Thyroid function abnormality detection system based on big data

Abstract

The invention relates to the technical field of health monitoring, in particular to a thyroid dysfunction detection system based on big data, which comprises a hormone data management module, a trend inflection point identification module, a section structure judgment module, a label dynamic updating module and a periodic evolution evaluation module. According to the method, continuous modeling is carried out on hormone concentration dynamic characteristics through multi-time-point data time sequence arrangement in hormone detection records, local extreme points of hormone values are recognized, key fluctuation inflection point positions are defined, an abnormal fluctuation range is recognized, and a reference interval ratio judgment mechanism is introduced to improve the abnormal recognition precision; the dynamic tracking of the function state change process is realized by recording the label evolution trajectory through the continuous state label comparison and updating process, the identification capability of potential function abnormality is enhanced by utilizing the quantitative evaluation of the individual state fluctuation stability, the identification efficiency and the monitoring timeliness of the key intervention individual are improved, and the monitoring accuracy is improved. And the continuity of thyroid function state change identification and the accuracy of intervention target identification are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of health monitoring, and particularly to a detection system for thyroid function abnormalities based on big data. Background Art

[0002] The technical field of health monitoring involves continuous or regular monitoring and evaluation of the health status of individuals or groups. The core matters include physiological parameter collection, health data recording, disease risk assessment, and health intervention suggestions. The overall technical field covers aspects such as biosensor signal acquisition, dynamic analysis of physiological data, and integration of clinical health information, and is widely used in health management scenarios such as chronic disease management, disease early warning, and remote medical services. Among them, traditional thyroid function abnormality detection refers to collecting blood samples from the subjects to be tested and performing laboratory chemiluminescence immunoassay to measure biochemical indicators such as thyroid-stimulating hormone, triiodothyronine, and thyroxine, and combining manual interpretation to judge the thyroid function status. It mainly relies on laboratory equipment to measure the hormone concentration and compare it with the standard reference value, and relies on the manually formulated interpretation rules for preliminary diagnosis. The process usually uses sample collection, chemical detection, and clinical experience rules for data interpretation and status evaluation, mainly based on single-point static data and artificially formulated diagnostic criteria, and it is difficult to make full use of the deep association analysis of historical data and multi-dimensional health information.

[0003] During the existing thyroid function abnormality detection process, the detection method mainly compares the hormone concentration value of a single blood sample with a static reference standard. It is easy to cause judgment errors in the case of insufficient sample time distribution or drastic state changes. Due to its high dependence on manual interpretation and fixed rules, it is difficult to adapt to the time series change characteristics of multiple detection data, and it is also difficult to establish an individualized dynamic trend model. When the individual health status evolves slowly or fluctuates periodically, it is impossible to capture potential abnormal signals in time. For example, in a state of fluctuating around certain critical values, it is easily misjudged as normal. In addition, it is impossible to perform linkage analysis on historical detection information and the current detection status, which limits the tracking of disease progression and the precise control of the intervention rhythm. Summary of the Invention

[0004] The purpose of the present invention is to solve the disadvantages existing in the prior art, and to propose a detection system for thyroid function abnormalities based on big data.

[0005] In order to achieve the above purpose, the present invention adopts the following technical scheme: The detection system for thyroid function abnormalities based on big data includes: The hormone data management module obtains the original detection records of TSH, FT3, and FT4 of the detected population, extracts the hormone name, concentration measurement value, and detection time node corresponding to each sample, and performs time sorting processing according to this to obtain an individual hormone time series linked list; The trend inflection point recognition module, based on the individual hormone time series linked list, performs a direction judgment operation on the concentration values corresponding to consecutive time nodes in each hormone sequence, marks local extreme points, and records the corresponding hormone names and detection time node information to obtain a hormone inflection point annotation list; The section structure discrimination module, based on the hormone inflection point annotation list, obtains the concentration change segments between the hormone URL and LRL in each inflection point path, counts the SDMA cycle lengths and occurrence frequencies of each type of trend segment in the section, calculates the trend correlation coefficients of the rising and falling trend sections, compares the recommended ratios and marks abnormal sections to obtain a structure anomaly discrimination result; The label dynamic update module, according to the structure anomaly discrimination result, makes a comparison and judgment on the ATA classification label in the current detection stage. If the abnormal section type is inconsistent with the existing label status and the continuous occurrence times exceed the quality control threshold, the current label status is updated to obtain a label status evolution trajectory.

[0006] As a further solution of the present invention, the individual hormone time series linked list includes hormone type distribution records, concentration change spans, and time node continuity detection results. The hormone inflection point annotation list includes extreme point distribution densities, hormone type correspondences, and inflection point time intervals. The structure anomaly discrimination result includes trend cycle statistical values, trend correlation deviation amounts, and start and end information of abnormal sections. The label status evolution trajectory includes label replacement times, hormone influence paths, and status change time axes.

[0007] As a further solution of the present invention, the hormone data management module includes: The hormone record extraction sub-module obtains the original detection records of the periodic thyroid function detection population, extracts the three hormone names of TSH, FT3, and FT4 contained in each sample, the corresponding hormone concentration measurement values and detection time nodes, screens the samples with hormone missing items in the records, and eliminates the corresponding records to generate a hormone measurement record set; The time sorting and processing sub-module, based on the hormone measurement record set, reads the LIS patient unique identifier and the corresponding detection time node in the record, sorts all hormone record items of the same patient according to the time node, constructs the hormone index change trajectory of the same patient at different times, and obtains the hormone time series sorting result; The individual data construction sub-module, according to the hormone time series sorting result, matches the hormone index type, concentration value, and detection time, assembles the sorted results, establishes a continuous structure of multi-time point index states, and generates an individual hormone time series linked list.

[0008] As a further solution of the present invention, the trend inflection point recognition module includes: The concentration trend judgment submodule calls the hormone concentration values corresponding to three consecutive adjacent time nodes in each hormone sequence based on the individual hormone time series chain table, judges the relative size relationship between the concentration values of each group of three points in turn, determines the potential inflection point candidate position, extracts the point pairs that meet the concentration difference change conditions, performs concentration difference calculation, judges whether the difference exceeds the concentration fluctuation threshold, retains the corresponding detected individual sequence, and obtains the hormone fluctuation trend analysis result; The inflection point condition identification submodule screens and judges the nodes that meet the relative size change direction conditions according to the hormone fluctuation trend analysis results, determines the local maximum point and the local minimum point, records the change direction type and location of the corresponding node, establishes state identification marks for all nodes that meet the conditions in the hormone sequence, and obtains the inflection point identification matrix; The inflection point information recording submodule extracts and assembles the hormone name, detection time node and hormone value corresponding to the node into structured information entries according to the node position and inflection point type information marked in the inflection point identification matrix, combines the individual number, hormone name, time node, and organizes them one by one to form a unified structure to generate a hormone inflection point annotation list.

