Machine learning based diabetes data processing analysis system

Abstract

The application relates to the technical field of medical data processing, in particular to a diabetes data processing and analyzing system based on machine learning. The system comprises a diabetes data acquisition module, which is used for acquiring medical record sample data of diabetes patients; a normal index analyzing module, which is used for analyzing the change trend and fluctuation of blood glucose time series data in each medical record sample data, and combining numerical characteristics to accurately quantify the final normal index of each medical record sample data; and a reference database construction module, which is used for classifying all medical record sample data based on multiple parameters in the medical record sample data, and in each category, screening medical record sample data with high normal degree and large reference value based on the aforementioned acquired final normal index, so as to construct a reference database. The application can accurately screen medical record sample data with high reference value, so that the constructed reference database can more accurately assist doctors in formulating treatment plans.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and more specifically to a diabetes data processing and analysis system based on machine learning. Background Technology

[0002] Diabetes mellitus is a chronic disease characterized by hyperglycemia, caused by an absolute or relative deficiency and impaired utilization of insulin, and can lead to various complications. As a chronic disease, diabetes treatment is typically a long-term process. Doctors adjust treatment and medication plans according to each patient's specific condition to ensure effectiveness. Over time, the volume of examination and treatment records for diabetic patients in hospital databases gradually increases. When determining treatment plans for diabetic patients, doctors usually select a portion of medical records from the hospital database as reference data.

[0003] When selecting reference data, current technologies typically screen data based on whether blood glucose levels are within the normal range. However, due to individual differences, the variations in different medical record samples and the various indicators within those samples can differ. Consequently, the reference value of different medical record samples will vary. Screening reference data solely based on whether blood glucose levels are within the normal range results in weak reference value and limited assistance to doctors in developing treatment plans. With the development of big data and machine learning technologies, advanced data analysis tools can be used to screen medical record samples in hospital databases, allowing for the selection of more valuable reference data. This provides stronger support for doctors in determining treatment plans for diabetic patients. Summary of the Invention

[0004] To address the issue that individual differences and variations in indicators across different medical record samples lead to varying reference value, and that relying solely on blood glucose levels for data selection results in weak data reliability and limited support for doctors in developing treatment plans, this invention aims to provide a machine learning-based diabetes data processing and analysis system. The specific technical solution employed is as follows:

[0005] This invention proposes a machine learning-based diabetes data processing and analysis system, the system comprising:

[0006] The diabetes data acquisition module is used to acquire medical record sample data of diabetic patients. The medical record sample data includes multiple parameters and blood glucose time series data.

[0007] The normal indicator analysis module is used to decompose the corresponding trend item time series data and seasonal item time series data from each blood glucose time series data; comprehensively analyze the overall change trend of the trend item time series data and the change of the seasonal item time series data corresponding to each blood glucose time series data to determine the initial normal indicators corresponding to each medical record sample data; based on the local data fluctuation and numerical characteristics of the data values ​​in each trend item time series data, the initial normal indicators are corrected to determine the final normal indicators of each medical record sample data.

[0008] The reference database construction module is used to classify all medical record sample data based on various parameters in the medical record sample data to obtain classification results; within each category, reference data is selected from the medical record sample data based on the final normal indicators, and a reference database is constructed based on the reference data.

[0009] Furthermore, the method for obtaining the initial normal indicators includes:

[0010] In the trend item time series data corresponding to each blood glucose time series data, a straight line is fitted to all data values, and the slope value of the fitted straight line is obtained. The slope value is normalized and negatively correlated to obtain the first normal factor.

[0011] In the seasonal time series data corresponding to each blood glucose time series data, all minimum points are obtained, and the data between two adjacent minimum points is taken as a change data segment;

[0012] In all the changing data segments, the second normal factor is determined based on the changes between the data values ​​in the changing data segments;

[0013] Based on the first normal factor and the second normal factor corresponding to each blood glucose time series data, the initial normal index corresponding to each medical record sample data is calculated, and both the first normal factor and the second normal factor are positively correlated with the initial normal index.

[0014] Furthermore, the method for obtaining the second normal factor includes:

[0015] Within each changing data segment, the difference between the maximum and minimum data values ​​is taken as the fluctuation range of each changing data segment;

[0016] Arrange the fluctuation amplitudes of all the changing data segments according to the time sequence of the changing data segments to obtain a sorted sequence. Then, perform backward difference on the values ​​in the sorted sequence to obtain a first-order difference sequence. Use the proportion of positive numbers in the first-order difference sequence as the second normal factor.

[0017] Furthermore, the method for obtaining the final normal indicator includes:

[0018] Based on the local data fluctuations in the trend data of each blood glucose time series, the initial normal indicators are initially corrected to obtain the corrected normal indicators for each medical record sample data.

[0019] Based on the numerical characteristics of the data values ​​in the trend items of each blood glucose time series data, the abnormal factors corresponding to each medical record sample data are determined.

[0020] The abnormal factors corresponding to each medical record sample data are negatively correlated and normalized. The product of the normalized values ​​and the corresponding corrected normal indicators is then normalized to obtain the final normal indicators for each medical record sample data.

