County tumor screening and referral platform based on regional medicine combination

Abstract

The embodiment of the invention provides a county tumor screening and referral platform based on regional medical conformity, relates to the technical field of medical information processing, and can solve the problems of data inconsistency, uneven quality and the like in multi-source medical data integration analysis. The county tumor screening and referral platform based on the regional medical coexistence comprises a multi-modal data acquisition module, a format processing module, an anomaly detection module, a cross validation module, an anomaly positioning module, an anomaly repair module, an anomaly recognition module and an evaluation module.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to a county-level tumor screening and referral platform based on a regional medical consortium. Background Technology

[0002] Cancer screening is a crucial area for regional medical consortia to promote the modernization of county-level medical services, directly impacting the effectiveness of residents' health protection and early disease intervention. Efficient and accurate screening and referral platforms can significantly improve the cancer diagnostic capabilities of primary hospitals, reduce the burden of medical expenses for patients, and optimize the allocation of medical resources. However, existing screening methods have revealed significant shortcomings in practical applications. Current technologies largely rely on the independent operation of single testing methods, lacking a systematic mechanism for integrating and analyzing data from multiple tests. This leads to biases in data flow between different stages, affecting the reliability of screening results. Furthermore, existing platforms have weak capabilities in monitoring data anomalies, making it difficult to detect potential problems in equipment or operation in real time, thus reducing the screening quality in primary hospitals.

[0003] In county-level cancer screening, the core challenge lies in ensuring the integrity and consistency of data from multiple testing methods. Firstly, the data generation and transmission processes of various methods, such as imaging examinations, blood biomarker testing, and genetic screening, involve multiple nodes. The significant differences in data formats and verification procedures between these nodes can easily lead to data loss or anomalies. This data integrity issue further exacerbates another challenge: how to promptly identify weaknesses such as equipment malfunctions or sample processing abnormalities within the complex and diverse data streams to ensure the accuracy of screening results. This continuous requirement, from data integration to anomaly identification, constitutes the technical bottleneck of regional medical consortium cancer screening platforms.

[0004] Therefore, how to automatically identify and resolve weaknesses in the data transmission and verification process by establishing a cross-validation mechanism for multimodal data in a regional medical consortium cancer screening platform has become a key issue in improving the quality of cancer screening in counties. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method that can solve problems such as data inconsistency and uneven quality in the integration and analysis of multi-source medical data.

[0006] To achieve the above objectives, the present invention provides a county-level tumor screening and referral platform based on a regional medical consortium.

[0007] This invention relates to a county-level cancer screening and referral platform based on a regional medical consortium. The platform enables cancer screening and referral within a county-level medical and health community. The platform includes: a multimodal data acquisition module for extracting raw datasets from cancer examination-related data from multiple sources and generating a standardized first dataset according to a preset unified acquisition standard; a format processing module for unifying the format of data from multiple sources in the first dataset using preset format conversion rules to obtain a second dataset with a consistent format; an anomaly detection module for identifying key node records in each data source of the second dataset; if anomalies are detected in the key node records, completion is performed to obtain a completed third dataset; and a cross-validation module for cross-validating multiple data sources in the third dataset. Cross-validation is used to determine the consistency verification results among various data sources; the anomaly location module is used to obtain the corresponding generation logs and transmission logs for data sources whose consistency verification results are lower than a preset threshold; it determines whether there are device anomalies or operational deviations in the generation logs and transmission logs to obtain anomaly source location information; the anomaly repair module is used to repair the abnormal data in the third dataset based on the anomaly source location information to obtain the repaired fourth dataset; the anomaly identification module is used to trigger a secondary verification process to obtain the final screening data results in response to the determination of potential undetected anomalies in the fourth dataset; the evaluation module is used to evaluate the accuracy indicators of each data source based on the final screening data results to determine the reliability level of each data source, and output a comprehensive analysis report based on the reliability level.

[0008] Optionally, the data sources for the tumor examination-related data include: imaging examination data, blood biomarker detection data, and gene screening data; the abnormalities include missing fields or abnormal values.

[0009] Optionally, the multimodal data acquisition module is further configured to: acquire raw data from imaging examination data, blood biomarker detection data, and gene screening data to form a raw dataset; adjust the format and fill missing values ​​of each data source in the raw dataset according to the unified acquisition standard to generate a preliminary processed dataset; normalize the data fields in the preliminary processed dataset using preset standardization processing rules to obtain a standardized dataset; filter out and remove outliers in the standardized dataset that are not within a preset range to obtain a corrected standardized dataset; classify the data features in the corrected standardized dataset using a support vector machine algorithm to obtain a classified feature dataset; integrate the classified feature dataset to generate a comprehensive feature dataset; and group the data in the comprehensive feature dataset to form a first dataset as a grouped dataset based on the distribution characteristics of each group of data.

[0010] Optionally, the format processing module is further configured to: identify format differences in the first dataset during transmission and obtain classification results of the format differences; convert the classification results of the format differences using the format conversion rules to obtain an intermediate dataset; map the fields of the intermediate dataset using a standardization tool to obtain a standardized dataset with consistent fields; if field values ​​in the standardized dataset are missing or abnormal, perform data cleaning using preset threshold judgment rules to obtain a cleaned complete dataset; apply a support vector machine algorithm to verify the data features in the cleaned complete dataset to determine whether the data features meet preset format consistency requirements; if the data features meet the format consistency requirements, merge the cleaned complete dataset to obtain a second dataset.

[0011] Optionally, the anomaly detection module is further configured to: obtain records from each data source in the second dataset; extract key node records during the generation and transmission process using a preset metadata template to obtain a node record set; if it is determined that there are missing fields in the node record set, fill in the missing fields using preset association field rules to obtain a completed field set; if it is determined that there are abnormal values ​​in the node record set that exceed a preset range using a statistical anomaly detection algorithm, obtain an anomaly value set from the abnormal values; complete the anomaly value set using historical data mean interpolation to obtain a repaired record set; and integrate the completed field set with the repaired record set to obtain a third dataset.

[0012] Optionally, the cross-validation module is further configured to: extract feature vectors from each data source in the third dataset using a feature extraction algorithm; perform dimensionality reduction on each feature vector using a principal component analysis algorithm to obtain a low-dimensional feature representation; perform data comparison on each low-dimensional feature representation through a cross-validation mechanism to calculate the similarity between each data source; if the similarity is lower than a preset threshold, re-extract features from the corresponding data source to obtain an updated feature vector; calculate a similarity matrix based on the updated feature vector to obtain the consistency degree between each data source; group each data source based on the consistency degree and a clustering analysis algorithm to obtain a grouping result; and generate a consistency verification result between each data source based on the grouping result.

