Oral ulcer image recognition autoimmune disease blood index correlation analysis system
By constructing a system for the correlation analysis of blood indicators in autoimmune diseases based on oral ulcer images, and using the dynamic evolution rate of lesions in the images to extrapolate blood indicators over time, the system solves the problem of asynchronous data between oral ulcer images and blood indicators in existing technologies, and achieves earlier and more accurate risk assessment and early warning of autoimmune diseases.
Patent Information
- Application Number
- CN202511616755.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-06
- Publication Date
- 2026-06-02
- Estimated Expiration
- 2045-11-06
AI Technical Summary
Existing technologies cannot achieve synchronous and dynamic correlation in multimodal time series data analysis of oral ulcer images and blood indicators, resulting in insufficient sensitivity and accuracy of risk assessment, especially in the early screening and disease management of autoimmune diseases, where there is lag and misjudgment.
By constructing a blood index correlation analysis system for oral ulcer image recognition of autoimmune diseases, including a time-series data acquisition module, a steady-state baseline construction module, a cross-modal synchronous analysis module, a deviation calculation module, and a disease risk assessment module, the system uses the dynamic evolution rate of image lesions to extrapolate blood indexes over time, generates predictive immune index data, and calculates the deviation to generate personalized risk assessment results.
It enables a keen capture of the user's own immune status, improves the sensitivity and accuracy of risk assessment, provides earlier identification of potential risks and personalized early warning information, and enhances the pertinence of health management and clinical decision-making.
Smart Images

Figure CN121416089B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical informatics technology, specifically to a system for image recognition of oral ulcers and correlation analysis of blood indicators in autoimmune diseases. Background Technology
[0002] Dynamic monitoring and assessment of oral ulcers are crucial for early screening and disease management of autoimmune diseases. Clinically, doctors typically need to combine local oral lesions with systemic immune blood markers for comprehensive judgment. In recent years, the development of mobile healthcare and home-based testing technologies has made it possible to continuously collect oral images and regularly test blood markers, providing a data foundation for dynamic disease analysis.
[0003] Several methods have been developed for automatic identification and classification using oral images, or for risk warnings based on deviations of single blood indicators from population baselines. These methods have achieved some degree of diagnostic assistance. However, these approaches face an inherent clinical challenge: the imaging changes of oral ulcers and the fluctuations in in vivo immune indicators are not synchronized in time, and the frequency and timing of their data collection are difficult to maintain in practice. For example, patients may have images taken at any time due to ulcer pain, but their blood tests follow a fixed re-examination cycle of weeks or even months. This results in a temporal "gap" between the available image data and blood data at any given decision point.
[0004] Current technologies for processing multimodal time-series data typically employ a simple "nearest point matching" strategy, correlating the most recent blood reports with the latest oral images. The drawback of this approach is that it ignores the "data window" between the last blood draw and the current image acquisition, during which the patient's immune status may have significantly evolved. Since the development of autoimmune diseases is a dynamic process, this static analysis based on outdated blood data cannot accurately reflect the intrinsic connection between the body's immune system and local lesions at any given moment, leading to decreased sensitivity in risk assessment and delayed warnings. The system may fail to promptly capture the "silent period" risk of immune system activation preceding the large-scale appearance of ulcers, or it may underestimate immune dysregulations hidden behind rapidly progressing ulcers that have not yet been detected by the latest blood tests. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a system for image recognition of oral ulcers and correlation analysis of blood indicators in autoimmune diseases.
[0006] To achieve the above objectives, the technical solution of the present invention is as follows:
[0007] This invention discloses a system for image recognition of oral ulcers and correlation analysis of blood indicators in autoimmune diseases, comprising:
[0008] The time-series data acquisition module is used to acquire the time series of images of oral ulcers and the time series of blood indicators related to autoimmunity from users.
[0009] The steady-state baseline construction module is used to calculate the steady-state baseline of the user's immune indicators based on the user's blood indicator time series.
[0010] The cross-modal synchronization analysis module is used for:
[0011] The dynamic evolution rate of the lesion in the oral ulcer image is obtained by time series quantization.
[0012] When the last sampling point of the blood index time series is earlier than the last sampling point of the oral ulcer image time series, the dynamic evolution rate of the image lesion is used as a guiding factor to extrapolate the time of the blood index time series, generating predictive immune index data synchronized with the last sampling point of the oral ulcer image time series.
[0013] The deviation calculation module is used to calculate the deviation between the predictive immune indicator data and the steady-state baseline of the immune indicator;
[0014] The disease risk assessment module is used to determine whether the deviation exceeds a preset individualized threshold, and to generate a risk assessment result for autoimmune diseases based on the assessment result.
[0015] The risk warning output module is used to output warning information containing the risk assessment results.
[0016] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0017] 1. This invention overcomes the problem of insufficient individual adaptability caused by relying on general population reference values in traditional methods by constructing a homeostatic baseline of immune indicators for each user. This allows the system to sensitively capture subtle but trending changes in the user's own immune status, rather than only issuing alerts when indicators exceed the generally normal range. This individualized dynamic baseline makes it possible to identify early abnormal fluctuations that are still within clinical reference values but have deviated from the user's own homeostasis, thereby shifting the risk assessment threshold forward and improving the sensitivity of early warning.
[0018] 2. This invention utilizes the dynamic evolution rate of lesions quantified from oral ulcer images as a guiding factor to extrapolate potentially lagging blood indicator time series, effectively bridging the data gap caused by the asynchronous acquisition time of two types of data in clinical practice. This allows the system to obtain a predictive immune status assessment that is synchronous with the current oral ulcer condition at the time of decision-making, rather than relying on outdated blood data. This enhances the real-time nature and accuracy of risk assessment results and solves the problem of delayed early warning in existing static correlation analysis methods.
[0019] 3. This invention calculates the deviation between predictive immune index data and individual homeostatic baseline, and generates risk assessment results based on this. This not only provides earlier warnings of potential risks, but also reveals the intrinsic dynamic relationship between local lesions and systemic immune status by the deep fusion and analysis of multimodal time-series data. This provides more targeted and forward-looking reference for users' self-management and doctors' clinical decision-making. Attached Figure Description
[0020] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:
[0021] Figure 1 This is a system module connection diagram of the present invention;
[0022] Figure 2 This is a flowchart of the system modules of the present invention;
[0023] Figure 3 This is a flowchart of the cross-modal synchronization analysis module of the present invention. Detailed Implementation
[0024] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.
