CN-121999487-A - Anti-nuclear antibody identification method and system for multi-algorithm collaborative judgment

CN121999487ACN 121999487 ACN121999487 ACN 121999487ACN-121999487-A

Abstract

The application discloses a method and a system for identifying an anti-nuclear antibody by multi-algorithm cooperative judgment, and relates to the technical field of disease detection, wherein the method comprises the steps of obtaining a target serum sample for detection to obtain an image to be identified; extracting features of an image to be identified to obtain a plurality of image features, interacting the plurality of image features to perform similar type feature iterative enhancement to obtain target image features, and calling a domain knowledge enhancement algorithm to identify the target image features to obtain an anti-nuclear antibody identification result. Solves the problems that the existing anti-nuclear antibody identification method is single, the identification accuracy is not high, and the clinical rapid diagnosis requirement is difficult to meet.

Inventors

ZHAO QIAN
Li Kuangfa
ZHANG HUISAN

Assignees

中国人民解放军空军军医大学

Dates

Publication Date: 20260508
Application Date: 20260116

Claims (8)

1. The anti-nuclear antibody identification method based on multi-algorithm cooperative judgment is characterized by comprising the following steps of: Obtaining a target serum sample, and detecting the target serum sample by using an immunofluorescence method to obtain an image to be identified; Extracting features of the image to be identified by utilizing a plurality of feature identification algorithms to obtain a plurality of image features, wherein the feature identification algorithms perform asynchronous reinforcement updating based on a dual-delay depth deterministic strategy; The image features are interacted to carry out the iterative enhancement of the same type of features, and target image features are obtained; and calling a domain knowledge enhancement algorithm to identify the target image characteristics to obtain an anti-nuclear antibody identification result.
2. The method for identifying the anti-nuclear antibody by multi-algorithm collaborative judgment according to claim 1, wherein the method for identifying the anti-nuclear antibody is characterized in that a plurality of feature identification algorithms are respectively utilized to respectively extract features of the image to be identified to obtain a plurality of image features, wherein the plurality of feature identification algorithms perform asynchronous reinforcement update based on a dual-delay depth deterministic strategy, and the method comprises the following steps: A plurality of strategy networks and a plurality of independent evaluation network groups of the feature recognition algorithms are respectively acquired, wherein each independent evaluation network group comprises a singular update evaluation network and a double update evaluation network; Acquiring a plurality of history image feature sets corresponding to N times of anti-nuclear antibody recognition occurring after a first moment of a plurality of feature recognition algorithms and a history target image feature set obtained after enhancement, wherein N is a positive integer; Analyzing the plurality of historical image feature sets and the historical target image feature set based on the dual-delay depth deterministic strategy, asynchronously updating the plurality of independent evaluation network groups, and determining a plurality of asynchronous intensive update schemes based on the plurality of updated independent evaluation network groups; And respectively carrying out asynchronous reinforcement updating on the plurality of feature recognition algorithms according to the plurality of asynchronous reinforcement updating schemes.
3. The method of claim 2, wherein resolving the plurality of historical image feature sets and the historical target image feature set based on the dual latency depth deterministic strategy, asynchronously updating the plurality of independent evaluation network groups, and determining a plurality of asynchronous intensive update schemes based on the plurality of updated independent evaluation network groups, comprises: carrying out same-batch mapping deviation recognition on the plurality of historical image feature sets and the historical target image feature set respectively to obtain a plurality of historical feature recognition deviation degree sets; Based on the anti-nuclear antibody recognition times corresponding to the plurality of historical feature recognition deviation degree sets, respectively carrying out asynchronous update on the singular update evaluation network and the even update evaluation network in the plurality of independent evaluation network groups in sequence to obtain a plurality of singular update weight sets and a plurality of even update weight sets; mapping and summarizing the plurality of singular update weight sets and the plurality of even update weight sets to obtain a plurality of asynchronous update weight sets; And integrating the plurality of asynchronous updating weight sets to perform balanced drift screening, determining a plurality of asynchronous strengthening weights, and taking the plurality of asynchronous strengthening weights as the plurality of asynchronous strengthening updating schemes.