[0013] Optionally, the anomaly localization module is further configured to: filter data sources whose consistency verification results are lower than a preset threshold to obtain an anomaly data set; extract corresponding generation logs and transmission logs from the anomaly data set to obtain a generation log dataset and a transmission log dataset; if the device operating parameters in the generation log dataset exceed a preset range, they are marked as device anomaly states, and device anomaly records are obtained; if the operation behavior in the generation log dataset or transmission log dataset deviates from the standard process, it is marked as operation deviation, and operation anomaly records are obtained; if transmission delay or packet loss rate in the transmission log dataset exceeds a preset standard, it is marked as transmission anomaly states, and transmission anomaly records are obtained; use the support vector machine algorithm to classify the device anomaly records and transmission anomaly records for anomaly patterns, and obtain device anomaly feature sets and transmission anomaly feature sets; use the random forest algorithm to perform integrated analysis on the device anomaly feature sets, transmission anomaly feature sets, and operation anomaly records to obtain the probability distribution of anomaly sources, and then obtain anomaly source localization information; generate anomaly source classification labels and location coordinates based on the anomaly source localization information, and output the localization results.

[0014] Optionally, the anomaly repair module is further configured to: determine the problem nodes in the third dataset based on the anomaly source location information; if the number of problem nodes exceeds a preset threshold, re-collect data from the data source using a data re-sampling method to obtain a first intermediate dataset; apply a correction algorithm to adjust the distribution of abnormal data in the first intermediate dataset to obtain a second intermediate dataset; extract features from the second intermediate dataset to generate anomaly data repair parameters, thereby obtaining a repair parameter set; perform numerical replacement on the abnormal data in the third dataset based on the repair parameter set to obtain a third intermediate dataset; if the anomaly detection results of the third intermediate dataset meet the requirements, then use the third intermediate dataset as the fourth dataset.

[0015] Optionally, the anomaly identification module is further configured to: acquire the original records of the fourth dataset; perform integrity and consistency checks on the original records to determine whether there are missing or inconsistent records, and obtain preliminary check results; use a support vector machine algorithm to extract features and label anomalies in the preliminary check results, thereby determining whether there is a potential risk of undetected anomalies; if there is a potential risk of undetected anomalies, transfer the labeled data to a secondary verification process, and obtain the anomaly confirmation result after secondary verification by comparing historical data and a preset range; distinguish normal data and abnormal data according to the anomaly confirmation result after secondary verification and the preset anomaly judgment logic, and obtain a classified dataset; use statistical tools to record the distribution characteristics of abnormal data in the classified dataset, and determine the category and impact range of abnormal data; perform correlation analysis on the category and impact range of abnormal data, and update the detection parameters of the anomaly identification module using preset consistency check rules to obtain an optimized module configuration; and analyze the processing effect of the fourth dataset according to the optimized module configuration, repeatedly perform integrity and consistency checks, and obtain the final screening data results.

[0016] Optionally, the evaluation module is further configured to: convert the final screening data results into structured data; perform data cleaning on the structured data to obtain a cleaned dataset; extract features from the cleaned dataset using a support vector machine algorithm, and calculate feature vectors for imaging examination data, blood biomarker detection data, and gene screening data respectively to obtain a feature set; calculate an accuracy index based on the feature set, and if the accuracy index is higher than a preset threshold, it is marked as a high-reliability feature, thereby obtaining a reliability feature set; classify the reliability feature set using a random forest algorithm to generate a reliability level for each data source; fuse the reliability levels of imaging examination data, blood biomarker detection data, and gene screening data using a weighted average method to obtain a comprehensive reliability level, and generate a comprehensive analysis report based on the comprehensive reliability level.

[0017] According to the technical solution of the present invention, one embodiment of the above invention has the following advantages or beneficial effects:

[0018] This invention achieves the fusion analysis of multi-source heterogeneous medical data by standardizing the collection, format unification, and anomaly detection of imaging examination data, blood biomarker detection data, and gene screening data. First, the original data is standardized and format unified. Then, a cross-validation mechanism is used to verify the consistency of multimodal data. Data quality is improved through anomaly source localization and data repair. Finally, the processed data undergoes secondary verification and comprehensive evaluation, outputting a reliability-graded analysis report. This invention effectively solves problems such as data inconsistency and uneven quality in the integration and analysis of multi-source medical data, improving the accuracy and reliability of medical data analysis. It provides more reliable data support for clinical diagnosis and medical research, and facilitates accurate tumor-related screening within counties based on regional medical consortia (regional healthcare communities), thereby facilitating the determination of treatment methods or referral strategies based on tumor-related screening results.

[0019] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0020] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein:

[0021] Figure 1 This is a schematic diagram of the components of the county-level tumor screening and referral platform based on a regional medical consortium in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the first execution step of the county-level tumor screening and referral platform based on a regional medical consortium in an embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram of the second execution step of the county-level tumor screening and referral platform based on a regional medical consortium in an embodiment of the present invention. Detailed Implementation

[0024] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0025] Figure 1 This is a schematic diagram of the components of the county-level tumor screening and referral platform based on a regional medical consortium in an embodiment of the present invention; Figure 2 This is a schematic diagram of the first execution step of the county-level tumor screening and referral platform based on a regional medical consortium in an embodiment of the present invention; Figure 3This is a schematic diagram of the second execution step of the county-level tumor screening and referral platform based on a regional medical consortium in an embodiment of the present invention. See [link / reference]. Figures 1-3 .

[0026] This invention provides a county-level tumor screening and referral platform based on a regional medical consortium. The platform can be implemented using a computer system and can realize county-level tumor screening and referral based on a regional medical and health consortium. The platform includes the following modules: a multimodal data acquisition module, a format processing module, an anomaly detection module, a cross-validation module, an anomaly localization module, an anomaly repair module, an anomaly identification module, and an evaluation module. The functions and execution steps of each module are as follows.

[0027] The multimodal data acquisition module extracts raw datasets from tumor examination-related data from multiple sources and generates a standardized first dataset according to a preset unified acquisition standard. The format processing module uses preset format conversion rules to unify the format of data from multiple sources in the first dataset, resulting in a second dataset with a consistent format. The anomaly detection module identifies key node records in each data source of the second dataset; if anomalies are detected in key node records, they are filled in to obtain a completed third dataset. The cross-validation module performs cross-validation on multiple data sources in the third dataset to determine the consistency verification results among the data sources. The anomaly localization module acquires the corresponding generation and transmission logs for data sources whose consistency verification results are below a preset threshold; it determines whether there are equipment malfunctions or operational deviations in the generation and transmission logs to obtain anomaly source localization information. The anomaly repair module repairs the abnormal data in the third dataset based on the anomaly source localization information, resulting in a repaired fourth dataset. The anomaly identification module triggers a secondary verification process to obtain the final screening data results in response to the detection of potential undetected anomalies in the fourth dataset. The evaluation module is used to assess the accuracy indicators of each data source based on the final screening data results to determine the reliability level of each data source, and output a comprehensive analysis report based on the reliability level.