[0025] Application Overview:
[0026] In existing technologies, early risk assessment of autoimmune diseases largely relies on independent blood marker analysis or clinical symptom observation, making it difficult to achieve dynamic correlation and synchronous analysis of multimodal data. Traditional methods cannot accurately assess real-time changes in the immune system when the timing of oral ulcer flare-ups and blood marker sampling is asynchronous, leading to delayed or distorted risk assessments. Existing systems lack quantitative analysis of the correlation between the evolution of lesions in images and changes in blood markers in time-series data. Especially during disease activity, this asynchrony significantly affects the timeliness of risk warnings, making it difficult to meet the needs of dynamic health management.
[0027] To address the aforementioned issues, this study discovered a potential correlation between the dynamic evolution rate of oral ulcer images and the changing trends of key immune indicators. A cross-modal coupling model between images and blood indicators can be established to achieve data synchronization and risk prediction. The study found that the dynamic evolution rate of lesions in images is sensitive to short-term changes, while blood indicators, although reflecting the intrinsic state, are collected discontinuously. Therefore, a method was proposed to extrapolate blood indicators over time using the image evolution rate as a guiding factor. Further verification through correlation analysis incorporated the mapping relationship between image dynamic rate and the deviation of blood indicators into the risk assessment model, forming a dynamic risk assessment system.
[0028] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0029] Example:
[0030] like Figure 1 As shown, the oral ulcer image recognition system for autoimmune diseases and its blood index correlation analysis system includes:
[0031] The time-series data acquisition module is used to acquire the time series of images of oral ulcers and the time series of blood indicators related to autoimmunity from users.
[0032] The steady-state baseline construction module is used to calculate the steady-state baseline of an individual's immune indicators based on the user's blood indicator time series.
[0033] The cross-modal synchronization analysis module is used for:
[0034] The dynamic evolution rate of the lesion in the image was obtained by quantizing the time series of oral ulcer images.
[0035] When the last sampling point of the blood index time series is earlier than the last sampling point of the oral ulcer image time series, the dynamic evolution rate of the image lesion is used as a guiding factor to extrapolate the blood index time series in time, generating predictive immune index data synchronized with the last sampling point of the oral ulcer image time series.
[0036] The deviation calculation module is used to calculate the deviation between the predictive immune indicator data and the steady-state baseline of the immune indicator.
[0037] The disease risk assessment module is used to determine whether the deviation exceeds a preset individualized threshold and to generate a risk assessment result for autoimmune diseases based on the assessment result.
[0038] The risk warning output module is used to output warning information containing the risk assessment results.
[0039] At the hardware level, this application relies on a hardware architecture integrating a data acquisition terminal, a central processing server, and an early warning output terminal. The data acquisition terminal mainly includes user-side smart terminals (such as smartphones equipped with dedicated image acquisition accessories or applications) and medical institution-side blood analyzers. These are responsible for acquiring sequential images of oral ulcers and blood sample test results, respectively, and transmitting them to the central processing server via a network. The central processing server consists of high-performance computing units and large-capacity storage devices. The early warning output terminal can be a doctor's workstation interface in the hospital or a user's smart terminal application.
[0040] like Figure 2 As shown, the working principle of this application is as follows: After the system starts, the time-series data acquisition module first retrieves or receives in real time the user's oral ulcer image time series and autoimmune-related blood index time series uploaded by the aforementioned hardware devices, both with precise timestamps, from the database. The calculation unit of the steady-state baseline construction module performs statistical analysis on the user's historical blood index time series, and by calculating its long-term trend and normal fluctuation range, establishes a personalized immune index steady-state baseline for the user. This baseline is stored in the system's storage device as an individualized benchmark for subsequent comparisons.
[0041] The core cross-modal synchronous analysis module first decodes and analyzes the time series of oral ulcer images. Using image processing algorithms (such as edge detection and pixel statistics), it quantifies the rate of change of ulcer area at each time interval, i.e., the dynamic evolution rate of the lesion in the image. When it detects that the last sampling point of the blood indicator time series is earlier than the last sampling point of the oral ulcer image time series, a time extrapolation procedure is triggered. At this time, the module inputs the dynamic evolution rate of the lesion in the image as a dynamic guiding factor into the extrapolation algorithm. Combining this with the historical trend of the blood indicator itself, it predicts and generates predictive immune indicator data synchronized with the latest oral ulcer image time point.
[0042] The embedded image processing model employs the classic U-Net convolutional neural network architecture for accurate ulcer region segmentation. Training of the image processing model is performed on a dataset with pixel-level annotations, containing approximately 10,000 samples that have undergone normalization (scaled to 512x512 pixels) and online data augmentation (including random rotation, flipping, and brightness adjustment). The Adam optimizer is used with an initial learning rate of 0.001, a batch size of 8, and 100 training epochs. The loss function is a linear combination of binary cross-entropy loss and Dice loss.
[0043] Subsequently, the deviation calculation module retrieves the stored individual immune indicator steady-state baseline and compares it with the predictive immune indicator data. It then calculates the specific deviation value using a preset mathematical formula (such as calculating Euclidean distance or relative percentage change). The disease risk assessment module compares the deviation with a preset individualized threshold for the user, logically determines the risk of autoimmune diseases, and generates a corresponding autoimmune disease risk assessment result (such as "low risk," "medium risk," or "high risk"). Finally, the risk warning output module drives the output device to clearly present the warning information containing this assessment result on the doctor's workstation or the user's mobile application interface in the form of a graphical report or text prompt.
[0044] Through the above technical solution, this application effectively integrates discrete and asynchronous multimodal medical data into a dynamic and synchronous risk assessment model, overcoming the inherent problem of time-inconsistency between imaging and laboratory data in clinical practice. This enables the provision of a more timely, accurate, and personalized risk warning service for autoimmune diseases, providing strong data support for users' health management and doctors' clinical decision-making.
[0045] This application further proposes that the system also includes a matrix construction module, used for:
[0046] Based on the time series of blood indicators, the ratios of predefined combinations of immune indicators are calculated to generate an immune ratio matrix. Specifically, the matrix construction module first selects predefined immune indicator pairs (such as CD4 / CD8, Th17 / Treg) from the blood indicator time series and calculates the raw ratios at each time point. Subsequently, to eliminate the influence of dimensions, the raw ratios are Z-score standardized to generate standardized immune indicator ratios. This ultimately generates a standardized immune ratio matrix. Standardized immune index ratios Calculated using the following formula:
[0047]
[0048] in, and These are the mean and standard deviation of the i-th indicator comparison value in the user's history, respectively. All standardized immune indicator ratios are arranged by row (indicator pair) and column (time point) to form an immune ratio matrix.
