An osteoporosis detection method and system based on data analysis and artificial intelligence

Abstract

The present application relates to the technical field of multi-cause intelligent analysis in medical diagnosis, and particularly relates to an osteoporosis detection method and system based on data analysis and artificial intelligence. The method comprises the following steps: S1: obtaining primary osteoporosis data and secondary osteoporosis data and determining synergistic data, and determining antagonistic masking data according to the synergistic data; S2: obtaining a first layer dominant right conversion sequence according to the synergistic data and the antagonistic masking data; S3: obtaining a synergistic data internal dominant relationship sequence and an antagonistic masking data internal dominant relationship sequence respectively according to the synergistic data and the antagonistic masking data, and generating a complete dominant relationship graph in combination with the first layer dominant right conversion sequence. The present application separates and analyzes the synergistic and antagonistic modes of primary and secondary osteoporosis, constructs a dynamic dominant right sequence, quantifies the evolution of mixed cause weights, and provides data support for individualized precise diagnosis and treatment.

Description

Technical Field

[0001] This invention relates to the field of multi-causal intelligent analysis technology in medical diagnosis, and in particular to a method and system for osteoporosis detection based on data analysis and artificial intelligence. Background Technology

[0002] Data analytics and artificial intelligence have become core technologies for improving the accuracy of medical diagnosis, deeply mining hidden information in medical data through pattern recognition and predictive modeling. In the field of osteoporosis detection, traditional methods rely on a single indicator, bone mineral density. However, existing detection technologies based on data analytics and artificial intelligence integrate imaging features and clinical biochemical indicators to construct multi-parameter risk assessment models, significantly improving the sensitivity of early screening and the accuracy of fracture risk prediction, thus laying the foundation for intelligent diagnosis of osteoporosis.

[0003] However, existing artificial intelligence methods have limitations in addressing the complex clinical reality of mixed primary and secondary osteoporosis. Current technologies often model multi-source data as a homogeneous whole, lacking the ability to analyze the intrinsic interaction mechanisms between the two types of etiologies. They fail to effectively distinguish between the two key interaction patterns of synergistic amplification and antagonistic masking, leading to biases in etiological attribution and dominant judgment, thus limiting their practical application value in personalized precision medicine.

[0004] Therefore, there is an urgent need for a new data analysis method that can resolve the complex interaction between primary and secondary osteoporosis, in order to overcome the shortcomings of existing technologies in analyzing mixed etiologies and to achieve accurate quantification of etiological contributions and dynamic identification of dominant factors. Summary of the Invention

[0005] To overcome the shortcomings of insufficient analysis of mixed etiologies, this invention provides an osteoporosis detection method and system based on data analysis and artificial intelligence.

[0006] The technical implementation scheme of the present invention is: an osteoporosis detection method based on data analysis and artificial intelligence, comprising the following steps: S1: Obtain primary osteoporosis data and secondary osteoporosis data and determine synergistic data, and determine antagonistic masking data based on synergistic data; S2: Obtain the first-layer dominance transition sequence based on collaborative data and antagonistic masking data; S3: Obtain the dominant relationship sequence within the collaborative data and the dominant relationship sequence within the antagonistic masking data respectively, and generate a complete dominant relationship map by combining the first-layer dominant power transfer sequence; S4: Correct the complete dominant relationship graph to obtain the final dominant relationship graph.

[0007] Preferably, the step of acquiring primary osteoporosis data and secondary osteoporosis data and determining synergistic data includes: The correlation between primary osteoporosis data and secondary osteoporosis data was calculated using the Pearson correlation coefficient. Primary osteoporosis data and corresponding secondary osteoporosis data that exhibited a synergistic amplification effect were labeled as synergistic data.

[0008] Preferably, determining the antagonistic masking data based on the cooperative data includes: From the primary osteoporosis data and the secondary osteoporosis data, synergistic data were removed, and the remaining data were marked as non-synergistic data; A linear model is used to calculate the data correlation between different dimensions of non-cooperative data and define it as the first correlation. A nonlinear model is used to calculate the data correlation between different dimensions of non-cooperative data and define it as the second correlation. Calculate the difference between the first correlation and the second correlation. If the difference is greater than or equal to a preset difference threshold, the non-cooperative data that produces the difference is marked as antagonistic masking data.

[0009] Preferably, obtaining the first-layer dominance transition sequence based on cooperative data and antagonistic masking data includes: The collaborative data and the antagonistic masking data are sorted in chronological order to obtain the time-series sorting results. Based on the time-series sorting results, the magnitude of change of collaborative data and antagonistic masking data within adjacent time windows is compared. The data type with the largest magnitude of change is determined as the dominant data type of the time window, and the conversion event of the dominant data type is recorded to obtain the first-level dominance conversion sequence.

