A data quality control method and device for a medical big data platform

Abstract

This application discloses a data quality control method and device for a medical big data platform. The method involves acquiring medical data to be processed, including at least medical record data, laboratory data, and image data. Quality control checks are performed on the medical data to obtain initial quality control results. When the initial quality control results indicate that the medical data is suspicious, abnormal feature detection is performed on the medical record data to obtain text anomaly probabilities. Abnormal feature detection is also performed on the image data to obtain image anomaly probabilities. The method checks whether the laboratory data is within a reference range. If the laboratory data is within the reference range, the final quality control result of the medical data is determined based on the text anomaly probabilities and image anomaly probabilities. By initially screening out suspicious data through quality control and then performing further quality control checks on these data, the method maintains high efficiency while ensuring the accuracy of results when processing massive amounts of data.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a data quality control method and device for a medical big data platform. Background Technology

[0002] With the rapid development of hospital informatization and regional medical big data platforms, the scale and complexity of medical data are constantly expanding, involving various information types such as electronic medical records, test results, imaging data, and surgical records. This data plays a crucial role in clinical decision-making and medical research; however, ensuring data quality and consistency has become a major challenge in the application of medical big data.

[0003] Currently, traditional methods for quality control of medical big data primarily rely on manual quality control. Manual quality control involves professionals randomly checking and reviewing data to identify obvious errors. While this method can effectively detect some non-compliant data, especially in cases of obvious errors, the sheer volume and complexity of medical big data makes manual quality control extremely inefficient. In particular, when faced with massive amounts of data, manual review cannot promptly and accurately identify problems. Summary of the Invention

[0004] This application provides a data quality control method and device for a medical big data platform to solve the problem that manual review cannot detect problems in the data in a timely and accurate manner when faced with massive amounts of data.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] The first aspect of this application provides a data quality control method for a medical big data platform, including:

[0007] Acquire medical data to be processed; the medical data to be processed includes at least medical record data, test data, and imaging data;

[0008] The medical data to be processed is subjected to quality control testing to obtain initial quality control results;

[0009] When the initial quality control result indicates that the medical data to be processed is suspicious, abnormal feature detection is performed on the medical record data to obtain the text abnormality probability;

[0010] Anomaly detection is performed on the image data to obtain the probability of image anomalies;

[0011] Check whether the test data is within the reference range;

[0012] If the test data is within the reference range, the final quality control result of the medical data to be processed is determined based on the text anomaly probability and the image anomaly probability.

[0013] Optionally, acquiring the medical data to be processed includes:

[0014] Collect initial medical data;

[0015] The initial medical data was subjected to quality testing;

[0016] Once the initial medical data passes quality control, it undergoes data anonymization to obtain the medical data to be processed.

[0017] Optionally, the step of performing quality control testing on the medical data to be processed to obtain initial quality control results includes:

[0018] The medical data to be processed is subjected to a data integrity check to obtain a first check result;

[0019] By calling reference data information, outlier detection is performed on the medical data to be processed to obtain a second detection result;

[0020] The consistency of the examination results in the medical data to be processed is verified to obtain the verification results; the examination results include at least the examination results of medical record data, the examination results of laboratory data, and the examination results of imaging data.

[0021] The initial quality control result is determined based on the first test result, the second test result, and the verification result.

[0022] Optionally, the step of performing abnormal feature detection on the medical record data to obtain the text abnormality probability includes:

[0023] The medical record data is input into a natural language processing model for text analysis to obtain the probability of text anomalies.

[0024] Optionally, the step of performing anomaly feature detection on the image data to obtain the image anomaly probability includes:

[0025] The image data is input into a deep learning image model to extract abnormal features and obtain the image anomaly probability.

[0026] Optionally, determining the final quality control result of the medical data to be processed based on the text anomaly probability and the image anomaly probability includes:

[0027] When the probability of image abnormality is a preset multiple of the probability of text abnormality, the medical data to be processed is marked as a potential quality control error.

[0028] Based on the text anomaly probability and the image anomaly probability, a conflict score is calculated;

[0029] When the conflict score value is greater than a preset threshold, the potential quality control error is changed to a quality control error and determined as the final quality control result of the medical data to be processed.