[0021] Furthermore, the method for obtaining the corrected normal indicators includes:

[0022] In each trend item time series data, the data between two adjacent minimum points is taken as a fluctuation data segment;

[0023] Analyze the continuity of fluctuations between all fluctuation data segments to determine the fluctuation continuity value;

[0024] In each fluctuation data segment, the product of the difference between the maximum and minimum values ​​and the length of each fluctuation data segment is used as the fluctuation factor; the fluctuation factors of all fluctuation data segments are averaged to obtain the fluctuation characteristic value.

[0025] Based on the fluctuation continuity value and fluctuation characteristic value, a correction factor is calculated for each medical record sample data, and both the fluctuation continuity value and the fluctuation characteristic value are negatively correlated with the correction factor;

[0026] The normalized value of the product of the correction factor and the initial normal index corresponding to each medical record sample data is used as the corrected normal index for each medical record sample data.

[0027] Furthermore, the method for obtaining the continuous value of the fluctuation includes:

[0028] The difference between the maximum values ​​of two adjacent fluctuation data segments at corresponding times is used as the time difference factor. The mean of all time difference factors in all fluctuation data segments is negatively correlated and mapped to the value of the fluctuation continuity.

[0029] Furthermore, the method for obtaining the abnormal factor includes:

[0030] In the trend data of each blood glucose time series, data values ​​greater than a preset threshold are taken as target values, and consecutive target values ​​in the time series are filtered out to obtain all subsequences;

[0031] The proportion of the target value in the time series data of each trend item is used as the quantity factor; the difference between the maximum value of all target values ​​and the preset threshold is used as the deviation factor; and the ratio of the maximum length of all subsequences to the minimum time value of all target values ​​is used as the time factor.

[0032] The normalized value of the product of the quantity factor, deviation factor, and time factor is taken as the anomaly factor.

[0033] Furthermore, the method for obtaining the classification result includes:

[0034] The parameters include at least age group, gender, presence or absence of complications, and type of diabetes;

[0035] Among all medical record sample data, medical record sample data with the same parameters are grouped into one category, thus obtaining the classification results of all medical record sample data.

[0036] Furthermore, the method for obtaining the reference data includes:

[0037] Within each category, all medical record sample data are sorted in descending order according to the final normal indicators to obtain a descending sequence;

[0038] In the descending sequence corresponding to each category, the first preset number of medical record samples are used as reference data.

[0039] Furthermore, the step of decomposing the corresponding trend term time series data and seasonal term time series data from each blood glucose time series data includes:

[0040] The STL time series decomposition algorithm is used to decompose each blood glucose time series data to obtain the trend item time series data and the seasonal item time series data corresponding to each blood glucose time series data.

[0041] The present invention has the following beneficial effects:

[0042] First, medical record samples and corresponding blood glucose time-series data of diabetic patients are obtained from the hospital database to provide a data foundation for subsequent data analysis. Since the medical record samples selected as reference data should be those with a high degree of normality, normality indicators are analyzed for the blood glucose time-series data corresponding to each medical record sample. Under normal circumstances, when the treatment plan formulated by the doctor is effective, the blood glucose time-series data of diabetic patients will gradually return to normal levels. Therefore, in this invention, the overall trend and fluctuation of each blood glucose time-series data can be analyzed. Given that the trend term can characterize the overall change trend of the data, and the seasonal term can characterize the fluctuation and change pattern of the data, each blood glucose time-series data is decomposed to analyze the overall change trend of the trend term time-series data and the changes in the seasonal term time-series data. Initial quantification of the normality of each medical record sample data is then performed to obtain initial normality indicators. Furthermore, given individual differences, even if blood glucose time-series data from different medical record samples show the same overall trend, they may exhibit varying degrees of fluctuation in local areas. Since the numerical characteristics of data values ​​in trend-related time-series data most directly reflect the normality of the data, for each trend-related time-series data, the local data fluctuations are combined with the numerical characteristics of the data values ​​to correct the initial normal indicators, resulting in the final normal indicators for each medical record sample. These final normal indicators not only characterize the normality of the medical record sample data but also reflect its reference value. Finally, to facilitate doctors in developing targeted treatment plans for different categories of medical record sample data, this invention categorizes all diabetic patient medical record sample data in the hospital database, obtaining classification results. Within each category, reference data is selected based on the final normal indicators to construct a reference database. This ensures that the reference data in the database covers different categories and has higher reference value, enabling doctors to develop more accurate treatment plans. Attached Figure Description

[0043] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a system block diagram of a diabetes data processing and analysis system based on machine learning, provided in one embodiment of the present invention.

