Medical examination information identification method and system based on AI

Abstract

The invention relates to the technical field of image recognition, in particular to a medical examination information recognition method and system based on AI, and the method comprises the steps: collecting DICOM transmission counting, screening a density deviation path, selecting a highest value path for locking as a main use, calculating the feature cosine similarity through effective nodes of the main use path, and selecting a disease code with the highest similarity to generate a pathological classification result. According to the method, medical examination efficiency is optimized through dynamic path adjustment and multi-modal verification, DICOM transmission identification abnormal paths are monitored in real time, path distribution prevention delay is optimized in combination with parameters such as loads, hop counts and success rates, meanwhile, data consistency is verified through Hash chain type comparison, multi-dimensional features are extracted to be matched with a standard pathological library in a cosine similarity mode, and therefore the accuracy of medical examination is improved. According to the method, real-time resource scheduling and cross-level data verification are fused, feature cross verification is embedded, noise interference is reduced, and processing stability and pathological classification precision in a high-concurrency scene are improved.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to an AI-based medical test information recognition method and system. Background Art

[0002] The field of image recognition technology encompasses technologies that discriminate, extract, and classify objects, features, or regions in images based on visual information. Its core lies in encoding, decoding, extracting features, and matching image signals through computing devices to accomplish tasks such as target detection, image segmentation, and morphological classification. It typically comprises steps such as data acquisition, feature representation, model building, and recognition decision-making, encompassing multiple dimensions including image preprocessing, feature modeling, and image semantic understanding. It is widely used in scenarios such as medical image analysis, security monitoring, and biometrics. Its key technical approach is based on image element learning and matching mechanisms within a convolutional neural network architecture.

[0003] Among them, the AI-based medical test information recognition method refers to the use of image recognition models to automatically parse and extract structural areas, numerical annotations and form formats in medical test images. The technical matters targeted include the positioning, qualitative and quantitative information recognition of target areas in medical images such as blood test forms and pathological section scans. Specifically, the trained classification and recognition model is used to partition the image pixel matrix, and the data location, name and value are annotated and extracted by combining regional morphological features and graphic identification patterns. The information dimensions used include image channel value combinations, edge distribution characteristics and color density distribution, and the multi-step parsing process of information elements is completed through a combination of logical rules.

[0004] Existing technologies have problems with weak dynamic resource adaptation capabilities and a single data verification dimension. Relying on preset service paths cannot respond to traffic fluctuations in real time. When there is a surge in test requests during peak periods, a single path is prone to overload, resulting in parsing delays. The transmission process lacks full-link data consistency verification. If pathological images have data anomalies due to network packet loss or tampering, traditional methods only rely on the model's noise resistance and cannot actively intercept erroneous inputs. Feature matching overly focuses on single-dimensional channel information and does not integrate multimodal features such as texture, grayscale, and contours. In complex scenarios with uneven staining or blurred tissue boundaries, classification accuracy is easily reduced due to local feature distortion. These defects restrict the processing efficiency and result reliability of high-load, high-noise medical testing scenarios. Summary of the Invention

[0005] In order to solve the technical problems existing in the prior art, the embodiments of the present invention provide a medical test information recognition method and system based on AI. The technical solution is as follows:

[0006] An AI-based medical test information recognition method includes the following steps:

[0007] S1: Collect DICOM transmission counts of medical testing nodes, calculate the absolute difference between the number of transmissions in the current window and the moving average of the time serial port, filter out transmission paths that exceed the preset difference range, and mark them as path density deviation status;

[0008] S2: Based on the path density deviation state, extract the candidate path load ratio, forwarding success ratio, and path hop average, normalize each data item, and generate a path real-time ranking parameter set;

[0009] S3: Call the service path index field of the microservice registry, select the highest value path code based on the real-time sorting parameter set of the path, overwrite the original field, lock the primary path, and generate the primary path index code;

[0010] S4: Collect lesion segmentation mask sequences and pathology classification code fields, generate hash values item by item according to the field order and original content, screen the hash character differences of the same-position fields before and after the task, and generate node verification failure flags for continuous fields that exceed the preset threshold;

[0011] S5: Call the node field in the main path index code that has not triggered the node verification failure flag, calculate the cosine similarity between the lesion characteristics of the corresponding image and the standard pathological characteristics, select the disease code corresponding to the highest similarity, and generate a medical examination pathology classification result.

[0012] As a further solution of the present invention, the path density deviation state is specifically a path identifier, a preset difference range, and a time window length; the path real-time sorting parameter set includes a load proportion weight, a forwarding success rate coefficient, and a hop mean factor; the main path index code is specifically the highest sorting parameter value, a service instance identifier, and a path lock time limit identifier; the node verification failure flag includes a difference field position index, a continuous field length threshold, and a trigger count counter; and the medical examination pathology classification result is specifically a similarity peak, a disease code matching item, and a feature index mapping relationship.