[0009] As a further solution of the present invention, the segment structure identification module includes: The reference segment extraction submodule reads the marked local maximum and local minimum nodes in each individual hormone sequence based on the hormone inflection point annotation list, extracts the time periods between adjacent inflection points as independent segments in turn, and makes a numerical range judgment on the hormone value in each segment. Combined with the standard upper limit URL and lower limit LRL provided by the NACB guidelines, it is determined whether it is completely within the standard reference interval of the hormone. If all are within the range, they are retained, otherwise they are removed. All standard segments of each patient under each hormone are continuously extracted and numbered cumulatively to obtain a standard trend segment sequence; The trend feature calculation submodule counts the start and end time nodes, trend direction and hormone value change data in each segment according to the standard trend segment sequence, and counts the cycle length and frequency of the two types of rising and falling trends respectively. Combined with the SDMA cycle analysis, the duration of each type of trend segment is counted respectively, and the number of occurrences of each type of trend under each hormone is counted to calculate the trend correlation coefficient; The structural state identification submodule calculates the ratio of the corresponding coefficients of the rising and falling trends of each segment based on the trend correlation coefficient, and compares it with the ratio interval recommended by the NACB guidelines. If the segment ratio is outside the interval, it is judged to be an abnormal trend state. The hormone type field and the start and end time node information corresponding to the abnormal segment are simultaneously extracted and combined to generate a structural abnormality judgment result.

[0010] As a further solution of the present invention, the tag dynamic update module includes: Based on the structural anomaly discrimination result, the label consistency discrimination sub-module extracts the hormone type, trend direction, start and end time of the abnormal section, obtains the ATA classification label status at the current detection stage, analyzes the trend characteristics corresponding to the label, and compares whether the label trend direction is consistent with the trend of the abnormal section. If the directions are opposite, it is determined as a label conflict event. The conflict times under each hormone dimension are counted to generate the label trend conflict times. Based on the label trend conflict times, the label trigger rule analysis sub-module records the cumulative conflict value of each hormone item for each individual, sets a judgment threshold with reference to the Westgard multi-rule quality control standard, makes a judgment according to the continuity of the conflict event on the time axis, evaluates whether the current label status meets the update condition, and if it is satisfied, marks the corresponding hormone category as triggered for change to obtain the label status update identifier. Based on the label status update identifier, the label evolution trajectory generation sub-module extracts the hormone name, original label, new label and corresponding time node of each change, constructs a label status change sequence, arranges the change records in chronological order, and establishes a label status evolution trajectory.

[0011] As a further solution of the present invention, the system further includes: Based on the label status evolution trajectory, the periodic evolution evaluation module counts the number of changes in the TDSS score, the duration of each state, and the TI-RADS structure change frequency before and after the label change of the detected individual in consecutive monitoring periods, evaluates the periodic trend stability characteristics of the current abnormal state of the detected individual. If the stability coefficient is within the intervention range recommended by the AACE / ACE thyroid guidelines, it is marked as a key intervention individual to obtain the thyroid function abnormality detection result. The thyroid function abnormality detection result includes the TDSS score fluctuation range, the TI-RADS structure abnormal movement frequency, and the key intervention individual identification marker result.

[0012] As a further solution of the present invention, the periodic evolution evaluation module includes: Based on the label status evolution trajectory, the status change statistics sub-module obtains the label change records of each detected individual in consecutive detection periods, extracts the label content of adjacent time points in time series, determines whether there is a TDSS score change event, and sequentially counts all individuals to generate the status change frequency and duration value. Based on the status change frequency and duration value, the structure change discrimination sub-module obtains the time nodes before and after each TDSS score change, correspondingly extracts the TI-RADS structure change records in the time period, and sequentially counts the TI-RADS level fluctuation times in all score change events to obtain the score structure synchronous change times. The stability screening marker sub-module synchronizes the number of changes according to the scoring structure, counts the TDSS change frequency, the duration of each state, and the number of TI-RADS changes of the current detected individual in consecutive cycles, calculates the relative deviation of the current individual's state feature vector from the interval boundary according to the recommended interval of the stability coefficient set by the AACE / ACE guidelines, compares the deviation threshold to determine whether the state meets the intervention target standard, and obtains the detection result of thyroid function abnormality.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, through the sequential arrangement of multi-timepoint data in hormone detection records and the continuous modeling of the dynamic characteristics of hormone concentrations, local extreme points of hormone values are identified to clarify the positions of key fluctuation inflection points. The structural analysis of trend segments identifies abnormal fluctuation ranges and introduces a reference interval ratio discrimination mechanism to improve the accuracy of abnormality identification. The dynamic tracking of the functional state change process is realized by recording the label evolution trajectory through the label comparison and update process of continuous states. By using the quantitative evaluation of the stability of individual state fluctuations, the ability to identify potential functional abnormalities is enhanced, the identification efficiency and monitoring timeliness of key intervention individuals are improved, and the continuity of thyroid function state change identification and the accuracy of intervention target identification are enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the hormone data management module of the present invention; Figure 3 is the flow chart of the trend inflection point identification module of the present invention; Figure 4 is the flow chart of the section structure discrimination module of the present invention; Figure 5 is the flow chart of the label dynamic update module of the present invention; Figure 6 is the flow chart of the periodic evolution evaluation module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0016] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0017] Please refer to Figure 1 , the big data-based thyroid function abnormality detection system includes: The hormone data management module obtains the original detection records of TSH (thyroid-stimulating hormone), FT3 (free triiodothyronine), and FT4 (free thyroxine) of the population undergoing periodic thyroid function tests, extracts the hormone name, concentration measurement value, and detection time node corresponding to each sample, and performs time sorting processing according to the LIS patient unique identifier (the laboratory information system patient ID format defined by the HL7 standard) to obtain an individual hormone time series linked list; The trend inflection point recognition module performs a direction judgment operation on the concentration values corresponding to consecutive time nodes in each hormone sequence based on the individual hormone time series linked list. If the current value is lower than the previous value and higher than the next value, it is recorded as a local maximum point. If the current value is higher than the previous value and lower than the next value, it is recorded as a local minimum point, and the corresponding hormone name and detection time node information are recorded to obtain a hormone inflection point annotation list; The section structure discrimination module, based on the hormone inflection point annotation list, obtains all concentration change segments between the upper limit (URL) and the lower limit (LRL) of the hormone value reference interval in each inflection point path, respectively counts the SDMA cycle length and occurrence frequency of each type of trend segment in the section (SDMA cycle analysis: a standard method for time series decomposition), calculates the trend correlation coefficient of the rising and falling trend sections, and compares it with the recommended ratio in the NACB guidelines (the Laboratory Medicine Practice Guidelines for the Diagnosis and Monitoring of Thyroid Diseases). If the ratio deviates from the interval range, the section is marked as an abnormal section, and the hormone type and start and end time indexes are recorded synchronously to obtain the structure abnormality discrimination result; The label dynamic update module makes a comparison and judgment on the ATA classification label (the international classification standard for thyroid diseases) at the current detection stage according to the structure abnormality discrimination result. If the type of the abnormal section is inconsistent with the existing label status and the number of consecutive occurrences exceeds the quality control threshold of the Westgard multi-rule (the multi-rule judgment system for clinical laboratory quality control), the current label status is updated, and the before and after changes of the label name, the updated hormone type, and the time point of status change are recorded to obtain the label status evolution trajectory; Based on the evolution trajectory of the label status, the periodic evolution evaluation module counts the number of changes in the TDSS score (Thyroid Disease Severity Quantification Assessment Tool) of the detected individual in consecutive monitoring periods, the duration of each status, and the frequency of TI-RADS (Thyroid Imaging Reporting and Data System) structure changes before and after label changes, and evaluates the periodic trend stability characteristics of the current abnormal status of the detected individual. If the stability coefficient is within the intervention range recommended by the AACE / ACE Thyroid Guidelines, it is marked as an individual requiring key intervention, and the thyroid function abnormality detection result is obtained.