[0030] A second aspect of this application provides a data quality control device for a medical big data platform, comprising:

[0031] An acquisition unit is used to acquire medical data to be processed; the medical data to be processed includes at least medical record data, test data, and image data.

[0032] The quality control and testing unit is used to perform quality control and testing on the medical data to be processed to obtain initial quality control results.

[0033] The first detection unit is used to perform abnormal feature detection on the medical record data and obtain the text abnormality probability when the initial quality control result indicates that the medical data to be processed is suspicious data.

[0034] The second detection unit is used to perform abnormal feature detection on the image data to obtain the probability of image abnormality.

[0035] The third detection unit is used to detect whether the test data is within the reference range;

[0036] The determining unit is used to determine the final quality control result of the medical data to be processed based on the text anomaly probability and the image anomaly probability if the test data is within the reference range.

[0037] Optionally, the acquisition unit is specifically used for:

[0038] Collect initial medical data;

[0039] The initial medical data was subjected to quality testing;

[0040] Once the initial medical data passes quality control, it undergoes data anonymization to obtain the medical data to be processed.

[0041] Optionally, the quality control and testing unit is specifically used for:

[0042] The medical data to be processed is subjected to a data integrity check to obtain a first check result;

[0043] By calling reference data information, outlier detection is performed on the medical data to be processed to obtain a second detection result;

[0044] The consistency of the examination results in the medical data to be processed is verified to obtain the verification results; the examination results include at least the examination results of medical record data, the examination results of laboratory data, and the examination results of imaging data.

[0045] The initial quality control result is determined based on the first test result, the second test result, and the verification result.

[0046] Optionally, the first detection unit is specifically used for:

[0047] The medical record data is input into a natural language processing model for text analysis to obtain the probability of text anomalies.

[0048] The technical solution provided in this application acquires medical data to be processed; the medical data to be processed includes at least medical record data, laboratory data, and imaging data; quality control testing is performed on the medical data to be processed to obtain initial quality control results; when the initial quality control results indicate that the medical data to be processed is suspicious, abnormal feature detection is performed on the medical record data to obtain the text anomaly probability; abnormal feature detection is performed on the imaging data to obtain the imaging anomaly probability; whether the laboratory data is within the reference range is checked; if the laboratory data is within the reference range, the final quality control result of the medical data to be processed is determined based on the text anomaly probability and the imaging anomaly probability. By initially screening out suspicious data through quality control and further conducting quality control testing on these data, high efficiency and accuracy of results can be maintained when processing massive amounts of data. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a data quality control method for a medical big data platform provided in this application embodiment;

[0051] Figure 2 This is a schematic diagram of the architecture of a data quality control device for a medical big data platform, provided as an embodiment of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] In this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0054] like Figure 1 The flowchart shown is a data quality control method for a medical big data platform provided in an embodiment of this application, including:

[0055] S101: Obtain medical data to be processed.

[0056] The medical data to be processed includes at least medical record data, test data, imaging data, and dynamic monitoring data generated by the patient outside the hospital (such as continuous monitoring data of blood pressure and blood sugar).

[0057] Optionally, in another embodiment of this application, the specific implementation of step S101 includes processes A1 to A3.

[0058] A1: Collect initial medical data.

[0059] Among them, a multi-channel data acquisition mechanism is established through interfaces with the Hospital Information System (HIS), Laboratory Information System (LIS), Picture Archiving and Communication System (PACS), and wearable devices to collect initial medical data.

[0060] A2: Perform quality checks on the initial medical data.

[0061] Understandably, before the data enters the medical big data platform, a lightweight quality control module is embedded at the source to perform quality checks on the initial medical data, which means quickly checking the data file format, field integrity, and upload latency.

[0062] For example, when a necessary label (such as examination site information) is detected missing in a DICOM image file, an alarm can be issued during the acquisition stage, rather than waiting until the subsequent processing stage.

[0063] In addition, during the quality inspection process, master index alignment and consistency comparison are performed on the same patient data from different medical institutions to ensure that subsequent quality control is carried out at the patient level, rather than targeting a single data item.