[0028] Preferably, the data sources for the tumor examination-related data include: imaging examination data, blood biomarker detection data, and gene screening data; the abnormalities include missing fields or abnormal values.

[0029] In one embodiment, the multimodal data acquisition module is further configured to: acquire raw data from imaging examination data, blood biomarker detection data, and gene screening data to form a raw dataset; adjust the format and fill missing values ​​of each data source in the raw dataset according to the unified acquisition standard to generate a preliminary processed dataset; normalize the data fields in the preliminary processed dataset using preset standardization processing rules to obtain a standardized dataset; filter out and remove outliers in the standardized dataset that are not within a preset range to obtain a corrected standardized dataset; classify the data features in the corrected standardized dataset using a support vector machine algorithm to obtain a classified feature dataset; integrate the classified feature dataset to generate a comprehensive feature dataset; and group the data in the comprehensive feature dataset to form a first dataset as a grouped dataset based on the distribution characteristics of each group of data.

[0030] In practical applications, the multimodal data acquisition module obtains raw data from imaging examinations, blood biomarkers, and genetic screening to form a raw dataset. For this raw dataset, the format of each data source is adjusted and missing values ​​are filled according to a unified acquisition standard, generating a preliminary processed dataset. Using preset standardization rules, the data fields in the preliminary processed dataset are normalized to obtain a standardized dataset. If outliers exist in the standardized dataset, data that does not meet the criteria is filtered out using a preset threshold range, resulting in a corrected standardized dataset. Based on the corrected standardized dataset, a support vector machine algorithm is applied to perform preliminary classification of data features, obtaining a classified feature dataset. For the classified feature dataset, combined with data integration methods, a multi-dimensional feature fusion method is used to generate a comprehensive feature dataset. Using the comprehensive feature dataset, a preset clustering method is used to group the data, determine the distribution characteristics of each group, and form the final grouped dataset.

[0031] For example, the multimodal data acquisition module first extracts the raw dataset from imaging examinations, such as lung imaging data obtained through CT scans. The unified acquisition standard is set as a resolution of no less than 512x512 pixels, a grayscale value range of 0 to 4095, and storage in DICOM format. During data preprocessing, an image normalization algorithm is used to linearly map the grayscale values ​​to the range of 0 to 1. The specific formula is Inorm = (I - Imin) / (Imax - Imin), where Imin is 0 and Imax is 4095, to ensure data consistency and form part of the standardized first dataset. Next, data was extracted from blood biomarker tests, such as detecting the concentration of the tumor marker AFP in serum. The collection standard was set at a detection range of 0.5 to 500 ng / mL. Data acquisition was performed using high-throughput enzyme-linked immunosorbent assay (ELISA), and outliers were processed. The Median Absolute Deviation (MAD) method was used to remove outliers, with the formula MAD = median(|X_i - median(X)|). If a value deviated from the median by more than 3 times the MAD, it was removed. The final standardized data was stored in CSV format and included in the first dataset. Subsequently, raw datasets were extracted from gene screening data, such as BRCA1 gene mutation data obtained through next-generation sequencing technology. The standard was set at a sequencing depth of not less than 30X and a coverage of not less than 95%. During data processing, the GATK tool was used for variant detection. The specific process included alignment to the reference genome hg38, using the BWA algorithm, controlling the error rate to below 0.01, and then filtering false positive variants through variant quality score recalibration (VQSR) to ensure data accuracy. Finally, the data was stored in VCF format and integrated into the first dataset. Using the above methods, the three types of data are standardized to form a unified first dataset, laying the foundation for subsequent analysis. Logically, image data can be used for structural abnormality detection, blood data can assist in diagnosis, and gene data can provide molecular-level evidence. The combination of the three improves diagnostic accuracy.

[0032] As a preferred embodiment, the format processing module is further configured to: identify format differences in the first dataset during transmission and obtain classification results of the format differences; convert the classification results of the format differences using the format conversion rules to obtain an intermediate dataset; map the fields of the intermediate dataset using a standardization tool to obtain a standardized dataset with consistent fields; if field values ​​in the standardized dataset are missing or abnormal, perform data cleaning using preset threshold judgment rules to obtain a cleaned complete dataset; apply a support vector machine algorithm to verify the data features in the cleaned complete dataset to determine whether the data features meet preset format consistency requirements; if the data features meet the format consistency requirements, merge the cleaned complete dataset to obtain a second dataset.

[0033] In practical applications, the format processing module identifies data format differences and performs preliminary analysis on various data sources using preset format conversion rules to obtain classification results of format differences. Based on the classification results, preset rules are applied to convert the formats of imaging examination, blood biomarker, and gene screening data to determine the intermediate dataset after conversion. For the intermediate dataset, a standardization tool is used to map data fields to obtain a standardized dataset with consistent fields. If field values ​​in the standardized dataset are missing or abnormal, data cleaning is performed using preset threshold judgment rules to obtain a cleaned complete dataset. Based on the cleaned complete dataset, a support vector machine algorithm is applied to verify the data features to determine whether the data features meet preset format consistency requirements. If the data features meet the format consistency requirements, the data integration module merges the cleaned complete dataset to obtain the final second dataset. By optimizing and adjusting the storage structure of the second dataset and using distributed storage technology for data fragmentation management, the efficiency of data storage is ensured.