[0018] The individual hormone time series linked list includes records of hormone type distribution, concentration change span, and detection results of time node continuity. The hormone inflection point annotation list includes extreme point distribution density, hormone type correspondence, and inflection point time interval. The structural abnormality discrimination result includes trend cycle statistical values, trend correlation deviation amounts, and start and end information of abnormal sections. The label status evolution trajectory includes the number of label replacements, hormone influence paths, and status change time axes. The thyroid function abnormality detection result includes the TDSS score fluctuation range, TI-RADS structure abnormal movement frequency, and key intervention individual identification marker result.

[0019] Please refer to Figure 2 , the hormone data management module includes: The hormone record extraction sub-module obtains the original test records of the population with periodic thyroid function tests, extracts the names of the three hormones TSH, FT3, and FT4 contained in each sample, the corresponding hormone concentration measurement values and test time nodes, screens the samples with missing hormone items in the records, and deletes the corresponding records to generate a hormone measurement record set; When obtaining the original test records of the population with periodic thyroid function tests, it is necessary to call the structured storage data table from the LIS system. The main fields include patient ID, test item name, test value, test time, department source, etc. In the example, from January 2023 to January 2024, a total of 1,250 patients' TSH, FT3, and FT4 hormone test records were collected. When screening these records for fields, it is necessary to focus on extracting the records with "test items" being TSH, FT3, and FT4. If there are other items such as TT4 and TT3, they will be excluded at this stage, and only the three target hormones will be retained. When extracting the concentration measurement values, the numerical format of the measurement value field such as "test result value" is standardized, and non-quantitative data such as "<0.01" is excluded and converted into the corresponding recognizable value of 0.01 nanomoles per liter or micro-international units per milliliter. The format of the test time node field is uniformly processed and converted into the YYYY-MM-DD hh:mm:ss timestamp format to ensure the consistency of subsequent sorting operations. At the same time, the hormone records of each patient are structurally matched. If there are missing items among the three hormones in the record at a certain time point, such as only having TSH and FT4 but no FT3 record, it is determined as hormone deficiency and the exclusion operation is performed to form a sample set in which each record completely contains the information of the three hormones. Considering a specific example, patient A recorded a TSH value of 3.12 μIU / mL, FT3 of 4.21 pmol / L, and FT4 of 13.8 pmol / L on March 12, 2023, 09:25, and the record was completely retained. However, patient B only recorded TSH of 1.45 μIU / mL and FT3 of 3.58 pmol / L on March 13, 2023, 10:03, and FT4 was missing, so it was excluded. After screening, a total of 948 complete records were formed, as shown in Table 1: Table 1 Hormone Measurement Record Table (All Three Indicators Complete)

[0020] Table 1 lists the hormone measurement record results after field screening, data format conversion, and integrity judgment in the hormone record extraction sub-module. This process obtains the hormone measurement record set.

[0021] Based on the hormone measurement record set, the time sorting processing sub-module reads the unique LIS patient identifier and the corresponding test time node in the record, sorts all the hormone record items of the same patient according to the time node, constructs the hormone index change trajectory of the same patient at different times, and obtains the hormone time sequence sorting result; Based on the hormone measurement record set, the unique identifier of the LIS patient in the record and the corresponding detection time node are called. The multiple hormone records of the same patient are sorted in ascending order of time to construct the detection time series of the individual. The sorting operation first groups the records of each patient. In each group of records, the detection time field is called, and the string timestamp is converted to an integer timestamp to ensure that the sorting logic is based on time sequence. During the sorting process, if there are multiple records with exactly the same time, the original order is maintained according to the record number order. Considering that patient A has 3 records, namely 2023-01-10, 2023-03-12, and 2023-06-25, and their corresponding TSH values are 2.85, 3.12, and 2.57 μIU / mL, FT3 is 3.97, 4.21, and 4.03 pmol / L, and FT4 is 13.2, 13.8, and 14.0 pmol / L. Then the time series formed after sorting is to arrange these three groups of values in sequence starting from 2023-01-10. Through this sequence, the changing trends of medium fluctuations in the TSH concentration, stable increase in FT3, and slow increase in FT4 of this patient can be established, providing a basis for the next linked list construction. After sorting, the number of records of each patient ranges from 2 to 8. According to the sorted hormone concentration changes and detection time positions, a dynamic sequence of each hormone item is established, and the hormone time series sorting result is obtained through this process.

[0022] The individual data construction sub-module matches the hormone index type, concentration value, and detection time according to the hormone time series sorting result, assembles the sorted result, establishes a continuous structure of the multi-time point index status, and generates an individual hormone time series linked list; According to the sorting results of hormone time series, the sorted hormone records are combined into a chained structure, matching the type fields, corresponding concentration values and detection time nodes of TSH, FT3, and FT4 in each record, and structured and assembled into a multi-node time series linked list. Each node represents a time point, and the concentration values of the three hormones, the record number and time label of this time point are embedded in the node. During the construction of the linked list, each patient is connected to the next node in chronological order to form a complete index evolution trajectory. Considering that patient B has 4 records, the detection times are January 15, 2023, April 17, 2023, July 10, 2023, and October 8, 2023. The TSH values are 2.30, 2.75, 2.95, and 2.50 μIU / mL respectively, the FT3 values are 3.80, 4.10, 4.05, and 4.00 pmol / L, and the FT4 values are 13.0, 13.5, 14.1, and 13.9 pmol / L. Then the values in the first node of its linked list structure are TSH: 2.30, FT3: 3.80, FT4: 13.0, corresponding to the time January 15, 2023. The second node corresponds to the time April 17, 2023, and so on. The length of the linked list for each patient is the number of their detections. A doubly linked list is used in the data structure for subsequent insertion or deletion of intermediate nodes. The head pointer and tail pointer of the linked list also need to be set in each linked list structure, and the total number of nodes is recorded for further analysis and processing. This processing result generates an individual hormone time series linked list.

