Dermatological disease pre-screening system

Abstract

The invention relates to the technical field of medical information, and discloses a dermatological disease pre-screening system. The system comprises a data acquisition module, a feature network construction module, a risk assessment space generation module, a real-time detection module and a prediction report generation module. The system constructs a unified data set by integrating long-term monitoring data of a patient; mining a multi-dimensional feature sequence to construct a feature change network reflecting dynamic association between features; generating a risk assessment space based on historical case distribution and delimiting an early warning area; projecting real-time data to the space, and obtaining a quantitative risk index by calculating the proximity to the early warning area; and finally generating a customized pre-screening report. According to the method, the disease evolution characteristics can be captured more essentially by analyzing the dynamic association rule between the characteristics; through interpretable risk space proximity analysis, visual monitoring and early warning of a risk evolution trajectory are realized, and the accuracy and foresight of disease risk identification are improved.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to a pre-screening system for dermatological diseases. Background Technology

[0002] The clinical management of chronic dermatological diseases heavily relies on long-term monitoring and risk assessment of disease activity. Currently, mainstream risk assessment techniques fall into two main categories: static analysis based on single visit data and simple trend analysis based on time series data. These methods generally treat each monitoring session as an independent event or focus only on isolated changes in individual indicators, failing to reveal the underlying evolutionary patterns of the disease as a whole.

[0003] The limitation of existing technologies lies in the singularity of their analytical dimensions. Static analysis cannot capture the dynamic process of disease progression from quantitative to qualitative change, while simple time-series analysis ignores the complex interactions and synergistic evolutionary relationships between different clinical features. These interconnected dynamic patterns are often key early signals of disease deterioration, but current technologies lack effective means to model and capture them. Furthermore, the decision-making process of current advanced machine learning prediction models is opaque, only outputting a risk probability. Clinicians find it difficult to understand the underlying logic and trace the source of risk, which severely limits the credibility and application value of the prediction results in clinical practice.

[0004] There is a need in this field for a technical solution that can break through existing analytical frameworks. This solution needs to be able to effectively mine dynamic correlation networks between multiple features from long-term, multi-dimensional monitoring data, thereby providing a more fundamental description of the disease state. It needs to transform risk assessment into an interpretable and visualized process, not only revealing the risk level but also clearly demonstrating the trajectory of risk evolution, providing intuitive and reliable quantitative evidence for early clinical intervention. Summary of the Invention

[0005] The purpose of this invention is to provide a pre-screening system for dermatological diseases to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a pre-screening system for dermatological diseases, the system comprising:

[0007] The data acquisition module is used to collect long-term medical monitoring data from dermatology patients and to integrate and process this data to form a unified health data set.

[0008] The feature network construction module is used to perform feature mining operations on the unified health data set to obtain multi-dimensional feature sequences, and to construct a feature change network based on the multi-dimensional feature sequences.

[0009] The risk assessment space generation module is used to generate a risk assessment space based on the feature change network, analyze the distribution pattern of historical disease cases in the risk assessment space, and delineate early warning feature areas according to the distribution pattern.

[0010] The real-time detection module is used to acquire real-time detection data of patients, extract instantaneous feature sets from them, project the instantaneous feature sets onto the risk assessment space, measure their proximity to the warning feature area, and obtain the current risk index.

[0011] The prediction and report generation module is used to combine the current risk indicators, the multi-dimensional feature sequences, and the feature change network to build a disease probability prediction system and generate customized pre-screening reports.

[0012] Preferably, the collection of long-term medical monitoring data from dermatology patients, and the integration and processing of this data to form a unified health data set, includes:

[0013] Collect medical records, image records, and physiological indicators of dermatology patients from medical platforms;

[0014] Perform quality checks on the collected data, including removing duplicate records and correcting formatting errors;

[0015] The processed data is archived according to disease type and time sequence to generate a unified health data set.