[0045] Figure 2 This is a flowchart of a method for obtaining initial normal indicators according to an embodiment of the present invention;

[0046] Figure 3 This is a flowchart of a method for obtaining a final normal indicator provided in an embodiment of the present invention;

[0047] Figure 4 This is a flowchart of a method for obtaining corrected normal indicators according to an embodiment of the present invention;

[0048] Figure 5 This is a schematic diagram of the system structure of a machine learning-based diabetes data processing and analysis system provided in one embodiment of the present invention. Detailed Implementation

[0049] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a machine learning-based diabetes data processing and analysis system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0051] The following description, in conjunction with the accompanying drawings, details a specific solution for a machine learning-based diabetes data processing and analysis system provided by this invention.

[0052] Please see Figure 1 The diagram illustrates a system block diagram of a machine learning-based diabetes data processing and analysis system according to an embodiment of the present invention. The system includes: a diabetes data acquisition module 101, a normal index analysis module 102, and a reference database construction module 103.

[0053] The diabetes data acquisition module 101 is used to acquire medical record sample data of diabetic patients. The medical record sample data includes multiple parameters and blood glucose time series data.

[0054] Diabetes is a chronic disease, and its treatment is usually a long-term process. During treatment, it requires long-term control of blood sugar levels and regular checkups, which helps doctors adjust the treatment plan promptly based on the specific situation to ensure treatment effectiveness. Over time, hospital databases accumulate a large amount of medical record data from diabetic patients. To improve efficiency, doctors often refer to treatment plans from these databases when treating diabetic patients. However, due to the sheer volume of data, it is often necessary to sift through it to select the most representative and valuable reference data.

[0055] With the development of machine learning technology, this embodiment of the invention analyzes and filters the medical record data of diabetic patients in the hospital database to determine more valuable reference data. By constructing a reference database with the filtered reference data, doctors can be provided with more accurate and representative references, thereby assisting doctors in formulating treatment plans.

[0056] First, obtain medical record sample data for all diabetic patients from the hospital database. This data should include multiple parameters and the patient's time-series blood glucose data. The time-series blood glucose data can be obtained by measuring at fixed time intervals. Given that diabetes includes type 1, type 2, and gestational diabetes, and that treatment plans differ for each type, and that diabetes can lead to complications such as retinopathy and neuropathy, and that the symptoms vary across different age groups (0-12 years, 13-19 years, 20-39 years, 40-59 years, and 60 years and older), the parameters in the medical record sample data should at least include age group, gender, presence or absence of complications, and type of diabetes.

[0057] It should be noted that the types and number of parameters can be adjusted according to the implementation scenario, and are not limited here; the measurement time interval of blood glucose time series data can be set to the time interval of follow-up visits for diabetic patients, and the specific time interval can be adjusted according to the implementation scenario, and is not limited here.

[0058] In this embodiment of the invention, the collection and acquisition of medical record sample data are authorized by the relevant users, and the process does not violate relevant laws and regulations, nor does it violate public order and good morals.

[0059] The normal indicator analysis module 102 is used to decompose the corresponding trend item time series data and seasonal item time series data from each blood glucose time series data; comprehensively analyze the overall change trend of the trend item time series data and the change of the seasonal item time series data corresponding to each blood glucose time series data to determine the initial normal indicators corresponding to each medical record sample data; based on the local data fluctuation and numerical characteristics of the data values ​​in each trend item time series data, the initial normal indicators are corrected to determine the final normal indicators of each medical record sample data.

[0060] The reference data should be selected from medical record samples of diabetic patients with good recovery. The recovery status of diabetic patients can be further characterized by their blood glucose time-series data. Therefore, for each blood glucose time-series data, this embodiment of the invention analyzes its normal condition to obtain a final normal index, which is used to evaluate the reference value of each medical record sample data. One of the criteria for evaluating the normal condition of blood glucose time-series data is to judge the trend and fluctuation of its data values. Since the trend term can characterize the overall trend of blood glucose level changes, and the seasonal term can characterize the periodic pattern and fluctuation of blood glucose time-series data, the data values ​​in the trend term time-series data can intuitively reflect the normal condition of the medical record sample data. Therefore, in this embodiment of the invention, the blood glucose time-series data is decomposed, and then the trend term time-series data and the seasonal term time-series data are further analyzed and processed. Combining the local data fluctuations and numerical characteristics of the data values ​​in the trend term time-series data, the final normal index of each medical record sample data is determined, providing a basis for the subsequent construction of the reference database.

[0061] First, a time series decomposition is performed on each blood glucose time series data point. Specifically, the STL time series decomposition algorithm is used to decompose each blood glucose time series data point, thereby obtaining the corresponding trend term time series data and seasonal term time series data for each blood glucose time series data point. It should be noted that the STL time series decomposition algorithm is a well-known technology, and the settings of various parameters, such as the length of the seasonal smoother and the length of the trend smoother, can be set according to the implementation scenario, and will not be elaborated here.

[0062] Trend data reflects the long-term trend of blood glucose levels, which helps doctors assess the effectiveness of treatment plans for diabetic patients, i.e., evaluate the normality of medical record data. At the same time, seasonal data can explain the detailed fluctuations in blood glucose levels, which also helps to evaluate the normality of medical record data. Therefore, by comprehensively analyzing the overall trend of trend data and seasonal data for each blood glucose time series, the initial normal indicators for each medical record data are determined.