[0023] Please refer to Figure 3 , the trend inflection point recognition module includes: The concentration trend judgment sub-module is based on the individual hormone time series linked list, calls the hormone concentration values corresponding to three consecutive adjacent time nodes in each hormone sequence, and sequentially judges the relative magnitude relationship between the concentration values for each group of three points to determine the candidate positions of potential inflection points. Extract the point pairs that meet the concentration difference change conditions for concentration difference calculation, judge whether the difference exceeds the concentration fluctuation threshold, retain the corresponding detection individual sequence, and obtain the hormone fluctuation trend analysis result; Based on the individual hormone time series linked list, a time series window is established for each hormone type of each patient. The concentration values corresponding to any three consecutive time nodes in the hormone sequence are sequentially extracted, and the magnitude relationship between the current value and the adjacent previous and next values is judged in turn. In each judgment, if there is a situation where the concentration values are exactly the same, this position is directly skipped to avoid interference from unchanged data. For a valid three-point combination for judgment, the adjacent detection time interval should not exceed 180 days, and at the same time, the complete hormone detection record of each individual should be no less than 3 times. Otherwise, this individual or this hormone type is skipped. The concentration value difference is calculated for the valid sequence to judge the actual change range of the hormone concentration. The concentration differences between all adjacent two detections in the hormone concentration sequence are extracted to form a set of difference values, and the average difference and standard deviation of this set are calculated to construct a reference threshold for hormone concentration change. The threshold is set as the average difference plus the standard deviation multiplied by 0.5. In actual implementation, for example, the data of 5 time points of the FT4 hormone concentration of a certain patient are 12.4, 13.1, 12.9, 13.3, 12.7 pmol / L, then the difference sequence is 0.7, 0.2, 0.4, 0.6, the mean value is 0.475, the standard deviation is 0.212, and the corresponding threshold is 0.475 + 0.106 = 0.581. If the difference between the first and last concentrations in a certain group of three points is less than 0.581, the trend direction judgment is no longer performed. If it is greater than this threshold, the direction judgment operation is executed, and the concentration trend is recorded as rising - falling, falling - rising, or other trends. In the whole process, this difference calculation and judgment operation are performed on all hormone sequences, and finally the hormone fluctuation trend analysis result is obtained.

[0024] The inflection point condition identification sub-module screens and judges the nodes that meet the relative magnitude change direction condition according to the hormone fluctuation trend analysis result, determines the local maximum point and local minimum point, records the change direction type and the position where the corresponding node is located, and establishes a status identification flag for all nodes in the hormone sequence that meet the conditions to obtain the inflection point identification matrix; According to the analysis results of hormone fluctuation trends, hormone sequence points that meet the requirements of the concentration value change direction are extracted. For each window composed of three points in each group, if the concentration value of the middle point shows a downward trend compared to the previous point and an upward trend compared to the subsequent point, it is recorded as a local maximum point. If the middle point is lower than the previous value and higher than the subsequent value, it is recorded as a local minimum point. To ensure that this judgment is representative in data fluctuations, only the data points whose corresponding differences meet more than 1.2 times the average difference threshold of the hormone are retained for significant inflection point screening. Taking the FT3 hormone value of a certain patient as an example, its 5 measurement values are 4.1, 4.4, 4.0, 4.5, 4.2 pmol / L, and the corresponding middle points are 4.4, 4.0, 4.5. Among them, 4.4 is higher than 4.1 and higher than 4.0, and the differences are 0.3 and 0.4 respectively. The average difference is 0.33, the standard deviation is 0.058, and 1.2 times is 0.396. Since the differences are all higher than this threshold, this point is marked as a valid maximum point. After judging other points in the same way, all inflection points that meet the direction and intensity conditions are marked with inflection point types and form a structure list. At the same time, a matrix representation is established according to the hormone name, individual ID, and position to obtain the inflection point identification matrix.

[0025] The inflection point information recording sub-module extracts and assembles the hormone name, detection time node, and hormone value corresponding to the node into a structured information entry according to the node position and inflection point type information marked in the inflection point identification matrix. Combining the individual number, hormone name, and time node, it is sorted item by item to form a unified structure and generate a hormone inflection point annotation list; According to the node data marked as local maximum or minimum in the inflection point identification matrix, they are respectively matched to the original individual hormone time series linked list, and the corresponding hormone name field, detection time field, and concentration value field are extracted and written into the newly created structured record table in sequence. Each record consists of five pieces of information: individual identification, hormone category, inflection point type, concentration value, and timestamp. When constructing the record, the combined primary key method is used to ensure the uniqueness and traceability of the inflection point information. In the example, a certain patient has an ID of A102, and the FT4 value measured on March 10, 2023 is 14.1 pmol / L, which is judged as a maximum point. The record format is [A102, FT4, maximum, 14.1, 2023-03-10]. This type of record is finally summarized into a table list, and a retrieval index is constructed through field combination for subsequent reference, as shown in Table 2: Table 2 Hormone Inflection Point Annotation Record Table

[0026] Table 2 lists the structured annotation information extracted and sorted during the hormone inflection point information recording process, and the records form a hormone inflection point annotation list.

[0027] Please refer to Figure 4 ., the section structure discrimination module includes: Based on the hormone inflection point annotation list, the reference segment extraction sub-module reads the marked local maximum and local minimum nodes in each individual hormone sequence, sequentially extracts the time periods between adjacent inflection points as independent segments, judges the numerical range of the hormone values within each segment, and combines the standard upper limit URL and lower limit LRL provided by the NACB guidelines to determine whether it is completely within the standard reference interval of the hormone. If all are within the range, it is retained; otherwise, it is excluded. Continuously extract all standard segments for each patient under each hormone and accumulate the numbers to obtain the standard trend segment sequence; Based on the hormone inflection point annotation list, extract the segments formed by adjacent maximum and minimum nodes in each individual hormone time series. First, confirm the integrity of the inflection point path and judge whether there is at least one pair of maximum and minimum nodes within each path. Then, sequentially extract their corresponding detection times and determine the start and end of the segment. Calculate whether all hormone values within the segment fall entirely within the standard reference interval of the corresponding hormone. The interval parameters are from the NACB guidelines. For example, for FT3, its upper and lower limits are 3.5 pmol / L and 6.5 pmol / L. In an actual case, the FT3 hormone value sequence of a patient is: 4.2, 4.6, 4.9, 4.7, 4.4 pmol / L, and its corresponding detection time is from January 15, 2023, to April 20, 2023. Since all values within the segment are within the standard interval, this segment is recognized as a standard segment. Record the hormone name, start and end nodes, concentration data, and trend direction of the segment as a temporary record of the trend segment. On this basis, sequentially traverse all hormones and all paths to extract all eligible time periods. Among them, if the hormone value sequence is 12.2, 12.5, 12.1, 11.9, 11.7 pmol / L, referring to the FT4 interval of 10.5–22.5 pmol / L, the data also meets the requirements, and continue to extract its corresponding segment. During this processing, all segments containing any out-of-bounds hormone values need to be skipped, and the remaining segments are uniformly numbered. For example: A102-FT4-20230201-20230410, to form a hormone trend path identifier, record its trend direction. Through the direction judgment rule, if the latter value is greater than the former value, it is an upward trend; otherwise, it is a downward trend. During this process, generate a standard trend segment list and use it as the input data for subsequent statistical and recognition processes to obtain the standard trend segment sequence.