[0064] A3: After the initial medical data passes the quality control, the initial medical data is anonymized to obtain the medical data to be processed.

[0065] Specifically, the initial medical data undergoes anonymization, automatically removing or encrypting strong identifiers such as names, ID numbers, and addresses, and adhering to the principle of minimization, retaining only the fields required for quality control. Simultaneously, time perturbation technology is used to obfuscate timestamps accurate to the second, ensuring the integrity of the quality control logic while effectively reducing the risk of data leakage.

[0066] S102: Perform quality control testing on the medical data to be processed to obtain initial quality control results.

[0067] The initial quality control results include: qualified data, questionable data (i.e., candidate error data), and significantly erroneous data.

[0068] Qualified data is data that passes the rule verification and can be directly used in downstream applications.

[0069] Significantly erroneous data, which is unrecoverable or logically contradictory (such as missing key fields or corrupted files), is directly removed and stored in the exception repository.

[0070] Suspicious data refers to data whose authenticity or error cannot be clearly determined and requires further analysis.

[0071] Optionally, in another embodiment of this application, the specific implementation of step S102 includes process B1 value process B4.

[0072] B1: Perform data integrity checks on the medical data to be processed to obtain the first test result.

[0073] The first test result is the result of whether the medical data to be processed is complete.

[0074] It is understandable that data integrity checks on medical data being processed may result in missing or abnormal data.

[0075] Furthermore, for missing or abnormal data that can be inferred or recovered, automatic correction will be prioritized. For example, logical errors in dates (discharge date earlier than admission date) will be automatically corrected using contextual information, and missing fields can be filled by inference from knowledge graphs or historical cases. Abnormal numerical values ​​(such as blood glucose value = 500) can be compared with historical distributions and marked as "suspected data entry error".

[0076] For example, data that cannot be directly corrected will be labeled as errors and highlighted in the quality control report. Labeling categories include "missing," "abnormal," "contradictory," and "suspected duplication," providing users with correction suggestions (such as recommending reasonable ranges or reference values).

[0077] For example, data with serious errors that cannot be repaired (such as completely missing key fields or corrupted image files) is removed. Removed data is not discarded directly but stored in an anomaly data warehouse for subsequent statistical analysis and model retraining.

[0078] B2: Call the reference data information to perform outlier detection on the medical data to be processed and obtain a second detection result.

[0079] The second test result is the result of whether there are outliers in the medical data to be processed.

[0080] Understandably, based on the patient's age in the medical data to be processed, reference data information is called to perform outlier detection on the medical data to be processed.

[0081] For example, when testing a patient who is 2 years old, reference data from a children's reference range library (such as the ranges of hemoglobin, white blood cells, etc., which are different from those of adults) is used to detect outliers in the medical data to be processed, thereby avoiding the mistaken identification of normal values ​​as abnormal.

[0082] B3: Perform consistency verification on the examination results in the medical data to be processed, and obtain the verification results.

[0083] The examination results should include at least the examination results of medical record data, examination results of laboratory data, and examination results of imaging data.

[0084] For example, if the diagnosis in the case data is "no abnormality" but the imaging data is missing or empty, it will be marked as a potential conflict and will be entered into the manual review queue in advance.

[0085] Understandably, the above content can help identify common "modal missing" or "intermodal contradictions" at an early stage, reducing the computational burden on subsequent deep models.

[0086] B4: Determine the initial quality control results based on the first test result, the second test result, and the verification result.

[0087] Understandably, if data passes quality control in the first, second, and validation stages without serious issues, it is marked as acceptable data. If serious errors are found at any stage of the testing process (such as missing key fields, obvious numerical anomalies, or uncorrectable data inconsistencies), the data is marked as erroneous data. Data with minor issues (such as fields that can be presumably filled, outliers that can be corrected using historical data, or minor modal missing or inconsistent data) is marked as suspicious data.

[0088] It should be noted that when reviewers confirm a certain error pattern (for example, a hospital's uploaded data contains a specific encoding error), the error pattern will automatically be converted into a new rule and added to the rule base, thereby enabling the rule engine to continuously learn and optimize, forming an "adaptive, learning rule engine".