[0049] Specifically, the steady-state baseline construction module calculates the steady-state baseline of an individual user's immune indicators based on historically generated immune ratio matrices; it extracts all historical immune ratio matrices from the user's stable health period and standardizes each ratio element (i.e., each...) in the matrix. Calculate its normal fluctuation range over a stable period (e.g., the past six months). The upper and lower limits can be set as the mean plus or minus 1.5 times the standard deviation, respectively. This normal fluctuation range constitutes the user's individualized immune indicator steady-state baseline. The preset threshold (1.5 times the standard deviation) is derived from the empirical value commonly used in clinical statistics to define "slight deviation," and it can be adjusted within the range of 0.5 to 2.0 according to the sensitivity requirements of different disease warnings.
[0050] The cross-modal synchronization analysis module extrapolates the immune ratio matrix over time to generate a predictive immune ratio matrix. When time extrapolation is required, the entire immune ratio matrix is extrapolated based on the dynamic evolution rate of the lesions in the image, generating a predictive immune ratio matrix synchronized with the latest oral ulcer image time point.
[0051] The deviation was calculated by comparing the predictive immune ratio matrix with the steady-state baseline of immune indicators. The predictive immune indicator data included the original indicator data and the immune ratio matrix data.
[0052] One feasible calculation method is to select several elements (such as the first three) in the prediction matrix that deviate most severely from their corresponding steady-state baseline range, and calculate the Euclidean norm of the deviation of these elements (such as the first three) as the comprehensive deviation.
[0053]
[0054] Where M is the number of elements with the most severe deviation selected (e.g., M=3). It is the deviation value of the k-th element.
[0055] If the predicted value If it exceeds the steady-state baseline range, then If it is within the range, then .
[0056] By introducing an immune ratio matrix, this application elevates risk assessment from the absolute level of a single indicator to the dimension of the intrinsic balance relationship within the immune network. This enables earlier and more accurate risk insights. Simultaneously, matrix-based calculations and baseline construction based on standardized ratios make the system more robust to individual physiological fluctuations, effectively reducing false alarms caused by slight fluctuations in a single test, and ultimately outputting more clinically valuable early warning information reflecting changes in the core state of the immune system.
[0057] This application further proposes that the specific steps for calculating the ratios of predefined combinations of immune indicators and generating an immune ratio matrix based on the time series of blood indicators include:
[0058] In the matrix construction module, the purpose of generating the immune ratio matrix is to transform the original, dimensionlessly variable blood indicators into standardized features that can effectively characterize the internal homeostasis of the immune system. First, based on an established medical knowledge base, the module selects at least two pairs of immune indicators from the time series of blood indicators that have synergistic or antagonistic significance in pathophysiology; these indicator pairs are not randomly combined, but are based on their known mechanisms of action in the occurrence and development of autoimmune diseases.
[0059] Typical predefined combinations include, but are not limited to: CD4 + T cells and CD8 + The ratio of T cells (CD4 / CD8) is a core indicator for assessing the balance between immune activation and suppression; the ratio of helper T cells 17 (Th17) to regulatory T cells (Treg) (Th17 / Treg) directly reflects the tension between pro-inflammatory and anti-inflammatory immune forces; and the ratio of immunoglobulin A to immunoglobulin G (IgA / IgG) can be used to indicate the bias of humoral immune responses. Multiple indicator pairs, rather than a single indicator, were selected to construct a multi-dimensional, cross-validated immune status assessment system.
[0060] For each sampling time point, the computing unit calculates the original ratio of these predefined index pairs one by one, performs necessary mathematical transformations on the original ratios, and the standardization process is as follows:
[0061] 1. Perform logarithmic transformations, such as calculating the natural logarithm to compress the extreme value range of the data, making its distribution closer to a normal distribution.
[0062] 2. Perform normalization processing, such as using Z-Score standardization, to eliminate dimensions and transform all ratios to the same scale, which facilitates subsequent matrix operations and comparisons.
[0063] Finally, all the ratios selected, calculated, and standardized as described above are organized into a structured immune ratio matrix in a uniform format. In this structured immune ratio matrix, each row represents a specific immune indicator pair (e.g., the first row is CD4 / CD8, and the second row is Th17 / Treg), and each column corresponds to a different sampling time point.
[0064] By generating an immune ratio matrix through the above steps, this application achieves an effective enhancement from raw blood indicators to higher-order immune characteristics. This method, by focusing on the relative proportions rather than absolute values between indicators, can more sensitively reveal imbalances in the homeostasis of the immune system. Simultaneously, the standardized processing enhances the stability and comparability of the data, laying a solid foundation for reliable, quantitative comparative analysis between different individuals and at different times within the same individual. This significantly improves the biological significance of the overall systemic risk assessment and the accuracy of clinical early warning.
[0065] like Figure 3 As shown, this application further proposes that the cross-modal synchronization analysis module is specifically used for:
[0066] First, the module reads the user's oral ulcer image time series from the storage unit. By invoking the embedded image processing model, it automatically analyzes the continuously acquired images in the sequence. By comparing the changes in ulcer area in consecutive images within the oral ulcer image time series, the daily rate of change of ulcer area is calculated as the dynamic evolution rate of the image lesion. Specifically, an edge detection-based algorithm (such as an active contour model or U-Net segmentation network) is used to accurately delineate the contour of the ulcer region in each image and calculate its pixel area. By comparing the ulcer area of images at adjacent time points (typically requiring an interval of 24 to 72 hours to conform to clinical observation logic), the module calculates the daily rate of change of ulcer area, which is then quantified as the dynamic evolution rate of the image lesion. This rate is a signed scalar; a positive value indicates that the ulcer is enlarging, and a negative value indicates that it is healing. Its absolute value reflects the intensity of the disease activity.
[0067] Subsequently, the module performs a time synchronization check. A time synchronization threshold window is then set. When the time difference between the last sampling point of the blood index time series and the last sampling point of the oral ulcer image time series exceeds the time synchronization threshold window, a time extrapolation operation is triggered; the time synchronization threshold window... The default value is typically set to 14 days. This is when the module detects the time difference between the last sampling point of the blood indicator time series and the last sampling point of the oral ulcer image time series. satisfy If the data is out of sync, a time extrapolation operation is immediately triggered. Otherwise, the last blood test data is used directly for subsequent analysis.