[0028] The trend feature calculation sub-module, according to the standard trend segment sequence, statistically analyzes the start and end time nodes, trend direction, and hormone value change data within each segment, respectively counts the cycle length and frequency for the upward and downward trends, combines the SDMA cycle analysis, respectively counts the number of days of each type of trend segment, and counts the number of occurrences of each type of trend under each hormone. Using the formula: ; Calculate the trend correlation coefficient , where is the th time node within the th segment, is the average time within this segment, is the th hormone concentration value, is the average hormone concentration within this segment, is the number of time nodes within this segment; According to the standard trend segment sequence, extract all the time nodes and hormone concentration values in each trend segment. Calculate the duration of the rising segment and the falling segment respectively according to the trend direction. The calculation method is the end date minus the start date plus one day. In the example, a rising trend time period is from March 1, 2023 to April 1, 2023, which is 32 days in total. If there are 3 rising trend segments in the FT3 sequence of the same patient, with durations of 28, 30, and 35 days respectively, then the cumulative frequency is 3, and the average rising trend cycle length is (28 + 30 + 35) ÷ 3 = 31 days. After the statistics are completed, continue with the trend consistency analysis. Extract all the time nodes and hormone concentration values in each segment, convert the time into integers according to the number of days. For example, from March 1, 2023 to March 3, 2023 is represented as the 1st day, the 2nd day, and the 3rd day, and the corresponding concentration values are 4.3, 4.5, and 4.8 pmol / L respectively. Take their average value and calculate the trend correlation coefficient according to the formula. For FT3, in a 3-day rising trend segment, its T values are [1, 2, 3], and its C values are [4.3, 4.5, 4.8], then: Time series (converted to day sequence number): ; Hormone concentration value (unit: pmol / L): ; Calculate the average time: ; Calculate the average hormone value: ; Substitute the above values into the formula and calculate step by step as follows: Calculate the numerator part of the covariance: ... ; Calculate the square of the time standard deviation: ; Calculate the square of the concentration standard deviation: ; Substitute into the formula to calculate the final value ; The calculated trend correlation coefficient of this trend segment is approximately 0.9932, belonging to the category of highly consistent linear changes. This result indicates that the FT3 concentration shows a highly linear upward trend over time, and the value close to 1 indicates that the fitting degree is very close to the ideal linear growth curve.

[0029] The trend correlation coefficient is used to measure the degree of linear change consistency of hormone concentration in the time series. Its specific significance lies in reflecting the stability and directional consistency of a certain hormone's change over time within a specific trend segment. When the coefficient value is close to 1, it indicates that the hormone concentration continuously increases or decreases approximately linearly over time, with a clear trend and small fluctuations; if the coefficient is far from 1, it indicates that there are non-linear fluctuations or repeated directions in the hormone value changes within this segment, and the trend is unstable, unable to form a clear upward or downward pattern. Therefore, the trend correlation coefficient can be used as an important numerical indicator to judge whether the hormone regulation process is disturbed and whether the trend segment is reliable, and as a quantitative basis for trend persistence in structure discrimination.

[0030] This formula essentially originates from the calculation logic of the Pearson correlation coefficient, which is used to measure the degree of linear correlation between two variables. The numerator part of the formula is the sum of the products of the time series deviation and the hormone concentration deviation, which is used to reflect whether the two variables deviate from their respective means synchronously at each time point. If both are greater than or less than their means, the product is positive, indicating a consistent trend; if one is large and the other is small, the product is negative, indicating the opposite direction. The larger the numerator value, the more consistent the trend; the denominator part is composed of the square root of the product of the sum of the squares of the time deviation and the sum of the squares of the hormone concentration deviation, which respectively measures the fluctuation degrees of the two variables themselves. The square root operation of their product can be regarded as a normalization process for the numerator, excluding the influence of the absolute value size and only retaining the change direction relationship between the two variables. The final result is between -1 and 1. The closer it is to 1, the stronger the linear consistency; the closer it is to -1, the completely opposite the linear direction; the closer it is to 0, the less obvious the linear relationship. This operation logic realizes information aggregation through summation, realizes the linkage of the offset direction through multiplication, and realizes numerical normalization through squaring and square root extraction, ensuring that trend consistency can be fairly compared for segments of different time lengths.

[0031] Based on the trend correlation coefficient, the structure state recognition sub-module calculates the ratio of the corresponding coefficients of the rising and falling trends in each segment, and compares it with the recommended ratio interval in the NACB guidelines. If the segment ratio is outside the interval, it is judged as an abnormal trend state, and the hormone type field and start and end time node information corresponding to the abnormal segment are synchronously extracted for combined marking to generate a structure anomaly discrimination result; Call the trend correlation coefficient, and statistically calculate the average of all correlation coefficients for each patient in the rising trend section and the falling trend section by hormone type. Suppose the correlation coefficients of the three rising trend segments of a patient's FT3 hormone are 0.9932, 0.9875, and 0.9780 respectively. After calculation, the average correlation coefficient is , similarly, the correlation coefficients of its three falling trend segments are 0.8560, 0.8443, and 0.8602 respectively, and the average correlation coefficient is . Continue to calculate the ratio of the two, and get . Compare this ratio with the ratio range of 0.85 - 1.15 recommended by the NACB guidelines for structural discrimination criteria. Since the result exceeds the upper limit by 0.05, it is determined as an abnormal structural section. Record the hormone type, trend type, trend ratio, determination conclusion, start and end times of this section in sequence, and summarize them into a structural abnormality marking table. The table needs to include fields "hormone type", "trend type", "ratio value", "determination result", "start time", and "end time" as follows: Table 3 Structural Abnormality Discrimination Record Table

[0032] As shown in Table 3, the trend consistency ratio of FT3 is higher than the upper limit of the reference range and is marked as an abnormal state. The start and end time nodes recorded in the table correspond to the continuous start and end intervals of the rising and falling trend segments, and finally obtain the structural abnormality discrimination result.