[0089] Furthermore, the results of all corrections and removals are fed back into the system's learning module to continuously optimize the quality control strategy. After initial quality control, the system marks suspicious data as candidate error data and sends it to the intelligent model for further processing. Through this initial quality control process, medical data is quickly and systematically filtered to identify obvious errors or missing data at minimal cost.

[0090] S103: When the initial quality control result indicates that the medical data to be processed is suspicious, perform abnormal feature detection on the medical record data to obtain the text abnormality probability.

[0091] Optionally, the specific implementation process for detecting abnormal features in medical record data to obtain the text anomaly probability is as follows: input the medical record data into a natural language processing model (such as the BERT model) for text analysis to obtain the text anomaly probability.

[0092] It should be noted that the natural language processing model used has been fine-tuned with medical corpora and is specifically designed to identify professional terminology, abbreviations, spelling variations, and unique clinical descriptions in medical documents. For example, terms such as "high blood sugar," "blood sugar ↑," and "GLU: 15mmol / L" can all be uniformly normalized to the same semantic representation. Unlike existing technologies that process sentences through general vectorization, this model incorporates a medical-specific vocabulary during text representation, giving diseases, symptoms, and test results higher discriminative power in the vector space. Simultaneously, the model also considers structured fields in medical records (such as test dates and department affiliations), ensuring that the text representation not only reflects semantics but also includes clinical contextual information. In this way, the model can accurately distinguish between descriptions like "patient has no obvious abnormalities" and "patient has no obvious head abnormalities," thereby avoiding erroneous quality control judgments due to semantic ambiguity.

[0093] In addition, when the medical data to be processed is suspicious, the model can be used to classify the suspicious data into subcategories such as "entry error, logical contradiction, extreme but reasonable value", and the reasonableness can be determined by combining the medical record context, historical distribution and patient characteristics (age, gender, disease background).

[0094] S104: Perform anomaly detection on the image data to obtain the probability of image anomalies.

[0095] Optionally, the specific implementation process of step S104 is as follows: input the image data into a deep learning image (e.g., CNN) model to extract abnormal features and obtain the image abnormality probability.

[0096] It should be noted that the deep learning image and natural language processing model in this application is an improvement on the traditional conditional probability model, mainly in the following aspects:

[0097] 1. Dynamic Weighting Mechanism: Different quality control indicators have varying degrees of importance; for example, the risk of drug dosage errors is higher than that of spelling errors. The system automatically adjusts the weights of various error types based on historical quality control results, allowing the model to focus more on issues with higher clinical risk. For instance, if frequent dosage errors are detected in recent reports, the detection priority of "dosage-related errors" will be automatically increased.

[0098] 2. Multidimensional Constraints: Traditional methods typically only consider the "possibility of a certain error occurring," while improved methods comprehensively consider contextual conditions such as disease type, examination items, and age group. For example, for reports of "pediatric patients," the system will more rigorously check drug dosage during quality control because the risk of dosage errors is higher in children.

[0099] 3. Continuous Feedback Learning: The model is not fixed at once, but continuously absorbs feedback results. Whenever a doctor confirms whether a quality control prompt is correct, the system records the feedback and adjusts subsequent probability judgments, thus making the model gradually more accurate during use.

[0100] S105: Whether the test data are within the reference range.

[0101] If the test data is within the reference range, proceed to step S106.

[0102] The process involves checking whether the test data falls within the reference range. If the test data is outside the reference range, especially if it exceeds the range (e.g., a blood glucose level of 20 mmol / L), the relevant descriptive fields in the case data are analyzed simultaneously. If contradictory descriptions such as "blood glucose level is within the normal range" still appear in the case data, they will be immediately identified as conflicting data. This process combines the aforementioned "conditional adaptive modeling," dynamically adjusting the abnormality thresholds for different populations (e.g., newborns, pregnant women, kidney dialysis patients) to ensure the clinical rationality of the comparison results and avoid false alarms.

[0103] S106: Determine the final quality control result of the medical data to be processed based on the text anomaly probability and the image anomaly probability.