[0016] Preferably, the step of performing feature mining operations on the unified health data set to obtain multi-dimensional feature sequences, and constructing a feature change network based on the multi-dimensional feature sequences, includes:

[0017] Filter parameters related to skin diseases from a unified health dataset;

[0018] Calculate the numerical changes of each parameter at different time points to form a multi-dimensional feature sequence;

[0019] The multi-dimensional feature sequence is converted into a graph model, where nodes represent feature values ​​and edges represent the time dependencies between features, thus constructing a feature change network.

[0020] Preferably, the step of generating a risk assessment space based on the feature change network, analyzing the distribution patterns of historical disease cases in the risk assessment space, and delineating early warning feature regions based on the distribution patterns includes:

[0021] The dimensions of the risk assessment space are determined based on the number of features in the feature-changing network.

[0022] Incorporate characteristic data points from historical disease cases into the risk assessment space;

[0023] Calculate the clustering degree of data points within the risk assessment space to obtain the distribution pattern;

[0024] A threshold value is set based on the distribution pattern, and areas with a clustering degree higher than the threshold value are marked as early warning characteristic areas.

[0025] Preferably, the calculation of the clustering degree of data points within the risk assessment space to obtain the distribution pattern is achieved through the following steps:

[0026] Obtain the coordinates of each historical disease case data point in the risk assessment space;

[0027] For each data point, calculate the number of neighboring data points within a preset radius, and use this as the local density value of that data point.

[0028] Sort the local density values ​​of all data points to generate density distribution curves;

[0029] Identify peak regions in the density distribution curve, where the peak regions correspond to areas of high data point aggregation.

[0030] The spatial range of the distribution pattern is determined based on the boundary coordinates of the highly concentrated areas.

[0031] Preferably, the step of acquiring real-time patient monitoring data, extracting an instantaneous feature set from it, projecting the instantaneous feature set onto the risk assessment space, measuring its proximity to the warning feature region, and obtaining the current risk indicator includes:

[0032] Real-time capture of dermoscopy images and descriptions of clinical symptoms in patients;

[0033] Feature values ​​corresponding to multi-dimensional feature sequences are extracted from real-time detection data to form an instantaneous feature set;

[0034] The instantaneous feature set is standardized and then mapped to the risk assessment space;

[0035] The minimum distance between the mapping point and the boundary of the warning feature area is calculated and used as the degree of proximity, which is then converted into the current risk indicator.

[0036] Preferably, the step of combining the current risk indicators, the multi-dimensional feature sequences, and the feature change network to construct a disease probability prediction system and generate a customized pre-screening report includes:

[0037] Input the current risk indicators and multi-dimensional feature sequences into the prediction model;

[0038] The predictive model uses a feature change network to learn the evolution of features and outputs a probability value of disease occurrence.

[0039] Customized pre-screening reports are generated based on probability values.

[0040] Preferably, the construction of the disease probability prediction system includes building a prediction model, which is achieved through the following steps:

[0041] Confirmed disease cases and healthy cases were selected from a unified health dataset as the training sample set;

[0042] Extract multi-dimensional feature sequences and feature variation network structure information for each sample in the training sample set;

[0043] Initialize the network parameters of the prediction model, using multi-dimensional feature sequences as input features and disease occurrence labels as output targets;

[0044] Use the edge connections in the feature variation network to define the time constraints between features;

[0045] The model parameters are iteratively optimized using the gradient descent algorithm to make the predicted output close to the actual disease occurrence label, thus completing the training of the prediction model.

[0046] Preferably, the prediction model utilizes a feature change network to learn the evolution patterns of features and outputs a disease occurrence probability value, which is achieved through the following steps:

[0047] Map the patient's immediate feature set to the corresponding node in the feature change network;

[0048] Tracing the temporal variation path of feature values ​​along the edge connection directions in the feature variation network;

[0049] Calculate the degree of matching between the change path of the instantaneous feature set and the feature path of disease cases in the training samples;

[0050] The probability value of disease occurrence is generated by weighting the disease occurrence labels based on the matching degree.

