Hospital examination result sharing method based on big data

Abstract

The invention discloses a hospital examination and inspection result sharing method based on big data, and relates to the technical field of hospital examination and inspection result sharing. Examination data and inspection data are respectively stored in independent databases, and the two types of data are intelligently linked by using a unique identity label of a patient and a time stamp; the format, the time sequence and the semantic information are accurately matched to form a continuous and visual comprehensive disease course record; introducing a deep learning model, carrying out natural language processing on an unstructured text in the inspection data, and automatically converting an image description and a diagnosis report into structured data and a predefined semantic tag; according to the method, the workload of manual structuring is reduced, distributed storage, dynamic updating and multi-dimensional index query modules are constructed based on a big data analysis platform, real-time retrieval and joint query are conducted on comprehensive disease course data records, meanwhile, an early warning report is generated through an anomaly detection algorithm, and a basis is provided for clinical decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of sharing hospital examination and test results, and particularly to a method for sharing hospital examination and test results based on big data. Background Art

[0002] Among hospital examination and test results, there are both unstructured information, including image materials, video records, subjective descriptions, etc., and semi-structured texts, including physical examination reports, diagnosis descriptions, etc.; while test data mainly consists of quantitative indicators and chart data; due to the disparate data types, it is difficult for the platform to achieve automatic linkage; there are differences in format, update frequency, and storage structure among images, reports, and numerical data. When facing the entire medical history of a patient, it is necessary to retrieve data from different systems separately, which is time-consuming and error-prone; secondly, historical data fails to form an effective continuous timeline.

[0003] We found in the process of result sharing that simply relying on manual checking and manual integration is difficult to fundamentally solve the problems of inconsistent data formats and insufficient information interconnection; for a large amount of historical data, there is a lack of intelligent linkage analysis means, and the utilization efficiency of patient information is relatively low. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides a method for sharing hospital examination and test results based on big data to solve the problems in the traditional solution that simply relying on manual checking and manual integration is difficult to fundamentally solve the problems of inconsistent data formats and insufficient information interconnection.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] An embodiment of the present invention provides a method for sharing hospital examination and test results based on big data, which includes:

[0008] Obtaining examination data and test data containing the unique identity identifier of the patient from the data centers of each hospital, where the examination data includes image materials, text descriptions, and diagnostic reports, and the test data includes quantitative indicators and chemical analysis data;

[0009] According to the unique identity identifier of the patient and the data generation time, storing the collected examination data and test data in a separately set first database and second database respectively, where the first database is used to store unstructured and semi-structured examination data, and the second database is used to store structured test data;

[0010] Using the patient's unique identity identifier and time stamp, the data stored in the first database and the second database are respectively connected and processed to form a comprehensive medical record data record, and the comprehensive medical record data record is matched and connected in terms of data format, time series and semantic information;

[0011] Based on the big data analysis platform, the comprehensive medical record data record is retrieved and output to form a continuous and intuitive disease evolution map.

[0012] As a preferred solution of the method for sharing hospital examination and test results based on big data according to the present invention, wherein: in the connection processing process, a unified interface protocol is used to preprocess the examination data and test data, and the preprocessing includes format conversion, encoding unification, data cleaning and time stamp correction, so as to ensure the consistency of the data of each hospital in terms of data type, time stamp and patient's unique identity identifier;

[0013] In the connection processing process, the unstructured text information in the examination data is structured and matched and connected with the test data according to the time series.

[0014] As a preferred solution of the method for sharing hospital examination and test results based on big data according to the present invention, wherein: in the connection processing process, a deep learning model is used to perform natural language processing on the unstructured text information in the examination data, convert it into structured data and predefined semantic tags, and match the structured examination data with the test data according to the patient's unique identity identifier and time stamp;

[0015] The deep learning model is constructed and trained based on the hospital's historical medical record data, automatically identifies and extracts image description and diagnosis report information, and generates corresponding structured data.