[0063] Preferably, in one embodiment of the present invention, the method for obtaining the initial normal indicators includes:

[0064] Please see Figure 2 The diagram illustrates a method flowchart for obtaining initial normal indicators in one embodiment of the present invention. The method includes the following steps:

[0065] Step S201: In the trend data corresponding to each blood glucose time series data, analyze the overall data change trend and quantify the first normal factor.

[0066] If the treatment plan is effective, then the trend data corresponding to the blood glucose time series should show an overall downward trend. Therefore, for each blood glucose time series, a linear fit is performed on all data values, and the slope of the fitted line is obtained. The slope value can intuitively represent the overall trend of the data. A positive slope value indicates that the blood glucose level is rising, and the blood glucose time series data is more abnormal; a negative slope value indicates that the blood glucose level is falling, that is, the blood glucose time series data should be closer to the normal level. Then, the slope value is normalized and negatively correlated to obtain the first normal factor. The formula model of the first normal factor includes:

[0067]

[0068] in, Indicates the first normal factor; Indicates the slope value; This represents the normalization function.

[0069] In the formula model for the first normal factor, since the slope value can be positive or negative, a larger positive slope value indicates that the blood glucose level in the blood glucose time series data is elevated, i.e., in an abnormal state. Conversely, a smaller negative slope value indicates that the blood glucose level in the blood glucose time series data is approaching a normal state. To facilitate subsequent calculations, the following formula is used... The function normalizes the slope value. A smaller normalized value indicates that the blood glucose time-series data is closer to a normal state. Therefore, the normalized value is used... By performing negative correlation mapping, where x represents the independent variable, logical relationship correction is achieved, thereby obtaining the first normal factor. The larger the first normal factor, the closer the blood glucose time series data is to the normal state, and the higher the normality of the medical record sample data corresponding to the blood glucose time series data.

[0070] Step S202: In the seasonal time series data corresponding to each blood glucose time series data, the data is segmented based on the fluctuation of the data value to obtain the change data segment.

[0071] Seasonal time-series data typically reflect patterns of data change over time, including periodic increases and decreases. To better understand the patterns of data change and quantify the normality of medical record sample data, the seasonal time-series data corresponding to each blood glucose time-series data can be segmented and analyzed.

[0072] In the seasonal time series data corresponding to each blood glucose time series data, all minimum points are obtained, and the data between two adjacent minimum points is taken as a change data segment. At this time, each change data segment represents a change fluctuation in the seasonal time series data.

[0073] Step S203: In all the changed data segments, determine the second normal factor based on the changes between the data values ​​in the changed data segments.

[0074] Under normal circumstances, blood glucose levels in trend-related time-series data should show a decreasing trend. Therefore, fluctuations in seasonal time-series data should exhibit synchronous changes, meaning the magnitude of the change gradually decreases over time. Thus, within each data segment, the difference between the maximum and minimum data values ​​is used as the fluctuation amplitude for that segment. The larger the amplitude, the greater the degree of fluctuation in the data segment.

[0075] Then, the fluctuation amplitudes of all the changing data segments are arranged according to the time sequence of the changing data segments to obtain a sorted sequence. The sorted sequence is then differentially divided backward to obtain a first-order difference sequence. At this time, the values ​​in the first-order difference sequence can reflect the change of the data fluctuation amplitude over time. When the value is negative, it means that the fluctuation amplitude is increasing over time, and the data is more abnormal; conversely, when the value is positive, it means that the fluctuation amplitude is decreasing over time, and the data is more likely to be normal. Finally, the proportion of positive numbers in the first-order difference sequence is used as the second normality factor. That is, the ratio of the number of positive numbers in the first-order difference sequence to the total number of all values ​​in the first-order difference sequence is used as the second normality factor. The larger the second normality factor, the higher the probability that the medical record sample data corresponding to the blood glucose time series data is normal.

[0076] Step S204: Merge the first normal factor and the second normal factor corresponding to each blood glucose time series data to obtain the initial normal indicators of the corresponding medical record sample data.

[0077] Based on the foregoing analysis, the larger the first normal factor, the closer the blood glucose time series data is to the normal state, and the higher the normality of the corresponding medical record sample data. The larger the second normal factor, the higher the probability that the corresponding medical record sample data is normal. Therefore, the product of the first normal factor and the second normal factor of each blood glucose time series data is used as the initial normal index of the corresponding medical record sample data, denoted as CZ. The larger the initial normal index, the higher the normality of the medical record sample data, and the greater its reference value.

[0078] At this point, a preliminary analysis of the blood glucose time-series data for each medical record sample has been conducted, and the initial normal indicators for each medical record sample can be obtained.