[0010] Preferably, the step of obtaining the dominance relationship sequence within the collaborative data and the dominance relationship sequence within the antagonistic masking data based on the collaborative data and the antagonistic masking data respectively, and generating a complete dominance relationship graph by combining the first-layer dominance transfer sequence, includes: Within the collaborative data, the Pearson correlation coefficient between primary osteoporosis data and secondary osteoporosis data is calculated in chronological order. The data component with the largest absolute value of the Pearson correlation coefficient is identified as the dominant data component within the time window of the collaborative data, and the corresponding transformation event is recorded to obtain the dominant relationship sequence within the collaborative data. Within the antagonistic masking data, the difference between the first and second correlation values ​​between primary osteoporosis data and secondary osteoporosis data is calculated in chronological order. The data component with the largest difference value is identified as the dominant data component of the time window within the antagonistic masking data, and the corresponding transformation event is recorded to obtain the dominant relationship sequence within the antagonistic masking data. By merging the first-level dominance transition sequence, the dominance relationship sequence within the collaborative data, and the dominance relationship sequence within the antagonistic masking data, a complete dominance relationship map between primary osteoporosis data and secondary osteoporosis data is generated.

[0011] Preferably, the step of modifying the complete dominance relationship graph to obtain the final dominance relationship graph includes: Extract the set of time windows corresponding to the unexpected phenomena and mark them as the unexpected time window set; Extract the time window set corresponding to the contradictory data phenomena and mark it as the contradictory time window set; The unexpected phenomenon refers to the nonlinear synergistic amplification effect between primary osteoporosis data and secondary osteoporosis data, which causes the rate of bone loss or fracture risk indicators to exceed the prediction range of independent factors. The contradictory data phenomenon refers to the conflict between the dominant conclusions indicated by primary osteoporosis data and secondary osteoporosis data within the same time window or anatomical site, resulting in conflicting data indicators. From the complete dominant relationship map, extract the primary osteoporosis data sequence and the secondary osteoporosis data sequence corresponding to the unexpected time window set, and mark the two extracted sequences as unexpected endpoint data sequences; From the complete dominant relationship graph, extract the primary osteoporosis data sequence and the secondary osteoporosis data sequence corresponding to the contradictory time window set, and mark the two extracted sequences as contradictory node data sequences; Align the unexpected endpoint data sequence with the contradictory node data sequence.

[0012] Preferably, the step of aligning the unexpected endpoint data sequence with the contradictory node data sequence includes: If the primary osteoporosis data in the contradictory node data sequence is the same as the primary osteoporosis data in the unexpected endpoint data sequence, then the primary osteoporosis data in the contradictory node data sequence is marked as a first-class aligned node. If the secondary osteoporosis data in the contradictory node data sequence is the same as the secondary osteoporosis data in the unexpected endpoint data sequence, then the secondary osteoporosis data in the contradictory node data sequence is marked as a second type of alignment node. For each alignment node, calculate the time span length of the alignment node in the unexpected endpoint data sequence. The time span length is equal to the difference between the time window number of the last occurrence of the alignment node in the unexpected endpoint data sequence and the time window number of the first occurrence. For each alignment node, calculate the frequency of the alignment node in the data sequence of conflicting nodes; The attribution fuzziness resolution coefficient is calculated based on the type of alignment node, the time span length of the alignment node, and the frequency of occurrence of the alignment node.

[0013] Preferably, the step of calculating the attribution fuzzification resolution coefficient based on the type of alignment node, the time span length of the alignment node, and the frequency of occurrence of the alignment node includes: For the first type of aligned nodes, the attribution fuzziness resolution coefficient is equal to the product of the time span length of the aligned node and the frequency of occurrence of the aligned node; For the second type of aligned nodes, the attribution fuzziness resolution coefficient is equal to the ratio of the time span length of the aligned node to the frequency of occurrence of the aligned node; The complete dominant relationship graph is corrected by using the attribution fuzziness resolution coefficient.

[0014] Preferably, the step of correcting the complete dominant relationship graph using attribution fuzziness resolution coefficients includes: For time windows with attribution ambiguity, if the corresponding primary or secondary osteoporosis data is an alignment node, the attribution ambiguity resolution coefficient of the corresponding alignment node is added to the dominance strength of the corresponding data in the dominance relationship graph. Based on the revised dominance strength, a final dominance relationship map between primary osteoporosis data and secondary osteoporosis data is regenerated.

[0015] Preferably, an osteoporosis detection system based on data analysis and artificial intelligence includes: The osteoporosis data collaborative analysis module is used to acquire primary osteoporosis data and secondary osteoporosis data, calculate Pearson correlation coefficients and label collaborative data, thereby identifying antagonistic masking data; The dominance transition sequence generation module is used to sort the cooperative data and antagonistic masking data in time sequence, compare the intensity change magnitude and record the transition events to obtain the first layer of dominance transition sequence. The dominant relationship graph construction module is used to calculate the Pearson correlation coefficient within the collaborative data and the difference in association within the antagonistic masking data, generate the internal dominant relationship sequence, and merge them into a complete dominant relationship graph. The graph correction and output module is used to extract unexpected and contradictory data sequences, calculate the attribution fuzziness resolution coefficient by aligning nodes, correct the dominance strength, and generate the final dominance relationship graph.