[0051] Preferably, the process of generating a customized pre-screening report based on probability values ​​is achieved through the following steps:

[0052] Set probability threshold ranges corresponding to different risk levels;

[0053] By comparing the probability value of disease occurrence with the probability threshold range, the patient's risk level is determined.

[0054] Select the appropriate template from the predefined report template library based on the risk level;

[0055] By filling in the patient's multidimensional characteristic sequence, current risk indicators, and risk level description into the corresponding fields in the template, a customized pre-screening report containing a risk summary and recommended measures is generated.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] By constructing a feature change network, the system elevates the analysis unit from independent feature points or linear sequences to the level of dynamic relationships between features. This method mines multi-dimensional feature sequences obtained from long-term monitoring, identifying synergistic changes, causal sequences, or strong coupling relationships of different clinical signs and parameters over time, and constructs a network model based on this. It can capture the more fundamental pathophysiological dynamics in the course of disease development. The system no longer judges whether a single feature exceeds the standard in isolation, but rather keenly identifies abnormal correlation patterns between multiple features. This relationship network-based analysis can reveal the essential evolution of disease states earlier, achieving modeling of the intrinsic driving forces of disease at the data level, and significantly improving the sensitivity and foresight of risk identification.

[0058] By generating a risk assessment space and calculating the proximity of real-time data to warning zones, the system transforms risk assessment into an interpretable, dynamic trajectory analysis process. This technology maps historical case patterns in a feature change network to a low-dimensional, measurable space and delineates warning zones based on the distribution of high-risk cases. This transforms risk assessment results from simple probability values ​​into a continuous and intuitive measure of the spatial distance between the current state and known hazard patterns. Clinicians can clearly observe the patient's current health status within this space and its trend towards the warning zone over time. This visualized and quantitative approximation analysis allows for the identification of the cumulative direction and speed of quantitative changes before clinically significant qualitative changes occur in the disease. This provides a clear window of opportunity and reliable decision-making basis for early intervention, enhancing the clinical operability and credibility of risk assessment results. Attached Figure Description

[0059] Figure 1 This is a schematic diagram illustrating the working principle of the dermatological disease pre-screening system described in this invention.

[0060] Figure 2 A flowchart for data acquisition and integration processing;

[0061] Figure 3 A flowchart for generating risk assessment space and delineating early warning areas;

[0062] Figure 4 Loss curve for training a pre-screening model for dermatological diseases;

[0063] Figure 5 Bar chart showing the distribution of pre-screened risk levels for dermatological diseases. Detailed Implementation

[0064] 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.

[0065] Please see Figure 1 This invention provides a pre-screening system for dermatological diseases. The system includes a data acquisition module, a feature network construction module, a risk assessment space generation module, a real-time detection module, and a prediction and report generation module. These modules work together to complete the disease pre-screening process. The data acquisition module is responsible for collecting long-term medical monitoring data of dermatological patients and integrating this data to form a unified health data set. The feature network construction module performs feature mining operations on the unified health data set to obtain multi-dimensional feature sequences and constructs a feature change network based on these sequences. The risk assessment space generation module generates a risk assessment space based on the feature change network, analyzes the distribution patterns of historical disease cases in the risk assessment space, and delineates warning feature areas based on these patterns. The real-time detection module acquires real-time patient monitoring data, extracts instantaneous feature sets from these data, projects these instantaneous feature sets onto the risk assessment space, measures their proximity to the warning feature areas, and obtains the current risk index. The prediction and report generation module combines the current risk index, multi-dimensional feature sequences, and feature change network to construct a disease probability prediction system and generate a customized pre-screening report.