[0034] Specifically, when dealing with format differences in the first dataset, the format processing module handles data from imaging examinations, blood biomarkers, and genetic screening, as these data have different original formats. These data need to be uniformly processed using preset format conversion rules to generate the second dataset. First, for imaging examination data, assuming the original data is stored in DICOM format with a resolution of 1024x1024 pixels and grayscale values ​​from 0 to 4095, a format conversion algorithm can be used to convert it to a unified PNG format, adjusting the resolution to 512x512 pixels and normalizing the grayscale values ​​to the range of 0 to 255. The specific algorithm is a linear mapping: new grayscale value = (old grayscale value / 4095) * 255. The image processing library automatically batch processes 1000 images to ensure output consistency. Next, for blood biomarker data, the raw data may be stored in CSV format, containing hemoglobin values ​​(in g / L) and white blood cell counts (in 10^9 / L), such as a hemoglobin value of 150.5 and a white blood cell count of 7.8. This data can be converted to JSON format, with unified units of mg / dL and 10^6 / L. The conversion rule is to multiply the hemoglobin value by 10, i.e., 150.5 becomes 1505.0, and the white blood cell count by 1000, i.e., 7.8 becomes 7800.0. The script automatically parses and generates a new file, ensuring that the numerical precision is preserved to one decimal place. Subsequently, for gene screening data, the raw data may be in TXT format, containing gene mutation site information such as "rs12345:AG". This data can be converted to XML format, with unified field names of "MutationSite" and "Genotype". Regular expressions are used to extract the site and genotype, automatically processing 5000 records to ensure data structure consistency. Finally, all processed data is integrated into a second dataset and stored in a unified database table structure. Image data is stored as binary fields, while blood and genetic data are stored as numerical and text fields. Automatic insertion is performed using SQL scripts to ensure cross-type data field alignment, such as associating image IDs with blood sample IDs, enabling data traceability, forming a complete logical chain, and outputting a second dataset of 10,000 records after verifying data integrity.

[0035] In one embodiment, the anomaly detection module is further configured to: obtain records from each data source in the second dataset; extract key node records during the generation and transmission process using a preset metadata template to obtain a node record set; if it is determined that there are missing fields in the node record set, fill in the missing fields using preset association field rules to obtain a completed field set; if it is determined that there are abnormal values ​​in the node record set that exceed a preset range using a statistical anomaly detection algorithm, obtain an anomaly value set from the abnormal values; complete the anomaly value set using historical data mean interpolation to obtain a repaired record set; and integrate the completed field set with the repaired record set to obtain a third dataset.

[0036] Specifically, the anomaly detection module obtains records of various data types from the second dataset, extracts key node records from the generation and transmission process using a preset metadata template, and obtains a node record set. For the node record set, if any fields are missing, the missing fields are filled in using preset rules for related fields, resulting in a filled field set. Numerical anomalies are detected from the node record set, using a statistical anomaly detection algorithm to determine if the values ​​exceed a preset threshold range, resulting in an outlier set. For the outlier set, the outlier values ​​are filled in using the mean interpolation method of historical data, generating a repaired record set. From the repaired record set, data transformation rules are used to integrate the filled fields and repaired values, generating a third dataset. For the third dataset, data integrity is verified; if missing fields or numerical anomalies still exist, the completion logic is repeated to obtain the final dataset. Key node records are extracted from the final dataset, and a data quality report is generated to confirm the integrity and consistency of the third dataset.

[0037] For example, for each data type in the second dataset, the anomaly detection module first extracts key node records from the generation and transmission process through the data parsing module. For instance, a data record may contain fields such as timestamp, device ID, and temperature value. Assuming a record is {Time: 2023-10-01 08:00:00, Device ID: DEV001, Temperature: 25.5}, ​​the system automatically compares it with the data dictionary and finds that the record lacks a humidity field, which is required by the data dictionary. Next, in the anomaly detection phase, the system uses a preset threshold range (temperature range 10.0 to 40.0) to verify the values. It finds that the temperature of 25.5 is within the normal range, but another record has a temperature of 45.0, exceeding the threshold, and is therefore determined to be an anomaly. Simultaneously, for missing fields, the system triggers data completion logic, calling historical data analysis algorithms to use the average humidity value (assumed to be 60.3%) of the same device DEV001 over the past 24 hours as the completion value, generating a completed record {Time: 2023-10-01 08:00:00, Device ID: DEV001, Temperature: 25.5, Humidity: 60.3}. For records with abnormal values, the system uses a smoothing algorithm to calculate a weighted average of the temperature 45.0 with the temperature values ​​at the preceding and following time points (assumed to be 26.0 and 27.0), resulting in (26.0 + 27.0 + 45.0) / 3 = 32.7, updating the record to a temperature of 32.7. Finally, the completed and corrected data forms a third dataset, which is stored in the database, and a verification report is generated, recording detailed logs of anomaly detection and completion, such as a 100% completion rate for missing fields and a 98.5% correction rate for abnormal values, ensuring data integrity and accuracy. By integrating with business systems, the third dataset can be directly used for subsequent environmental monitoring and analysis, ensuring business continuity. For example, after temperature anomaly correction, false high-temperature alarms are avoided, and after humidity is supplemented, complete environmental assessment model calculations are supported, forming data-driven decision support.

[0038] In an optional technical solution, the cross-validation module is further configured to: extract feature vectors from each data source in the third dataset using a feature extraction algorithm; perform dimensionality reduction on each feature vector using a principal component analysis algorithm to obtain a low-dimensional feature representation; perform data comparison on each low-dimensional feature representation through a cross-validation mechanism to calculate the similarity between each data source; if the similarity is lower than a preset threshold, re-extract features from the corresponding data source to obtain an updated feature vector; calculate a similarity matrix based on the updated feature vector to obtain the consistency degree between each data source; group each data source based on the consistency degree and a clustering analysis algorithm to obtain a grouping result; and generate a consistency verification result between each data source based on the grouping result.

[0039] Specifically, the cross-validation module acquires multimodal features from a third dataset and processes the imaging examination, blood biomarker, and gene screening data using feature extraction algorithms to obtain feature vectors for each data type. Principal component analysis (PCA) is then used to reduce the dimensionality of the feature vectors, resulting in low-dimensional feature representations. Cross-validation is then used to compare these low-dimensional feature representations and calculate the similarity matrix between data types. If the similarity between any data type in the similarity matrix is ​​below a preset threshold, feature extraction is performed again for that data type to obtain an updated feature vector. Based on the updated feature vectors, the similarity matrix is ​​recalculated to determine the consistency level between data types. Based on the consistency level, clustering analysis is used to group the data types, yielding the grouping results. Finally, based on the grouping results, validation results for the consistency level between data types are generated, and the final result is output.