[0016] As a preferred solution of the method for sharing hospital examination and test results based on big data according to the present invention, wherein: the original examination data and test data of each hospital data center are standardized, and format conversion, encoding unification, data cleaning and time stamp correction are performed. The process is expressed as:

[0017]

[0018] Wherein, represents the i-th original examination data, F represents the format conversion function for converting the original data into a unified format, U represents the encoding unification function for converting the data encoding into a predetermined standard, C represents the data cleaning function for removing noise and error information, T represents the time stamp correction function for adjusting the time stamp in the data to a unified standard, and b i represents the preprocessed examination data, represents the j-th original test data, cj Denote the preprocessed inspection data;

[0019] The T calculation process is defined as:

[0020]

[0021] Where x represents the data to be corrected, and Δt(x) represents the offset calculated based on the reference time standard, and this offset is further defined as:

[0022] Δt(x) = α · (t ref - t(x)),

[0023] Where t(x) represents the original time stamp in data x, and t ref represents the reference time standard, and α is an adjustment factor.

[0024] As a preferred solution of the method for sharing hospital inspection and test results based on big data according to the present invention, wherein: for the unstructured text information in the preprocessed inspection data, the deep learning model is applied for structured conversion to generate predefined semantic labels, and this process is completed by the cooperation of the encoder E and the decoder G, and its expression is:

[0025] d i = E(T(b i ))), e i = G(d i ),

[0026] Where T(b i ) represents the unstructured text extracted from the inspection data b i , E represents the encoder, which is used to convert text information into a structured feature vector, d i represents the structured feature vector output by the encoder, G represents the decoder, which is used to generate predefined semantic labels according to the feature vector, and e i represents the generated semantic label;

[0027] The model training constructs a training set based on the historical medical record dataset H and is optimized using a combined loss function. The loss function is defined as:

[0028]

[0029] Where represents the semantic label generation loss, represents the feature vector reconstruction loss, λ represents the balance factor, represents the overall loss function;

[0030] The semantic label generation loss is calculated using cross-entropy, and its formula is:

[0031]

[0032] Among them, y k represents the true label distribution, represents the predicted label distribution, and k represents the dimension of the label;

[0033] Feature vector reconstruction loss is described by the mean square error and is expressed as:

[0034]

[0035] Among them, represents the feature vector reconstructed by the decoder, and d i represents the original feature vector;

[0036] After combining the preprocessed data and the model output, a structured inspection data record is formed

[0037]

[0038] Among them, the structured inspection data record includes the preprocessed data b i , the structured feature vector d i and the semantic label e i .

[0039] As a preferred solution of the method for sharing hospital inspection and test results based on big data according to the present invention, wherein: cross-modal data connection is performed based on the patient's unique identity identifier and the time stamp, and the connection formula is:

[0040]

[0041] Among them, a(b i ) represents the patient's unique identity identifier included in the inspection data b i , a(c j ) represents the patient's unique identity identifier included in the test data c j , f(b i ) represents the time stamp of the inspection data b i , f(c j ) represents the time stamp of the test data c j , δ represents the allowable time difference threshold, and R represents the set of finally generated comprehensive course data records;

[0042] The time stamp extraction function f(x) is defined as:

[0043]

[0044] Among them, \(t(x)\) represents the original time stamp in data \(x\), and \(\Delta t'(x)\) represents the secondary adjustment term obtained based on the correction mechanism, and its calculation formula is:

[0045] \(\Delta t'(x)=\beta\cdot(t\) std -t(x))

[0046] Among them, \(t\) std represents the standard time, and \(\beta\) is the adjustment coefficient.

[0047] As a preferred solution of the method for sharing hospital examination and test results based on big data according to the present invention, wherein: the big data analysis platform process includes:

[0048] A distributed storage module for storing the comprehensive medical record data records formed after connection processing;

[0049] A data processing module that uses a distributed computing framework to dynamically update data, construct an index structure, and perform anomaly detection;

[0050] A multi-dimensional index query module for jointly retrieving the comprehensive medical record data records based on the data index structure and outputting a disease evolution map and an anomaly warning report.

[0051] As a preferred solution of the method for sharing hospital examination and test results based on big data according to the present invention, wherein: in the step of constructing the index structure, a data index structure is established by using the patient's unique identity identifier, time stamp, predefined semantic label, and numerical range in the test data;

[0052] In the retrieval output process, based on the big data analysis platform, the comprehensive medical record data records are queried and compared. The query is a joint retrieval based on the time stamp, semantic label, and numerical range, and a disease evolution map and an anomaly warning report are generated.