[0066] Example 1: See Figure 2In a specific implementation, the data acquisition module collects long-term medical monitoring data from dermatology patients and integrates and processes it to form a unified health dataset. The feature network construction module performs feature mining operations on the unified health dataset to obtain multi-dimensional feature sequences and construct a feature change network. In the specific implementation, the data acquisition module collects medical records, image records, and physiological indicators from dermatology patients from a medical platform. These data sources include electronic health record systems and digital imaging devices, and the collected data may contain structured text, numerical measurements, and high-resolution image files. Quality checks are performed on the collected data, including removing duplicate records and correcting format errors. Removing duplicate records is achieved by comparing patient identifiers and timestamps, and correcting format errors involves standardizing date formats and numerical units. The processed data is archived according to disease type and chronological order. Disease type classification is based on the International Classification of Diseases (ICD), and chronological order archiving is based on the data collection time points, generating a unified health dataset. In some embodiments, the unified health dataset is stored in the form of database tables, with each table entry associated with a unique patient identifier. Optionally, the quality check operations also include missing value imputation, filling continuous variables with the mean or median.

[0067] In its implementation, the feature network construction module filters parameters related to skin diseases from a unified health dataset. This parameter selection is based on a predefined feature list from a medical knowledge base, including lesion size, color histogram, and biomarker concentration. For each parameter, its numerical changes at different time points are calculated, forming a multi-dimensional feature sequence. The calculation of numerical changes involves time series analysis for each parameter. At the point of time and The change is defined as:

[0068] ;

[0069] in: This represents the value of the i-th parameter. This represents the j-th time point. Representing the changes, the multi-dimensional feature sequence is composed of the sequences of changes in all parameter terms. The multi-dimensional feature sequence is converted into a graphical model, where nodes represent feature values ​​and edges represent temporal dependencies between features. A feature change network is constructed, with node generation based on discretized binning of feature values ​​and edge connections based on conditional probability calculations of temporally adjacent feature values. In some embodiments, temporal dependencies are quantized using transition matrices, representing the evolution pattern of feature values ​​at consecutive time points. Optionally, the graphical model uses a directed graph structure, with the direction indicating the flow of time.

[0070] Example 2: See Figure 3In a specific implementation, the risk assessment space generation module generates a risk assessment space based on a feature change network, analyzes the distribution patterns of historical disease cases within the risk assessment space, and delineates warning feature areas based on these patterns. Specifically, the dimension of the risk assessment space is determined by the number of features in the feature change network. The number of features corresponds to the total number of node types in the network, and the number of dimensions equals the number of features, with each dimension representing a specific feature axis. Feature data points from historical disease cases are placed into the risk assessment space. Each historical disease case is represented by a feature value vector of its multi-dimensional feature sequence at a specific time point, and this vector is directly mapped to a coordinate point in the risk assessment space. The degree of clustering of data points within the risk assessment space is calculated to obtain a distribution pattern, which describes the concentration of historical disease cases in the feature space. A threshold value is set based on the distribution pattern, and areas with a clustering degree higher than the threshold value are marked as warning feature areas. The threshold value is determined based on the statistical quantile of the clustering degree.

[0071] In practice, the degree of clustering of data points within the risk assessment space is calculated by obtaining the coordinate values ​​of each historical disease case data point in the risk assessment space. These coordinate values ​​are numerical vectors containing the values ​​of each feature dimension. For each data point, the number of neighboring data points within a preset radius is calculated as the local density value of that data point. The preset radius is adaptively determined by the spatial range and the overall density of the data points. The local density values ​​of all data points are sorted to generate a density distribution curve. The density distribution curve uses the local density value as the x-axis and the frequency of data points with that local density value as the y-axis. Peak regions in the density distribution curve are identified. These peak regions correspond to highly clustered areas of data points and are determined by finding points where the first derivative of the curve is zero and the second derivative is negative. Based on the boundary coordinates of the highly clustered areas, the spatial range of the distribution pattern is determined. The boundary coordinates are defined by the minimum and maximum values ​​of the data points in each dimension within the highly clustered areas.

[0072] In some embodiments, local density values The calculation uses the kernel density estimation method, and the formula is expressed as:

[0073] ;

[0074] in: It is the local density value of data point k. It is the total number of historical disease case data points in the risk assessment space. and These are the coordinate vectors of data point k and data point m, respectively. This represents the Euclidean distance between two coordinate vectors. The bandwidth parameter is used to control the smoothness of the kernel function. It is a kernel function, such as the Gaussian kernel function.