[0068] Finally, in the case of triggered extrapolation, a weighted linear extrapolation algorithm based on the dynamic evolution rate of lesions in the image is used to extrapolate the time series of blood indicators, generating predictive immune indicator data synchronized with the last sampling point of the oral ulcer image. The extrapolation step size does not exceed half of the time synchronization threshold window. Through the weighted linear extrapolation algorithm, the module ultimately generates predictive immune indicator data synchronized with the latest oral ulcer image time point. Synchronized predictive immune marker data.
[0069] The weighted linear extrapolation algorithm incorporates image information as a guiding factor into the extrapolation process, assigning weights to historical data points in the time series of blood indicators. The weight allocation follows these principles: data points closer to the present time have higher weights (time decay effect), while also considering the current dynamic evolution rate of lesions in the image. The larger the value (indicating active lesions), the higher the weighting of recent data points. Active lesions suggest that the immune status may be undergoing rapid changes.
[0070] Let the historical value of the immune indicator (or an element in the immune ratio matrix) to be extrapolated be... The corresponding time point is The current (last blood sample) time is The image to which extrapolation needs to be performed is the last sampling time. Extrapolation step size And the system forcibly limits (That is, for a 14-day window, the extrapolation should not exceed 7 days) to ensure the reliability of the forecast and avoid introducing excessive errors due to over-extrapolation.
[0071] The predicted value is calculated using the following weighted linear model. The formula is as follows:
[0072]
[0073] The slope m is not calculated directly from historical data, but is determined by the following formula:
[0074]
[0075] It is a time decay weight. (Attenuation coefficient); It is a preset coupling coefficient (e.g., 0.1 to 0.3) used to adjust the strength of the extrapolation effect of the image evolution rate.
[0076] When the ulcer expands rapidly The slope of the trend will be appropriately amplified; when the ulcer is healing The trend slope will be suppressed or reversed.
[0077] Through the aforementioned sophisticated time-series alignment and extrapolation mechanism, this module provides an image-calibrated estimate that reflects the most probable state of the current immune system. This enhances the system's practicality in real-world clinical scenarios and improves the timeliness and accuracy of early warning signals, providing a solid and synchronous data foundation for core risk assessment.
[0078] This application further proposes that the system also includes a trajectory coupling analysis module, used for:
[0079] The system database retrieves time series data on the dynamic evolution rate of lesions and blood indicators (or immune ratio matrix time series) from images collected multiple times in the user's history. Based on these time series data, the system aligns and standardizes them on a unified time axis to construct a cross-modal time series trajectory that can synchronously reflect the activity of local lesions and the systemic immune status.
[0080] Identify predefined abnormal trajectory patterns and calculate trajectory anomaly confidence.
[0081] The module's pattern recognition engine internally stores several predefined abnormal trajectory patterns summarized from a large number of retrospective clinical studies. These patterns describe the typical spatiotemporal relationships between image and indicator changes in the pre- or early stages of autoimmune diseases. For example:
[0082] Image-led pattern: Characterized by a persistently elevated or recurringly abnormal activity (evolution rate) of oral ulcers, while systemic blood parameters remain near baseline for a considerable period. This pattern may indicate that local mucosal immune abnormalities precede systemic, dramatic changes that can be detected by blood tests.
[0083] Indicator-led pattern: In contrast to the former, this pattern is characterized by a sustained worsening of key hematologic markers (such as specific autoantibody titers or inflammatory factor levels), while the oral ulcers themselves remain relatively quiescent or show only slight fluctuations. This may indicate that a systemic immune dysregulation has been initiated but has not yet triggered a severe pathological response in the oral cavity.
[0084] Parallel deterioration pattern: characterized by a high degree of synchronicity and positive correlation between the activity of image lesions and the deterioration of systemic immune indicators, indicating that the local and systemic pathological processes are closely coupled and the disease may be in a rapid progression phase.
[0085] The disease risk assessment module employs a weighted fusion strategy to integrate deviation assessment results and trajectory anomaly confidence scores. The module calculates the similarity between the current user's cross-modal trajectory and these predefined abnormal patterns, and combines this with the continuity and steepness of the trajectory's own changes to comprehensively calculate a trajectory anomaly confidence score. This trajectory anomaly confidence score is a value between 0 and 1, quantifying the probability that the current trajectory deviates from a normal healthy trajectory pattern. The confidence threshold is set to 0.7. This value was determined through receiver operating characteristic (ROC) curve analysis of historical retrospective datasets. On the validation set, this threshold balances the sensitivity and specificity of the warning, effectively identifying real risks while controlling false positive warnings to an acceptable level. This threshold can be fine-tuned between 0.6 and 0.8 to adapt to the risk tolerance requirements of different clinical application scenarios.
[0086] The specific judgment strategy is as follows: when the deviation exceeds the preset individualization threshold, it indicates that the immune status at the current moment is clearly abnormal, and a high-risk judgment result is directly generated. At this time, trend information is used as auxiliary confirmation.
[0087] When the deviation does not exceed the preset individualization threshold but the confidence level of the trajectory abnormality exceeds the confidence threshold, it indicates that although the current immune indicators are acceptable, the trend of their linkage with oral lesions has shown a highly suspicious abnormal pattern, generating a medium-risk warning result, suggesting that close observation is needed and early intervention may be necessary.
[0088] When neither of them exceeds the corresponding threshold, a low-risk judgment result is generated.
[0089] By introducing trajectory coupling analysis, this application can effectively identify "latent" risks in the very early stages of disease, manifesting only as abnormalities in the temporal relationship between local and systemic parameter changes, as well as "progressive" risks that, although a single test indicator does not reach the alarm threshold, continue to deteriorate. This dual verification mechanism reduces the false positive and false negative rates of the system, provides a more comprehensive and forward-looking risk stratification, and offers crucial decision-making basis for implementing truly early intervention and personalized health management.
[0090] This application further proposes that the predefined abnormal trajectory patterns include at least one of the following:
[0091] A leading-edge pattern exists where the rate of dynamic progression of lesions in images worsens before changes in blood parameters over time series. In several consecutive observation periods, the lesion progression rate quantified from a user's oral ulcer images shows a continuous worsening trend, such as accelerated ulcer enlargement or significantly slowed healing. However, the concurrently collected blood parameter time series (such as key elements in the immune ratio matrix) remain within a steady-state baseline established based on the ulcer's own history, or show only very slight fluctuations below the threshold. This discrepancy of "local activity, systemic calm" may correspond to a very early stage of disease in clinical practice, where abnormal immune attack is initially limited to susceptible target organs such as the oral mucosa, and has not yet caused drastic changes in immune parameters that are routinely detectable in systemic circulation. Identifying this pattern allows for the screening of potentially high-risk individuals among users whose blood parameters remain "normal."