[0053] As a preferred solution of the method for sharing hospital examination and test results based on big data according to the present invention, wherein: on the big data analysis platform, a multi-dimensional index structure is constructed for the set \(R\) of the comprehensive medical record data records:

[0054] For each record \(r\in R\), a composite key is constructed, and its expression formula is:

[0055] \(k(r)=(a(r),\tau(r),e(r),I(r))\)

[0056] Among them, a(r) represents the unique patient identification contained in record r, which is derived from examination data or test data; τ(r) represents the time stamp of record r, and its value is obtained by weighted calculation based on the time stamps of examination data and test data; e(r) represents the predefined semantic label in record r, which is automatically extracted by a deep learning model; I(r) represents the numerical interval corresponding to the test data in record r.

[0057] τ(r) is defined as:

[0058] τ(r) = ω·f(b i ) + (1 - ω)·f(c j ),

[0059] Among them, f(b i ) represents the corrected time stamp in examination data b i ; f(c j ) represents the corrected time stamp in test data c j ; ω is the weight coefficient, and its value range is within [0, 1], which is used to balance the contributions of the two parts of data to the time stamp.

[0060] For the numerical interval parameter I(r), it is further expressed as:

[0061] I(r) = [min{v(c j )}, max{v(c j )}],

[0062] Among them, v(c j ) represents the index value in test data c j .

[0063] As a preferred solution of the hospital examination and test result sharing method based on big data described in the present invention, among them: after the composite key is constructed, the distributed storage module S is used to save the index set composed of all composite keys, which is expressed as IDX:

[0064] IDX = {k(r): r ∈ R},

[0065] The distributed storage module S and the data processing module P run in coordination. When a new record r' arrives, the dynamic update formula of the index set is:

[0066] IDX ← IDX ∪ {k(r')},

[0067] In the joint retrieval stage, when defining the patient identification a * , time range [t1, t2], semantic label e * and numerical interval I * for the query condition, the query operation is expressed as:

[0068]

[0069] After the data processing module P performs multi-dimensional joint matching on the query set Q using distributed computing, it sorts the results in ascending order according to the time stamp τ(r) to form a continuous disease evolution graph, and the sorting result is denoted as H:

[0070] H = Sort(Q,τ);

[0071] For the anomaly detection part, the anomaly detection function A(·) is used to perform statistical analysis on the numerical intervals of each record in the query results. The anomaly detection function is:

[0072]

[0073] Among them, φ(I(r)) represents the index value after statistical conversion of the numerical interval of record r, and μ(Q) represents the average value of φ(I(r)) of all records in the query set Q;

[0074] The anomaly threshold Δ is set as:

[0075] Δ = κ·σ(Q),

[0076] Among them, σ(Q) represents the standard deviation of the index values in the query set Q, and κ is a constant adjustment coefficient,

[0077] If the detection result satisfies A(Q)>Δ, an anomaly warning is triggered and a report is generated.

[0078] The beneficial effects of the present invention are as follows: The present invention stores the inspection data and the test data in independent databases respectively, and uses the unique patient identity identifier and time stamp to intelligently link the two types of data, and performs precise matching in terms of format, time series and semantic information to form a continuous and intuitive comprehensive disease course record; introduces a deep learning model to perform natural language processing on the unstructured text in the inspection data, and automatically converts the image description and diagnosis report into structured data and predefined semantic labels; reduces the workload of manual structuring, relies on the big data analysis platform, constructs a distributed storage, dynamic update and multi-dimensional index query module, performs real-time retrieval and joint query on the comprehensive disease course data records, and at the same time uses the anomaly detection algorithm to generate a warning report to provide a basis for clinical decision-making. Overcomes the limitations of traditional sharing methods and improves the intelligence and clinical application value of data sharing. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0080] Figure 1 It is a flowchart of the method for sharing hospital examination and test results of the present invention. DETAILED DESCRIPTION

[0081] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0082] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0083] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0084] Example 1, reference Figure 1 This embodiment provides a method for sharing hospital examination and test results based on big data, comprising the following steps:

[0085] Step S1, obtaining the examination data and test data containing the patient's unique identity from each hospital data center, the examination data including imaging data, text description and diagnosis report, and the test data including quantitative indicators and chemical analysis data;

[0086] Step S2, based on the patient's unique identity and the data generation time, the collected examination data and test data are stored in a first database and a second database that are independently set, respectively, the first database is used to store unstructured and semi-structured examination data, and the second database is used to store structured test data;