[0075] In some embodiments, the selection of the preset radius range is based on the statistical distribution of distances between data points, for example, taking a certain proportion of the median distances between all data points. It can be understood that the preset radius affects the granularity of local density estimation.

[0076] Optionally, peak region identification employs a peak detection algorithm, which locates peaks by comparing the local density value of each point with the local density values ​​of its neighboring points. Optionally, the boundary coordinates of highly clustered regions can be obtained by calculating the region's convex hull or axial bounding box.

[0077] Example 3: In a specific implementation, the real-time detection module acquires real-time patient detection data, extracts an immediate feature set from it, and projects the immediate feature set onto the risk assessment space to measure its proximity to the warning feature region, thereby obtaining the current risk index. In the specific implementation, the real-time detection module captures the patient's dermoscopic images and clinical symptom descriptions in real time. The dermoscopic images are acquired through a digital imaging device, and the clinical symptom descriptions are obtained through structured forms or natural language input. The real-time detection data includes pixel data and text information. Feature values ​​corresponding to the multi-dimensional feature sequence are extracted from the real-time detection data to form the immediate feature set. The feature extraction process involves image processing algorithms to extract lesion size and color features, and text analysis techniques to parse symptom keywords, ensuring that the feature dimensions of the immediate feature set are consistent with the historical multi-dimensional feature sequence. The immediate feature set is standardized and then mapped to the risk assessment space. The standardization process uses the Z-score method to scale the feature values ​​to zero mean and unit variance. The mapping operation is achieved by placing the standardized feature vectors as coordinate points into the risk assessment space. The minimum distance between the mapping point and the boundary of the warning feature area is calculated and used as the degree of proximity. This distance is then converted into the current risk index. The minimum distance calculation is based on a spatial geometric algorithm. The current risk index maps the distance value to the risk score range through a linear or nonlinear function.

[0078] In specific implementations, the minimum distance between the mapping point and the boundary of the early warning feature region is calculated using a spatial search algorithm. The boundary of the early warning feature region is defined by a set of boundary points or a mathematical expression, and the minimum distance is the minimum value among the distances from the mapping point to all points on the boundary. In some embodiments, the minimum distance... The calculation formula is expressed as:

[0079] ;

[0080] in: It is the minimum distance between the mapping point and the boundary of the warning feature area. It is the coordinate vector of the mapping point in the risk assessment space. It is the coordinate vector of any point on the boundary of the warning feature area. It is the set of points at the boundary of the warning feature area. Representing vectors and The Euclidean distance between them. This formula provides the mathematical basis for distance measurement. In some embodiments, the boundary of the warning feature region is generated from the distribution of historical data points using a convex hull algorithm, and the set of boundary points is stored as a list of coordinates. During the generation of the warning feature region boundary, the specific implementation of the convex hull algorithm is based on the distribution of historical disease case data points in the risk assessment space. This algorithm is used to calculate the minimum convex polygon boundary from the set of data points. In practice, firstly, the coordinate values ​​of all historical disease case data points are extracted from the highly clustered areas identified within the risk assessment space. These coordinate values ​​represent the positions of the multi-dimensional feature sequence in space. Then, the convex hull algorithm is applied to sort and iteratively process the data points, gradually constructing the convex hull boundary. The algorithm determines the convex hull vertices by comparing the polar angles or distance relationships of the data points. Finally, the coordinates of the convex hull vertices are stored as a list of boundary points. This list is used to define the geometric range of the warning feature region and is integrated with the distance calculation operation in the real-time detection module.

[0081] Optionally, standardization can be performed using a min-max scaling method to normalize feature values ​​to a specified range. Alternatively, the proximity metric can be converted into a current risk indicator using a sigmoid function to invert the distance value into a risk value, resulting in a smaller distance indicating a higher risk.