[0040] The cross-validation module uses multimodal feature analysis to verify the consistency of imaging examination, blood biomarker, and genetic screening data. First, the module extracts features from the imaging data; for example, it uses a convolutional neural network (CNN) to extract gray-level co-occurrence matrix features from brain MRI images of 1000 patients, calculating texture parameters such as contrast (mean 0.85, standard deviation 0.12) and entropy (mean 2.3, standard deviation 0.4). Blood biomarker data is analyzed using high-throughput mass spectrometry to detect the concentrations of 100 biomarkers, such as C-reactive protein (CRP, mean 5.2 mg / L, standard deviation 1.1) and interleukin-6 (IL-6, mean 3.8 pg / mL, standard deviation 0.9). Genetic screening data is analyzed using genome-wide association studies (GWAS) to screen 20 key disease-related sites from 5000 single nucleotide polymorphisms (SNPs) and calculate their allele frequencies (e.g., the A allele frequency at the rs12345 site is 0.65). To achieve cross-validation, a 5-fold cross-validation algorithm was employed. The dataset was divided into a training set (80%) and a test set (20%). A Support Vector Machine (SVM) classification model was constructed for each fold of data, with a Radial Basis Function (RBF) kernel, parameter C set to 1.0, and γ set to 0.01. The model's accuracy in predicting diseases based on imaging features was 0.88, for blood biomarkers 0.82, and for genetic data 0.79. To assess consistency, Pearson correlation coefficients were calculated for the prediction results of the three classes: 0.75 for imaging and blood biomarkers, 0.68 for imaging and genetic data, and 0.71 for blood biomarkers and genetic data. Further chi-square tests were used to analyze the consistency among the three, with a p-value of 0.03, indicating a significant correlation. To enhance business relevance, patient clinical symptom data (e.g., headache frequency of 10 times per month, standard deviation 2.5) was incorporated, and a logistic regression model was used to integrate multimodal features, outputting a comprehensive consistency score of 0.90. Validation results indicate a high degree of consistency among the data, supporting the reliability of disease diagnosis.

[0041] In an optional implementation, the anomaly localization module is further configured to: filter data sources whose consistency verification results are below a preset threshold to obtain an anomaly data set; extract corresponding generation logs and transmission logs from the anomaly data set to obtain a generation log dataset and a transmission log dataset; if the device operating parameters in the generation log dataset exceed a preset range, they are marked as device anomaly states, and device anomaly records are obtained; if the operation behavior in the generation log dataset or transmission log dataset deviates from the standard process, it is marked as operation deviation, and operation anomaly records are obtained; if transmission delay or packet loss rate in the transmission log dataset exceeds a preset standard, it is marked as transmission anomaly states, and transmission anomaly records are obtained; use a support vector machine algorithm to classify the device anomaly records and transmission anomaly records for anomaly patterns, and obtain a device anomaly feature set and a transmission anomaly feature set; use a random forest algorithm to perform integrated analysis on the device anomaly feature set, transmission anomaly feature set, and operation anomaly records to obtain the probability distribution of the anomaly source, and then obtain the anomaly source localization information; generate the anomaly source classification label and location coordinates based on the anomaly source localization information, and output the localization result.

[0042] Specifically, the anomaly localization module first obtains the consistency verification results and filters data groups with consistency levels below a preset threshold to obtain an anomaly data group set. It then extracts the corresponding generation logs and transmission logs from the anomaly data group set to generate a log dataset and a transmission log dataset. If the device operating parameters in the generation log dataset exceed a preset range, it is marked as a device anomaly state, resulting in a device anomaly record. If the operation behavior in the generation log dataset deviates from the standard process, it is marked as an operation deviation record, resulting in an operation anomaly record. If the transmission delay or packet loss rate in the transmission log dataset exceeds a preset standard, it is marked as a transmission anomaly state, resulting in a transmission anomaly record. If unauthorized operations exist in the transmission log dataset, they are marked as operation deviation records and merged into the operation anomaly record. The support vector machine algorithm is used to classify the device anomaly records and transmission anomaly records, analyze anomaly patterns, and obtain device anomaly feature sets and transmission anomaly feature sets. The random forest algorithm is used to integrate and analyze the device anomaly feature sets, transmission anomaly feature sets, and operation anomaly records, calculating the probability distribution of the anomaly source and obtaining anomaly source localization information. Based on the anomaly source localization information, a classification label and location coordinates of the anomaly source are generated, and the localization result is output.

[0043] For example, regarding the processing of consistency verification results, the anomaly localization module first automatically filters data groups with a consistency level below a preset threshold. Assuming the threshold is set at 85%, the system scan finds a data group with a consistency level of only 72.3%, below the threshold, and automatically marks it as an anomaly data group, recording its unique identifier ID as D20231001. Next, the system retrieves the corresponding generation and transmission logs from the database based on this ID. The logs show that the data was generated on October 1, 2023, at 10:00:23, transmitted at 10:00:45, with 3 retries during transmission and a data packet size of 2.5MB. Subsequently, the system calls an anomaly detection algorithm to analyze the logs. The algorithm uses time-series-based anomaly detection, calculating the average transmission delay to be 15 seconds. This data group's delay is 22 seconds, exceeding the standard deviation by 1.5 times, and is therefore determined to be a transmission anomaly. Meanwhile, the system compared the device's operational status logs and found that the CPU utilization of the generating device reached as high as 98% at 10:00:20, exceeding the normal range of 80%. Based on this, the rule engine deduced that the data generation deviation might be due to excessive device load. Ultimately, the system generated anomaly source location information, indicating that the anomaly was caused by excessive device load and transmission latency, with the device ID E123 and the transmission node N456. The results were stored in the anomaly report database, and the associated business process was data quality monitoring, automatically triggering subsequent optimization tasks, such as device load balancing scheduling, to ensure that data consistency was improved to above the threshold. Through this process, the system achieved fully automated analysis from anomaly identification to source location, forming a complete technical closed loop.

[0044] Optionally, the anomaly repair module is further configured to: determine the problem nodes in the third dataset based on the anomaly source location information; if the number of problem nodes exceeds a preset threshold, re-collect data from the data source using a data re-sampling method to obtain a first intermediate dataset; apply a correction algorithm to adjust the distribution of abnormal data in the first intermediate dataset to obtain a second intermediate dataset; extract features from the second intermediate dataset to generate anomaly data repair parameters, thereby obtaining a repair parameter set; perform numerical replacement on the abnormal data in the third dataset based on the repair parameter set to obtain a third intermediate dataset; if the anomaly detection results of the third intermediate dataset meet the requirements, then use the third intermediate dataset as the fourth dataset.

[0045] Specifically, the anomaly repair module first obtains anomaly source location information from the third dataset and identifies problem nodes using an anomaly detection algorithm. If the number of problem nodes exceeds a preset threshold, a data resampling method is used to re-collect data from the original data source, resulting in a first intermediate dataset. For the first intermediate dataset, a correction algorithm is applied to adjust the numerical distribution of the anomaly data, resulting in a second intermediate dataset. By extracting features from the second intermediate dataset, repair parameters for the anomaly data are generated, resulting in a repair parameter set. Based on the repair parameter set, the anomaly data in the third dataset is numerically replaced, resulting in a third intermediate dataset. If the anomaly detection results of the third intermediate dataset meet a preset threshold, it is used as the fourth dataset to complete data repair. By evaluating the quality of the fourth dataset, a result output is generated to determine the final data repair effect.