[0087] Step S3, using the patient's unique identity and time stamp, the data stored in the first database and the second database are linked to form a comprehensive medical history data record, which is matched and linked in terms of data format, time series and semantic information;

[0088] During the connection process, a unified interface protocol is used to preprocess the inspection data and test data. The preprocessing includes format conversion, encoding unification, data cleaning, and time stamp correction to ensure the consistency of data among hospitals in terms of data type, time stamp, and unique patient identification;

[0089] During the connection process, the unstructured text information in the inspection data is structured and matched and connected with the test data according to the time series;

[0090] During the connection process, a deep learning model is used to perform natural language processing on the unstructured text information in the inspection data, convert it into structured data and predefined semantic tags, and match the structured inspection data with the test data based on the unique patient identification and time stamp;

[0091] The deep learning model is constructed and trained based on the historical medical record data of the hospital, automatically identifies and extracts image description and diagnosis report information, and generates corresponding structured data;

[0092] The original inspection data and test data in the data centers of each hospital are standardized, and format conversion, encoding unification, data cleaning, and time stamp correction are performed. The process is expressed as:

[0093]

[0094] Among them, represents the i-th piece of original inspection data, F represents the format conversion function, which is used to convert the original data into a unified format, U represents the encoding unification function, which is used to convert the data encoding into a predetermined standard, C represents the data cleaning function, which is used to eliminate noise and error information, T represents the time stamp correction function, which is used to adjust the time stamp in the data to a unified standard, b i represents the preprocessed inspection data, represents the j-th piece of original test data, c j represents the preprocessed test data;

[0095] The calculation process of T is defined as:

[0096]

[0097] Among them, x represents the data to be corrected, Δt(x) represents the offset calculated according to the reference time standard, and this offset is further defined as:

[0098] Δt(x) = α·(t ref - t(x)),

[0099] Among them, t(x) represents the original time stamp in the data x, t ref represents the reference time standard, and α is the adjustment factor;

[0100] For the unstructured text information in the preprocessed inspection data, a deep learning model is applied for structured conversion to generate predefined semantic labels. This process is jointly completed by the encoder E and the decoder G, and its expression is:

[0101] d i = E(T(b i ))), e i = G(d i ),

[0102] where T(b i ) represents the unstructured text extracted from the inspection data b i . E represents the encoder, which is used to convert text information into a structured feature vector. d i represents the structured feature vector output by the encoder. G represents the decoder, which is used to generate predefined semantic labels based on the feature vector. e i represents the generated semantic label;

[0103] The model training constructs a training set based on the historical medical record dataset H and is optimized using a combined loss function. The loss function is defined as:

[0104]

[0105] where represents the semantic label generation loss, represents the feature vector reconstruction loss, and λ represents the balance factor, represents the overall loss function;

[0106] The semantic label generation loss is calculated using cross-entropy, and its formula is:

[0107]

[0108] where y k represents the true label distribution, represents the predicted label distribution, and k represents the dimension of the label;

[0109] The feature vector reconstruction loss is described using the mean squared error and is expressed as:

[0110]

[0111] where represents the feature vector reconstructed by the decoder, and d i represents the original feature vector;

[0112] After combining the preprocessed data and the model output, a structured inspection data record is formed

[0113]

[0114] Among them, the structured inspection data record contains the preprocessed data b i , the structured feature vector d i and the semantic label e i ;

[0115] Perform cross-modal data connection based on the patient's unique identity identifier and time stamp, and the connection formula is:

[0116]

[0117] Among them, a(b i ) represents the patient's unique identity identifier contained in the inspection data b i , a(c j ) represents the patient's unique identity identifier contained in the test data c j , f(b i ) represents the time stamp of the inspection data b i , f(c j ) represents the time stamp of the test data c j , δ represents the allowable time difference threshold, and R represents the set of comprehensive course data records finally generated;

[0118] The time stamp extraction function f(x) is defined as:

[0119]

[0120] Among them, t(x) represents the original time stamp in the data x, and Δt′(x) represents the secondary adjustment term obtained based on the correction mechanism, and its calculation formula is:

[0121] Δt′(x) = β · (t std -t(x)),

[0122] Among them, t std represents the standard time, and β is the adjustment coefficient;