[0033] Please refer to Figure 5 , the label dynamic update module includes: The label consistency discrimination sub-module, based on the structural abnormality discrimination result, extracts the hormone type, trend direction, and start and end times of the abnormal section, obtains the ATA classification label status at the current detection stage, analyzes the trend characteristics corresponding to the label, and compares whether the label trend direction is consistent with the trend of the abnormal section. If the directions are opposite, it is determined as a label conflict event, counts the number of conflicts under each hormone dimension, and generates the label trend conflict count; Based on the results of structural anomaly discrimination, first extract the hormone types marked as abnormal, their corresponding trend directions, and the time range of abnormal sections for each individual in each detection stage. Using the individual ID as an index, traverse the structural anomaly discrimination table, extract the hormone type field, the start and end times of the abnormal section, and the trend direction label for each abnormal record respectively, and establish a hormone trend mapping table. Subsequently, obtain the assigned ATA classification label for this individual within this time range. For example, if individual A205 is marked as having an abnormal downward trend of TSH in the section from February 1, 2023, to March 15, 2023, and is in the ATA "Type II hyperthyroidism" label state within this section, then obtain the trend direction corresponding to "Type II hyperthyroidism" from the label dictionary, which is an upward trend in TSH concentration, and then perform a trend direction comparison operation to determine whether the trend corresponding to the current ATA label is consistent with the structural anomaly trend direction. This process uses the symbol judgment method of the trend direction. If the abnormal trend is in the negative direction (downward) while the label trend is in the positive direction (upward), it is recorded as an inconsistent event. Process each abnormal record in turn, count the cumulative number of trend conflict events under each hormone type, and further group by individual to classify and aggregate the conflict counts under each hormone dimension to obtain a conflict statistics record with the following structure: Table 4 Statistics Table of Label Trend Conflicts

[0034] As shown in Table 4, there are 3 trend direction conflicts of TSH hormone in individual A205, and the number of label trend conflicts is initially obtained.

[0035] The label trigger rule analysis sub-module records the cumulative conflict value under each hormone item for each individual according to the number of label trend conflicts, sets a judgment threshold with reference to the Westgard multi-rule quality control standard, makes a judgment based on the continuity of conflict events on the time axis, evaluates whether the current label state meets the update condition, and if so, marks the corresponding hormone category as triggered for change to obtain the label state update identifier; According to the number of label trend conflicts, count the number of trend conflict events accumulated by each individual during the current detection stage for each hormone type, and perform a threshold judgment on these quantities. The judgment criteria are based on the Westgard quality control rule system. In this embodiment, the 10× rule and the R4s rule are selected as the judgment benchmarks. The R4s rule is defined as the deviation exceeding the limit between two consecutive adjacent detections, while the 10× rule is that the data deviates from one side of the baseline for ten consecutive times. To maintain the system accuracy, the conflict count threshold in this embodiment is set to 3, that is, when the same hormone continuously appears 3 times of trend conflicts, the label update judgment is triggered. Subsequently, extract the duration of the abnormal section and the concentration deviation amplitude corresponding to each conflict record for intensity quantification analysis. For example, if the TSH concentration in a certain abnormal section drops from 4.3 μIU / mL to 2.6 μIU / mL, the deviation amplitude is 1.7 μIU / mL, and the duration is 35 days. Further, extract the trend consistency coefficient of 0.88 in the same patient record. Then, it can be comprehensively judged that this conflict event is substantial in three dimensions: time length, fluctuation intensity, and trend stability. At this time, the cumulative event count is 3, that is, the trigger condition is met, and the corresponding hormone label status is set to "to be updated". At the same time, record the hormone type involved in the conflict section, the corresponding stage time point, and the conflict value structure to provide basic data for subsequent label chain changes and obtain the label status update identifier.

[0036] The label evolution trajectory generation sub-module extracts the hormone name, original label, new label, and corresponding time node of each change according to the label status update identifier, constructs a label status change sequence, arranges the change records in chronological order, and establishes a label status evolution trajectory. According to the label status update identifier, obtain all label change events marked as "to be updated" status, extract the original ATA classification label, newly generated replacement label, hormone type, individual ID, and time point information in each event, and arrange them in chronological order to construct a time evolution structure table, form a label evolution chain record item, use the time stamp as the main index, and establish a label update sequence node. This node includes the content of the label status change before and after. For example, for a certain patient A205, the previous label was "Type II hyperthyroidism" before March 20, 2023, and after the conflict accumulation, the system marked it as "Type III hypothyroidism", then record an evolution path change event. On this basis, aggregate all evolution records according to hormone classification, and generate an evolution trajectory table with the structure of [individual ID, hormone name, original label, new label, change time], as follows: Table 5 Label Status Evolution Trajectory Table

[0037] Referring to Table 5, the constructed label update chain can directly reflect the hormone function evolution process and generate a label status evolution trajectory.

[0038] Please refer to Figure 6, the periodic evolution evaluation module includes: Based on the label status evolution trajectory, the status change statistics sub-module obtains the label change records of each detected individual within consecutive detection cycles, extracts the label content at adjacent time points in time series, determines whether there is a TDSS score change event, and sequentially performs statistics on all individuals to generate the status change frequency and duration values; According to the structural status evolution trajectory, extract the TDSS label sequence of each individual within all detection cycles, arrange them in chronological order and identify consecutive label paragraphs, count the cycle length of each status maintenance period, and at the same time perform directional marking on the level changes of the TDSS status labels. If the status level increases, it is regarded as structural upgrade, if it decreases, it is regarded as structural degradation, and if it remains unchanged, it is regarded as a stable state. During this process, count the total number of label switches and their change type distributions. For example, if the TDSS label sequence of a patient's 6 detections is T0, T1, T1, T2, T1, T2, then 4 structural changes are identified: T0→T1, T1→T2, T2→T1, T1→T2, among which 2 are upgrades, 1 is degradation, and 1 is repeated upgrade. Further count the number of continuous time periods and the mean value of each type of status such as T1 and T2, and at the same time calculate the density index of status change (that is, the number of structural changes per unit time), so as to construct the status change statistical result of each individual, and summarize it into a data set with the structure of [individual ID, status sequence, number of changes, average continuous cycle, change density], and obtain the status change statistical matrix.

[0039] Based on the status change frequency and duration values, the structural change discrimination sub-module obtains the time nodes before and after each TDSS score change, correspondingly extracts the TI-RADS structural change records within the time period, and sequentially counts the number of TI-RADS level fluctuations in all score change events to obtain the number of synchronous score structure changes; According to the status change statistical matrix, individually identify the structural feature differences of different types of status changes, extract the status pairs before and after each structural upgrade or degradation process, and combine the difference values between TDSS levels to evaluate the change amplitude. For example, although both T0→T2 and T1→T2 belong to upgrades, their amplitudes are different. The former has a level span of 2 and the latter has a span of 1. Extract this value as the change intensity parameter, and combine the change density index to judge whether the change is drastic. Further extract the mean value of the continuous cycle under each change direction as a stability reference. If the average status change density of an individual is greater than 0.4 times / cycle and the average upgrade or degradation amplitude is greater than 1.2 levels, then mark it as an individual with drastic structural changes. On the contrary, if the density is less than 0.2 times / cycle and the average change amplitude is less than 1 level, then classify it as a structurally stable individual, and set the individuals in the middle interval to a transitional state. Finally, form a record table with four fields: individual ID, hormone type, structural change amplitude, status change density, and marking type, to form the structural change recognition result.