[0016] Beneficial Effects: This invention employs a separate synergistic and antagonistic analysis of primary and secondary osteoporosis data to identify and label two core interaction patterns: synergistic amplification and antagonistic masking. This constructs a first-layer dominance transfer sequence characterizing the dynamic game relationship between the two. Further, temporal dominance analysis is implemented within each pattern, and the preliminary analysis map is precisely corrected by integrating the cross-influence quantification results of unexpected phenomena and contradictory data phenomena. Ultimately, a dominance relationship map that fully reflects the dynamic evolution of mixed etiological weights is generated. This invention changes the traditional analytical paradigm that treats multiple etiologies as a static, homogeneous mixture, achieving a layer-by-layer deconstruction and quantification of complex interaction mechanisms. This provides a reliable data-driven decision-making basis for individualized etiological diagnosis, dynamic staging of disease course, and the formulation of precision treatment strategies for clinical osteoporosis. Attached Figure Description

[0017] Figure 1 This is a flowchart of the osteoporosis detection method based on data analysis and artificial intelligence according to the present invention; Figure 2 This is a structural diagram of the osteoporosis detection system based on data analysis and artificial intelligence of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0019] Example 1: A method for osteoporosis detection based on data analysis and artificial intelligence, such as... Figure 1 As shown, it includes the following steps: S1-1: Obtain primary osteoporosis data and secondary osteoporosis data and identify synergistic data, including: The correlation between primary osteoporosis data and secondary osteoporosis data was calculated using the Pearson correlation coefficient. Primary osteoporosis data and corresponding secondary osteoporosis data that exhibited a synergistic amplification effect were labeled as synergistic data.

[0020] It should be noted that this step aims to identify combinations of risk factors that can mutually exacerbate the condition from diagnostic data. Primary osteoporosis data typically comes from bone mineral density tests such as T-scores and Z-scores obtained from DXA scans, as well as skeletal geometry parameters; secondary osteoporosis data includes laboratory biochemical indicators such as vitamin D levels and parathyroid hormone concentrations, and medication records such as the dosage and duration of glucocorticoid treatment. The consistency of these two types of data over time is quantified using the Pearson correlation coefficient. The correlation represented by this coefficient essentially reflects the dynamic coupling strength of primary and secondary factors: a high correlation means that their trends are synchronized and they can jointly drive bone loss; a low correlation indicates that they are relatively independent. The synergistic amplification effect specifically refers to the fact that when two types of causative factors coexist, the resulting rate of bone loss or fracture risk is significantly higher than the simple sum of the expected values ​​of the independent effects of each factor. This effect usually manifests as a strong positive correlation at the data level. In this invention, a Pearson correlation coefficient greater than 0.7, or another statistical significance threshold determined based on the distribution of a specific clinical dataset, such as p < 0.01, is defined as a strong positive correlation, and this serves as the basis for determining synergistic data. Labeling data pairs with such strong correlation characteristics as synergistic data lays the foundation for further in-depth analysis of the core links that play a synergistic aggravating role in the etiological interaction network, thereby overcoming the limitations of traditional diagnostic methods that treat multiple etiologies as isolated factors and fail to capture their combined effects.

[0021] S1-2: Determine antagonistic masking data based on cooperative data, including: From the primary osteoporosis data and the secondary osteoporosis data, synergistic data were removed, and the remaining data were marked as non-synergistic data; A linear model is used to calculate the data correlation between different dimensions of non-cooperative data and define it as the first correlation. A nonlinear model is used to calculate the data correlation between different dimensions of non-cooperative data and define it as the second correlation. Calculate the difference between the first correlation and the second correlation. If the difference is greater than or equal to a preset difference threshold, the non-cooperative data that produces the difference is marked as antagonistic masking data.

[0022] It is important to note that, to identify data interaction patterns beyond the scope of synergy, the identified synergistic data must first be excluded from the primary and secondary osteoporosis datasets. This step aims to focus subsequent analysis on the remaining data containing antagonistic or masking relationships, i.e., non-synergistic data. For this data, the linear and non-linear association characteristics need to be evaluated in parallel. The first association obtained by the linear model reflects the most direct, proportional trend between variables; in contrast, the second association extracted by the non-linear model captures a more complex and dynamic interaction between variables. Calculating the difference between the first and second associations essentially measures the degree of deviation of the complexity of the data relationship from a simple linear model. An increase in this difference value usually indicates the presence of a strong non-linear interference mechanism between the data; for example, severe spinal degeneration, a secondary factor, may mask the true extent of primary bone loss in radiographic measurements. This indication signifies a significant mutual interference effect in the data combination, where traditional linear analysis can easily lead to misleading conclusions. The linear model used is a multiple linear regression model, with independent and dependent variables representing data sequences of different dimensions from the non-synergistic data. The nonlinear model selected is the random forest regression model, as it can effectively capture complex nonlinear relationships. The preset difference threshold is determined as follows: On historical data, a large number of data pairs with clear, non-antagonistic masking relationships are calculated, such as those from healthy controls or patients with a clear single etiology, to determine the difference between linear and nonlinear associations. A specific high percentile of this difference value distribution, such as the 95th percentile, is set as the threshold. If the difference value of the current data pair exceeds this threshold, it is considered to exceed the interpretable range of a conventional linear relationship, indicating a significant antagonistic or masking effect. This step, by actively detecting and labeling such complex interference patterns, directly addresses the shortcomings of existing diagnostic techniques in analyzing mutual masking phenomena between etiologies, establishing an analytical foundation for clearly separating the mutually confusing influencing components in mixed etiologies.

[0023] S2: Obtain the first-layer dominance transition sequence based on cooperative data and antagonistic masking data, including: The collaborative data and the antagonistic masking data are sorted in chronological order to obtain the time-series sorting results. Based on the time-series sorting results, the magnitude of change of collaborative data and antagonistic masking data within adjacent time windows is compared. The data type with the largest magnitude of change is determined as the dominant data type of the time window, and the conversion event of the dominant data type is recorded to obtain the first-level dominance conversion sequence.