[0123] Specifically, the data here is first subjected to format conversion, encoding unification, data cleaning, and time stamp correction, and the core function T is used to perform secondary correction on the time deviation to meet the unified time standard. The deep learning model is used to process the unstructured text information in the inspection data, and the structured feature vector and predefined semantic label are generated through the encoder and decoder, and the combined loss function is used to train the model; the structured inspection record and the preprocessed test data are matched according to the patient identity identifier and time stamp to form a comprehensive course data record;

[0124] The time marker extraction in cross-modal matching also adopts the quadratic expansion method to ensure the time consistency in the data connection process;

[0125] Step S4: Based on the big data analysis platform, retrieve and output the comprehensive course data records to form a continuous and intuitive disease evolution map;

[0126] The process of the big data analysis platform includes:

[0127] The distributed storage module is used to store the comprehensive course data records formed after connection processing;

[0128] The data processing module uses a distributed computing framework to dynamically update the data, construct the index structure, and detect anomalies;

[0129] The multi-dimensional index query module is used to jointly retrieve the comprehensive course data records based on the data index structure and output the disease evolution map and anomaly warning report;

[0130] In the index structure construction step, use the patient's unique identity identifier, time marker, predefined semantic label, and numerical range in the test data to establish a data index structure;

[0131] During the retrieval and output process, based on the big data analysis platform, query and compare the comprehensive course data records. The query is a joint retrieval based on the time marker, semantic label, and numerical range, and generate a disease evolution map and an anomaly warning report;

[0132] On the big data analysis platform, construct a multi-dimensional index structure for the comprehensive course data record set R:

[0133] For each record r ∈ R, construct a composite key, and its expression formula is:

[0134] k(r) = (a(r), τ(r), e(r), I(r)),

[0135] where a(r) represents the patient's unique identity identifier contained in record r, and this identifier is derived from examination data or test data; τ(r) represents the time marker of record r, and its value is obtained by weighted calculation based on the time markers of the examination data and test data; e(r) represents the predefined semantic label in record r, which is automatically extracted by the deep learning model; I(r) represents the numerical range corresponding to the test data in record r;

[0136] τ(r) is defined as:

[0137] τ(r) = ω · f(b i ) + (1 - ω) · f(c j ),

[0138] Among them, f(b i ) represents the corrected time stamp in the inspection data b i ; f(c j ) represents the corrected time stamp in the inspection data c j ; ω is the weight coefficient, and its value range is within [0, 1], which is used to balance the contributions of the two parts of data to the time stamp;

[0139] For the numerical interval parameter I(r), it is further expressed as:

[0140] I(r) = [min{v(c j )}, max{v(c j )}],

[0141] where v(c j ) represents the index value in the inspection data c j ;

[0142] After constructing the composite key, use the distributed storage module S to save the index set composed of all composite keys, denoted as IDX:

[0143] IDX = {k(r): r ∈ R},

[0144] The distributed storage module S and the data processing module P run in coordination. When a new record r' arrives, the dynamic update formula for the index set is:

[0145] IDX ← IDX ∪ {k(r')},

[0146] In the joint retrieval stage, when defining the patient identifier a * , time range [t1, t2], semantic label e * and numerical interval I * for the query condition, the query operation is expressed as:

[0147]

[0148] After the data processing module P performs multi-dimensional joint matching on the query set Q using distributed computing, the results are sorted in ascending order according to the time stamp τ(r) to form a continuous disease evolution map, and the sorting result is denoted as H:

[0149] H = Sort(Q, τ);

[0150] For the anomaly detection part, use the anomaly detection function A(·) to perform statistical analysis on the numerical intervals of each record in the query results. The anomaly detection function is:

[0151]

[0152] Among them, φ(I(r)) represents the index value after statistical conversion of the numerical range of record r, and μ(Q) represents the average value of φ(I(r)) of all records in the query set Q;

[0153] Set the anomaly threshold Δ as:

[0154] Δ = κ·σ(Q),

[0155] where σ(Q) represents the standard deviation of the index values in the query set Q, and κ is a constant adjustment coefficient.