[0046] For example, the anomaly repair module first uses statistical analysis and machine learning algorithms to detect and locate abnormal data in the third dataset based on the anomaly source location information. For instance, if the third dataset contains 1000 records, each record has three features: temperature, pressure, and flow rate, with values ​​of [20.5, 101.3, 50.2], etc., the Z-score method is used to detect anomalies. The mean and standard deviation of each feature are calculated. For example, the mean temperature is 22.0, and the standard deviation is 2.0. A threshold of 3 is set. If a record has a temperature of 28.5, the Z-score is (28.5-22.0) / 2.0 = 3.25, which exceeds the threshold and is marked as an anomaly node. After locating the anomaly, data resampling or correction algorithms are used to repair the problematic node. If the anomaly node has a missing value, such as a record with a flow rate value of null, data can be resampled using time series interpolation. Based on the flow rate values ​​at five time points before and after [50.2, 49.8, null, 50.1, 49.9], the average value of 50.0 is calculated to fill the missing value. If the anomaly is an outlier, such as a stress value of 200.0 far exceeding the normal range [90.0, 110.0], the K-nearest neighbor regression algorithm is used to select the stress values ​​of the 5 nearest neighbor samples [101.3, 100.8, 101.5, 100.9, 101.2], and the weighted average is used to obtain a corrected value of 101.1. The corrected data forms a fourth dataset, such as [20.5, 101.1, 50.0], to ensure data integrity and consistency. The analysis process needs to verify the repair effect by calculating the statistical characteristics of the fourth dataset, such as the stress standard deviation decreasing from 5.0 before repair to 1.2, indicating a more stable data distribution. Logically, anomaly detection ensures accurate location, resampling and correction algorithms ensure reasonable repair, and statistical verification confirms the effectiveness of the repair, forming a rigorous data processing chain. If business scenarios are involved, such as industrial monitoring, the corrected dataset can be further used for predictive model training to ensure stable system operation.

[0047] In one embodiment, the anomaly identification module is further configured to: acquire the original records of the fourth dataset; perform integrity and consistency checks on the original records to determine whether there are missing or inconsistent records, and obtain preliminary check results; use a support vector machine algorithm to extract features and label anomalies in the preliminary check results, thereby determining whether there is a potential risk of undetected anomalies; if there is a potential risk of undetected anomalies, transfer the labeled data to a secondary verification process, and obtain the anomaly confirmation result after secondary verification by comparing historical data and a preset range; distinguish normal data and abnormal data according to the anomaly confirmation result after secondary verification and the preset anomaly judgment logic, and obtain a classified dataset; use statistical tools to record the distribution characteristics of abnormal data in the classified dataset, and determine the category and impact range of abnormal data; perform correlation analysis on the category and impact range of abnormal data, and update the detection parameters of the anomaly identification module using preset consistency check rules to obtain an optimized module configuration; and analyze the processing effect of the fourth dataset according to the optimized module configuration, repeatedly perform integrity and consistency checks, and obtain the final screening data results.

[0048] Specifically, the anomaly detection module first acquires the original records of the fourth dataset and performs preliminary checks on completeness and consistency characteristics. By comparing the records against preset completeness standards and consistency rules, it determines whether any records are missing or inconsistent, obtaining preliminary check results. Based on these results, a support vector machine (SVM) algorithm is used to extract features and label anomalies. The module's processing performance is evaluated to determine if there is a potential risk of undetected anomalies. If such a risk exists, the labeled data is transferred to a secondary verification process. By comparing historical data with preset threshold ranges, the data consistency characteristics are checked, obtaining a secondary verification anomaly confirmation result. Based on the secondary verification anomaly confirmation result and the anomaly judgment logic, the final screened data is classified to distinguish between normal and abnormal data, resulting in a classified dataset. For the classified dataset, the data processing results are analyzed, and statistical tools are used to record the distribution characteristics of the abnormal data, determining the specific categories and impact ranges of the abnormal data. By performing correlation analysis on the categories and impact ranges of the abnormal data and combining the consistency check rules, the detection parameters of the anomaly detection module are updated, resulting in an optimized module configuration. Based on the optimized module configuration, a new round of processing effect analysis is performed on the fourth dataset, and the integrity and consistency characteristics are verified cyclically to obtain the final data screening results.

[0049] For example, in the integrity and consistency feature analysis of the fourth dataset, the anomaly detection module first scans the records in the dataset using a data integrity check algorithm. Assuming the dataset contains 10,000 records, it checks whether any fields in each record are missing. For instance, it finds that 5% of the records are missing the key field "User ID," meaning 500 records are incomplete. The integrity score is therefore 0.95, which is lower than the preset threshold of 0.98, triggering consistency analysis. Next, a consistency detection algorithm is used, based on a rule-based matching method, to check whether field values ​​conform to a predefined format. For example, the "Date" field should be in YYYY-MM-DD format. It finds that 200 records have a date format of MM-DD-YYYY, accounting for 2%, with a consistency score of 0.98, close to the threshold of 0.99. However, combined with the integrity score, the overall score is 0.93, which is lower than the anomaly detection module's threshold of 0.95, indicating a potential anomaly. In the anomaly detection module, the K-means clustering algorithm is applied to divide the data into 5 clusters based on feature vectors. The Euclidean distance from each record to the cluster center is calculated. 300 records were found to have a distance exceeding twice the standard deviation of the average distance, and were marked as suspected anomalies, accounting for 3%, triggering a secondary verification process. The secondary verification process uses a random forest classification model to reclassify these 300 records. Combining historical data features, the anomaly probability is predicted, and 150 records with an anomaly probability exceeding 0.8 are confirmed as anomalous data. The final screening result is: out of 10,000 records, 150 are anomalous and 9,850 are normal. To form a rigorous logical chain, the anomalous data will be sent to subsequent business modules for feature extraction and analysis of the causes of the anomalies. For example, if the anomalous records are concentrated in a certain time period, it may be related to system input errors. An anomaly report is generated, and the scores and anomaly ratio data are automatically entered into the system log for subsequent optimization of thresholds and algorithm models, ensuring system adaptability.

[0050] In this embodiment of the invention, the evaluation module is further configured to: convert the final screening data results into structured data; perform data cleaning on the structured data to obtain a cleaned dataset; extract features from the cleaned dataset using a support vector machine algorithm, and calculate feature vectors for imaging examination data, blood biomarker detection data, and gene screening data respectively to obtain a feature set; calculate an accuracy index based on the feature set, and if the accuracy index is higher than a preset threshold, it is marked as a high reliability feature, thereby obtaining a reliability feature set; classify the reliability feature set using a random forest algorithm to generate a reliability level for each data source; fuse the reliability levels of imaging examination data, blood biomarker detection data, and gene screening data using a weighted average method to obtain a comprehensive reliability level, and generate a comprehensive analysis report based on the comprehensive reliability level.