[0092] A leading indicator pattern emerges where changes in blood parameters over time precede the deterioration of the dynamic evolution rate of lesions in images. In this pattern, the system's blood parameters show a clear deterioration trend first, such as a sustained increase in the concentration of specific autoantibodies or an imbalance in the ratio of key immune cells. Meanwhile, the dynamic evolution rate of the lesions shown in the user's oral ulcer images remains relatively stable, without synchronous signs of activity. This reveals that the source or main battleground of immune dysregulation is at the systemic level, with oral manifestations being only one of the subsequent or secondary affected organs. It may also indicate a delay in an individual's perception or reporting of oral ulcers. Even if the current oral condition is acceptable, the system can issue an early warning based on the deteriorating trend of the immune system, suggesting the need for more comprehensive examination or intervention to prevent the subsequent outbreak of local lesions.
[0093] A parallel pattern of synchronous deterioration between the dynamic evolution rate of lesions in imaging and the time series deterioration of blood indicators was observed. Under this parallel deterioration pattern, the deterioration of the dynamic evolution rate of lesions in imaging and the deterioration of blood indicators over time exhibited a high degree of synchronicity and positive correlation. That is, when ulcer activity significantly increased, inflammatory markers or abnormal immune markers in the blood also rose synchronously; conversely, when ulcers tended to heal, these markers decreased accordingly. This strongly coupled trajectory indicates a close linkage between local lesions and systemic immune status, usually predicting an active or acute phase of the disease. The significance of identifying this parallel pattern lies in its ability to strongly confirm that currently observed oral abnormalities are not isolated events, but rather a clear component of systemic immune disease activity. This provides strong evidence from two independent dimensions to support high-risk assessment, increasing the certainty of early warning and the urgency of clinical action.
[0094] By precisely defining and identifying these three core abnormal trajectory patterns, this application transcends the limitations of traditional methods that view various indicators in isolation. Based on the pattern recognition capability of temporal relationships, it enhances the judgment of early hidden signals, development paths, and activity intensity of diseases, providing key decision-making basis for achieving earlier, more accurate, and more predictive risk stratification and intervention for autoimmune diseases.
[0095] This application further proposes that the system also includes a self-learning module, which continuously runs in the system background. This module compares each historical risk assessment result (high risk, moderate risk, low risk) generated by the system with the user's actual health status subsequently confirmed by clinical gold standard methods (e.g., a diagnosis made by a rheumatologist based on an updated comprehensive examination). It records the error between the risk assessment result and the subsequent actual test results. The generated error includes not only a binary result of whether the assessment is correct or not, but may also include information on the degree of continuous error. For example, for a high-risk assessment, it records the disease activity score of the subsequent diagnosis; for deviation calculations, it records the numerical difference between predictive immune indicator data and subsequent actual blood test results. All these error data, along with the corresponding input data (such as the dynamic evolution rate of lesions in images), are recorded. and predictive immune indicator data Both the timestamp and the data are encrypted and stored in a specific log database of the system.
[0096] When the cumulative number of valid error samples recorded for a user reaches a preset number (e.g., N = 20 times), the module's model update unit will be automatically triggered. This preset number is set based on statistical principles to ensure that the amount of data used for updates is sufficient for meaningful statistical inference, while avoiding model oscillations due to insufficient samples. At this point, the module will initiate an optimization process, the core objective of which is to adjust the dynamic evolution rate of image lesions. and predictive immune indicator data The relative weights used in the final calculation of the overall deviation.
[0097] In one specific embodiment, the module utilizes accumulated error samples to solve for a set of optimal weight coefficients by minimizing an objective function (such as mean squared error). The overall deviation D is calculated using the following formula:
[0098]
[0099] Where: D is the overall deviation used for risk assessment.
[0100] It is a function that maps the dynamic evolution rate of image lesions to a scalar risk contribution.
[0101] It is a function for calculating the deviation of predictive immune indicator data from their individual homeostatic baseline.
[0102] and These are the weights assigned to image information and blood indicator information, respectively, and can be set to [value] in the initial state. .
[0103] The self-learning module recalculates based on historical error data. and The value of . For example, if error analysis shows that, in a specific user case, the deterioration of image information always precedes abnormalities in blood indicators by a significant margin, and that image-based warnings are more accurate, then the optimization algorithm will assign after iteration . A relatively higher weighting. Conversely, if the predictive power of blood indicators is found to be very stable and reliable, the weighting will be increased accordingly. The weights are then adjusted. The optimization process can be based on gradient descent or other numerical optimization algorithms to ensure that the system's overall prediction error for historical data is minimized under the new weights. The updated weights will replace the old weights and be applied to the user's subsequent risk assessments.
[0104] The module's weight optimization process employs gradient descent. The target loss function is the mean squared error between the calculated comprehensive deviation D and the subsequent actual clinical diagnosis (which can be quantified as a binary or continuous label). The optimization process uses mini-batch gradient descent, with the batch size dynamically adjusted based on the number of available error samples, but not less than 5. The base learning rate is set to 0.01, and an exponential decay strategy is adopted (a decay coefficient of 0.9 every 10 iterations). The maximum number of optimization iterations is set to 50, or the process terminates early when the loss function value decreases by less than 1e-5 over three consecutive iterations to avoid overfitting.
[0105] By introducing a self-learning module, this application achieves personalized calibration of the core algorithm weights by quantifying and correcting its own historical errors. This enables the system to better adapt to the unique correlation patterns and time differences between oral ulcer manifestations and systemic immune status for different users. This improves the long-term accuracy and individualized adaptability of the system, effectively reducing misjudgments caused by individual differences or different disease phenotypes, and ensuring that risk assessment conclusions become increasingly reliable with data accumulation.
[0106] This application further proposes that the self-learning module also includes:
[0107] Statistics show the distribution of time intervals in multiple user cases where the dynamic evolution rate of image lesions was significantly correlated with the actual rate of change of subsequent blood indicators.
[0108] Based on the central tendency of the time interval distribution, adjust the default size of the preset window value used when performing time extrapolation.