[0082] Example 4: In specific implementation, building a disease probability prediction system includes constructing a prediction model. This model is constructed by selecting confirmed disease cases and healthy cases from a unified health dataset as training sample sets. Confirmed disease cases have clear disease diagnosis labels, while healthy cases are screened and confirmed to be free of the target dermatological disease. Multi-dimensional feature sequences and feature variation network structure information are extracted from each sample in the training sample set. The multi-dimensional feature sequences serve as the model input features, and the feature variation network structure information is used to define the association constraints between features. The network parameters of the prediction model are initialized, using the multi-dimensional feature sequences as input features and the disease occurrence label as the output target. The disease occurrence label is a binary classification variable. The edge connections in the feature variation network are used to define the time constraints between features, which are reflected in the model's loss function. The model parameters are iteratively optimized using a gradient descent algorithm to make the predicted output approximate the true disease occurrence label, thus completing the training of the prediction model.

[0083] In practical implementation, referring to Table 1, the prediction model utilizes a feature change network to learn the evolution patterns of features and outputs a disease occurrence probability value, which is a continuous value between 0 and 1. The patient's immediate feature set is mapped to corresponding nodes in the feature change network; this mapping process is based on feature identifier matching. The temporal change path of feature values ​​is traced along the edge connection directions in the feature change network; this temporal change path consists of a series of feature nodes connected in chronological order. The matching degree between the change path of the immediate feature set and the feature paths of disease cases in the training samples is calculated; this matching degree calculation involves comparing the similarity of the path sequences. A disease occurrence probability value is generated by weighting the disease occurrence label based on the matching degree; the weights are proportional to the matching degree.

[0084] Table 1: Structure of the Training Sample Set

[0085] In some embodiments, path matching degree The calculation formula is expressed as:

[0086] ;

[0087] in: It is the path matching degree. It is the length of the path being compared. It is the feature node of the instantaneous feature set change path at step l. It is the feature node of the training sample feature path at step l. This is a comparison function that returns 1 if the feature values ​​of two nodes are equal within the allowable error range, and 0 otherwise. This formula quantifies the similarity between two paths. In some embodiments, the gradient descent algorithm employs stochastic gradient descent or the Adam optimizer to minimize the prediction error.

[0088] Optionally, the disease occurrence probability value can be generated using a weighted voting mechanism based on matching degree, where the voting weight of each training sample is its path matching degree. Optionally, the edge weights in the feature change network can be used to adjust the importance of different feature evolutions during path tracing. It can be understood that the disease occurrence probability value provides a quantitative basis for generating customized pre-screening reports.

[0089] See Figure 4This figure illustrates the changing trends of training and validation losses during the training process of the prediction model. The horizontal axis represents the number of training iterations, and the vertical axis represents the loss value. The loss value decreases rapidly in the early stages of training, indicating that the model quickly learns the feature patterns. As iterations progress, the loss value gradually converges, and the gap between the training and validation losses remains small, demonstrating that the model does not suffer from overfitting or underfitting and exhibits good generalization ability. This figure is a core visualization result of the disease probability prediction system construction process: the model learns the correlation between multi-dimensional feature sequences and feature change networks, gradually optimizing parameters to improve prediction accuracy. The stable convergence of the loss curve proves that the model can effectively mine the evolutionary logic of dermatological disease features, providing algorithmic performance support for subsequently outputting reliable disease occurrence probabilities.

[0090] Example 5: In specific implementation, the patient's immediate feature set is mapped to corresponding nodes in the feature change network. The mapping process is achieved by matching node labels with feature identifiers. Each node stores statistical information of historical feature values. The temporal change path of feature values ​​is traced along the edge connection direction in the feature change network. The temporal change path is obtained by recursively traversing the starting node according to the edge pointers, and the path records the node visit sequence. The matching degree between the change path of the immediate feature set and the feature path of disease cases in the training samples is calculated. The matching degree calculation involves comparing the similarity of feature values ​​of the two paths at corresponding nodes. Based on the matching degree, a disease occurrence probability value is generated by weighted averaging of the disease occurrence label. The weight of each training sample in the weighted average is its path matching degree, and the disease occurrence label is a binary variable.