[0024] It should be noted that the time window is set according to the clinical monitoring cycle, such as monthly or quarterly. Its core function is to provide a unified time-series benchmark for quantitative analysis, thereby dividing the continuous course of disease into comparable stages to characterize the dynamic evolution of etiological interaction patterns. To characterize the evolutionary trajectory of different etiological interaction patterns over time, synergistic and antagonistic masking data are first integrated and sorted according to their corresponding time windows. The resulting time-series sorting results provide a unified time scale for observing the intensity changes of the two key interaction relationships. Based on this result, by examining the magnitude of changes in synergistic and antagonistic masking data within adjacent time windows, the most active data types with the most significant changes in influence during a specific period can be identified. The magnitude of change mentioned here directly reflects the fluctuation intensity of the corresponding data type between adjacent time points; an increase in the magnitude value means that the influence of this type is in a period of rapid increase or decrease. Within each time window, the data type with the largest magnitude of change is identified as the dominant data type. This determination reveals the core interaction mechanism that influences the development of the disease during that period. When the dominant data type changes between different time windows, a dominance transition event occurs, such as a shift in the disease-driving mechanism from a synergistic amplification effect to an antagonistic masking effect. All such events are systematically recorded and ultimately aggregated into a first-level dominance transition sequence. The comparison of the changes in synergistic and antagonistic masking data within adjacent time windows is implemented as follows: First, both synergistic and antagonistic masking data are standardized, such as using Z-score standardization to eliminate the influence of dimensions, and representative indicators of the overall strength of each data type within each time window are calculated, such as the average Pearson correlation coefficient for synergistic data and the average difference value for antagonistic masking data. Then, the absolute changes in the strength indicators of synergistic and antagonistic masking data between adjacent time windows are calculated. Finally, the magnitudes of these two absolute changes are compared, and the one with the larger change value is determined as the dominant data type during that transition period. This sequence constitutes a high-level dynamic framework for describing the waxing and waning of the power of synergistic and antagonistic interaction modes, thus breaking through the limitation of treating the etiological role as static and unchanging in the existing diagnostic assessment, and providing key temporal evidence for realizing the dynamic stage division and mechanism interpretation of complex osteoporosis course.

[0025] S3: Obtain the dominance relationship sequence within the collaborative data and the dominance relationship sequence within the antagonistic masking data based on the collaborative data and the antagonistic masking data respectively, and generate a complete dominance relationship graph by combining it with the first-layer dominance transfer sequence, including: Within the collaborative data, the Pearson correlation coefficient between primary osteoporosis data and secondary osteoporosis data is calculated in chronological order. The data component with the largest absolute value of the Pearson correlation coefficient is identified as the dominant data component within the time window of the collaborative data, and the corresponding transformation event is recorded to obtain the dominant relationship sequence within the collaborative data. Within the antagonistic masking data, the difference between the first and second correlation values ​​between primary osteoporosis data and secondary osteoporosis data is calculated in chronological order. The data component with the largest difference value is identified as the dominant data component of the time window within the antagonistic masking data, and the corresponding transformation event is recorded to obtain the dominant relationship sequence within the antagonistic masking data. By merging the first-level dominance transition sequence, the dominance relationship sequence within the collaborative data, and the dominance relationship sequence within the antagonistic masking data, a complete dominance relationship map between primary osteoporosis data and secondary osteoporosis data is generated.

[0026] It should be noted that, to deeply analyze the underlying dominant mechanisms of the two interaction modes, synergy and antagonistic masking, further temporal analysis is required within each mode. For synergistic data, Pearson correlation coefficients between primary and secondary data are continuously calculated along the time axis. Within each time window, the data component with the largest absolute value of the correlation coefficient is identified as the dominant data component, indicating that this component bears primary responsibility for the synergistic amplification effect during that period. For example, if secondary data, such as specific biochemical indicators, exhibit the strongest synchronicity with the bone loss process within a certain period, it is marked as the dominant party in that synergistic phase. By tracking the temporal changes of such dominant markers, a sequence of dominant relationships within the synergistic data is obtained, thereby revealing the evolution of the core driving force behind the synergistic effect.

[0027] For antagonistic masking data, the analysis focuses on the temporal calculation of the difference between the first and second correlation values. The magnitude of the difference directly measures the strength of the masking effect; therefore, the data component with the largest difference value is identified as the dominant data component, which can identify the key factors causing measurement distortion or effect cancellation. Assuming that at a certain period, the imaging artifacts caused by a specific secondary lesion have the most prominent masking effect on true bone density, the data corresponding to that lesion is considered dominant. Recording the succession process of these dominant components forms the dominant relationship sequence within the antagonistic masking data.

[0028] Ultimately, the first-level dominance transfer sequence, depicting the alternation of victory and defeat in macroscopic interaction patterns, is integrated with the two sequences described above, depicting the changes in microscopic dominant factors within various patterns. This integration is not a simple splicing, but rather a hierarchical alignment and fusion based on a unified time benchmark. The resulting complete dominance relationship map can comprehensively and at multiple scales present a panoramic view of the dynamic game and dominance transfer of primary and secondary osteoporosis data throughout the entire disease course. This map overcomes the shortcomings of existing diagnostic models, such as coarse analysis of complex etiological interaction structures and limited attribution capabilities, providing a core basis for realizing personalized dynamic staging of the disease course and the formulation of precise intervention strategies.