[0109] The module anonymizes and retrieves complete time-series data of a large number of user cases from the system database. For each eligible user case, the module calculates the dynamic evolution rate of its image lesions. Time series and subsequent actual rate of change of blood indicators The correlation between time series. This can be obtained by calculating the rate of change between two consecutive blood test results (e.g., a key element in an immune ratio matrix). The module's task is to find the time interval at which the image sequence leads the blood sequence when the two sequences reach their strongest correlation. Specifically, the module systematically shifts the image sequence relative to the blood sequence by different time lengths (e.g., 0 days, 3 days, 7 days, 14 days, etc.) and calculates the correlation coefficient (e.g., Pearson correlation coefficient) between the two sequences after each shift. For each user case, the shift amount that maximizes the correlation coefficient (and is statistically significant, e.g., p-value less than 0.05) is recorded as the optimal lead time for image changes to precede blood changes in that case. .
[0110] Optimal lead time when a sufficient number of user cases are collected Afterwards (e.g., when the sample size M > 100), the module begins performing distributional statistical analysis. It will determine the optimal lead time for all these user cases. The values form a dataset, and the central tendency of that dataset is calculated. The most commonly used indicator is the median. The median was chosen over the mean because it is less sensitive to outliers (such as a user leading for an exceptionally long period) and better reflects the general patterns of most users. This calculated median represents the typical time span over which changes in oral ulcer images statistically lead changes in blood indicators within the current user group and data context.
[0111] Finally, the self-learning module updates the system's global parameters based on this analysis. It will use the time synchronization threshold window for the time extrapolation operation. The default value is adjusted to a function based on this central tendency. Time synchronization threshold window. The default initial value is set to 14 days. This is longer than the recommended interval for routine outpatient blood tests (usually 7-14 days), avoiding frequent extrapolation due to short-term fluctuations; at the same time, it is significantly shorter than the typical progression cycle of most autoimmune diseases from local activity to systemic outbreak (usually several weeks to several months), ensuring timely warning. In systemic applications, this threshold can be adjusted within the range of 7 to 21 days according to clinical preferences.
[0112] A preferred implementation is to set the new default window value. Set as the median lead time plus a buffer margin B, its calculation formula can be expressed as:
[0113]
[0114] The buffer margin B can be set as follows: A certain percentage (e.g., 20%), or a fixed number of days (e.g., 3 days), is used to accommodate normal fluctuations in individuals around the typical value. For example, if statistics show that the group median leads by a certain time... =10 days, buffer margin B=3 days, then the system will The default value has been updated from the initial 14 days to 13 days. This update process can be performed automatically on a regular basis (such as quarterly) to ensure that system parameters can evolve with the accumulated data.
[0115] Through this dynamic optimization mechanism based on population data, this application avoids prematurely abandoning the use of valuable historical data due to an excessively short window, and also prevents unreliable predictions based on outdated information due to an excessively long window. From a global perspective, it improves the timeliness and scientific rigor of the entire system's cross-modal synchronous analysis, making the timing of early warnings more aligned with the natural pathophysiological processes of disease development.
[0116] This application further proposes that the risk warning output module is specifically used for:
[0117] The risk assessment results are linked to key driving factors extracted from the cross-modal synchronous analysis module and the deviation calculation module. Upon receiving the risk assessment results from the disease risk assessment module, it does not immediately output them. Instead, it first initiates an information fusion and attribution process. The module then extracts the core dynamic parameters used in this calculation from the cross-modal synchronous analysis module, primarily the dynamic evolution rate of image lesions. The module extracts one or more key driving factors that cause the deviation to exceed the limit, along with their positive and negative signs. Specifically, these driving factors refer to the specific indicators or indicator pairs that contribute the most to the overall deviation D in the immune ratio matrix or original immune indicators, such as an abnormal decrease in the "CD4 / CD8 ratio" or an abnormal increase in the "IgA / IgG ratio." The module's logical processing unit deeply binds the risk assessment results with these extracted key driving factors, forming a structured early warning information object.
[0118] Based on this information object, the module generates highly specific and scenario-based warning content. Specifically, when the risk assessment result is high risk, the output warning information includes:
[0119] Based on specific predictive immune marker data that led to the deviation exceeding the limit, the specific retesting recommendations are not simply a general suggestion to "retest blood," but rather clearly indicate which specific immune markers should be prioritized for retesting. For example, the warning message might explicitly state: "The system has detected a possible significant deviation in your CD4 / CD8 immune ratio, and it is recommended to prioritize retesting T lymphocyte subset analysis within 3 to 5 days," thereby achieving precise guidance of testing resources and improving the efficiency of clinical decision-making.
[0120] This is a predicted lesion progression indicator generated based on the dynamic evolution rate of lesions in images. The predicted lesion progression indicator is based on the dynamic evolution rate of lesions in images obtained from the cross-modal synchronous analysis module. To make calculations. For example, if the current dynamic evolution rate of the lesion in the image... A positive and large value may generate a message such as, "Based on the current ulcer enlargement trend, without intervention, the ulcer area is expected to increase by approximately 20% within the next week." This estimate is not a precise prediction, but rather a linear extrapolation based on the current rate, designed to intuitively convey the activity and urgency of the lesion to the user or doctor.
[0121] Through the aforementioned mechanism, this application enhances the practical value and guiding significance of early warning information by revealing the specific driving factors behind the risks and providing targeted action suggestions. This enables users to not only be aware of the existence of risks but also understand their sources and potential short-term consequences. This not only increases user trust and compliance with the system but also provides doctors with a clear and focused reference path for subsequent diagnosis and intervention, thereby optimizing the overall efficiency of the transformation from risk warning to clinical action and forming a closed-loop management system based on human-machine collaboration.
[0122] This application further proposes that, when the user's historical immune ratio matrix data is insufficient, the steady-state baseline construction module is specifically used for:
[0123] When the module detects that the amount of historical immunity ratio matrix data for a user is lower than the minimum statistical requirement (e.g., fewer than 10 valid data points), ( When a user is identified as having a certain age, gender, and medical history (e.g., whether they have a confirmed autoimmune disease or are in an immunosuppressed state), the system will initiate a baseline initialization process. Based on the user's age, gender, and medical history (e.g., whether they have a confirmed autoimmune disease or are in an immunosuppressed state), the system will match the corresponding stratified immune ratio baseline from a pre-stored population database as the initial steady-state baseline for immune indicators. The population database is derived from large-scale epidemiological surveys or historical health checkup data of healthy individuals, and the statistical ranges for each immune indicator ratio have been pre-calculated based on different attribute combinations (e.g., "30-40 years old, female, no relevant medical history"), typically using the mean as the baseline. and standard deviation Representation. The module will match the user with the stratum that best matches their attributes and assign the mean of that stratum. This serves as the initial steady-state baseline for the user's immune indicators. .