4. The method for identifying an anti-nuclear antibody for collaborative determination according to claim 3, wherein integrating the plurality of sets of asynchronously updated weights for equalization drift screening, determining a plurality of asynchronously enhanced weights, comprises: extracting a first asynchronous update weight set from a plurality of asynchronous update weight sets, and extracting multi-point-position anisotropic weights from the first asynchronous update weight set to obtain a multi-point-position asynchronous update weight set; respectively carrying out equalization drift screening on each point position asynchronous updating weight in a multi-point position asynchronous updating weight set, and determining a first point position asynchronous strengthening weight set and a first point position asynchronous strengthening weight reliability coefficient set; and taking the point position asynchronous reinforcement weight corresponding to the maximum value in the first point position asynchronous reinforcement weight reliability coefficient set as a first asynchronous reinforcement weight, and adding the first asynchronous reinforcement weight into a plurality of asynchronous reinforcement weights.
5. The method for identifying an anti-nuclear antibody for collaborative judgment according to claim 4, wherein the step of performing equalization drift screening on each of a plurality of sets of asynchronous update weights to determine a first set of asynchronous reinforcement weights and a first set of asynchronous reinforcement weight reliability coefficients comprises: according to a preset screening bandwidth, iterating each point location asynchronous updating weight in the multi-point location asynchronous updating weight set, and determining a plurality of iterated point location asynchronous updating weights; Respectively counting the number of distances from the plurality of iteration point positions to the asynchronous updating weights to be smaller than or equal to a preset screening bandwidth to obtain a reliability coefficient of the plurality of iteration point positions; Respectively counting the number of distances from the plurality of point location asynchronous updating weights in the multi-point location asynchronous updating weight set to be smaller than or equal to a preset screening bandwidth to obtain a plurality of point location asynchronous updating weight reliability coefficients; When the reliability coefficient of the asynchronous updating weights of the multiple iteration points is larger than or equal to that of the multiple point positions, the corresponding asynchronous updating weights of the multiple iteration points are used as multiple stage screening results, and the multiple stage screening results are repeatedly screened and compared with the reliability coefficient according to a preset screening bandwidth until the preset iteration times are met, so that the asynchronous updating weights of the multiple target points and the asynchronous updating reliability coefficients of the multiple target points are obtained; And summarizing the plurality of target point position asynchronous updating weights and the plurality of target point position asynchronous updating weight reliability coefficients to obtain a first point position asynchronous strengthening weight set and a first point position asynchronous strengthening weight reliability coefficient set.
6. The method for identifying an anti-nuclear antibody by multi-algorithm collaborative judgment according to claim 1, wherein the step of performing the same type of feature iterative enhancement on the plurality of image features to obtain the target image features comprises the steps of: respectively carrying out image feature pairwise non-repeated interaction on the plurality of image features to obtain a plurality of interaction combinations; extracting a first interaction combination from the plurality of interaction combinations, performing similar feature iterative enhancement on the first interaction combination to obtain a first interaction enhancement combination, and adding the first interaction enhancement combination into an interaction enhancement combination set; Dividing the interaction enhancement combination set which belongs to one image feature to obtain a plurality of interaction enhancement image feature clusters; And respectively carrying out intra-cluster mean processing on the plurality of interaction enhanced image feature clusters to obtain target image features.
7. The method for identifying an anti-nuclear antibody by multi-algorithm collaborative judgment according to claim 6, wherein performing iterative enhancement of the same type of feature on the first interaction combination to obtain a first interaction enhancement combination comprises: performing similar feature similarity analysis on the first interaction combination to construct a similar adjacency matrix; and carrying out picture convolution enhancement on each image characteristic in the first interaction combination based on the similar adjacency matrix to obtain the first interaction enhancement combination.
8. An anti-nuclear antibody recognition system for multi-algorithm collaborative judgment, characterized in that it is used for executing an anti-nuclear antibody recognition method for multi-algorithm collaborative judgment according to any one of claims 1-8, comprising: The sample acquisition module is used for acquiring a target serum sample, detecting the target serum sample by using an immunofluorescence method, and acquiring an image to be identified; The feature extraction module is used for extracting features of the image to be identified by utilizing a plurality of feature recognition algorithms to obtain a plurality of image features, wherein the feature recognition algorithms perform asynchronous reinforcement updating based on a dual-delay depth deterministic strategy; The feature iteration module is used for interacting the plurality of image features to perform the same type of feature iteration enhancement to obtain target image features; and the result acquisition module is used for calling a domain knowledge enhancement algorithm to identify the target image characteristics so as to obtain an anti-nuclear antibody identification result.

