatial deviation: The distance from V7 to its class center C_B is 3.2 units. Calculate its deviation from the historical maximum distance boundary of State_B (2.8 units): 3.2 - 2.8 = 0.4 units. Compare whether this deviation exceeds the distance deviation limit of 3.5 units (i.e., determine if 0.4 units < 3.5 units). Note that the historical maximum distance boundary defines the "radius" of the normal distribution of this class. The current distance of 3.2 units has exceeded the historical boundary of 2.8 units, but a simple difference is used in the deviation calculation, and the overall spatial distribution is considered abnormal only when the difference exceeds the limit.

[0111] Anomaly Identification and Aggregation: The anomaly detection module comprehensively evaluates the limit comparison results of three deviation indicators.

[0112] The deviation of the core feature value (0.4mm) is less than the thickness deviation limit (1.0mm), and this dimension does not exceed the standard.

[0113] The time trend deviation (0.15 mm / month) exceeds the slope deviation limit (0.1 mm / month), indicating that this dimension exceeds the standard.

[0114] The overall spatial deviation (0.4 units) is less than the distance deviation limit (3.5 units), so this dimension is within the acceptable range.

[0115] Because the time trend deviation index exceeded the tolerance limit, the system determined that although the new data point T7 was classified into the State_B category by the clustering algorithm, its displayed rate of cortical thinning (-0.3 mm / month) was much faster than the highest rate in the historical records of this category (-0.15 mm / month), and the deviation of 0.15 mm / month was greater than the tolerable limit of 0.1 mm / month. Therefore, the module marked the T7 data point as an "abnormal state data point". The information of this point (including the original image, the extracted feature vector V7, the assigned label State_B, the calculated feature values, the identified abnormal index and its deviation) was recorded and added to an accumulating abnormal dataset. If the system also found similar out-of-limit points in the same patient's most recent examinations (e.g., T5 or T6), or detected out-of-limit points in the same category of data from other patients, the abnormal dataset will contain multiple data points. The system continuously monitors the number of data points in the abnormal dataset. Once the number reaches a preset trigger threshold (e.g., two consecutive abnormalities in the same patient or three out of ten patients showing the same category of abnormality), the system automatically triggers a preset data verification process. The verification process may include: automatically sending an alert to the attending physician indicating that Zhang's renal function indicators are deteriorating rapidly and require retesting; automatically queuing the data for secondary confirmation by a higher-level image analysis algorithm; or notifying the data management module to prepare to label the batch of data points for possible clustering rule updates or new category discovery. All abnormal data points and their metadata are output to a dedicated area in the renal disease status data repository for long-term storage and associated with the corresponding patient records and examination time points. The system continues to run, awaiting processing the next newly collected data.

[0116] Example 5: The physical architecture of the kidney disease status data storage system includes a dedicated server cluster and network infrastructure. Medical imaging equipment, such as CT scanners or ultrasound diagnostic instruments in hospital radiology departments, transmits the acquired raw image data to an image acquisition interface deployed in the data center via standard medical image communication protocols. This interface is compatible with multiple vendors' devices, with an embedded protocol parser recognizing formats such as DICOM and HL7, and automatically extracting necessary patient identification, examination serial numbers, and device type identification information. The interface server is configured with dual network isolation: the front end is configured with dedicated firewall rules for the medical device network segment, allowing only encrypted communication; the back end connects to the data processing module via an internal high-speed network. When network latency or data packet loss occurs during device transmission, the interface automatically activates a data buffering mechanism to temporarily store the received fragments and request retransmission. After preliminary metadata verification, the complete received image sequence is converted into an internally unified data format, and a precise reception timestamp and processing pipeline identifier are appended to form a continuous time-series medical image data set. This set is published to a message queue service, awaiting consumption by downstream data processing engines.

[0117] The data processing engine is deployed on a containerized computing platform with horizontal scalability. It mainly consists of a feature extraction unit, a clustering analysis unit, and an anomaly detection unit. The feature extraction unit subscribes to image datasets in a message queue. During unit initialization, a predefined image segmentation strategy configuration file is loaded, and grid division rules are set based on the kidney's anatomical structure. A processing thread pool executes spatial domain feature calculation tasks in parallel: each thread acquires a single image patch, applies the gray-level co-occurrence matrix algorithm to calculate 14 types of texture statistics, calls the Sobel operator for convolution to calculate edge intensity histograms, and combines morphological processing to identify cortical regions and measure thickness pixel values. The time-series analysis module runs independently, querying the unit's internal cache to obtain feature values ​​from three consecutive past examinations of the same spatial location and calculating short-term trend indicators. The feature fusion component connects spatial feature values ​​and temporal trend values ​​according to a preset dimensional mapping table, serializing them into binary feature vector objects and attaching timestamps and spatial coordinate metadata. After the multi-dimensional feature vector set is generated, it is pushed to a distributed event stream and marked as ready for the clustering task.