[0079] When analyzing the trend of blood glucose levels during the process of obtaining initial normal indicators, only the overall trend of the trend items in the blood glucose time series data was analyzed. However, due to individual differences, different medical record samples may have consistent overall trends, but local fluctuations may differ. At the same time, while data fluctuations occur, some fluctuations may be abnormally severe, showing an aggravating trend compared to the initial or average state of blood glucose levels. This means that there are large data values ​​in the trend item time series data, which also indicates that the normality of the medical record sample data is low. Therefore, the initial normal indicators of the medical record sample data are corrected by combining the local data fluctuations in the trend item time series data and the numerical characteristics of the data values, so as to determine the final normal indicators for each medical record sample data.

[0080] Preferably, in one embodiment of the present invention, the method for obtaining the final normal indicators includes:

[0081] Please see Figure 3 The diagram illustrates a method flowchart for obtaining the final normal indicator in one embodiment of the present invention. The method includes the following steps:

[0082] Step S301: Based on the local data fluctuations in the trend data of each blood glucose time series data pair, the initial normal indicators are initially corrected to obtain the corrected normal indicators corresponding to each medical record sample data.

[0083] Please see Figure 4 The diagram illustrates a method flowchart for obtaining corrected normal indicators according to an embodiment of the present invention. The method includes the following steps:

[0084] Step S3011: Divide the trend data corresponding to each blood glucose time series data to obtain the fluctuation data segment.

[0085] In trend term time series data, the data between two adjacent minimum points is considered as a fluctuation data segment, and each fluctuation data segment corresponds to one data fluctuation in the trend term time series data.

[0086] Step S3012: Analyze the continuity of fluctuations between all fluctuation data segments and determine the fluctuation continuity value.

[0087] When data fluctuations are more continuous, it indicates more frequent increases and decreases in blood glucose levels, suggesting a higher degree of abnormality and a lower degree of normality. Therefore, the continuity of fluctuations between data segments is considered here. This can be characterized by the temporal difference between adjacent data segments: the difference between the maximum values ​​of two adjacent data segments is taken as the time difference factor. In this case, each pair of adjacent data segments in the time series has a time difference factor, and the smaller the time difference factor, the more continuous the fluctuation. Therefore, the value obtained by negatively correlating the mean of all time difference factors is taken as the fluctuation continuity value. The formula model for the fluctuation continuity value includes:

[0088]

[0089] in, Indicates a continuous fluctuation value; Indicates the number of time difference factors; This represents the value of the i-th time difference factor; This indicates the preset first constant.

[0090] In the formula model of continuous fluctuation value, since the smaller the time difference factor, the larger the continuous fluctuation value, the mean of the time difference factor between all fluctuation data segments is used as the denominator to correct the logical relationship and obtain the continuous fluctuation value. The larger the continuous fluctuation value, the more continuous the fluctuation of the data. The more frequent the rise and fall of blood glucose level, the lower the normality of the corresponding medical record sample data.

[0091] It should be noted that the first constant is preset. The purpose of this is to prevent the denominator from being 0. It can be 0.001. The specific value can be adjusted according to the implementation scenario and is not limited here.

[0092] In other embodiments of the present invention, the formula model for the continuous value of fluctuation can also be: ;use To achieve negative correlation mapping, where, Let x represent an exponential function with the natural constant e as the base, and let x represent the independent variable.

[0093] Step S3013: Determine the fluctuation characteristic value by combining the differences between data values ​​in each fluctuation data segment and the length of the fluctuation data segment.

[0094] Not only can the continuity of fluctuations be used as a standard to measure whether medical record sample data is normal, but the magnitude and duration of fluctuations can also be used as indicators to measure the normality of medical record sample data.

[0095] Therefore, in each fluctuation data segment, the product of the difference between the maximum and minimum values ​​and the length of each fluctuation data segment is used as the fluctuation factor. The larger the difference between the maximum and minimum values, the greater the amplitude of the fluctuation, indicating a more severe degree of recurrence in blood glucose levels. At the same time, the longer the fluctuation data segment, the longer the duration of the fluctuation. Therefore, the larger the fluctuation factor of the fluctuation data segment, the lower the normality of the corresponding medical record sample data is considered.

[0096] Finally, the volatility factors of all volatility data segments are averaged to obtain the volatility characteristic value, denoted as BT. Similarly, the larger the volatility characteristic value, the lower the normality of the corresponding medical record sample data, and the lower its reference value.

[0097] Step S3014: Correct the initial normal indicator based on the continuous value of fluctuation and the characteristic value of fluctuation to obtain the corrected normal indicator.

[0098] Based on the analysis in the preceding steps, it is known that the larger the continuous fluctuation value, the lower the normality of the corresponding medical record sample data; similarly, the larger the fluctuation characteristic value, the lower the normality of the corresponding medical record sample data, and therefore the lower its reference value. Therefore, by combining the continuous fluctuation value and the fluctuation characteristic value, a correction factor is calculated for each medical record sample data, and the initial normality index is initially corrected. The corrected normality index for each medical record sample data is obtained. The formula model for the corrected normality index includes:

[0099]

[0100] in, This represents the corrected normal index for each medical record sample data; This represents the initial normal indicators for each medical record sample; This represents the continuous fluctuation value corresponding to each medical record sample data; This represents the fluctuation characteristic value corresponding to each medical record sample data; This represents an exponential function with the natural constant e as its base. This represents the normalization function.