[0104] Understandably, conflict scores are calculated based on the probability of text anomalies and the probability of image anomalies, and the final quality control result of the medical data to be processed is determined based on the conflict scores.

[0105] Optionally, in another embodiment of this application, the specific implementation of determining the final quality control result of the medical data to be processed based on the text anomaly probability and the image anomaly probability in step S106 includes processes C1 to C3.

[0106] C1: When the probability of image abnormality is a preset multiple, the medical data to be processed will be marked as a potential quality control error.

[0107] Optionally, the preset multiplier may include, but is not limited to, 0.2x.

[0108] Specifically, when the probability of image anomaly is a preset multiple of the probability of text anomaly, it indicates that the probability of image anomaly is very high while the probability of text anomaly is very low. This suggests that the image findings do not match the textual conclusions, and the medical data to be processed is marked as a potential quality control error. For example, the image indicates the presence of a significant lesion, while the text describes the image as normal.

[0109] C2: The conflict score is calculated based on the text anomaly probability and the image anomaly probability.

[0110] It should be noted that the specific form of the conflict score value calculated based on the text anomaly probability and the image anomaly probability is shown in formula (1).

[0111] (1)

[0112] In formula (1), For conflict score, This represents the probability of image abnormalities. This represents the probability of text anomalies.

[0113] It can be seen that the first term in formula (1) measures the magnitude of the disagreement between the two parties; the second term weights the severity of the abnormal event (for example, the disagreement when both parties think it is "very likely to be abnormal" is more severe than the disagreement when both parties think it is "very likely to be normal").

[0114] C3: When the conflict score is greater than the preset threshold, the potential quality control error is changed to a quality control error and determined as the final quality control result of the medical data to be processed.

[0115] Where T is a preset threshold, when When a potential quality control error is identified, it is changed to a quality control error and identified as the final quality control result of the medical data to be processed. A modal consistency alarm is triggered, prompting manual review.

[0116] It's important to note that potential quality control errors are not discarded directly but instead enter the subsequent intelligent correction suggestion stage. First, errors are categorized according to the type of contradiction and their severity is assessed. For example, minor errors might be imprecise wording (e.g., mistakenly writing "normal" instead of "mildly abnormal"); moderate errors might be contradictory location labeling (e.g., inconsistent labeling on the left and right sides); while severe errors might contradict the presence or absence of lesions, seriously affecting diagnosis. For cases with obvious imaging features, the system automatically generates correction suggestions (e.g., changing "normal image" to "abnormal lesion visible on image") and marks them as "awaiting manual confirmation." Furthermore, potential quality control errors are stored in an anomaly data warehouse for secondary model training. The system records error patterns and updates rules and model weights, thereby improving the accuracy of subsequent contradiction detection.

[0117] Furthermore, the final quality control results are output and feedback through a multi-level process. First, data deemed normal is directly stored and analyzed on the medical big data platform. Abnormal data with low confidence levels is marked as "candidate error data" and pushed to the manual review stage. Data with serious conflicts or obvious errors triggers an alarm and is recorded in the quality control log.

[0118] Regarding the visual feedback mechanism, the final quality control results are output through a visual interface, intuitively displaying abnormal fields, conflict patterns, and error locations. For example, if a CT image detects a lesion but the report describes it as "normal," the interface will highlight the conflict area and the corresponding text content. Furthermore, it supports returning the final quality control results to the hospital information system via a standard interface, allowing clinical staff to receive feedback on their original work platform. For candidate erroneous data, manual reviewers can mark "Confirmed as erroneous" or "No modification required" on the interface. This mark will automatically enter the feedback channel and serve as a training sample for subsequent model optimization.

[0119] After outputting and providing feedback, manually verified error samples and conflict patterns are automatically added to the quality control knowledge base, forming an "error case library." For example, if multiple instances of "test values ​​exceeding the standard but the text report describing them as normal" occur, this pattern will be extracted into a rule template for subsequent quality control calls. Simultaneously, dynamic updates to the rule library and model parameters are supported. If a certain type of error occurs frequently, the detection threshold will be automatically adjusted or the rule conditions expanded to improve the detection rate. To achieve self-evolutionary learning, the accumulated verification results are used to retrain the model, gradually enabling it to recognize new types of errors. For example, when new inspection methods (such as gene sequencing data) are introduced to the platform, it can quickly adapt to this data quality control requirement with minimal manual annotation.