[0013] As a further solution of the present invention, the step of obtaining the path density deviation state is:

[0014] S101: Collecting the DICOM file transfer counts forwarded to the medical inspection node in the microservice cluster, recording the number of transmissions according to the time window sequence, and generating transmission number time series data;

[0015] S102: Based on the transmission timestamp time series data, a time window sequence is applied to calculate a moving average of the transmission timestamps to generate a moving average transmission timestamp;

[0016] S103: calling the moving average transmission times, calculating the absolute difference between the current window transmission times and the corresponding moving average, comparing the absolute difference with a preset difference range threshold, screening out-of-limit transmission paths and marking them as path density deviation states.

[0017] As a further solution of the present invention, the steps for obtaining the path real-time sorting parameter set are:

[0018] S201: Based on the path density deviation state, extract the ratio of the current processing number to the maximum processing capacity load, the ratio of the number of successes to the number of attempts, and the average number of path hops, and combine them into a path performance indicator set;

[0019] S202: Perform linear scaling on each data item in the path performance indicator set, unify the load ratio, success ratio, and hop count average to the same dimensional range, and generate a scaled path performance parameter set;

[0020] S203: Based on the scaled path performance parameter set, the three scaled parameters are multiplied to generate a comprehensive ranking coefficient, and the path real-time ranking parameter set is output.

[0021] As a further solution of the present invention, the step of obtaining the primary path index code is:

[0022] S301: Call the service path index field of the microservice registry, extract the index codes of all current paths and their corresponding real-time path sorting parameter sets, establish a mapping relationship between the path codes and the sorting parameters, and generate a path index parameter association set;

[0023] S302: Based on the path index parameter association set, use the formula:

[0024]

[0025] Traverse all paths and calculate the comprehensive priority score of each path;

[0026] in, represents the comprehensive priority score of path i, represents the real-time sorting parameter of path i, represents the mean number of hops of path i, H max Represents the system's preset maximum hop count threshold. represents the load ratio of path i, Represents the average value of the real-time ranking parameters of all paths;

[0027] S303: The path comprehensive priority score is called, the path code with the highest score is selected, the original service path index field is overwritten and locked in an unmodifiable state, and a primary path index code is generated.

[0028] As a further solution of the present invention, the step of obtaining the node verification failure flag is:

[0029] S401: Collect the lesion segmentation mask sequence and pathology classification code field, extract the original content item by item in the order of the fields, calculate the hash value of each field using the SHA-256 algorithm, and generate a hash sequence set;

[0030] S402: Based on the hash sequence set, use the formula:

[0031]

[0032] Traverse the same-bit fields at the beginning and end of the chains before and after the task, calculate and obtain the difference coefficient of each field, and generate a difference coefficient set;

[0033] Among them, D j is the difference coefficient of the jth field, and Represents the decimal value converted from the hexadecimal string of the hash value of the fields before and after the task. and The length of the original content of the field. and Generate timestamp difference for the field;

[0034] S403: calling the difference coefficient set, comparing the difference coefficient of each field with a preset difference threshold, screening an interval where three consecutive field coefficients exceed the threshold, marking all fields in the interval, and generating a node verification failure flag.

[0035] As a further solution of the present invention, the steps for obtaining the pathological classification results of the medical examination are:

[0036] S501: calling the node field of the primary path index code that has not triggered the node verification failure flag, extracting the gray value matrix of the lesion area in the field image, calculating the gray value standard deviation and co-occurrence matrix contrast within the neighborhood of each pixel point, and constructing a texture feature matrix;

[0037] S502: Based on the texture feature matrix, use the formula:

[0038]

[0039] Traverse all disease codes in the standard pathology feature library, calculate and obtain the similarity value of each disease code, and generate a disease similarity set;

[0040] Among them, S r is the similarity value of the rth disease, F r,m is the texture feature value of the mth dimension of the node field, C std,mis the mean of the m-th dimension feature of the r-th disease in the standard pathological feature library, E r,m is the curvature consistency difference between the node field edge contour coordinates and the standard contour in the mth dimension;

[0041] S503: calling the disease similarity set, screening disease codes with similarity values exceeding a preset threshold, selecting the code corresponding to the highest similarity value, and generating a medical examination pathology classification result.

[0042] An AI-based medical test information recognition system, comprising:

[0043] The transmission density monitoring module collects the DICOM transmission counts of medical testing nodes, calculates the absolute difference between the number of transmissions in the current window and the moving average of the time serial port, filters out the transmission paths that exceed the preset difference range, and marks them as path density deviation status;

[0044] A path parameter normalization module extracts the load ratio, forwarding success ratio, and path hop average of candidate paths based on the path density deviation state, normalizes each data item, and generates a path real-time ranking parameter set;

[0045] The primary path index module calls the service path index field of the microservice registry, selects the highest value path code based on the real-time sorting parameter set of the path, overwrites the original field and locks the primary path, and generates the primary path index code;

[0046] The field verification module collects the lesion segmentation mask sequence and pathology classification code fields, generates hash values item by item based on the field sequence and the original content, screens the hash character differences of the same-position fields before and after the task, and generates a node verification failure flag for consecutive fields that exceed the preset threshold;

[0047] The pathological feature matching module calls the node field in the main path index code that has not triggered the node verification failure flag, calculates the cosine similarity between the lesion features of the corresponding image and the standard pathological features, selects the disease code corresponding to the highest similarity, and generates the medical examination pathology classification result.