[0156] If the detection result satisfies A(Q) > Δ, then trigger an anomaly warning and generate a report;

[0157] Specifically, a multi-dimensional index structure is constructed on the big data analysis platform here. The comprehensive course data records are composed of composite keys based on the unique patient identity identifier, time stamp, predefined semantic tags, and numerical ranges in the test data. Under the management of the distributed storage module S, the data processing module P is used to achieve real-time dynamic update to ensure the consistency of the index and the data. The joint retrieval generates a continuous disease condition evolution map through multi-condition matching and sorting. At the same time, a statistical anomaly detection method is adopted to judge and output a warning report according to the set anomaly threshold;

[0158] In summary, this embodiment realizes data standardization through the separate storage and unified preprocessing of inspection data and test data; uses semantic parsing and time series matching to intelligently connect unstructured inspection data with structured test data to form a continuous and intuitive comprehensive course record.

[0159] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for sharing hospital examination and test results based on big data, characterized in that: including obtaining examination data and test data containing the unique patient identification from the data centers of each hospital, where the examination data includes imaging materials, text descriptions, and diagnostic reports, and the test data includes quantitative indicators and chemical analysis data; storing the collected examination data and test data in a separately established first database and second database respectively according to the unique patient identification and the data generation time, where the first database is used to store unstructured and semi-structured examination data, and the second database is used to store structured test data; using the unique patient identification and time stamps to perform a joining process on the data stored in the first database and the second database respectively to form a comprehensive medical record data record, where the comprehensive medical record data record is matched and joined in terms of data format, time series, and semantic information; retrieving and outputting the comprehensive medical record data record based on a big data analysis platform to form a disease evolution map.

2. The method for sharing hospital examination and test results based on big data according to claim 1, wherein: During the joining process, a unified interface protocol is used to preprocess the examination data and test data, and the preprocessing includes format conversion, encoding unification, data cleaning, and time stamp correction; During the joining process, the unstructured text information in the examination data is structured and matched and joined with the test data according to the time series.

3. The method for sharing hospital examination and test results based on big data according to claim 2, wherein: During the joining process, a deep learning model is used to perform natural language processing on the unstructured text information in the examination data, convert it into structured data and predefined semantic tags, and match the structured examination data and test data according to the unique patient identification and time stamps; The deep learning model is constructed and trained based on the historical medical record data of the hospital, automatically identifies and extracts image description and diagnostic report information, and generates corresponding structured data.

4. The method for sharing hospital examination and test results based on big data according to claim 2, wherein: Perform standardization processing on the original examination data and test data of the data centers of each hospital, including format conversion, encoding unification, data cleaning, and time stamp correction. The process is expressed as: Among them, represents the i-th original inspection data, F represents a format conversion function for converting the original data into a unified format, U represents an encoding unification function for converting the data encoding into a predetermined standard, C represents a data cleaning function for removing noise and error information, T represents a time mark correction function for adjusting the time mark in the data to a unified standard, b i represents the preprocessed inspection data, represents the j-th original test data, c j represents the preprocessed test data; The T calculation process is defined as: where x represents the data to be corrected, and Δt(x) represents the offset calculated according to the reference time standard. This offset is further defined as: Δt(x) = α·(t ref - t(x)), Among them, t(x) represents the original time mark in data x, t ref represents the reference time standard, and α is the adjustment factor.

5. The method for sharing hospital examination and test results based on big data according to claim 4, wherein: For the unstructured text information in the examination data after preprocessing, apply a deep learning model for structured conversion to generate predefined semantic tags. This process is jointly completed by an encoder E and a decoder G, and its expression is: d i = E(T(b i )),e i = G(d i ), Among them, T(b i ) represents the unstructured text extracted from the inspection data b i . E represents the encoder, which is used to convert text information into structured feature vectors. d i represents the structured feature vectors output by the encoder. G represents the decoder, which is used to generate predefined semantic labels based on the feature vectors. e i represents the generated semantic label; The model training constructs a training set based on the historical medical record data set H and is optimized using a combined loss function. The loss function is defined as: Among them, represents the semantic label generation loss, represents the feature vector reconstruction loss, and λ represents the balance factor, represents the overall loss function; Semantic label generation loss It is calculated using cross-entropy, and its formula is: Among them, y k represents the true label distribution, represents the predicted label distribution, and k represents the dimension of the label; Feature vector reconstruction loss Described by the mean square error and expressed as: Among them, represents the feature vector reconstructed by the decoder, and d i represents the original feature vector; After combining the preprocessed data and the model output, a structured inspection data record is formed Among them, the structured inspection data record contains preprocessed data b i , structured feature vector d i and semantic label e i .