[0051] Specifically, the evaluation module first acquires screening data results, extracting raw datasets from imaging examinations, blood biomarkers, and genetic screening to obtain structured data. The structured data is then cleaned using a preset threshold; outliers are removed if data integrity falls below this threshold, resulting in a cleaned dataset. A support vector machine (SVM) algorithm is used to extract features from the cleaned dataset, calculating feature vectors for each of the imaging examination, blood biomarker, and genetic screening data to obtain feature sets. An accuracy index is calculated based on the feature sets; features with an accuracy index higher than a preset threshold are marked as high-reliability features, resulting in a reliability feature set. The reliability feature set is then classified using a random forest algorithm, generating reliability levels for each data type, resulting in classification results. The classification results are then combined using a weighted average method to merge the reliability levels from imaging examinations, blood biomarkers, and genetic screening to obtain a comprehensive reliability level. Finally, an analysis report is generated based on the comprehensive reliability level, outputting the reliability level and index evaluation results for each data type, resulting in the final report.

[0052] For example, the evaluation module comprehensively evaluates the final screening data, using imaging examinations, blood biomarkers, and genetic screening data as objects. It employs a multi-dimensional accuracy index analysis method to determine the reliability level of each type of data and generates a comprehensive analysis report. First, for imaging examination data, taking CT scan results as an example, assuming the dataset contains 1000 samples, of which 600 are confirmed cases, with a false positive rate of 5% and a false negative rate of 3%. By calculating precision (the proportion of correctly identified confirmed cases), i.e., 600 / (600+50) = 0.92, and recall (600 / (600+30) = 0.95), combined with the F1 score (2*Precision*Recall / (Precision+Recall)), a score of 0.935 is obtained, indicating that the reliability of the imaging examination is high, and the level is set as Grade A. Secondly, for blood biomarker data, taking the tumor marker AFP value as an example, with a threshold of 20 ng / ml, a total sample size of 800, 300 outliers, a false positive rate of 8%, and a false negative rate of 6%, ROC curve analysis yielded an AUC value of 0.88, indicating strong discriminative ability, and the reliability level was assigned to Grade B. Next, for gene screening data, taking BRCA1 gene mutation detection as an example, with a sample size of 500, 200 mutation-positive cases, a false positive rate of 2%, and a false negative rate of 4%, logistic regression model prediction yielded an accuracy of 0.94. Combined with confusion matrix analysis, the sensitivity was 0.96, the specificity was 0.93, and the reliability level was assigned to Grade A. Finally, considering the reliability levels of the three data types, a weighted average method was used, with weights of 0.4 for imaging, 0.3 for blood biomarkers, and 0.3 for gene screening. The overall reliability score was calculated as 0.4*0.935 + 0.3*0.88 + 0.3*0.94 = 0.919, and the overall reliability level was assigned to Grade A. Through the above algorithms and numerical analysis, a complete logical chain from single-item assessment to comprehensive evaluation is formed, ensuring that the results are scientific and traceable, and can also be linked to subsequent disease risk prediction models, providing data support for precision medicine.

[0053] In the technical solution of this invention, the fusion analysis of multi-source heterogeneous medical data is achieved by standardizing the collection, format unification, and anomaly detection of imaging examination data, blood biomarker detection data, and gene screening data. This invention first standardizes and unifies the format of the raw data, then verifies the consistency of multimodal data through a cross-validation mechanism, and improves data quality through anomaly source localization and data repair. Finally, the processed data undergoes secondary verification and comprehensive evaluation, outputting a reliability-graded analysis report. This invention effectively solves the problems of data inconsistency and uneven quality in the integration and analysis of multi-source medical data, improving the accuracy and reliability of medical data analysis. It provides more reliable data support for clinical diagnosis and medical research, and facilitates accurate tumor-related screening within counties based on regional medical consortia (regional healthcare communities), thereby facilitating the determination of treatment methods or referral strategies based on tumor-related screening results.

[0054] It should be noted that the collection, analysis, use, transmission, and storage of user personal information involved in the technical solution of this invention all comply with the provisions of relevant laws and regulations, are used for legal and reasonable purposes, and are not shared, disclosed, or sold outside of these legal uses, and are subject to supervision and management by regulatory authorities. Necessary measures should be taken to prevent unauthorized access to such personal information data, ensure that personnel authorized to access personal information data comply with the provisions of relevant laws and regulations, and ensure the security of user personal information. Once this user personal information data is no longer needed, the risk should be minimized by restricting or even prohibiting data collection and / or deleting the data. When used, including in certain related applications, user privacy is protected by de-identifying the data, for example, by removing specific identifiers (e.g., date of birth), controlling the amount or specificity of stored data (e.g., collecting location data at the city level rather than the specific address level), controlling how the data is stored, and / or other de-identification methods.

[0055] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A county tumor screening and referral platform based on regional medical community, characterized in that, The platform realizes county tumor screening and referral based on regional medical and health communities; the platform comprises: A multi-modal data acquisition module configured to extract an original data set from tumor examination related data of multiple data sources, and generate a first data set after standardization according to a preset unified acquisition standard; A format processing module configured to perform format uniform processing on data of multiple data sources in the first data set by using a preset format conversion rule, and obtain a second data set with consistent formats; An anomaly detection module configured to determine key node records in each data source of the second data set; if it is determined that there is an abnormal situation in the key node records, the key node records are completed to obtain a third data set after completion; A cross-validation module configured to cross-validate multiple data sources in the third data set to determine consistency verification results between the data sources; An anomaly positioning module configured to, for a data source with a consistency verification result lower than a preset threshold, obtain corresponding generation logs and transmission logs; determine whether there is equipment anomaly or operation deviation in the generation logs and the transmission logs to obtain anomaly source positioning information; An anomaly repair module configured to repair abnormal data in the third data set according to the anomaly source positioning information to obtain a fourth data set after repair; An anomaly identification module configured to, in response to a determination that there is a potential undetected anomaly in the fourth data set, trigger a secondary verification process to obtain a final screening data result; An evaluation module configured to evaluate accuracy indicators of each data source based on the final screening data result to determine a reliability level of each data source, and output a comprehensive analysis report according to the reliability level.

2. The county tumor screening and referral platform of claim 1, wherein, The data sources in the tumor examination related data include: imaging examination data, blood marker detection data, and genetic screening data; and the abnormal situation includes field missing or numerical anomaly.