[0048] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0049] In the present invention, the efficiency of medical examination processing is optimized through dynamic path adjustment and multimodal verification mechanism. The number of DICOM file transmissions is monitored in real time, abnormal paths are identified based on the moving average deviation, and normalized sorting parameters are generated by combining the node load ratio, hop mean and success rate. The service path allocation is dynamically optimized to prevent task delays caused by sudden traffic or node failures. The hash chain comparison technology is used to verify the consistency of the lesion segmentation mask and the classification field, screen the tampering risk data, synchronously extract texture, grayscale and contour features, and integrate the standard pathology library for cosine similarity matching. This method combines real-time resource scheduling with cross-level data verification, embeds multi-dimensional feature cross-validation in transmission anomaly detection, reduces the impact of noise interference and local distortion, and improves processing stability and pathology classification accuracy in high-concurrency scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the method of the present invention;

[0051] Figure 2 This is a flow chart for obtaining the path density deviation state of the present invention;

[0052] Figure 3 This is a flow chart for obtaining a real-time path sorting parameter set according to the present invention;

[0053] Figure 4 This is a flow chart for obtaining the main path index code of the present invention;

[0054] Figure 5 A flowchart for obtaining a node verification failure flag according to the present invention;

[0055] Figure 6 This is a flow chart for obtaining the pathological classification results of medical examinations according to the present invention. DETAILED DESCRIPTION

[0056] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0057] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0058] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0059] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0060] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0061] See also Figure 1 The present invention provides a technical solution: a medical test information recognition method based on AI, comprising the following steps:

[0062] S1: Collect the DICOM file transfer counts forwarded to the medical inspection node in the microservice cluster, calculate the moving average of the transmission times based on the time window sequence, extract the absolute difference between the current window transmission times and the moving average, filter out the transmission paths whose absolute difference exceeds the preset difference range, and mark them as path density deviation status;

[0063] S2: Based on the path density deviation status, the ratio of the current processing number to the maximum processing capacity load, the ratio of the number of successes to the number of attempts, and the average number of path hops are extracted. Each data item is linearly scaled to the same dimensional range to generate a real-time path sorting parameter set;

[0064] S3: Calls the service path index field of the microservice registry, selects the highest-valued path code based on the real-time path sorting parameter set, overwrites the original field, locks the primary path, and generates the primary path index code.

[0065] S4: Collect the lesion segmentation mask sequence and pathology classification code field, generate hash values item by item according to the field order and the original content, calculate the character difference between the hash values of the same-position fields at the beginning and end of the chain before and after the task, filter out the continuous fields whose character difference exceeds the preset difference threshold, and generate a node verification failure flag;

[0066] S5: Call the node field in the main path index code that has not triggered the node verification failure flag, construct the texture feature matrix, grayscale histogram and edge contour coordinates of the lesion in the corresponding node field image, calculate the cosine similarity with the standard pathology feature library, select the disease code corresponding to the highest similarity, and generate the medical examination pathology classification result.

[0067] The path density deviation status specifically includes the path identifier, preset difference range, and time window length. The path real-time sorting parameter set includes the load proportion weight, forwarding success rate coefficient, and hop count average factor. The main path index code specifically includes the highest sorting parameter value, service instance identifier, and path lock time limit identifier. The node verification failure flag includes the difference field position index, continuous field length threshold, and trigger count counter. The medical examination pathology classification result specifically includes the similarity peak, disease code matching item, and feature index mapping relationship.

[0068] See also Figure 2 , the steps to obtain the path density deviation state are:

[0069] S101: Collecting the DICOM file transfer counts forwarded to the medical inspection node in the microservice cluster, recording the number of transmissions according to the time window sequence, and generating transmission number time series data;

[0070] Collect the DICOM file transfer counts forwarded to the medical laboratory node in the microservice cluster. Set the time window to a fixed 5-minute period and count the number of transmissions for each path in each window. For example, path A is transmitted 12 times, 15 times, and 18 times in three consecutive windows, and path B is transmitted 8 times, 10 times, and 9 times. Associate the window number with the number of transmissions and store it as structured data. For example, the sequence of path A is stored as {window 1:12, window 2:15, window 3:18}, and that of path B is {window 1:8, window 2:10, window 3:9}. By traversing the transmission logs of all paths, extracting the window granularity counts, and generating transmission time series data, the window length is set according to the system processing delay requirements. For example, a 5-minute window can balance real-time performance and fluctuation sensitivity.

[0071] S102: Based on the time series data of the number of transmissions, a moving average of the number of transmissions is calculated using a time window sequence to generate a moving average number of transmissions;

[0072] The data of the current window and the two previous windows are selected to calculate the moving average. For example, path A is transmitted 21 times in window 4, and the moving average of its first two windows (window 2: 15, window 3: 18) is (15+18+21) / 3=18 times. If the number of transmissions of window 5 is 24, the moving average window is updated to window 3 (18), window 4 (21), and window 5 (24). The calculation result is (18+21+24) / 3=21 times. Each time the window slides, the earliest window data is removed and the latest window data is added. The moving average transmission times are dynamically updated. For new paths without previous windows, the current window data is used to fill in the initial stage.