[0124] After acquiring new user data, a weighted average algorithm is used to update the initial steady-state baseline of the immune indicator. The weight of each user's historical data increases with the amount of data, until it completely replaces the baseline of the stratified immune ratio. The update algorithm employs a weighted average strategy, the core idea of which is to update the baseline based on the user's own data... Through continuous accumulation, the baseline of the population is gradually lowered. The contribution of individual data, and accordingly improve the role of individual data in building new baselines. The weights in the equation. This process can be represented by the following formula:
[0125]
[0126] in, This is the current baseline value (the initial steady-state baseline of immune indicators upon the first update). ), This refers to newly added user data. The key is the weighting. The design of is not a fixed value, but a function that increases with the amount of valid user data n. A preferred implementation is to let Where n is the number of data points for the user, and K is a preset smoothing factor (e.g., K = 10). When n = 1 (i.e., the first update), The new baseline is largely determined by the initial population baseline; as n increases to 5, The influence of individual data increases significantly; when n is large enough (e.g. )hour, Approaching 1 means that the steady-state baseline at this point is... It is almost entirely determined by the user's own historical data, enabling a smooth transition and replacement from a group benchmark to a fully individualized baseline.
[0127] Through the above technical solution, this application effectively overcomes the data scarcity problem faced by personalized medicine models in the initial stage. It ensures that the system has basic assessment capabilities from the beginning of new user use, avoiding system stagnation due to the lack of available baselines, and through data-driven weight transfer, ensures that the baseline can increasingly accurately reflect the user's unique immune physiological characteristics over time.
[0128] The following is a specific embodiment of a blood index correlation analysis system for oral ulcer image recognition of autoimmune diseases:
[0129] A 25-year-old female user experienced recurrent oral ulcers (4 times in the past 3 months). The system collected images of the oral ulcers from the user's smartphone (once daily, automatically timestamped) and simultaneously retrieved her blood test data from the past year (once every 3 months, including CD4). + T cells, CD8 + (Immune indicators such as T cells, IgA, and IgG). When a user uses the system for the first time, there is no historical baseline data, and an individual's immune homeostasis baseline needs to be initialized based on the population database.
[0130] The user took an oral cavity image on the third day of an ulcer flare-up. The system calculated the ulcer area to be 8.2 mm using the U-Net segmentation algorithm. 2 Two days later (the fifth day of the attack), another photograph was taken, and the affected area had increased to 12.5 mm. 2 Continuous image comparison showed a daily rate of change of ulcer area (dynamic evolution rate of lesion in images) of +2.15 mm. 2 / day (positive value indicates expansion).
[0131] The last blood test was conducted one month before the onset of the illness, and the data showed CD4 count. + T cells = 680 cells / μL, CD8 + T cells = 420 cells / μL, IgA = 2.3 g / L, IgG = 11.5 g / L. The system calculates the immune ratios based on predefined combinations: CD4 / CD8 = 680 / 420 ≈ 1.62; IgA / IgG = 2.3 / 11.5 ≈ 0.20. These ratios are Z-score standardized (based on the mean μ and standard deviation σ of the population database) to generate an immune ratio matrix.
[0132] Indicator Pair Standardized ratio (1 month before onset) CD4 / CD8 ( / 0.32 ≈ 0.38 IgA / IgG ( / 0.05 ≈ 0.40
[0133] Due to insufficient historical data for the user (only one blood test), the system initialized the steady-state baseline by calling the population stratification baseline (data from healthy women aged 25).
[0134] CD4 / CD8 baseline range: [1.45, 1.75] (corresponding to normalized ratios [-0.16, 0.78]);
[0135] IgA / IgG baseline range: [0.15, 0.25] (corresponding to standardized ratios [-0.60, 1.40]);
[0136] As user data accumulates (with two additional blood tests added within the next 6 months), the system gradually increases the weight of individual data through a weighted average algorithm (initially 0.3, and completely replaces the population baseline after 5 data tests).
[0137] The user's latest oral ulcer image (day 5 of onset) was taken 35 days after the last blood test, exceeding the system's preset time synchronization threshold window (default 14 days). This triggered time extrapolation: Image lesion dynamic evolution rate guidance: based on the ulcer's daily enlargement rate (+2.15mm). 2 The blood index trend slope was adjusted using a weighted linear extrapolation algorithm with the factor ( / day) as the extrapolation factor.
[0138] Extrapolation calculation (taking the CD4 / CD8 ratio as an example): The CD4 / CD8 ratios in the last three tests were 1.58 (3 months ago) and 1.62 (1 month ago), with an original change rate of 0.02 / 2 months = 0.01 / month. The ulcer enlargement rate is positive. The system amplifies the trend slope through a coupling coefficient α = 0.2, and the adjusted slope m = 0.01 × (1 + 0.2 × 2.15) = 0.0143 / month. Extrapolating to the current time point (35 days ≈ 1.17 months): the predicted CD4 / CD8 ratio = 1.62 + 0.0143 × 1.17 ≈ 1.64, which, after standardization, is (1.64 - 1.50) / 0.32 ≈ 0.44.
[0139] Similarly, the predicted IgA / IgG ratio is calculated to generate a predictive immune ratio matrix:
[0140] Indicator Pair Predicted standardized ratio (current time point) CD4 / CD8 0.44 IgA / IgG 0.52
[0141] The system selects the two indicators with the most significant deviations (CD4 / CD8, IgA / IgG) from the predicted immune ratio matrix and calculates the Euclidean distance as the overall deviation: (Note: The steady-state baseline mean is standardized to 0, and the deviation threshold is set to 0.65.)
[0142] The system constructs a cross-modal trajectory of image lesion rate and blood parameters, and discovers the user's current ulcer enlargement rate (+2.15mm). 2The trend of the CD4 / CD8 ratio (predicted daily) and the CD4 / CD8 ratio (continuously rising) shows a parallel deterioration pattern, with an anomaly confidence level of 0.82 (higher than the confidence threshold of 0.7). The deviation (0.68) > the individualized threshold (0.65), triggering a high-risk judgment; combined with the trajectory anomaly confidence level (0.82), the system outputs a "high-risk warning".
[0143] The system generates structured early warning information, including:
[0144] Specific retest recommendation: "The predicted CD4 / CD8 ratio deviates from the steady-state baseline. It is recommended to retest T lymphocyte subsets within 3 days."