[0101] In the formula model for correcting normality indicators, a higher initial normality indicator for medical record sample data indicates a higher degree of normality and thus greater reference value. Conversely, a higher fluctuation continuous value and fluctuation characteristic value corresponding to the medical record sample data will decrease its degree of normality and reference value. Therefore, the product of the fluctuation continuous value and fluctuation characteristic value corresponding to the medical record sample data is negatively correlated and normalized to obtain the correction factor. The smaller the correction factor, the lower the reference value of the medical record sample data should be. Therefore, the correction factor corresponding to each medical record sample data is multiplied by the initial normal index, and the product is normalized to obtain the corrected normal index. The larger the corrected normal index, the higher the normality of the medical record sample data and the greater its reference value.

[0102] Step S302: Based on the numerical characteristics of the data values ​​in the trend item of each blood glucose time series data, determine the abnormal factors corresponding to each medical record sample data.

[0103] First, abnormal data points are identified. In the trend data of each blood glucose time series, data values ​​exceeding a preset threshold are used as target values ​​(abnormal data points). Target values ​​characterize data points in the trend data that tend towards abnormality and form the basis for measuring the anomalous factors in medical record sample data. It should be noted that since the data value at the starting position in the trend data can characterize the initial state of the data, comparing each data value with the data value at the starting time better reflects the degree of abnormality when screening abnormal data points. Therefore, in this embodiment of the invention, the preset threshold is set as the data value at the starting position in the trend data.

[0104] Then, sequentially consecutive target values ​​are filtered out to obtain all subsequences. This helps identify persistent abnormal patterns, rather than just isolated anomalous data points. For example, if the target values ​​correspond to times 4, 7, 8, and 10, then the data values ​​corresponding to times 7 and 8 can form a subsequence. It should be noted that in practical applications, if there are no sequentially consecutive target values, that is, if there is no subsequence of length greater than 1, a single target value can also be considered as a subsequence.

[0105] The number of target values ​​can be used as an indicator of data anomalies. Therefore, the proportion of target values ​​in each trend item's time series data is used as a quantity factor, which is the ratio of the total number of target values ​​to the total number of data values ​​in the trend item's time series data. The larger the quantity factor, the more target values ​​there are in the trend item's time series data, indicating a higher degree of data anomaly. The difference between the maximum value of all target values ​​and a preset threshold is used as a deviation factor. The larger the deviation factor, the greater the extent to which the maximum target value exceeds the preset threshold, and the higher the degree of anomaly. The length of the subsequence can characterize the duration of the anomaly. Therefore, the ratio of the maximum length of all subsequences to the minimum value of all target values ​​is used as a time factor. The minimum value of the target value represents the starting position of the initial abnormal fluctuation in the data. The earlier the abnormal fluctuation occurs, the more severe the data anomaly. Therefore, the larger the time factor, the higher the degree of data anomaly.

[0106] Finally, the product of the quantity factor, bias factor, and time factor is normalized and used as the outlier factor, denoted as YY. A larger outlier factor indicates a higher degree of data anomalousness. It should be noted that the normalization method used here can be... Normalization function.

[0107] Step S303: Combine the abnormal factors and corrected normal indicators corresponding to each medical record sample data to calculate the final normal indicators.

[0108] The outlier factor corresponding to each medical record sample data represents the abnormality of the medical record sample data. Therefore, the corrected normal index of the medical record sample data is further corrected using the outlier factor to obtain the final normal index. The formula model of the final normal index includes:

[0109]

[0110] in, This represents the final normal indicators for each medical record sample; This represents the corrected normal index for each medical record sample data; This represents the outlier factor corresponding to each medical record sample data; This represents an exponential function with the natural constant e as its base. This represents the normalization function.

[0111] In the formula model of the final normal index, based on step S302, it is known that the larger the abnormal factor, the higher the degree of abnormality of the medical record sample data. Therefore, the abnormal factor corresponding to each medical record sample data is negatively correlated and normalized. The reason for performing negative correlation mapping and normalization is to correct the logical relationship. At this time, the corrected value is... The larger the value, the higher the normality of the medical record sample data. This value is then multiplied by the corresponding corrected normality index, and the resulting product is normalized to obtain the final normality index for each medical record sample data. The larger the final normality index, the higher the normality of the medical record sample data and the greater its reference value.

[0112] Thus, after all the aforementioned steps, the final normal indicators of each diabetic patient's medical record sample data in the hospital database can be obtained. The higher the final normal indicator, the closer the blood glucose level of the medical record sample data is to the normal level, and the more valuable the treatment plan in the medical record sample data will be.

[0113] The reference database construction module 103 is used to classify all medical record sample data based on various parameters in the medical record sample data to obtain classification results; in each category, reference data is selected from the medical record sample data based on the final normal indicators, and a reference database is constructed based on the reference data.