[0040] The stability screening marker sub-module synchronizes the number of changes according to the scoring structure, counts the TDSS change frequency, the duration of each state, and the number of TI-RADS changes of the current detected individual in consecutive cycles. According to the recommended interval of the stability coefficient set by the AACE / ACE guidelines (the recommended threshold is 0.6–0.8), the formula is used: ; Calculate the relative deviation degree of the current individual's state feature vector from the interval boundary , compare with the deviation threshold to determine whether the state meets the intervention target standard, and obtain the detection result of thyroid function abnormality. Among them, represents the TDSS change frequency of the th detected individual, is the central frequency value of the recommended interval, is the difference between the upper and lower limits of the recommended interval, represents the average duration of the th detected individual's state, is the reference value of the recommended state duration, is the reference interval span, represents the TI-RADS change frequency of the th detected individual, is the recommended reference frequency, is the width of its recommended fluctuation range; According to the number of synchronized changes in the scoring structure, first extract the number of TDSS state changes of each individual during the detection period. This value is defined as the total number of switches of different structural annotation states of the individual in the detection sequence. For example, if a patient's TDSS states are "T0, T0, T1, T1, T2, T1" in six detections, then the number of changes is 3 times, corresponding to three changes: T0→T1, T1→T2, T2→T1. Subsequently, count the duration of each stable state, that is, the number of consecutive cycles of the same structural state. In the above example, they are 2, 2, 1, 1 respectively. Calculate their average value, which is the average duration of the individual's state. Then extract the number of changes in the TI-RADS grade annotation during the detection period. If an individual's TI-RADS grade is "3, 3, 4a, 4a, 4a, 4b", then the number of changes is 2 times (3→4a, 4a→4b). The above three data constitute the stability state feature vector of the current individual, including the TDSS change frequency F, the average state duration S, and the TI-RADS change frequency R, and calculate the relative deviation degree from the recommended interval boundary. This deviation degree is used to quantify the degree of compliance of the individual with the recommended value in terms of functional state stability. The recommended reference value is based on the stable state interval set in AACE / ACE, and the center of the set interval is respectively , , , the corresponding interval width is , , . If the current state characteristic value of a certain patient is , , , then substitute it into the deviation formula: ; Thus, the deviation of this individual is obtained as 0.5556. Then compare it with the deviation threshold of 0.05 to judge whether it meets the intervention target standard. Since the deviation is much greater than 0.05, it is marked as an unstable state and not included in the stable group list. Finally, extract the hormone type, deviation, marked status, and time field of all individuals to establish a thyroid function abnormality detection record form. The field structure is as follows: Table 6 Thyroid Function Abnormality Detection Result Table

[0041] As shown in Table 6, the deviation values in the table are obtained through formula calculation. The closer the deviation is to 0, the closer the individual state is to the recommended interval. The finally obtained marked status field is the target result of this section, that is, the thyroid function abnormality detection result.

[0042] The relative deviation is used to measure the overall deviation degree of the current individual relative to the recommended reference interval under multi-dimensional state characteristics. Its specific meaning is: by quantifying the normalized offset degrees of key state indicators such as "TDSS frequency", "state duration", and "TI-RADS change times", it reflects whether the state of this individual is in a stable structural state within the recommended range. When the relative deviation value is lower, it means that this individual is close to the medical recommended interval in all reference dimensions, the state fluctuation is small, and the structural characteristics are stable; on the contrary, when the deviation value is higher, it indicates that at least one or more dimension indicators are far from the recommended boundary, and there is a tendency of structural instability or abnormal fluctuation. Therefore, as a key evaluation indicator for multi-dimensional feature stability screening, the relative deviation can uniformly quantify the overall offset characteristics of the individual state and assist in the formulation of subsequent state marking and intervention suggestions.

[0043] The operation logic of this deviation calculation formula is based on the idea of multi-dimensional feature distance measurement. By separately measuring the relative deviation degrees of individual state feature values on each reference dimension, and then performing weighted averaging on the deviation values of each dimension to comprehensively reflect the overall stability deviation degree. The absolute value calculation of each term in the formula is used to remove the influence of positive and negative directions, only retaining the deviation amplitude itself. The denominator part is the span of the recommended interval for each dimension, which plays a role of normalization, enabling parameters with different dimensions to be comparable, and avoiding a certain feature dominating the overall deviation degree due to a large numerical scale. After summing and averaging the three normalized deviation terms, it is essentially equivalent to treating the deviation degrees of an individual on the three dimensions of "TDSS frequency", "state duration", and "number of TI-RADS changes" equally, and comprehensively evaluating its overall consistency with the recommended value. If the deviations of all dimensions are small, the overall deviation degree is low, indicating that the individual state is close to the reference interval. On the contrary, if the deviation of any one dimension is large, even if the deviations of the remaining dimensions are small, the average deviation degree will be increased, thereby reflecting that there are stability problems in a certain aspect of this individual. Therefore, through operations such as absolute value, normalization, summation, and mean value, this formula constructs a stability deviation measurement model that takes into account information in all aspects, is quantifiable, and comparable.

[0044] The above is only a preferred embodiment of the present invention, and it is not intended to limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A thyroid function abnormality detection system based on big data, characterized in that, The system includes: The hormone data management module obtains the original test records of TSH, FT3, and FT4 of the tested population, extracts the hormone name, concentration measurement value, and test time node corresponding to each sample, and performs time sorting processing to obtain an individual hormone time series linked list; The trend inflection point recognition module, based on the individual hormone time series linked list, performs a direction judgment operation on the concentration values corresponding to consecutive time nodes in each hormone sequence, marks local extreme points, and records the corresponding hormone name and test time node information to obtain a hormone inflection point annotation list; The section structure discrimination module, based on the hormone inflection point annotation list, obtains the concentration change segments between the hormone URL and LRL in each inflection point path, counts the SDMA cycle length and occurrence frequency of each type of trend segment in the section, calculates the trend correlation coefficient of the rising and falling trend sections, compares the recommended ratio and marks abnormal sections to obtain a structure abnormality discrimination result; The label dynamic update module, according to the structure abnormality discrimination result, makes a comparison and judgment on the ATA classification label in the current test stage. If the abnormal section type is inconsistent with the existing label status and the consecutive occurrence times exceed the quality control threshold, the current label status is updated to obtain a label status evolution track.