Description

Anti-nuclear antibody identification method and system for multi-algorithm collaborative judgment Technical Field The application relates to the technical field of disease detection, in particular to an anti-nuclear antibody identification method and system for multi-algorithm cooperative judgment. Background Anti-nuclear antibody (ANA) detection is a core test for screening and diagnosing autoimmune diseases, and by adding patient serum to cells cultured in vitro (most commonly used human laryngeal carcinoma HEp-2 cells), if the serum contains autoantibodies capable of binding to cell nucleus (or cytoplasm and mitotic device) antigens, and adding fluorescent labeled anti-human IgG, a characteristic bright green fluorescent pattern can be seen under a fluorescent microscope, so that the possibility of autoimmune diseases such as systemic lupus erythematosus, sjogren's syndrome, scleroderma, mixed connective tissue diseases and the like is suggested, and the test is called a window of autoimmune diseases. With the continuous development of medical technology, the detection of antinuclear antibodies is increasingly important in the diagnosis, condition monitoring, prognosis evaluation and the like of autoimmune diseases. The core contradiction of traditional artificial ANA detection is that the detection is highly dependent on experience of a tester and cannot be standardized, so that the result variation is large, the efficiency is low and the traceability is difficult. In order to solve the defects, the medical aspect promotes the rapid development of the automatic fluorescence imaging combined with the AI algorithm auxiliary interpretation technology. After an artificial intelligent algorithm enters anti-nuclear antibody (ANA) Indirect Immunofluorescence (IIF) interpretation, obvious advantages that a traditional artificial microscope is difficult to replace are shown on clinic, hospitalization routine, large-scale physical examination, quality control and remote consultation, latest domestic multi-center verification shows that the overall interpretation accuracy of ANA-AI intelligent agents based on deep learning on various single and partial mixed nuclear types can reach 97.34%, the overall interpretation accuracy is highly consistent with three-main-stage technician blind method rechecking results, human errors are obviously reduced, full-automatic film production is combined with AI second-level image analysis, the average turnaround time of single specimen is shortened to be within 18 s from 5-8 min, laboratory flux is improved by 16-26 times, the efficiency of an input link is improved by 90%, thousands of physical examination or emergency peaks of daily examination can be easily handled, the algorithm outputs a fluorescent index and a probability value for each serum, automatic archiving of all original images and interpretation parameters is realized, and traceable requirements of results are met. However, the existing identification method often depends on a single algorithm, and is difficult to comprehensively and accurately capture complex features of the anti-nuclear antibody in an image, so that conditions of missed detection (false negative) and false detection (false positive) are easy to occur in scenes such as low-titer fine particle modes, rare subtypes, mixed nuclear types, abnormal cell sheet quality, high background/blood source interference, microscopic imaging condition drift and the like, and the accuracy of a diagnosis result is low. On the other hand, the existing feature extraction and recognition processes lack an effective synergistic mechanism, so that the efficiency is low when a large number of serum sample images are processed, and the requirement of clinical rapid diagnosis is difficult to meet. Disclosure of Invention The embodiment of the application solves the technical problems that the existing anti-nuclear antibody identification method is single, the identification accuracy is low and the clinical rapid diagnosis requirement is difficult to meet by providing the anti-nuclear antibody identification method and the system for multi-algorithm collaborative judgment. The technical scheme for solving the technical problems is as follows: in a first aspect, the present application provides a method for identifying an antinuclear antibody by cooperative judgment of multiple algorithms, the method comprising: Obtaining a target serum sample, and detecting the target serum sample by using an immunofluorescence method to obtain an image to be identified; Extracting features of the image to be identified by utilizing a plurality of feature identification algorithms to obtain a plurality of image features, wherein the feature identification algorithms perform asynchronous reinforcement updating based on a dual-delay depth deterministic strategy; The image features are interacted to carry out the iterative enhancement of the same type of features, and target image features are