[0114] Because the hospital database contains a large number of medical record samples, they are classified based on various parameters in the medical record samples. This results in medical record samples in each category having relatively similar characteristics. Then, within each category, more valuable medical record samples are selected as reference data based on the final normal indicators. This makes the reference data more comprehensive, and the reference database built based on the reference data will be more representative and have higher reference value.

[0115] First, the medical record sample data is classified based on various parameters to obtain the classification results.

[0116] Preferably, in one embodiment of the present invention, the method for obtaining the classification result includes:

[0117] In the diabetes data acquisition module 101, when acquiring medical record sample data, the medical record sample data contains multiple parameters, such as age group, gender, presence or absence of complications, and type of diabetes. All medical record sample data are classified based on these parameters.

[0118] In all medical record samples, those with identical parameters are grouped together. For example, age range 0-12 years is labeled 1, age range 13-19 years is labeled 2, age range 20-39 years is labeled 3, age range 40-59 years is labeled 4, and age range 60 years and above is labeled 5; gender is labeled 1 for male and 2 for female; comorbidities are labeled 1 and no complications are labeled 2; type 1 diabetes is labeled 1, type 2 diabetes is labeled 2, and gestational diabetes is labeled 3. If a medical record sample contains parameters of 60 years and above, male, comorbidities, and type 1 diabetes, then that medical record sample corresponds to the sequence [5,1,1,1]. Based on this method, the sequence corresponding to each medical record sample can be obtained, thus grouping medical record samples with the same sequence into one category.

[0119] After obtaining the classification results, reference data can be filtered out in each category based on the final normal indicators.

[0120] Preferably, the method for obtaining reference data includes:

[0121] The higher the normal indicator, the higher the normality of the medical record sample data and the greater its reference value. Therefore, in each category, all medical record sample data are arranged in descending order according to the final normal indicator to obtain a descending sequence.

[0122] Then, from the descending sequence corresponding to each category, the first preset number of medical record samples are selected as reference data. It should be noted that the preset number is set to one-quarter of the total number of medical record samples in the corresponding category, rounded up. The specific value can be adjusted according to the implementation scenario and is not limited here.

[0123] After obtaining all the reference data, a reference database can be built, containing medical record samples of different categories. The reference database supports categorized searches, and the search results can display diabetes-related data and treatment plans from the reference data, thus helping doctors to develop new treatment plans based on those plans.

[0124] In summary, the first step is to acquire medical record sample data and corresponding blood glucose time-series data of diabetic patients from the hospital database, providing a data foundation for subsequent data analysis. Since the medical record sample data selected as reference data should be of a high degree of normality, the normality indicators for the blood glucose time-series data corresponding to each medical record sample are analyzed. Under normal circumstances, when the treatment plan formulated by the doctor is effective, the blood glucose time-series data of diabetic patients will gradually return to normal levels. Therefore, in this invention, the overall trend and fluctuation of each blood glucose time-series data can be analyzed. Given that the trend term can characterize the overall change trend of the data, and the seasonal term can characterize the fluctuation and change pattern of the data, each blood glucose time-series data is decomposed to analyze the overall change trend of the trend term time-series data and the changes in the seasonal term time-series data. The normality of each medical record sample data is initially quantified to obtain initial normality indicators. Furthermore, given individual differences, even if blood glucose time-series data from different medical record samples show the same overall trend, they may exhibit varying degrees of fluctuation in local areas. Since the numerical characteristics of data values ​​in trend-related time-series data most directly reflect the normality of the data, for each trend-related time-series data, the local data fluctuations are combined with the numerical characteristics of the data values ​​to correct the initial normal indicators, resulting in the final normal indicators for each medical record sample. These final normal indicators not only characterize the normality of the medical record sample data but also reflect its reference value. Finally, to facilitate doctors in developing targeted treatment plans for different categories of medical record sample data, this invention categorizes all diabetic patient medical record sample data in the hospital database, obtaining classification results. Within each category, reference data is selected based on the final normal indicators to construct a reference database. This ensures that the reference data in the database covers different categories and has higher reference value, enabling doctors to develop more accurate treatment plans.

[0125] Please see Figure 5The diagram illustrates a system architecture of a machine learning-based diabetes data processing and analysis system according to an embodiment of the present invention. The system includes a processor 500, a memory 501, a bus 502, and a communication interface 503. The processor 500, communication interface 503, and memory 501 are connected via the bus 502. The memory 501 may include a high-speed random access memory, the bus 502 may be an ISA bus, PCI bus, or EISA bus, etc., and the processor 500 may be an integrated circuit chip with signal processing capabilities.