2. The thyroid function abnormality detection system based on big data according to claim 1, characterized in that, The individual hormone time series linked list includes hormone type distribution records, concentration change spans, and time node continuity detection results. The hormone inflection point annotation list includes extreme point distribution density, hormone type correspondence, and inflection point time interval. The structure abnormality discrimination result includes trend cycle statistical values, trend correlation deviation amounts, and abnormal section start and end information. The label status evolution track includes label replacement times, hormone influence paths, and status change time axes.

3. The thyroid function abnormality detection system based on big data according to claim 1, characterized in that The hormone data management module includes: The hormone record extraction sub-module obtains the original test records of the population with periodic thyroid function tests, extracts the three hormone names of TSH, FT3, and FT4 contained in each sample, the corresponding hormone concentration measurement values and test time nodes, screens the samples with missing hormone items in the records, and eliminates the corresponding records to generate a hormone measurement record set; The time sorting processing sub-module, based on the hormone measurement record set, reads the LIS patient unique identifier and the corresponding test time node in the record, sorts all hormone record items of the same patient according to the time node, constructs the hormone index change trajectory of the same patient at different times, and obtains the hormone time series sorting result; The individual data construction sub-module matches the hormone index type, concentration value, and test time according to the hormone time series sorting result, assembles the sorted results, establishes a continuous structure of multi-time point index states, and generates an individual hormone time series linked list.

4. The thyroid function abnormality detection system based on big data according to claim 1, characterized in that The trend inflection point recognition module includes: The concentration trend judgment sub-module, based on the individual hormone time series linked list, calls the hormone concentration values corresponding to three consecutive adjacent time nodes in each hormone sequence, sequentially judges the relative magnitude relationship between the concentration values for each group of three points, determines the candidate positions of potential inflection points, extracts the point pairs that meet the concentration difference change conditions for concentration difference calculation, judges whether the difference exceeds the concentration fluctuation threshold, retains the corresponding detected individual sequences, and obtains the hormone fluctuation trend analysis result; The inflection point condition recognition sub-module, according to the hormone fluctuation trend analysis result, screens and judges the nodes that meet the relative magnitude change direction condition, determines the local maximum points and local minimum points, records the change direction type and the position where the corresponding nodes are located, and establishes a status identifier flag for all the nodes that meet the conditions in the hormone sequence, obtaining the inflection point identification matrix; The inflection point information recording sub-module, according to the node position and inflection point type information marked in the inflection point identification matrix, extracts and assembles the hormone name, detection time node, and hormone value corresponding to the node into structured information entries, combines the individual number, hormone name, and time node, and organizes them item by item into a unified structure to generate a hormone inflection point annotation list.

5. The thyroid function abnormality detection system based on big data according to claim 1, characterized in that The section structure discrimination module includes: The reference section extraction sub-module, based on the hormone inflection point annotation list, reads the marked local maximum and local minimum nodes in each individual hormone sequence, sequentially extracts the time periods between adjacent inflection points as independent sections, judges the numerical range of the hormone values within each section, and combines the standard upper limit URL and lower limit LRL provided by the NACB guidelines to determine whether it is completely within the standard reference range of the hormone. If it is all within the range, it is retained; otherwise, it is excluded. Continuously extract all the standard sections for each patient under each hormone and accumulate the numbers to obtain the standard trend section sequence; The trend feature calculation sub-module, according to the standard trend section sequence, statistics the start and end time nodes, trend direction, and hormone value change data within each section, respectively counts the cycle length and frequency for the rising and falling trends, combines the SDMA cycle analysis, respectively counts the number of continuous days of each type of trend segment, and counts the number of occurrences of each type of trend under each hormone, and calculates the trend correlation coefficient; The structure state recognition sub-module, based on the trend correlation coefficient, calculates the ratio of the corresponding coefficients of the rising and falling trends in each section, and compares it with the recommended ratio interval in the NACB guidelines. If the section ratio is outside the interval, it is judged as a trend abnormal state, and synchronously extracts the hormone type field and start and end time node information corresponding to the abnormal section for combined marking to generate the structure abnormality discrimination result.

6. The thyroid function abnormality detection system based on big data according to claim 1, characterized in that, The label dynamic update module includes: The label consistency discrimination sub-module, based on the structure abnormality discrimination result, extracts the hormone type, trend direction, and start and end time of the abnormal section, obtains the ATA classification label status at the current detection stage, analyzes the trend characteristics corresponding to the label, and compares whether the label trend direction is consistent with the trend of the abnormal section. If the directions are opposite, it is judged as a label conflict event, and counts the number of conflicts in each hormone dimension to generate the label trend conflict count; The label trigger rule analysis sub-module records the cumulative conflict value for each hormone item of each individual according to the number of label trend conflicts, sets a judgment threshold with reference to the Westgard multi-rule quality control standard, makes a judgment based on the continuity of conflict events on the time axis, evaluates whether the current label status meets the update condition, and if it is satisfied, marks the corresponding hormone category as triggered for change to obtain a label status update identifier. The label evolution trajectory generation sub-module extracts the hormone name, original label, new label, and corresponding time node of each change according to the label status update identifier, constructs a label status change sequence, arranges the change records in chronological order, and establishes a label status evolution trajectory.

7. The thyroid function abnormality detection system based on big data according to claim 1, characterized in that The system further includes: Based on the label status evolution trajectory, the periodic evolution evaluation module counts the number of changes in the TDSS score, the duration of each state, and the change frequency of the TI-RADS structure before and after label change in the continuous monitoring period of the detected individual, evaluates the periodic trend stability characteristics of the current abnormal state of the detected individual. If the stability coefficient is within the intervention interval recommended by the AACE / ACE thyroid guidelines, it is marked as an individual for key intervention to obtain the detection result of thyroid function abnormality. The detection result of thyroid function abnormality includes the fluctuation range of the TDSS score, the abnormal change frequency of the TI-RADS structure, and the identification result of the individual for key intervention.

8. The thyroid function abnormality detection system based on big data according to claim 7, characterized in that The periodic evolution evaluation module includes: Based on the label status evolution trajectory, the status change statistics sub-module obtains the label change records of each detected individual in the continuous detection period, extracts the label content at adjacent time points in time series, determines whether there is a TDSS score change event, and makes statistics on all individuals in turn to generate the status change frequency and duration value. Based on the status change frequency and duration value, the structure change discrimination sub-module obtains the time nodes before and after each TDSS score change, extracts the TI-RADS structure change records in the corresponding time period, and counts the number of TI-RADS level fluctuations in all score change events in turn to obtain the number of synchronous changes in the score and structure. Based on the number of synchronous changes in the score and structure, the stability screening and marking sub-module counts the TDSS change frequency, the duration of each state, and the number of TI-RADS changes of the current detected individual in the continuous period, calculates the relative deviation degree of the current individual's state feature vector from the interval boundary according to the recommended interval of the stability coefficient set by the AACE / ACE guidelines, and compares the deviation threshold to determine whether the state meets the intervention target standard to obtain the detection result of thyroid function abnormality.

Images