[0073] S103: Call the moving average transmission count, calculate the absolute difference between the current window transmission count and the corresponding moving average, compare the absolute difference with a preset difference range threshold, filter out the transmission paths that exceed the limit and mark them as path density deviation status;

[0074] The number of transmissions in the current window is obtained. For example, if window 5 of path A transmits 24 times, the corresponding moving average is 21 times. The absolute difference is calculated as |24-21|=3 times. The preset difference range threshold is 5 times. The judgment rule is to mark an abnormality when the difference exceeds the threshold. For example, if window 5 of path C transmits 30 times, the moving average is 20 times, and the difference is 10 times, which exceeds the threshold by 5 times, it is marked as a path density deviation state. The threshold is set based on the standard deviation of historical transmission fluctuations. For example, the difference distribution of the past 100 windows is statistically analyzed, and the 95% quantile (such as 5 times) is taken as the threshold to avoid misjudgment of occasional fluctuations.

[0075] See also Figure 3 ,The steps to obtain the path real-time sorting parameter set are:

[0076] S201: Based on the path density deviation state, extract the ratio of the current processing number to the maximum processing capacity load, the ratio of the number of successes to the number of attempts, and the average number of path hops, and merge them into a path performance indicator set;

[0077] Obtain the real-time processing count of the path marked as path density deviation. For example, the current processing count of path D is 85, and its preset maximum processing capacity is 100. The calculated load ratio is 85 / 100 = 0.85. The number of successful transmissions is 82, and the number of attempts is 85. The calculated success ratio is 82 / 85≈0.965. At the same time, the average number of path hops is extracted as 4 hops (the path passes through 4 intermediate nodes). The three indicators are combined into a path performance indicator set. For example, the indicator set of path D is [0.85, 0.965, 4]. The data source is the real-time monitoring interface of the microservice cluster and is updated every 5 seconds.

[0078] S202: Perform linear scaling on each data item in the path performance indicator set, unify the load ratio, success ratio, and hop count average to the same dimensional range, and generate a scaled path performance parameter set;

[0079] Linear scaling is performed on the path performance indicator set. The original range of the load ratio and success ratio is set to 0-1, without scaling. The original range of the hop count average is 1-8 hops (based on the maximum hop count of the network topology). The scaling formula is: hop count scaling value = (8-current hop count) / (8-1). For example, if the average hop count of path D is 4, the scaling value is (8-4) / 7≈0.571. If the hop count of path E is 2, the scaling value is (8-2) / 7≈0.857. The scaled path performance parameter set is generated, for example, [0.85, 0.965, 0.571] for path D and [0.78, 0.92, 0.857] for path E, ensuring that all parameters are in the range of 0-1.

[0080] S203: Based on the scaled path performance parameter set, the three scaled parameters are multiplied to generate a comprehensive ranking coefficient, and the real-time ranking parameter set of the path is output;

[0081] The comprehensive ranking coefficient is defined as load ratio × success ratio × (1-hop scaling value). (1-hop scaling value) converts the hop count into a positive indicator (the lower the hop count, the higher the value). For example, the coefficient of path D = 0.85 × 0.965 × (1-0.571) = 0.85 × 0.965 × 0.429 ≈ 0.354, and the coefficient of path E = 0.78 × 0.92 × (1-0.857) = 0.78 × 0.92 × 0.143 ≈ 0.103. Paths are sorted from high to low by coefficient to generate a real-time ranking parameter set. For example, the ranking value of path D is 0.354, and that of path F is 0.298. These paths are used to prioritize paths with high load, high success rate, and low hop count.

[0082] See also Figure 4 , the steps to obtain the main path index code are:

[0083] S301: Call the service path index field of the microservice registry, extract the index codes of all current paths and their corresponding path real-time sorting parameter sets, establish a mapping relationship between the path codes and the sorting parameters, and generate a path index parameter association set;

[0084] Extract the coding identifiers of all paths in the service path index table, for example, path A is coded as PA001, path B is coded as PB002, and path C is PC003. Synchronously obtain the path real-time sorting parameter set corresponding to each path, for example, the parameter of path A is 0.36, path B is 0.135, and path C is 0.28. Construct a mapping table of the coding and parameters and store it as a key-value pair structure {PA001:0.36, PB002:0.135, PC003:0.28}. The mapping table is updated every 30 seconds. The data source is the real-time monitoring module of the registry. The latest parameters are pulled through the API interface to generate a path index parameter association set. For example, the mapping table contains 10 path data, among which the parameter of path D coded PD004 is 0.42.