[0029] S4: Revise the complete dominant relationship graph to obtain the final dominant relationship graph, including: Extract the set of time windows corresponding to the unexpected phenomena and mark them as the unexpected time window set; Extract the time window set corresponding to the contradictory data phenomena and mark it as the contradictory time window set; The unexpected phenomenon refers to the nonlinear synergistic amplification effect between primary osteoporosis data and secondary osteoporosis data, which causes the rate of bone loss or fracture risk indicators to exceed the prediction range of independent factors. Unexpected phenomena are identified by using an ensemble learning model to perform regression prediction on multiple time-series data points, and by identifying significant deviations between the predicted values ​​and the actual observed values. The contradictory data phenomenon refers to the conflict between the dominant conclusions indicated by primary osteoporosis data and secondary osteoporosis data within the same time window or anatomical site, resulting in conflicting data indicators. The contradictory data phenomenon is addressed by employing a dual-branch neural network based on an attention mechanism for joint representation and consistency calculation of multi-source data. Conflicts are quantified and identified by calculating the difference or distance between the output representations of the two branches. From the complete dominant relationship map, extract the primary osteoporosis data sequence and the secondary osteoporosis data sequence corresponding to the unexpected time window set, and mark the two extracted sequences as unexpected endpoint data sequences; From the complete dominant relationship graph, extract the primary osteoporosis data sequence and the secondary osteoporosis data sequence corresponding to the contradictory time window set, and mark the two extracted sequences as contradictory node data sequences; Align the unexpected endpoint data sequence with the contradictory node data sequence.

[0030] It should be noted that the starting point for revising the complete dominant relationship map lies in systematically locating and extracting data segments containing specific clinical significance. To this end, it is necessary to construct separate time window sets corresponding to unexpected phenomena and contradictory data phenomena. The generation of the unexpected phenomenon set relies on a pre-trained ensemble learning prediction model. This model learns the evolution patterns of bone metabolism indicators from a large amount of historical disease data, thereby making a baseline prediction of the current bone loss trend in patients. When the actual collected monitoring data, such as fracture risk scores or bone mineral density decline rates, show a sustained and significant prediction deviation, an unexpected phenomenon is identified, and the corresponding time period is included in this set. The criteria for determining significant deviation are: the actual observed value exceeds the model's prediction interval, such as the 95% confidence interval, or the absolute value of the prediction error is greater than the 95th percentile of the historical prediction error distribution.

[0031] The identification of contradictory data sets employs a dual-branch neural network architecture, which can process primary and secondary data streams in parallel and evaluate the consistency of their deep representations using an internal attention mechanism. Once contradictory data from the same period or the same anatomical site are found to contradict each other on the dominant conclusion, it signifies the occurrence of contradictory data phenomena, and the relevant time window is included in the set.

[0032] Based on the aforementioned set, the next step is to precisely locate and extract the corresponding raw data streams from the dominant relationship graph. Specifically, from the time period covered by the unexpected time window set, primary osteoporosis data and secondary osteoporosis data arranged chronologically are extracted, together forming an unexpected endpoint data sequence. This sequence carries the core data interaction information that leads to the deviation of disease progression from conventional expectations. Similarly, from the time period covered by the contradictory time window set, the extracted primary and secondary data constitute a contradictory node data sequence, which centrally reflects the data conflict ontology that leads to dilemmas in clinical judgment. Traditional analysis methods often struggle to properly handle such unconventional data patterns or mitigate their impact. The ensemble learning model, for example, employs a gradient boosting decision tree model, whose input is the primary and secondary data features from multiple time points, and whose output is a predicted value for bone loss rate or fracture risk. The model is trained on a large-scale historical dataset containing clear outcome indicators. The attention-based dual-branch neural network has one branch processing primary osteoporosis data and the other processing secondary osteoporosis data. Each branch contains a feature extraction layer and an attention layer to generate weighted representations of its respective data. Finally, consistency is quantified by calculating the cosine or Euclidean distance between the output representations of the two branches. If this distance is greater than a preset conflict threshold calibrated using historical data, contradictory data is identified. This invention, by selectively extracting and separating these key sequences, creates the necessary conditions for subsequent steps to specifically quantify their impact, thereby calibrating and optimizing the preliminary analysis map.

[0033] Data alignment is performed between the unexpected endpoint data sequence and the contradictory node data sequence, including: If the primary osteoporosis data in the contradictory node data sequence is the same as the primary osteoporosis data in the unexpected endpoint data sequence, then the primary osteoporosis data in the contradictory node data sequence is marked as a first-class aligned node. If the secondary osteoporosis data in the contradictory node data sequence is the same as the secondary osteoporosis data in the unexpected endpoint data sequence, then the secondary osteoporosis data in the contradictory node data sequence is marked as a second type of alignment node. For each alignment node, calculate the time span length of the alignment node in the unexpected endpoint data sequence. The time span length is equal to the difference between the time window number of the last occurrence of the alignment node in the unexpected endpoint data sequence and the time window number of the first occurrence. For each alignment node, calculate the frequency of the alignment node in the data sequence of conflicting nodes; The attribution fuzziness resolution coefficient is calculated based on the type of alignment node, the time span length of the alignment node, and the frequency of occurrence of the alignment node.