6. The method for sharing hospital examination and test results based on big data according to claim 5, wherein: Perform cross-modal data joining based on the unique patient identification and the time stamps. The joining formula is: Among them, a(b i ) represents checking the unique patient identification contained in data b i ; a(c j ) represents checking the unique patient identification contained in test data c j ; f(b i ) represents the time stamp of checking data b i ; f(c j ) represents the time stamp of test data c j ; δ represents the allowable time difference threshold; R represents the set of finally generated comprehensive course of disease data records; The time stamp extraction function f(x) is defined as: where t(x) represents the original time stamp in the data x, and Δt′(x) represents the secondary adjustment term obtained based on the correction mechanism. Its calculation formula is: Δt′(x) = β·(t std - t(x)), Among them, t std represents the standard time, and β is the adjustment coefficient.

7. The method for sharing hospital examination and test results based on big data according to claim 1, characterized in that: The big data analysis platform process includes: a distributed storage module for storing the comprehensive medical record data record formed after the joining process; a data processing module that uses a distributed computing framework to perform dynamic update, index structure construction, and anomaly detection on the data; The multi-dimensional index query module is used to perform a joint retrieval on the comprehensive medical record data based on the data index structure, and output a disease evolution graph and an abnormal warning report.

8. The method for sharing hospital examination and test results based on big data according to claim 1, wherein: In the step of constructing the index structure, a data index structure is established by using the patient's unique identity identifier, time stamp, predefined semantic tags, and the numerical range in the test data. In the process of retrieval and output, based on the big data analysis platform, the comprehensive medical record data is queried and compared. The query is a joint retrieval based on the time stamp, semantic tags, and numerical range, and a disease evolution graph and an abnormal warning report are generated.

9. A method for sharing hospital examination and test results based on big data according to claim 8, characterized in that: On the big data analysis platform, a multi-dimensional index structure is constructed for the set R of the comprehensive medical record data: Construct a composite key for each record r ∈ R, and its expression formula is: k(r) = (a(r), τ(r), e(r), I(r)), where a(r) represents the patient's unique identity identifier included in record r, and this identifier is derived from the examination data or test data; τ(r) represents the time stamp of record r, and its value is obtained by weighted calculation based on the time stamps of the examination data and test data; e(r) represents the predefined semantic tags in record r, which are automatically extracted by the deep learning model; I(r) represents the numerical range corresponding to the test data in record r. τ(r) is defined as: τ(r) = ω·f(b i ) + (1 - ω)·f(c j ) Among them, f(b i ) represents the corrected time stamp in the inspection data b i , f(c j ) represents the corrected time stamp in the inspection data c j , ω is the weight coefficient, and its value range is within [0, 1], which is used to balance the contributions of the two parts of data to the time stamp; For the numerical range parameter I(r), it is further expressed as: I(t) = [min{v(c j )}, max{v(c j )}], Among them, v(c j ) represents the index value in the test data c j .

10. A method for sharing hospital examination and test results based on big data according to claim 9, characterized in that: After the composite key is constructed, the index set composed of all composite keys is saved by using the distributed storage module S, denoted as IDX: IDX = {k(r): r ∈ R}, The distributed storage module S and the data processing module P run in coordination. When a new record r' arrives, the dynamic update formula of the index set is: IDX ← IDX ∪ {k(r')}, In the combined retrieval stage, a patient identifier a is defined for the query condition * , time range [t1, t2], semantic label e * and numerical interval I * When this is the case, the query operation is expressed as: After the data processing module P performs multi-dimensional joint matching on the query set Q by using distributed computing, the results are sorted in ascending order according to the time stamp τ(r) to form a continuous disease evolution graph, and the sorting result is denoted as H: H = Sort(Q, τ); For the abnormal detection part, an abnormal detection function A(·) is used to perform statistical analysis on the numerical ranges of each record in the query results. The abnormal detection function is: where φ(I(r)) represents the index value after statistical conversion of the numerical range of record r, and μ(Q) represents the average value of φ(I(r)) of all records in the query set Q. Set the abnormal threshold Δ as: Δ = κ · σ(Q), where σ(Q) represents the standard deviation of the index values in the query set Q, and κ is a constant adjustment coefficient. If the detection result satisfies A(Q) > Δ, an abnormal warning is triggered and a report is generated.

Images