[0126] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0127] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0128] 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 principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine learning-based diabetes data processing and analysis system, characterized in that, The system includes: The diabetes data acquisition module is used to acquire medical record sample data of diabetic patients. The medical record sample data includes multiple parameters and blood glucose time series data. The normal indicator analysis module is used to decompose the corresponding trend item time series data and seasonal item time series data from each blood glucose time series data; comprehensively analyze the overall change trend of the trend item time series data and the change of the seasonal item time series data corresponding to each blood glucose time series data to determine the initial normal indicators corresponding to each medical record sample data; based on the local data fluctuation and numerical characteristics of the data values ​​in each trend item time series data, the initial normal indicators are corrected to determine the final normal indicators of each medical record sample data. The reference database construction module is used to classify all medical record sample data based on various parameters in the medical record sample data to obtain classification results; within each category, reference data is selected from the medical record sample data based on the final normal indicators, and a reference database is constructed based on the reference data; The method for obtaining the initial normal indicators includes: In the trend item time series data corresponding to each blood glucose time series data, a straight line is fitted to all data values, and the slope value of the fitted straight line is obtained. The slope value is normalized and negatively correlated to obtain the first normal factor. In the seasonal time series data corresponding to each blood glucose time series data, all minimum points are obtained, and the data between two adjacent minimum points is taken as a change data segment; In all the changing data segments, the second normal factor is determined based on the changes between the data values ​​in the changing data segments; Based on the first normal factor and the second normal factor corresponding to each blood glucose time series data, the initial normal index corresponding to each medical record sample data is calculated, and both the first normal factor and the second normal factor are positively correlated with the initial normal index. The methods for obtaining the second normal factor include: Within each changing data segment, the difference between the maximum and minimum data values ​​is taken as the fluctuation range of each changing data segment; Arrange the fluctuation amplitudes of all the changing data segments according to the time sequence of the changing data segments to obtain a sorted sequence. Then, perform backward difference on the values ​​in the sorted sequence to obtain a first-order difference sequence. Use the proportion of positive numbers in the first-order difference sequence as the second normal factor.

2. The diabetes data processing and analysis system based on machine learning according to claim 1, characterized in that, The method for obtaining the final normal indicators includes: Based on the local data fluctuations in the trend data of each blood glucose time series, the initial normal indicators are initially corrected to obtain the corrected normal indicators for each medical record sample data. Based on the numerical characteristics of the data values ​​in the trend items of each blood glucose time series data, the abnormal factors corresponding to each medical record sample data are determined. The abnormal factors corresponding to each medical record sample data are negatively correlated and normalized. The product of the normalized values ​​and the corresponding corrected normal indicators is then normalized to obtain the final normal indicators for each medical record sample data.

3. The diabetes data processing and analysis system based on machine learning according to claim 2, characterized in that, The method for obtaining the corrected normal indicators includes: In each trend item time series data, the data between two adjacent minimum points is taken as a fluctuation data segment; Analyze the continuity of fluctuations between all fluctuation data segments to determine the fluctuation continuity value; In each fluctuation data segment, the product of the difference between the maximum and minimum values ​​and the length of each fluctuation data segment is used as the fluctuation factor; the fluctuation factors of all fluctuation data segments are averaged to obtain the fluctuation characteristic value. Based on the fluctuation continuity value and fluctuation characteristic value, a correction factor is calculated for each medical record sample data, and both the fluctuation continuity value and the fluctuation characteristic value are negatively correlated with the correction factor; The normalized value of the product of the correction factor and the initial normal index corresponding to each medical record sample data is used as the corrected normal index for each medical record sample data.

4. The diabetes data processing and analysis system based on machine learning according to claim 3, characterized in that, The method for obtaining the continuous value of the fluctuation includes: The difference between the maximum values ​​of two adjacent fluctuation data segments at corresponding times is used as the time difference factor. The mean of all time difference factors in all fluctuation data segments is negatively correlated and mapped to the value of the fluctuation continuity.

5. A machine learning-based diabetes data processing and analysis system according to claim 2, characterized in that, The method for obtaining the abnormal factors includes: In the trend data of each blood glucose time series, data values ​​greater than a preset threshold are taken as target values, and consecutive target values ​​in the time series are filtered out to obtain all subsequences; The proportion of the target value in the time series data of each trend item is used as the quantity factor; the difference between the maximum value of all target values ​​and the preset threshold is used as the deviation factor; and the ratio of the maximum length of all subsequences to the minimum time value of all target values ​​is used as the time factor. The normalized value of the product of the quantity factor, deviation factor, and time factor is taken as the anomaly factor.

6. The diabetes data processing and analysis system based on machine learning according to claim 1, characterized in that, The methods for obtaining the classification results include: The parameters include at least age group, gender, presence or absence of complications, and type of diabetes; Among all medical record sample data, medical record sample data with the same parameters are grouped into one category, thus obtaining the classification results of all medical record sample data.

7. The diabetes data processing and analysis system based on machine learning according to claim 1, characterized in that, The method for obtaining the reference data includes: Within each category, all medical record sample data are sorted in descending order according to the final normal indicators to obtain a descending sequence; In the descending sequence corresponding to each category, the first preset number of medical record samples are used as reference data.

8. The diabetes data processing and analysis system based on machine learning according to claim 1, characterized in that, The process of decomposing the corresponding trend term time series data and seasonal term time series data from each blood glucose time series data includes: The STL time series decomposition algorithm is used to decompose each blood glucose time series data to obtain the trend item time series data and the seasonal item time series data corresponding to each blood glucose time series data.

Images