[0034] It's important to note that the core purpose of data alignment is to identify the common driving factors between unexpected and contradictory data phenomena. The identification process unfolds based on data type: if a primary osteoporosis data point within a contradictory endpoint sequence also appears in an unexpected endpoint sequence, that data point is labeled as a first-type alignment node. This labeling signifies that the primary data point is identified as a dual key factor simultaneously involved in both diagnostic conflict and abnormal disease progression. Similarly, if the same secondary osteoporosis data point appears in both sequences, it is labeled as a second-type alignment node, indicating that the secondary factor has a similar combined effect.

[0035] To quantify the influence of these alignment nodes in different phenomena, appropriate metrics need to be used for measurement. For unexpected endpoint data sequences, the focus is on the time span of the alignment nodes. This length directly reflects the duration of the data's activity during the period of unexpected risk; a longer span generally indicates a more stable fundamental role. For contradictory node data sequences, the measurement shifts to the frequency of alignment node occurrence. The frequency directly correlates with how often the data induces diagnostic inconsistencies; a high frequency indicates a persistent source of conflict.

[0036] The formulation of these two differentiated processing strategies stems from consideration of the essential differences between the two types of phenomena. Unexpected phenomena represent the abnormal accumulation of risk, a continuous process; therefore, a time span is used to characterize the persistence of the impact. Contradictory phenomena are more often manifested as discrete, sporadic conflicts of indicators; therefore, frequency of occurrence is used to measure their recurrence. Existing technical frameworks typically view various data anomalies in isolation, lacking the ability for correlation analysis and cross-quantification. This step, by performing precise data alignment and applying targeted indicator quantification, successfully constructs a key analytical link connecting the two types of anomalies and analyzing the common data bridge behind them, thus laying a solid foundation for implementing precise map correction in subsequent steps.

[0037] The attribution fuzzification resolution coefficient is calculated based on the type of alignment node, the time span of the alignment node, and the frequency of occurrence of the alignment node, including: For the first type of aligned nodes, the attribution fuzziness resolution coefficient is equal to the product of the time span length of the aligned node and the frequency of occurrence of the aligned node; For the second type of aligned nodes, the attribution fuzziness resolution coefficient is equal to the ratio of the time span length of the aligned node to the frequency of occurrence of the aligned node; The complete dominant relationship graph is corrected by using the attribution fuzziness resolution coefficient.

[0038] It should be noted that the calculation of the attribution fuzziness resolution coefficient serves the need for quantitative assessment of the aforementioned cross-influence nodes. Since the time span and frequency of occurrence have different dimensions, direct calculation leads to bias. Therefore, these two indicators need to be standardized before calculation to eliminate the influence of dimensions, for example, by using the max-min normalization method to scale each indicator to the [0, 1] interval. Different calculation formulas were designed for different types of alignment nodes to match the different patterns presented by primary and secondary data in the interaction.

[0039] When processing the first type of alignment nodes—primary osteoporosis data that simultaneously associates with unexpected and contradictory phenomena—the solution coefficient is defined as the product of the time span and the frequency of occurrence. The idea behind this calculation is that primary data, as the underlying cause, has both temporal persistence and recurring contradictory manifestations, both of which amplify the resulting diagnostic complexity. Therefore, an increase in the coefficient indicates that the data point, due to its long-term activity and repeated creation of diagnostic conflicts, significantly exacerbates the overall attribution ambiguity.

[0040] For the second type of alignment nodes, namely secondary osteoporosis data, the solution coefficients are obtained by dividing the time span by the frequency of occurrence. Ratio calculations are used to assess the relative relationship between the secondary factor's persistent influence and its frequency of causing discrete conflicts. A high ratio tends to reveal the presence of this secondary factor as a stable background source of interference; while a low ratio points to an intervention factor more likely to cause occasional but prominent diagnostic contradictions. By implementing this discriminative quantification strategy, this step transforms the cross-interference patterns that are difficult to handle in traditional analysis into scalar indicators for subsequent precise atlas calibration, thus directly addressing the challenge of insufficient quantitative tools in the analysis of mixed etiologies.

[0041] The complete dominant relationship graph is corrected using attribution fuzziness resolution coefficients, including: For time windows with attribution ambiguity, if the corresponding primary or secondary osteoporosis data is an alignment node, the attribution ambiguity resolution coefficient of the corresponding alignment node is added to the dominance strength of the corresponding data in the dominance relationship graph. Based on the revised dominance strength, a final dominance relationship map between primary osteoporosis data and secondary osteoporosis data is regenerated.