[0118] The clustering analysis unit listens to the feature vector event stream. Upon startup or rule update, the unit loads the current version of the cluster centroid set and associated grouping rule configuration file from the kidney disease state data repository. The rule file defines the specific form of the distance calculation formula (such as weighted Euclidean distance or cosine distance), the initial weight vectors for each feature dimension, the neighborhood search radius threshold, and other parameters. The iterative grouping process is completed within an in-memory computing framework: in the initialization phase, feature vectors are loaded in batches into the distributed dataset; a Map operation is performed to calculate the distance from each vector to all centroids; in the Reduce phase, vectors are assigned according to the nearest neighbor principle; the driver module recalculates the geometric center coordinates of the clusters based on the assignment results. After each iteration, the engine calls the key feature recognition submodule to scan the temporary grouping results and count the indexes of frequently occurring feature dimensions. This index list and the current iteration number are written to the state storage. The unit retrieves reference centroid vectors with the same state label from the historical storage dataset through the database connector and calculates the difference metric. Based on the difference metric and centroid displacement, adjustment parameters are generated, and the radius threshold and feature weight vector in the distance formula are updated in real time before the next iteration.

[0119] The anomaly monitoring unit runs as an independent microservice instance. The unit continuously receives two types of data input: first, snapshot data of the latest kidney disease status category set periodically pushed by the kidney disease status data repository (including coordinates of various centroids, historical feature distribution boundary values, and tolerance limit definitions); second, real-time data streams output by the feature extraction and classification modules from newly acquired images (including new feature vectors, assigned status labels, and attribution distance values). The unit maintains a dynamic boundary model for categories. When a new status category label is received, it immediately queries the local cache to obtain the reference distribution parameters for that category. Multiple comparison logics are executed in parallel: numerical feature dimensions are checked to see if they exceed the historical value range; trend features are calculated to determine the difference between the current value and the historical maximum rate of change; and the overall spatial distance is compared to the deviation value of the historical maximum distance for the same category. The result judgment module applies a preset logical rule tree: if any feature dimension exceeds the limit or the overall deviation score exceeds the threshold, it is marked as an anomaly. All anomalous data points are associated with the original image ID, processing time, and anomaly dimension list, compressed, and batch-written to the anomaly table partition of the repository.

[0120] The distributed storage nodes employ a hybrid storage architecture, configured with a physical server cluster to support NoSQL databases and object storage services. The underlying distributed file system provides volume management, and the database nodes include a three-replica Cassandra ring for high availability. The repository is partitioned according to data characteristics: raw image sequences are stored in object storage buckets with time-partitioned indexes; feature vector sets are stored in a wide-table database, with the primary key generated by patient ID + timestamp hash; cluster centroid sets and historical grouping rules use a configuration storage area, managed with a version control mechanism; abnormal data point records are written to a time-series database and automatically sharded by time. The data index structure is automatically reconstructed after rule updates: when a clustering analysis unit completes a grouping rule update, the system publishes a rule change event. The index management service listens for the event and parses the feature weight vectors in the new rules. The service extracts the indexes of dimensions with weight values ​​higher than the average, triggering a feature vector data index reconstruction operation for that patient group. A separate B+ tree index is created in the database for high-weight dimensions to optimize query efficiency under subsequent rules. An inverted index is established in the raw image storage area to associate all derived feature vector records. The storage system's operational status is monitored in real time. Data writes undergo pre-written logs and checksum verification. In the event of node failure, data rebalancing is automatically triggered to maintain the continuous availability of the storage service. The audit module records full lifecycle operation logs, comprehensively tracking all events and time consumption from the start of image equipment transmission to the final storage status. Logs are uniformly aggregated into a centralized analysis platform to support system optimization decisions.