[0085] S302: Based on the path index parameter association set, the formula is used:

[0086]

[0087] Traverse all paths and calculate the comprehensive priority score of each path;

[0088] in, represents the comprehensive priority score of path i, represents the real-time sorting parameter of path i, represents the mean number of hops of path i, H max Represents the system's preset maximum hop count threshold. represents the load ratio of path i, Represents the average value of the real-time ranking parameters of all paths;

[0089] Based on the path index parameter association set, path A (PA001), path C (PC003), and path D (PD004) are selected as examples to extract real-time sorting parameters. Average hop count Load ratio System default H max =8,

[0090] Calculate the numerator part of path A:

[0091] 0.36×(1-3 / 8)+(1-0.75)=0.36×0.625+0.25=0.475;

[0092] Calculate the denominator:

[0093] Real-time sorting parameter average

[0094]

[0095] The square of the difference between each parameter and the mean:

[0096] (0.36-0.353) 2 =0.00005;

[0097] (0.28-0.353) 2 =0.0053;

[0098] (0.42-0.353) 2 =0.0045;

[0099] Sum of variances: 0.00005 + 0.0053 + 0.0045 = 0.00985;

[0100] Standard Deviation:

[0101] Denominator:

[0102] final:

[0103] The results show that the comprehensive priority score of path A is 0.4726. A higher score indicates that the path performs better in terms of load, success rate, hop count, and other dimensions. This value will be directly used for subsequent path ranking comparison.

[0104] Similarly, calculate path C:

[0105] molecular:

[0106] 0.28×(1-4 / 8)+(1-0.85)=0.28×0.5+0.15=0.29;

[0107] Denominator: (shares the same denominator as path A);

[0108] final:

[0109] The results show that path C has a lower score of 0.2886 because its average number of hops is higher (4 hops) and its load ratio is higher (0.85), and its overall performance is weaker than other paths.

[0110] Path D:

[0111] molecular:

[0112] 0.42×(1-2 / 8)+(1-0.68)=0.42×0.75+0.32=0.635;

[0113] Denominator:

[0114] final:

[0115] The results show that the path with a D score of 0.6318 currently has the highest score. This path has the lowest average hop count (2 hops) and the highest real-time ranking parameter (0.42), significantly outperforming other paths in overall priority. This value will serve as the core basis for selecting the primary path.

[0116] The generated path comprehensive priority score set {PA001: 0.4726, PC003: 0.2886, PD004: 0.6318}.

[0117] S303: Call the path comprehensive priority score, select the path code with the highest score, overwrite the original service path index field and lock it in an unmodifiable state, and generate the primary path index code;

[0118] The path comprehensive priority score set {PA001: 0.4726, PC003: 0.2886, PD004: 0.6318} is called, and the scores are compared. Path PD004 (score 0.6318) with the highest score is selected. The identifier of path PD004 in the original index field of the microservice registry is overwritten, and the field status is changed to locked (uneditable). The primary path index code PD004 is generated. The lock takes effect at 2025-04-15 15:00:00 and lasts until the next scoring cycle (30 seconds later). During this period, all traffic is forcibly routed to path D. If path E has the highest score of 0.72 in the next cycle, the primary code is updated to PE005 and the lock is re-set.

[0119] See also Figure 5 , the steps to obtain the node verification failure flag are:

[0120] S401: Collect the lesion segmentation mask sequence and pathology classification code field, extract the original content item by item in the order of the fields, calculate the hash value of each field using the SHA-256 algorithm, and generate a hash sequence set;

[0121] Collect the lesion segmentation mask sequence and pathology classification code field, extract the original content of three consecutive fields in a certain task, for example, field 1 is "liver nodule_diameter 5mm", field 2 is "classification T1 N0M0", field 3 is "malignant probability 0.72", and use the SHA-256 algorithm to generate hash values for each.

[0122] The hash value of field 1 is:

[0123] 6b86b273ff34fce19d6b804eff5a3f5747ada4eaa22f1d49c01e52ddb7875b4b;

[0124] Field 2 is:

[0125] d4735e3a265e16eee03f59718b9b5d03019c07d8b6c51f90da3a666eec13ab35;

[0126] Field 3 is:

[0127] 4e07408562bedb8b60ce05c1decfe3ad16b72230967de01f640b7e4729b49fce;

[0128] Convert each two digits of the hexadecimal hash value to decimal and add them up. For example, the first two digits "6b" of the hash value of field 1 are converted to 107, and the cumulative total is 28495. At the same time, add the hash values of field 2 and field 3 respectively to generate the hash sequence set {field 1: 28495, field 2: 37210, field 3: 19876}.

[0129] S402: Based on the hash sequence set, use the formula:

[0130]

[0131] Traverse the same-bit fields at the beginning and end of the chains before and after the task, calculate and obtain the difference coefficient of each field, and generate a difference coefficient set;

[0132] Among them, D j is the difference coefficient of the jth field, and Represents the decimal value converted from the hexadecimal string of the hash value of the fields before and after the task. and The length of the original content of the field. and Generate timestamp difference for the field;

[0133] Select the original content of field 3 "malicious probability 0.72", the hash value before the task After the task is tampered with, it is changed to "malicious probability 0.85", and the hash value Original content length Characters, tampered Characters (content changes but length remains unchanged), timestamp difference seconds (generated 10 seconds after the task starts), (5 seconds delay after tampering), substitute into the formula:

[0134]

[0135] The result shows that the difference coefficient of field 3 is 544.54. Since it is lower than the preset threshold of 800, it is determined that the current field has not been significantly tampered with. However, it is necessary to combine the adjacent fields to determine the continuity anomaly. If the content length increases after tampering (for example, it is changed to "malicious probability 0.85_high risk", ),but:

[0136]

[0137] The result shows that the difference coefficient of field 3 is still 583.92. Although it does not exceed the threshold, it is necessary to link field 4 (D4 = 1220.8 after tampering) and field 5 (D5 = 1185.6) to determine whether the continuous tampering condition is met, and generate the difference coefficient set {field 3: 544.54, field 4: 1220.8, field 5: 1185.6}.