[0120] It should be noted that the following effects can be achieved through the implementation of steps S101 to S106:

[0121] 1. By combining artificial intelligence models and rule engines, this application has achieved automated and real-time quality control of medical data, significantly reducing the workload of manual review of each item, improving quality control efficiency, and better adapting to the processing needs of large-scale medical big data.

[0122] 2. It can simultaneously process electronic medical record text, structured test values, and medical image data, achieving cross-modal consistency detection and identifying semantic conflicts and logical contradictions that cannot be detected by quality control of single data types.

[0123] 3. The text quality control submodule, based on medical knowledge graphs and pre-trained language models, can not only perform surface-level format verification but also understand medical semantics and clinical logic, uncovering deeper issues such as "contradictions between diagnosis and chief complaint" and "inconsistencies between imaging conclusions and medical records." Combined with statistical modeling and anomaly detection algorithms, it can not only identify out-of-range values ​​but also distinguish between data entry errors and genuine anomalies, reducing false alarm rates.

[0124] 4. It can be quickly configured and expanded according to different medical institutions, departments, and standards, and has good cross-institutional generalization capabilities, facilitating large-scale promotion and application. Quality control results are fed back to the model, forming a closed-loop learning process that continuously improves the accuracy and robustness of quality control, enabling the system to adapt to constantly changing medical knowledge and data distribution.

[0125] like Figure 2 The diagram shown is an architectural schematic of a data quality control device for a medical big data platform provided in an embodiment of this application. The data quality control device includes: an acquisition unit 100, a quality control detection unit 200, a first detection unit 300, a second detection unit 400, a third detection unit 500, and a determination unit 600.

[0126] The acquisition unit 100 is used to acquire medical data to be processed; the medical data to be processed includes at least medical record data, test data and image data.

[0127] The acquisition unit 100 is specifically used for: collecting initial medical data; performing quality checks on the initial medical data; and after the initial medical data passes the quality check, performing data anonymization processing on the initial medical data to obtain the medical data to be processed.

[0128] The quality control testing unit 200 is used to perform quality control testing on the medical data to be processed and obtain initial quality control results.

[0129] The quality control testing unit 200 is specifically used for: performing data integrity testing on the medical data to be processed to obtain a first test result; calling reference data information to perform outlier detection on the medical data to be processed to obtain a second test result; performing consistency verification on the examination results in the medical data to be processed to obtain a verification result; the examination results include at least the examination results of medical record data, the examination results of laboratory data, and the examination results of imaging data; and determining the initial quality control result based on the first test result, the second test result, and the verification result.

[0130] The first detection unit 300 is used to perform abnormal feature detection on medical record data and obtain the text abnormality probability when the initial quality control result indicates that the medical data to be processed is suspicious data.

[0131] The first detection unit 300 is specifically used to: input medical record data into a natural language processing model for text analysis and obtain the probability of text anomalies.

[0132] The second detection unit 400 is used to detect abnormal features in the image data and obtain the probability of image abnormality.

[0133] The second detection unit 400 is specifically used to: input image data into a deep learning image model to extract abnormal features and obtain the probability of image abnormality.

[0134] The third detection unit 500 is used to detect whether the test data is within the reference range.

[0135] Unit 600 is used to determine the final quality control result of the medical data to be processed based on the text anomaly probability and the image anomaly probability if the test data is within the reference range.

[0136] The determination unit 600 is specifically used to: mark the medical data to be processed as a potential quality control error when the probability of image abnormality is a preset multiple of the probability of text abnormality; calculate the conflict score value based on the probability of text abnormality and the probability of image abnormality; and change the potential quality control error to a quality control error when the conflict score value is greater than a preset threshold, and determine it as the final quality control result of the medical data to be processed.