[0121] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0122] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for storing the status data of kidney disease patients based on image data processing, characterized in that, Including the following steps: Acquire a collection of continuous time-series medical image data of patients with kidney disease; The continuous time-series medical image data set is processed using an image feature extraction algorithm to generate a multi-dimensional feature vector set. A feature clustering algorithm is used to perform iterative grouping operations on the multi-dimensional feature vector set to form a set of kidney disease state categories; In each iteration of the grouping operation, key feature sets are identified, and the adjustment parameters of the feature clustering algorithm are calculated based on the historical stored data set. The grouping rules of the feature clustering algorithm are updated according to the adjusted parameters, and the updated grouping rules are synchronized to the kidney disease status data repository. The feature clustering algorithm is used to perform iterative grouping operations on the multi-dimensional feature vector set to form a set of kidney disease state categories, including: Initialize the set of cluster centers for the feature clustering algorithm; In each iteration, the similarity distance between the multi-dimensional feature vector and the cluster center is calculated, and the feature vector is assigned to the nearest cluster center. Update the cluster center positions and generate new cluster groups based on the allocation results; Repeat the iteration until the change in cluster centroids is below a stable threshold, then output the set of kidney disease status categories and store it in the database; The step of identifying key feature sets and calculating adjustment parameters for the feature clustering algorithm based on historical stored data sets during each iteration grouping operation includes: Extract frequently occurring feature vectors from the current iterative grouping operation phase as the key feature set; Retrieve reference feature vectors for the same kidney disease state category from the historical stored dataset; Calculate the difference measure between the key feature set and the reference feature vector; Based on the difference metric and the cluster centroid stability factor, adjustment parameters are generated for the update step; The step of updating the grouping rules of the feature clustering algorithm according to the adjusted parameters and synchronizing the updated grouping rules to the kidney disease status data repository includes: Analyze and adjust the cluster radius correction factor and weight allocation values ​​in the parameters; Modify the distance calculation rules of the feature clustering algorithm to incorporate a clustering radius correction factor; Adjust the weighting values ​​in the importance score of the feature vectors and reconstruct the grouping rules; Write the restructured grouping rules into the kidney disease status data repository, overriding the old version of the rules.

2. The method for storing the status of kidney disease patients based on image data processing according to claim 1, characterized in that, The acquisition of the continuous time-series medical image data set of patients with kidney disease includes: Multiple frames of images of the kidney region of patients with kidney disease are acquired using medical imaging equipment to form an initial image sequence; Image enhancement processing, including contrast adjustment and noise filtering, is performed on the initial image sequence to generate a preprocessed image set; Extract timestamp information from the preprocessed image set, align image frames from different acquisition times, and generate a time-synchronized image sequence; The time-synchronized image sequence is input as a continuous time-series medical image data set into subsequent processing steps.

3. The method for storing the status of kidney disease patients based on image data processing according to claim 1, characterized in that, The applied image feature extraction algorithm processes the continuous time series medical image data set to generate a multi-dimensional feature vector set, including: The continuous time-series medical image data set is divided into multiple image block regions; A spatial domain feature extractor is used to analyze the texture features and edge distribution of each image patch region to generate a primary feature matrix. The initial feature matrix is ​​processed using the time series analysis module to calculate the feature change trend value; The primary feature matrix and feature change trend values ​​are combined to generate a multi-dimensional feature vector set, which is then output to the feature clustering algorithm.

4. The method for storing the status of kidney disease patients based on image data processing according to claim 1, characterized in that, Also includes: Read the latest set of kidney disease status categories from the kidney disease status data repository; Receive newly collected medical image data from patients with kidney disease; The updated feature clustering algorithm is applied to process newly acquired medical image data of kidney disease patients to generate new status category labels; Compare the new status category labels with the stored historical status categories to identify abnormal status data points; If the number of abnormal data points exceeds a preset threshold, the data verification process is triggered.

5. The method for storing the status of kidney disease patients based on image data processing according to claim 4, characterized in that, The updated feature clustering algorithm is used to process newly acquired medical image data of kidney disease patients, generating new status category labels, including: The newly collected medical image data of kidney disease patients are input into the image feature extraction algorithm to generate new feature vectors; Calculate the membership degree between the new feature vector and the cluster center using the updated grouping rules; Assign new state category labels based on the maximum membership degree; The new status category label is transmitted to the abnormal status detection module.

6. The method for storing the status of kidney disease patients based on image data processing according to claim 4, characterized in that, The step of comparing the new state category label with the stored historical state categories to identify abnormal state data points includes: Obtain the boundary values ​​of the distribution range of historical state categories; Calculate the degree of deviation between the new state category label and the distribution range boundary value; If the deviation exceeds the tolerance limit, it is marked as an abnormal data point; Aggregate abnormal state data points into an abnormal dataset and output it to the repository update module.

7. The method for storing the status of kidney disease patients based on image data processing according to claim 1, characterized in that, Also includes: Construct a data storage system for kidney disease status, including an image acquisition interface and a data processing engine; The image acquisition interface receives a set of continuous time-series medical image data transmitted by the medical imaging equipment; The data processing engine executes image feature extraction algorithms and feature clustering algorithms; After each iteration of the grouping operation, the data processing engine writes the set of kidney disease status categories to the distributed storage nodes; Distributed storage nodes adjust the data index structure based on the updated grouping rules.

Images