[0138] S403: Call the difference coefficient set, compare the difference coefficient of each field with the preset difference threshold, select the interval where three consecutive field coefficients exceed the threshold, mark all fields in the interval, and generate a node verification failure flag;

[0139] The difference coefficients of fields 3-5 are detected. The continuous field interval is traversed. Field 4 (1220.8) and field 5 (1185.6) both exceed the threshold of 800, but field 3 (544.54) does not exceed the threshold. It is determined that the condition of three consecutive fields exceeding the threshold is not met. The subsequent field 6 (1290.2), field 7 (1315.7), and field 8 (1288.3) are detected. The coefficients of the three fields all exceed the threshold. This result shows that fields 6-8 constitute a continuous abnormal interval, triggering the node verification failure flag generation condition. Fields 6-8 are marked as an abnormal interval and a node verification failure flag {start bit: 6, end bit: 8} is generated. After the flag is triggered, the system performs the following operations:

[0140] Data isolation: Block the transmission link of fields 6-8. For example, in the lung cancer pathology system, field 6 (lymph node metastasis number 3) is tampered with to "5". After isolation, it is prevented from entering the TNM staging calculation module.

[0141] Data rollback: Restore the original values of fields 6-8 (e.g., "Number of lymph node metastases 3_check code a1 b2c") from the backup database, overwriting the tampered data.

[0142] Audit trail: records the tampering timestamp (2025-04-15 14:30:00), operator ID (User_X), and original data hash value for security audits;

[0143] For example, in a breast cancer diagnosis scenario, field 7 "Hormone receptor positivity rate 70%" was tampered with to "30%". The system locked the abnormal interval through an invalidation flag, and after rolling back, ensured that the treatment plan was generated based on correct data, and at the same time initiated an early warning notification to the quality management platform.

[0144] See also Figure 6 ,The steps for obtaining the pathological classification results of medical examinations are:

[0145] S501: Calling the node field in the primary path index code that has not triggered the node verification failure flag, extracting the grayscale value matrix of the lesion area in the field image, calculating the grayscale standard deviation and co-occurrence matrix contrast within the neighborhood of each pixel point, and constructing a texture feature matrix;

[0146] The node field in the main path index code that does not trigger the node verification failure flag is called, and the grayscale value matrix of the lesion area in a liver cancer case image is extracted. The grayscale value set {85, 88, 90, 82, 86, 84, 89, 87, 83} of the neighborhood (3×3 window) of the pixel coordinate (x=120, y=80) is collected. The grayscale standard deviation of this neighborhood is calculated to be 2.89. The standard deviation calculation is completed by traversing all pixels to generate a grayscale standard deviation matrix. At the same time, based on the same image area, the contrast parameter of the grayscale co-occurrence matrix (direction 0°, distance 1 pixel) is calculated to be 12.5, the energy parameter is 0.15, and the homogeneity parameter is 0.92. The grayscale standard deviation and co-occurrence matrix parameters of all pixels are integrated to construct a texture feature matrix. For example, in the liver cancer case, the matrix contains the grayscale standard deviation of 1200×800 pixels and the three sets of parameters of contrast, energy, and homogeneity for each pixel.

[0147] S502: Based on the texture feature matrix, use the formula:

[0148]

[0149] Traverse all disease codes in the standard pathology feature library, calculate and obtain the similarity value of each disease code, and generate a disease similarity set;

[0150] Among them, S r is the similarity value of the rth disease, F r,m is the texture feature value of the node field m, C std,m is the mean of the m-th dimension feature of the r-th disease in the standard pathological feature library, E r,m is the curvature consistency difference between the node field edge contour coordinates and the standard contour in the mth dimension;

[0151] The texture feature value of a node field of a lung cancer case is selected, and the first dimension feature is set to contrast 12.5, the second dimension is energy 0.15, and the third dimension is homogeneity 0.92. The mean value of the first dimension feature of lung cancer in the standard pathological feature library is 11.8, the second dimension is 0.13, and the third dimension is 0.89. The curvature radius of the node field edge contour coordinate at the sampling point (x=50, y=60) is 8.2mm, and the curvature radius of the standard lung cancer contour at the same position is 7.5mm. The curvature consistency difference E r,1 =|8.2-7.5|=0.7, substitute into the formula:

[0152]

[0153] Step-by-step calculation:

[0154] molecular:

[0155] 147.5+0.0195+0.8228=148.3423;

[0156] Denominator:

[0157]

[0158] S r =148.3423 / (12.53×12.54)≈0.942;

[0159] The result shows that the similarity value of lung cancer is 0.942, which exceeds the preset threshold of 0.90. It is determined that the current node field feature is highly matched with the standard feature of lung cancer and needs to be included in the disease similarity set. The lung cancer coding similarity value in the disease similarity set is 0.942, the gastric cancer coding is 0.821, and the colon cancer coding is 0.785 to generate a disease similarity set.