[0137] In summary, by screening out suspicious data through preliminary quality control and then conducting further quality control tests on these data, we can maintain high efficiency and ensure the accuracy of results when processing massive amounts of data.

[0138] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. In particular, for system or system embodiments, since they are fundamentally similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. Components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0139] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0140] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data quality control method for a medical big data platform, characterized in that, include: Acquire medical data to be processed; the medical data to be processed includes at least medical record data, test data, and imaging data; The medical data to be processed is subjected to quality control testing to obtain initial quality control results; When the initial quality control result indicates that the medical data to be processed is suspicious, abnormal feature detection is performed on the medical record data to obtain the text abnormality probability; Anomaly detection is performed on the image data to obtain the probability of image anomalies; Check whether the test data is within the reference range; If the test data is within the reference range, the final quality control result of the medical data to be processed is determined based on the text anomaly probability and the image anomaly probability.

2. The method of claim 1, wherein, The acquisition of medical data to be processed includes: Collect initial medical data; The initial medical data was subjected to quality testing; Once the initial medical data passes quality control, it undergoes data anonymization to obtain the medical data to be processed.

3. The method of claim 1, wherein, The process of performing quality control testing on the medical data to be processed to obtain initial quality control results includes: The medical data to be processed is subjected to a data integrity check to obtain a first check result; By calling reference data information, outlier detection is performed on the medical data to be processed to obtain a second detection result; The consistency of the examination results in the medical data to be processed is verified to obtain the verification results; the examination results include at least the examination results of medical record data, the examination results of laboratory data, and the examination results of imaging data. The initial quality control result is determined based on the first test result, the second test result, and the verification result.

4. The method of claim 1, wherein, The step of performing abnormal feature detection on the medical record data to obtain the text abnormality probability includes: The medical record data is input into a natural language processing model for text analysis to obtain the probability of text anomalies.

5. The method according to claim 1, characterized in that, The step of performing anomaly feature detection on the image data to obtain the image anomaly probability includes: The image data is input into a deep learning image model to extract abnormal features and obtain the image anomaly probability.

6. The method of claim 1, wherein, The determination of the final quality control result of the medical data to be processed based on the text anomaly probability and the image anomaly probability includes: When the probability of image abnormality is a preset multiple of the probability of text abnormality, the medical data to be processed is marked as a potential quality control error. Based on the text anomaly probability and the image anomaly probability, a conflict score is calculated; When the conflict score value is greater than a preset threshold, the potential quality control error is changed to a quality control error and determined as the final quality control result of the medical data to be processed.

7. A data quality control device for a medical big data platform, characterized in that, include: An acquisition unit is used to acquire medical data to be processed; the medical data to be processed includes at least medical record data, test data, and image data. The quality control and testing unit is used to perform quality control and testing on the medical data to be processed to obtain initial quality control results. The first detection unit is used to perform abnormal feature detection on the medical record data and obtain the text abnormality probability when the initial quality control result indicates that the medical data to be processed is suspicious data. The second detection unit is used to perform abnormal feature detection on the image data to obtain the probability of image abnormality. The third detection unit is used to detect whether the test data is within the reference range; The determining unit is used to determine the final quality control result of the medical data to be processed based on the text anomaly probability and the image anomaly probability if the test data is within the reference range.

8. The device according to claim 7, characterized in that, The acquisition unit is specifically used for: Collect initial medical data; The initial medical data was subjected to quality testing; Once the initial medical data passes quality control, it undergoes data anonymization to obtain the medical data to be processed.

9. The device according to claim 7, characterized in that, The quality control and testing unit is specifically used for: The medical data to be processed is subjected to a data integrity check to obtain a first check result; By calling reference data information, outlier detection is performed on the medical data to be processed to obtain a second detection result; The consistency of the examination results in the medical data to be processed is verified to obtain the verification results. The examination results include at least the examination results of medical record data, the examination results of laboratory data, and the examination results of imaging data; The initial quality control result is determined based on the first test result, the second test result, and the verification result.

10. The device according to claim 7, characterized in that, The first detection unit is specifically used for: The medical record data is input into a natural language processing model for text analysis to obtain the probability of text anomalies.

Images