[0042] It should be noted that the final correction step aims to accurately calibrate specific challenging areas identified in the complete dominance relationship graph. The criteria for determining attribution ambiguity zones are as follows: within a given time window, if the dominance difference within collaborative data is less than a preset collaborative ambiguity threshold, and / or the dominance difference within antagonistic masking data is less than a preset antagonistic ambiguity threshold, and / or the dominance difference between layers is less than a preset inter-layer ambiguity threshold. These ambiguity thresholds are determined by analyzing the distribution of differences between clear and difficult cases with clearly defined etiologies in historical data, selecting a boundary value that effectively distinguishes the two. For example, by plotting a histogram or cumulative distribution curve of the difference distribution between clear and difficult cases, the boundary value is set to minimize the sum of the misclassification rates for both types of cases; or by using a machine learning classifier, such as a univariate decision tree, to find the optimal split point on the difference features. This operation focuses on the time window where attribution ambiguity zones exist and is only triggered when the data points within the window belong to already labeled alignment nodes. The fundamental reason for triggering the correction is that these aligned node data have been proven to simultaneously drive unexpected risks and cause data contradictions, making them a key source of confusion that makes it difficult to clearly determine the contribution of etiology during this period. Integrating the attribution ambiguity resolution coefficients calculated for these nodes into the original dominance strength values ​​of the corresponding data in the attribution map through summation essentially transforms the quantitatively assessed cross-interference strength into a direct numerical adjustment to the preliminary analysis results. This adjustment aims to offset the ambiguity effect caused by complex interactions on dominance judgment, thereby approximating a more accurate causal weight allocation within this time window.

[0043] For other regular portions of the atlas that do not exhibit such specific interference, their initial analysis results are retained without correction. Subsequently, the atlas is reconstructed based on the dominance intensity values ​​corresponding to all time windows—including both corrected and uncorrected ones. The regenerated final dominance relationship atlas integrates global analysis conclusions with localized targeted calibrations, thus outputting a complete view that clearly reveals the dynamic interaction and dominance shifts between primary and secondary factors throughout the disease course. The dominance intensity applied to the corresponding data in the atlas is specifically a weighted correction operation. For example, let the original dominance intensity value be... The corresponding attribution fuzzy resolution coefficient is (Already normalized), then the corrected intensity value or ,in or A correction coefficient, for example, set to 0.5, is used to adjust the correction magnitude. The process of regenerating the final dominant relationship map refers to rerunning the map construction logic based on the dominant intensity values ​​corrected for all time windows. This includes re-evaluating the dominant type and dominant component, or directly updating the intensity labels on the original map's data structure, and then rendering the output using the same visualization rules. Ultimately, this method provides clinical osteoporosis detection not merely with a simple yes / no judgment, but with a dynamic visualization report that deeply analyzes the etiology and evolution of osteoporosis, thereby supporting more accurate individualized diagnostic and treatment decisions.

[0044] Example 2: Based on Example 1, an osteoporosis detection system based on data analysis and artificial intelligence, such as... Figure 2 As shown, it includes: The osteoporosis data collaborative analysis module is used to acquire primary osteoporosis data and secondary osteoporosis data, calculate Pearson correlation coefficients and label collaborative data, thereby identifying antagonistic masking data; The dominance transition sequence generation module is used to sort the cooperative data and antagonistic masking data in time sequence, compare the intensity change magnitude and record the transition events to obtain the first layer of dominance transition sequence. The dominant relationship graph construction module is used to calculate the Pearson correlation coefficient within the collaborative data and the difference in association within the antagonistic masking data, generate the internal dominant relationship sequence, and merge them into a complete dominant relationship graph. The graph correction and output module is used to extract unexpected and contradictory data sequences, calculate the attribution fuzziness resolution coefficient by aligning nodes, correct the dominance strength, and generate the final dominance relationship graph.

[0045] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting osteoporosis based on data analysis and artificial intelligence, characterized in that, Includes the following steps: S1: Obtain primary osteoporosis data and secondary osteoporosis data and determine synergistic data, and determine antagonistic masking data based on synergistic data; S2: Obtain the first-layer dominance transition sequence based on collaborative data and antagonistic masking data; S3: Obtain the dominant relationship sequence within the collaborative data and the dominant relationship sequence within the antagonistic masking data respectively, and generate a complete dominant relationship map by combining the first-layer dominant power transfer sequence; S4: Correct the complete dominant relationship graph to obtain the final dominant relationship graph.

2. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 1, characterized in that, The acquisition of primary osteoporosis data and secondary osteoporosis data, and the determination of synergistic data, includes: The correlation between primary osteoporosis data and secondary osteoporosis data was calculated using the Pearson correlation coefficient. Primary osteoporosis data and corresponding secondary osteoporosis data that exhibited a synergistic amplification effect were labeled as synergistic data.

3. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 1, characterized in that, The determination of antagonistic masking data based on collaborative data includes: From the primary osteoporosis data and the secondary osteoporosis data, synergistic data were removed, and the remaining data were marked as non-synergistic data; A linear model is used to calculate the data correlation between different dimensions of non-cooperative data and define it as the first correlation. A nonlinear model is used to calculate the data correlation between different dimensions of non-cooperative data and define it as the second correlation. Calculate the difference between the first correlation and the second correlation. If the difference is greater than or equal to a preset difference threshold, the non-cooperative data that produces the difference is marked as antagonistic masking data.

4. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 1, characterized in that, The process of obtaining the first-layer dominance transition sequence based on cooperative data and antagonistic masking data includes: The collaborative data and the antagonistic masking data are sorted in chronological order to obtain the time-series sorting results. Based on the time-series sorting results, the magnitude of change of collaborative data and antagonistic masking data within adjacent time windows is compared. The data type with the largest magnitude of change is determined as the dominant data type of the time window, and the conversion event of the dominant data type is recorded to obtain the first-level dominance conversion sequence.

5. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 1, characterized in that, The process involves obtaining the dominance relationship sequence within the collaborative data and the dominance relationship sequence within the antagonistic masking data based on the collaborative data and the antagonistic masking data respectively, and then combining this with the first-layer dominance transfer sequence to generate a complete dominance relationship graph, including: Within the collaborative data, the Pearson correlation coefficient between primary osteoporosis data and secondary osteoporosis data is calculated in chronological order. The data component with the largest absolute value of the Pearson correlation coefficient is identified as the dominant data component within the time window of the collaborative data, and the corresponding transformation event is recorded to obtain the dominant relationship sequence within the collaborative data. Within the antagonistic masking data, the difference between the first and second correlation values ​​between primary osteoporosis data and secondary osteoporosis data is calculated in chronological order. The data component with the largest difference value is identified as the dominant data component of the time window within the antagonistic masking data, and the corresponding transformation event is recorded to obtain the dominant relationship sequence within the antagonistic masking data. By merging the first-level dominance transition sequence, the dominance relationship sequence within the collaborative data, and the dominance relationship sequence within the antagonistic masking data, a complete dominance relationship map between primary osteoporosis data and secondary osteoporosis data is generated.

6. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 1, characterized in that, The process of revising the complete dominant relationship graph to obtain the final dominant relationship graph includes: Extract the set of time windows corresponding to the unexpected phenomena and mark them as the unexpected time window set; Extract the time window set corresponding to the contradictory data phenomena and mark it as the contradictory time window set; The unexpected phenomenon refers to the nonlinear synergistic amplification effect between primary osteoporosis data and secondary osteoporosis data, which causes the rate of bone loss or fracture risk indicators to exceed the prediction range of independent factors. The contradictory data phenomenon refers to the conflict between the dominant conclusions indicated by primary osteoporosis data and secondary osteoporosis data within the same time window or anatomical site, resulting in conflicting data indicators. From the complete dominant relationship map, extract the primary osteoporosis data sequence and the secondary osteoporosis data sequence corresponding to the unexpected time window set, and mark the two extracted sequences as unexpected endpoint data sequences; From the complete dominant relationship graph, extract the primary osteoporosis data sequence and the secondary osteoporosis data sequence corresponding to the contradictory time window set, and mark the two extracted sequences as contradictory node data sequences; Align the unexpected endpoint data sequence with the contradictory node data sequence.

7. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 6, characterized in that, The step of aligning the unexpected endpoint data sequence with the contradictory node data sequence includes: If the primary osteoporosis data in the contradictory node data sequence is the same as the primary osteoporosis data in the unexpected endpoint data sequence, then the primary osteoporosis data in the contradictory node data sequence is marked as a first-class aligned node. If the secondary osteoporosis data in the contradictory node data sequence is the same as the secondary osteoporosis data in the unexpected endpoint data sequence, then the secondary osteoporosis data in the contradictory node data sequence is marked as a second type of alignment node. For each alignment node, calculate the time span length of the alignment node in the unexpected endpoint data sequence. The time span length is equal to the difference between the time window number of the last occurrence of the alignment node in the unexpected endpoint data sequence and the time window number of the first occurrence. For each alignment node, calculate the frequency of the alignment node in the data sequence of conflicting nodes; The attribution fuzziness resolution coefficient is calculated based on the type of alignment node, the time span length of the alignment node, and the frequency of occurrence of the alignment node.

8. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 7, characterized in that, The calculation of the attribution fuzzification resolution coefficient based on the type of alignment node, the time span length of the alignment node, and the frequency of occurrence of the alignment node includes: For the first type of aligned nodes, the attribution fuzziness resolution coefficient is equal to the product of the time span length of the aligned node and the frequency of occurrence of the aligned node; For the second type of aligned nodes, the attribution fuzziness resolution coefficient is equal to the ratio of the time span length of the aligned node to the frequency of occurrence of the aligned node; The complete dominant relationship graph is corrected by using the attribution fuzziness resolution coefficient.

9. The osteoporosis detection method based on data analysis and artificial intelligence according to claim 8, characterized in that, The method of using attribution fuzziness resolution coefficients to correct the complete dominant relationship graph includes: For time windows with attribution ambiguity, if the corresponding primary or secondary osteoporosis data is an alignment node, the attribution ambiguity resolution coefficient of the corresponding alignment node is added to the dominance strength of the corresponding data in the dominance relationship graph. Based on the revised dominance strength, a final dominance relationship map between primary osteoporosis data and secondary osteoporosis data is regenerated.

10. An osteoporosis detection system based on data analysis and artificial intelligence, used to implement the osteoporosis detection method based on data analysis and artificial intelligence as described in any one of claims 1-9, characterized in that, include: The osteoporosis data collaborative analysis module is used to acquire primary osteoporosis data and secondary osteoporosis data, calculate Pearson correlation coefficients and label collaborative data, thereby identifying antagonistic masking data; The dominance transition sequence generation module is used to sort the cooperative data and antagonistic masking data in time sequence, compare the intensity change magnitude and record the transition events to obtain the first layer of dominance transition sequence. The dominant relationship graph construction module is used to calculate the Pearson correlation coefficient within the collaborative data and the difference in association within the antagonistic masking data, generate the internal dominant relationship sequence, and merge them into a complete dominant relationship graph. The graph correction and output module is used to extract unexpected and contradictory data sequences, calculate the attribution fuzziness resolution coefficient by aligning nodes, correct the dominance strength, and generate the final dominance relationship graph.

Images