[0160] S503: Calling the disease similarity set, screening the disease codes whose similarity values exceed a preset threshold, selecting the code corresponding to the highest similarity value, and generating a medical examination pathology classification result;

[0161] The disease similarity set {lung cancer code: 0.942, liver cancer code: 0.915, gastric cancer code: 0.821} is called, and the similarity values of all disease codes are traversed. The preset similarity threshold is set to 0.90 (based on the statistics of 1000 clinical pathology data, the 98% quantile is 0.89, and the redundancy is increased to 0.90 to reduce the false matching rate). The lung cancer code (0.942) and liver cancer code (0.915) that exceed the threshold are screened, and the gastric cancer code (0.821) is excluded. The similarity values of lung cancer and liver cancer codes are compared, and the difference Δ = 0.942-0.915 = 0.027 is calculated. It is determined whether Δ is greater than the preset fluctuation tolerance of 0.02 (based on the standard deviation of fluctuation of the same type of disease codes in historical data σ = 0. 015, setting the tolerance to σ×1.5≈0.02). Since Δ=0.027>0.02, the lung cancer code is determined to be the highest similarity item, and the ICD-10 standard field (C34.901) of the lung cancer code in the standard pathology feature library is matched to generate the medical examination pathology classification result. For example, the liver nodule image of a patient is calculated to have a similarity of 0.942 with the standard feature of lung cancer and 0.915 with liver cancer. The system outputs the C34.901 code, triggering the multidisciplinary consultation process for lung cancer and simultaneously pushing a list of targeted treatment drugs (such as osimertinib). If the similarity value does not exceed the threshold (such as 0.895 for liver cancer code), the secondary verification process is started, and the texture feature matrix of the adjacent node field is called to recalculate the similarity until the threshold condition is met.

[0162] An AI-based medical test information recognition system, the system comprising:

[0163] The transmission density monitoring module collects the DICOM transmission counts of medical testing nodes, calculates the absolute difference between the number of transmissions in the current window and the moving average of the time serial port, filters out the transmission paths that exceed the preset difference range, and marks them as path density deviation status;

[0164] The path parameter normalization module extracts the candidate path load ratio, forwarding success ratio, and path hop average based on the path density deviation status, normalizes each data item, and generates a real-time path ranking parameter set;

[0165] The primary path index module calls the service path index field of the microservice registry, selects the highest value path code based on the real-time path sorting parameter set, overwrites the original field, locks the primary path, and generates the primary path index code;

[0166] The field verification module collects the lesion segmentation mask sequence and pathology classification code fields, generates hash values item by item based on the field sequence and the original content, screens the hash character differences of the same-position fields before and after the task, and generates a node verification failure flag for consecutive fields that exceed the preset threshold;

[0167] The pathological feature matching module calls the node field in the main path index code that has not triggered the node verification failure flag, calculates the cosine similarity between the lesion features of the corresponding image and the standard pathological features, selects the disease code corresponding to the highest similarity, and generates the medical examination pathology classification results.

[0168] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A medical test information recognition method based on AI, characterized in that: The following steps are involved: S1: Collect DICOM transmission counts of medical testing nodes, calculate the absolute difference between the number of transmissions in the current window and the moving average of the time serial port, filter out transmission paths that exceed the preset difference range, and mark them as path density deviation status; S2: Based on the path density deviation state, extract the candidate path load ratio, forwarding success ratio, and path hop average, normalize each data item, and generate a path real-time ranking parameter set; S3: Call the service path index field of the microservice registry, select the highest value path code based on the real-time sorting parameter set of the path, overwrite the original field, lock the primary path, and generate the primary path index code; S4: Collect lesion segmentation mask sequences and pathology classification code fields, generate hash values item by item according to the field order and original content, screen the hash character differences of the same-position fields before and after the task, and generate node verification failure flags for continuous fields that exceed the preset threshold; S5: Call the node field in the main path index code that has not triggered the node verification failure flag, calculate the cosine similarity between the lesion characteristics of the corresponding image and the standard pathological characteristics, select the disease code corresponding to the highest similarity, and generate a medical examination pathology classification result.

2. The AI-based medical test information recognition method according to claim 1, characterized in that: The path density deviation state specifically includes a path identifier, a preset difference range, and a time window length; the path real-time sorting parameter set includes a load proportion weight, a forwarding success rate coefficient, and a hop count mean factor; the primary path index code specifically includes the highest sorting parameter value, a service instance identifier, and a path lock time limit identifier; the node verification failure flag includes a difference field position index, a continuous field length threshold, and a trigger count counter; and the medical examination pathology classification result specifically includes a similarity peak, a disease code matching item, and a feature index mapping relationship.