[0145] The lesion progression indicates that "the ulcer is currently enlarging by 2.15 mm per day." 2 Without intervention, the affected area is expected to reach 23mm within 5 days. 2 ".
[0146] The user sought medical attention as prompted, and a follow-up examination showed that the actual CD4 / CD8 ratio was 1.66 (standardized 0.50), with an error of less than 12% compared to the predicted value (0.44). Based on the system's early warning and the follow-up examination results, the doctor made an early diagnosis of mild autoimmune cheilitis and promptly initiated anti-inflammatory treatment to prevent the condition from progressing.
[0147] This embodiment effectively bridges the temporal asynchrony between oral ulcer images and blood indicators through multimodal time-series data fusion and dynamic baseline adjustment, providing an innovative solution for early warning of autoimmune diseases by linking "local lesions-systemic immunity" analysis.
[0148] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.
Claims
1. A system for image recognition and correlation analysis of blood indicators in autoimmune diseases involving oral ulcers, characterized in that: include: The time-series data acquisition module is used to acquire the time series of images of oral ulcers and the time series of blood indicators related to autoimmunity from users. The steady-state baseline construction module is used to calculate the steady-state baseline of the user's immune indicators based on the user's blood indicator time series. The cross-modal synchronization analysis module is used for: The dynamic evolution rate of the lesion in the oral ulcer image is obtained by time series quantization. When the last sampling point of the blood index time series is earlier than the last sampling point of the oral ulcer image time series, the dynamic evolution rate of the image lesion is used as a guiding factor to extrapolate the time of the blood index time series, generating predictive immune index data synchronized with the last sampling point of the oral ulcer image time series. The deviation calculation module is used to calculate the deviation between the predictive immune indicator data and the steady-state baseline of the immune indicator; The disease risk assessment module is used to determine whether the deviation exceeds a preset individualized threshold, and to generate a risk assessment result for autoimmune diseases based on the assessment result. The risk warning output module is used to output warning information containing the risk assessment results; Specifically, the cross-modal synchronization analysis module is used for: The daily rate of change of ulcer area is calculated as the dynamic evolution rate of the lesion in the image by comparing the changes in ulcer area in consecutive images in the time series of oral ulcer images. A time synchronization threshold window is set. When the time difference between the last sampling point of the blood index time series and the last sampling point of the oral ulcer image time series exceeds the time synchronization threshold window, a time extrapolation operation is triggered. A weighted linear extrapolation algorithm based on the dynamic evolution rate of the lesions in the image is used to extrapolate the time series of blood indicators to generate predictive immune indicator data synchronized with the last sampling point of the oral ulcer image, wherein the extrapolation step size does not exceed 1 / 2 of the time synchronization threshold window.
2. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 1, characterized in that: The system also includes a matrix construction module, used for: Based on the time series of blood indicators, calculate the ratio of predefined combinations of immune indicators and generate an immune ratio matrix; The steady-state baseline construction module calculates the steady-state baseline of the immune indicators of the individual user based on the historically generated immune ratio matrix. The cross-modal synchronous analysis module extrapolates the immune ratio matrix over time to generate a predictive immune ratio matrix. The deviation is calculated by comparing the predictive immune ratio matrix with the steady-state baseline of the immune indicators.
3. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 2, characterized in that: The specific steps for calculating the ratios of predefined combinations of immune indicators and generating an immune ratio matrix based on the blood indicator time series include: Select at least two pairs of immune indicators from the blood indicator time series; Calculate the ratio of each immune indicator pair; The calculated ratios are logarithmically or normally processed to generate the immune ratio matrix.
4. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 1, characterized in that: The system also includes a trajectory coupling analysis module, used for: Based on the dynamic evolution rate of the lesions in the images collected multiple times and the time series of the blood indicators, a cross-modal time series trajectory is constructed; Identify predefined abnormal trajectory patterns and calculate trajectory anomaly confidence. The disease risk assessment module employs a weighted fusion strategy to integrate deviation assessment results and trajectory anomaly confidence levels. When the deviation exceeds a preset individualization threshold, a high-risk judgment result is generated directly; When the deviation does not exceed the preset individualization threshold but the trajectory anomaly confidence exceeds the confidence threshold, a medium-risk warning result is generated; When neither of them exceeds the corresponding threshold, a low-risk judgment result is generated.
5. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 4, characterized in that: The predefined abnormal trajectory pattern includes at least one of the following: The image-preceding pattern shows that the rate of dynamic evolution of lesions in the image deteriorates before the time-series changes of the blood indicators. The time-series changes of blood indicators precede the deterioration of the dynamic evolution rate of the image lesions in an indicator-leading pattern. The image lesion dynamic evolution rate is synchronously deteriorated in a parallel mode with the blood index time series.
6. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 1, characterized in that: The system also includes a self-learning module for: Record the error between the risk assessment result and the subsequent actual test result; When the number of recorded error samples reaches a preset number, the weights of the image lesion dynamic evolution rate and the predictive immune index data in calculating the deviation are adjusted according to the error.
7. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 6, characterized in that: The self-learning module also includes: Statistics show that in multiple user cases, the time intervals between the dynamic evolution rate of the image lesion and the actual change rate of subsequent blood indicators are related. Based on the central tendency of the time interval distribution, adjust the default size of the preset window value used when performing the time extrapolation.
8. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 1, characterized in that: The risk warning output module is specifically used for: The risk assessment results are linked to key driving factors extracted from the cross-modal synchronization analysis module and the deviation calculation module; When the risk assessment result is high risk, the output warning information includes: a special retest suggestion generated based on specific predictive immune indicator data that caused the deviation to exceed the limit, and an estimated lesion progression prompt generated based on the dynamic evolution rate of the lesion in the image.
9. The oral ulcer image recognition and autoimmune disease blood index correlation analysis system according to claim 2, characterized in that: When the user's historical immunity ratio matrix data is insufficient, the steady-state baseline construction module is specifically used for: Based on the user's age, gender, and past medical history, the corresponding population stratification immune ratio baseline is matched from the pre-stored population database as the initial steady-state baseline for immune indicators. After acquiring new user data, a weighted average algorithm is used to update the initial steady-state baseline of the immune index, wherein the weight of the user's own historical data increases with the increase of data volume until it completely replaces the baseline of the population stratified immune ratio.
Citation Information
Patent Citations
Monitoring imaging system for oral care
CN119048875A
Gene sequence variations with utility in determining the treatment of disease
WO2000050639A2