3. The county cancer screening and referral platform of claim 2, wherein, The multi-modal data acquisition module is further configured to: Obtain original data from the imaging examination data, the blood marker detection data, and the genetic screening data to form the original data set; Adjust the format and fill in missing values of each data source in the original data set according to the unified acquisition standard to generate a preliminary processed data set; Perform normalization operation on data fields in the preliminary processed data set by using a preset standardization processing rule to obtain a standardized data set; Filter out and remove abnormal values in the standardized data set that are not in a preset range to obtain a corrected standardized data set; Classify data features in the corrected standardized data set by using a support vector machine algorithm to obtain a classified feature data set; Integrate the classified feature data set to generate a comprehensive feature data set; Group data in the comprehensive feature data set, and form a first data set as a grouped data set according to distribution characteristics of each group of data.

4. The rural tumor screening and referral platform of claim 3, wherein, The format processing module is further configured to: Identify format differences of the first data set in a transmission link to obtain a classification result of the format differences; Convert the classification result of the format differences by using the format conversion rule to obtain an intermediate data set; Map fields of the intermediate data set by using a standardization tool to obtain a standardized data set with consistent fields; and If there is a missing or abnormal field value in the standardized data set, data cleaning is performed through a preset threshold judgment rule to obtain a complete data set after cleaning; The application of support vector machine algorithm to the data characteristics in the complete data set after cleaning judges whether the data characteristics meet the preset format consistency requirement; If the data characteristics meet the format consistency requirement, the complete data set after cleaning is merged to obtain a second data set.

5. The rural tumor screening and referral platform of claim 4, wherein, The anomaly detection module is also used to: Obtain the records of each data source from the second data set, and extract the key node records in the generation and transmission process using a preset metadata template to obtain a node record set; If it is judged that there is a missing field in the node record set, the missing field is completed through a preset association field rule to obtain a completed field set; If it is judged that there is an abnormal value exceeding the preset range in the node record set using a statistical-based anomaly detection algorithm, an abnormal value set is obtained from the abnormal value; The abnormal value set is completed by a historical data mean interpolation method to obtain a repaired record set; The completed field set and the repaired record set are integrated to obtain a third data set.

6. The rural tumor screening and referral platform of claim 5, wherein, The cross-validation module is also used to: Extract feature vectors of each data source in the third data set using a feature extraction algorithm; Perform dimension reduction processing on each feature vector using a principal component analysis algorithm to obtain a low-dimensional feature representation; Data comparison is performed on each low-dimensional feature representation through a cross-validation mechanism to calculate the similarity between each data source; If the similarity is lower than a preset threshold, the corresponding data source is re-extracted to obtain an updated feature vector; A similarity matrix is calculated based on the updated feature vector to obtain the consistency degree between each data source; Grouping is performed on each data source based on the consistency degree and a clustering analysis algorithm to obtain a grouping result; A consistency verification result between each data source is generated based on the grouping result.

7. The rural tumor screening and referral platform of claim 6, wherein, The anomaly positioning module is also used to: Filter data sources with a consistency verification result lower than a preset threshold to obtain an abnormal data group set; Extract corresponding generation logs and transmission logs from the abnormal data group set to obtain a generation log data set and a transmission log data set; If the device operating parameters in the generation log data set exceed the preset range, they are marked as device abnormal states to obtain device abnormal records; If the operation behaviors in the generation log data set or the transmission log data set deviate from the standard process, they are marked as operation deviations to obtain operation abnormal records; If transmission delay or packet loss rate in the transmission log data set is higher than a preset standard, it is marked as a transmission abnormal state to obtain transmission abnormal records; Support vector machine algorithm is used to classify the device abnormal records and the transmission abnormal records to obtain a device abnormal feature set and a transmission abnormal feature set; Random forest algorithm is used to perform integrated analysis on the device abnormal feature set, the transmission abnormal feature set, and the operation abnormal records to obtain a probability distribution of the abnormal source, and then obtain abnormal source positioning information; A classification label and a position coordinate of the abnormal source are generated based on the abnormal source positioning information to output a positioning result.

8. The rural tumor screening and referral platform of claim 7, wherein, The anomaly repair module is also used to: Determine a problem node in the third data set based on the abnormal source positioning information; If the number of problem nodes exceeds the preset threshold, data is re-collected from the data source in a data resampling manner to obtain a first intermediate data set; an abnormal data distribution in the first intermediate data set is adjusted by applying a correction algorithm to obtain a second intermediate data set; feature extraction is performed on the second intermediate data set to generate an abnormal data repair parameter, and then a repair parameter set is obtained; numerical replacement is performed on the abnormal data in the third data set according to the repair parameter set to obtain a third intermediate data set; If the abnormal detection result of the third intermediate data set meets the requirements, the third intermediate data set is taken as the fourth data set.

9. The rural tumor screening and referral platform of claim 8, wherein, The abnormality identification module is also used to: Obtain the original records of the fourth data set, and perform integrity and consistency verification on the original records to determine whether there are missing or inconsistent records to obtain a preliminary verification result; Support vector machine algorithm is used to extract features and label abnormal points of the preliminary verification result, and then it is determined whether there is a potential risk of undiscovered abnormality; If there is a potential risk of undiscovered abnormality, the labeled data is transferred to a secondary verification process, and the abnormal confirmation result after secondary verification is obtained by comparing historical data and a preset range; According to the abnormal confirmation result after secondary verification and the preset abnormality judgment logic, normal data and abnormal data are distinguished to obtain a classified data set; The statistical tool is used to record the abnormal data distribution characteristics in the classified data set to determine the category and impact range of the abnormal data; The category and impact range of the abnormal data are associated and analyzed, and the detection parameters of the abnormality identification module are updated using the preset consistency verification rule to obtain an optimized module configuration; According to the optimized module configuration, the processing effect of the fourth data set is analyzed, and the integrity and consistency verification are repeatedly performed to obtain the final screening data result.

10. The rural tumor screening and referral platform of claim 9, wherein, The evaluation module is also used to: Convert the final screening data result into structured data; Data cleaning is performed on the structured data to obtain a cleaned data set; Support vector machine algorithm is used to extract features from the cleaned data set, and feature vectors are calculated for imaging examination data, blood marker detection data and genetic screening data to obtain a feature set; According to the feature set, an accuracy index is calculated, and if the accuracy index is higher than a preset threshold, it is marked as a high reliability feature, thereby obtaining a reliability feature set; The reliability feature set is classified by a random forest algorithm, and reliability levels are generated for each data source; The reliability levels of imaging examination data, blood marker detection data and genetic screening data are fused by weighted average to obtain a comprehensive reliability level, and a comprehensive analysis report is generated based on the comprehensive reliability level.

Images