[0091] In practice, customized pre-screening reports are generated based on probability values ​​by setting probability threshold ranges corresponding to different risk levels. These threshold ranges are divided based on the disease incidence probability distribution of historical samples. The patient's risk level is determined by comparing the disease incidence probability value with the probability threshold range; this comparison uses an interval matching algorithm. A corresponding template is selected from a predefined report template library that stores text template files corresponding to different risk levels. The patient's multi-dimensional feature sequence, current risk indicators, and risk level description are filled into the corresponding fields in the template to generate a customized pre-screening report containing a risk summary and recommended measures. Field filling is completed through string replacement. In some embodiments, the disease incidence probability value... The calculation formula is expressed as: ;

[0092] in: It is the probability value of disease occurrence. It is the total number of disease cases in the training sample. It is the path matching weight of the s-th training sample. This represents the disease occurrence label of the s-th training sample. This formula implements probability estimation based on similarity. In some embodiments, the probability threshold range is divided into three levels: low risk, medium risk, and high risk, each corresponding to a different probability value range. This risk level division facilitates risk stratification management.

[0093] Optionally, the templates in the report template library are defined using Extensible Markup Language (XML) format and include reserved field tags for data population. Optionally, the output format of the customized pre-screened report can be either Portable Document format or Hypertext Markup Language (HTML) format. It can be understood that customized pre-screened reports provide structured information support for medical decision-making.

[0094] See Figure 5 This chart, presented as a bar chart, displays the number of cases corresponding to different risk levels. The data distribution shows that low-risk cases account for the largest proportion, while high-risk cases are relatively fewer, reflecting the reality in clinical practice where most patients with chronic dermatological diseases have stable conditions, while a minority require intensive intervention. This chart is a core visualization result of the risk stratification management process: the system classifies patients into different risk levels based on a threshold range of disease incidence probability. This distribution data not only validates the rationality of the risk assessment system but also provides a basis for the precise allocation of clinical resources. Although high-risk cases are fewer in number, they require priority for close monitoring and early intervention, while low-risk cases can be managed with routine follow-up strategies. This chart also intuitively demonstrates the risk stratification capability of the pre-screening system and provides key data support for the risk level description field in customized reports.

[0095] 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 pre-screening system for dermatological diseases, characterized in that, The system includes: The data acquisition module is used to collect long-term medical monitoring data from dermatology patients and to integrate and process this data to form a unified health data set. The feature network construction module is used to perform feature mining operations on the unified health data set to obtain multi-dimensional feature sequences, and to construct a feature change network based on the multi-dimensional feature sequences. The risk assessment space generation module is used to generate a risk assessment space based on the feature change network, analyze the distribution pattern of historical disease cases in the risk assessment space, and delineate early warning feature areas according to the distribution pattern. The real-time detection module is used to acquire real-time detection data of patients, extract instantaneous feature sets from them, project the instantaneous feature sets onto the risk assessment space, measure their proximity to the warning feature area, and obtain the current risk index. The prediction and report generation module is used to combine the current risk indicators, the multi-dimensional feature sequences, and the feature change network to build a disease probability prediction system and generate customized pre-screening reports.

2. The dermatological disease pre-screening system according to claim 1, characterized in that, The collection of long-term medical monitoring data from dermatology patients, and the integration and processing of this data to form a unified health data set, includes: Collect medical records, image records, and physiological indicators of dermatology patients from medical platforms; Perform quality checks on the collected data, including removing duplicate records and correcting formatting errors; The processed data is archived according to disease type and time sequence to generate a unified health data set.

3. The dermatological disease pre-screening system according to claim 2, characterized in that, The step of performing feature mining on the unified health data set to obtain multi-dimensional feature sequences, and constructing a feature change network based on the multi-dimensional feature sequences, includes: Filter parameters related to skin diseases from a unified health dataset; Calculate the numerical changes of each parameter at different time points to form a multi-dimensional feature sequence; The multi-dimensional feature sequence is converted into a graph model, where nodes represent feature values ​​and edges represent the time dependencies between features, thus constructing a feature change network.