3. The AI-based medical examination information recognition method according to claim 1, characterized in that: The steps for obtaining the path density deviation state are: S101: Collecting the DICOM file transfer counts forwarded to the medical inspection node in the microservice cluster, recording the number of transmissions according to the time window sequence, and generating transmission number time series data; S102: Based on the transmission timestamp time series data, a time window sequence is applied to calculate a moving average of the transmission timestamps to generate a moving average transmission timestamp; S103: calling the moving average transmission times, calculating the absolute difference between the current window transmission times and the corresponding moving average, comparing the absolute difference with a preset difference range threshold, screening out-of-limit transmission paths and marking them as path density deviation states.

4. The AI-based medical test information recognition method according to claim 1, characterized in that: The steps for obtaining the path real-time sorting parameter set are: S201: Based on the path density deviation state, extract the ratio of the current processing number to the maximum processing capacity load, the ratio of the number of successes to the number of attempts, and the average number of path hops, and combine them into a path performance indicator set; S202: Perform linear scaling on each data item in the path performance indicator set, unify the load ratio, success ratio, and hop count average to the same dimensional range, and generate a scaled path performance parameter set; S203: Based on the scaled path performance parameter set, the three scaled parameters are multiplied to generate a comprehensive ranking coefficient, and the path real-time ranking parameter set is output.

5. The AI-based medical test information recognition method according to claim 1, characterized in that: The steps for obtaining the primary path index code are as follows: S301: Call the service path index field of the microservice registry, extract the index codes of all current paths and their corresponding real-time path sorting parameter sets, establish a mapping relationship between the path codes and the sorting parameters, and generate a path index parameter association set; S302: Based on the path index parameter association set, use the formula: Traverse all paths and calculate the comprehensive priority score of each path; in, represents the comprehensive priority score of path i, represents the real-time sorting parameter of path i, represents the mean number of hops of path i, H max Represents the system's preset maximum hop count threshold. represents the load ratio of path i, Represents the average value of the real-time ranking parameters of all paths; S303: The path comprehensive priority score is called, the path code with the highest score is selected, the original service path index field is overwritten and locked in an unmodifiable state, and a primary path index code is generated.

6. The AI-based medical test information recognition method according to claim 1, characterized in that: The steps for obtaining the node verification failure flag are: S401: Collect the lesion segmentation mask sequence and pathology classification code field, extract the original content item by item in the order of the fields, calculate the hash value of each field using the SHA-256 algorithm, and generate a hash sequence set; S402: Based on the hash sequence set, use the formula: Traverse the same-bit fields at the beginning and end of the chains before and after the task, calculate and obtain the difference coefficient of each field, and generate a difference coefficient set; Among them, D j is the difference coefficient of the jth field, and Represents the decimal value converted from the hexadecimal string of the hash value of the fields before and after the task. and The length of the original content of the field. and Generate timestamp difference for the field; S403: calling the difference coefficient set, comparing the difference coefficient of each field with a preset difference threshold, screening an interval where three consecutive field coefficients exceed the threshold, marking all fields in the interval, and generating a node verification failure flag.

7. The AI-based medical test information recognition method according to claim 1, characterized in that: The steps for obtaining the medical examination pathology classification results are as follows: S501: calling the node field of the primary path index code that has not triggered the node verification failure flag, extracting the gray value matrix of the lesion area in the field image, calculating the gray value standard deviation and co-occurrence matrix contrast within the neighborhood of each pixel point, and constructing a texture feature matrix; S502: Based on the texture feature matrix, use the formula: Traverse all disease codes in the standard pathology feature library, calculate and obtain the similarity value of each disease code, and generate a disease similarity set; Among them, S r is the similarity value of the rth disease, F r,m is the texture feature value of the mth dimension of the node field, C std,m is the mean of the m-th dimension feature of the r-th disease in the standard pathological feature library, E r,m is the curvature consistency difference between the node field edge contour coordinates and the standard contour in the mth dimension; S503: calling the disease similarity set, screening disease codes with similarity values exceeding a preset threshold, selecting the code corresponding to the highest similarity value, and generating a medical examination pathology classification result.

8. An AI-based medical test information recognition system, characterized in that: The system is used in the AI-based medical test information recognition method according to any one of claims 1 to 7, and the system comprises: The transmission density monitoring module collects the DICOM transmission counts of medical testing nodes, calculates the absolute difference between the number of transmissions in the current window and the moving average of the time serial port, filters out the transmission paths that exceed the preset difference range, and marks them as path density deviation status; A path parameter normalization module extracts the load ratio, forwarding success ratio, and path hop average of candidate paths based on the path density deviation state, normalizes each data item, and generates a path real-time ranking parameter set; The primary path index module calls the service path index field of the microservice registry, selects the highest value path code based on the real-time sorting parameter set of the path, overwrites the original field and locks the primary path, and generates the primary path index code; The field verification module collects the lesion segmentation mask sequence and pathology classification code fields, generates hash values item by item based on the field sequence and the original content, screens the hash character differences of the same-position fields before and after the task, and generates a node verification failure flag for consecutive fields that exceed the preset threshold; The pathological feature matching module calls the node field in the main path index code that has not triggered the node verification failure flag, calculates the cosine similarity between the lesion features of the corresponding image and the standard pathological features, selects the disease code corresponding to the highest similarity, and generates the medical examination pathology classification result.

Images