4. The dermatological disease pre-screening system according to claim 3, characterized in that, The process of generating a risk assessment space based on the feature change network, analyzing the distribution patterns of historical disease cases within the risk assessment space, and delineating early warning feature regions based on the distribution patterns includes: The dimensions of the risk assessment space are determined based on the number of features in the feature-changing network. Incorporate characteristic data points from historical disease cases into the risk assessment space; Calculate the clustering degree of data points within the risk assessment space to obtain the distribution pattern; A threshold value is set based on the distribution pattern, and areas with a clustering degree higher than the threshold value are marked as early warning characteristic areas.

5. A dermatological disease pre-screening system according to claim 4, characterized in that, The calculation of the clustering degree of data points within the risk assessment space to obtain the distribution pattern is achieved through the following steps: Obtain the coordinates of each historical disease case data point in the risk assessment space; For each data point, calculate the number of neighboring data points within a preset radius, and use this as the local density value of that data point. Sort the local density values ​​of all data points to generate density distribution curves; Identify peak regions in the density distribution curve, where the peak regions correspond to areas of high data point aggregation. The spatial range of the distribution pattern is determined based on the boundary coordinates of the highly concentrated areas.

6. The dermatological disease pre-screening system according to claim 1, characterized in that, The process of acquiring real-time patient monitoring data, extracting an instantaneous feature set from it, projecting the instantaneous feature set onto the risk assessment space, measuring its proximity to the warning feature region, and obtaining the current risk indicator includes: Real-time capture of dermoscopy images and descriptions of clinical symptoms in patients; Feature values ​​corresponding to multi-dimensional feature sequences are extracted from real-time detection data to form an instantaneous feature set; The instantaneous feature set is standardized and then mapped to the risk assessment space; The minimum distance between the mapping point and the boundary of the warning feature area is calculated and used as the degree of proximity, which is then converted into the current risk indicator.

7. A dermatological disease pre-screening system according to claim 6, characterized in that, The system combines the current risk indicators, the multi-dimensional feature sequences, and the feature change network to construct a disease probability prediction system and generate a customized pre-screening report, including: Input the current risk indicators and multi-dimensional feature sequences into the prediction model; The predictive model uses a feature change network to learn the evolution of features and outputs a probability value of disease occurrence. Customized pre-screening reports are generated based on probability values.

8. A dermatological disease pre-screening system according to claim 7, characterized in that, The construction of the disease probability prediction system includes building a prediction model, which is achieved through the following steps: Confirmed disease cases and healthy cases were selected from a unified health dataset as the training sample set; Extract multi-dimensional feature sequences and feature variation network structure information for each sample in the training sample set; Initialize the network parameters of the prediction model, using multi-dimensional feature sequences as input features and disease occurrence labels as output targets; Use the edge connections in the feature variation network to define the time constraints between features; The model parameters are iteratively optimized using the gradient descent algorithm to make the predicted output close to the actual disease occurrence label, thus completing the training of the prediction model.

9. A dermatological disease pre-screening system according to claim 8, characterized in that, The prediction model utilizes a feature change network to learn the evolution patterns of features and outputs a disease occurrence probability value, which is achieved through the following steps: Map the patient's immediate feature set to the corresponding node in the feature change network; Tracing the temporal variation path of feature values ​​along the edge connection directions in the feature variation network; Calculate the degree of matching between the change path of the instantaneous feature set and the feature path of disease cases in the training samples; The probability value of disease occurrence is generated by weighting the disease occurrence labels based on the matching degree.

10. A dermatological disease pre-screening system according to claim 7, characterized in that, The process of generating a customized pre-screening report based on probability values ​​is achieved through the following steps: Set probability threshold ranges corresponding to different risk levels; By comparing the probability value of disease occurrence with the probability threshold range, the patient's risk level is determined. Select the appropriate template from the predefined report template library based on the risk level; By filling in the patient's multidimensional characteristic sequence, current risk indicators, and risk level description into the corresponding fields in the template, a customized pre-screening report containing a risk summary and recommended measures is generated.

Images