A method and system for risk recalibration based on site-level endoscopic pathology mismatch

Abstract

This invention discloses a risk recalibration method based on site-level endoscopic-pathological mismatch, comprising the following steps: S1: For a fixed gastric site of a patient, collect site-level data from multiple follow-ups, wherein the site-level data includes endoscopic abnormality scores, pathological examination results, adequacy information, localization accuracy information, and follow-up time series information; S2: Based on multiple follow-up data of the same site, calculate the endoscopic-pathological sequence mismatch degree of the site; S3: Determine the reliability level of the current test result of the site based on the endoscopic-pathological sequence mismatch degree of the site; S4: In system-level risk assessment, when the current pathological examination result of the site is used as the risk assessment input, adjust the weight of the pathological examination result of the site according to the reliability level.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a risk recalibration method and system based on site-level endoscopic pathology mismatch. Background Technology

[0002] Gastric mucosal atrophy, intestinal metaplasia, and dysplasia are important stages in the development of gastric cancer. Current clinical management typically employs a combination of long-term endoscopic follow-up, histological stratification, and risk stratification. Existing research and guidelines generally emphasize the importance of high-quality endoscopic observation, fixed gastric site sampling, standardized biopsy pathways, and stratified follow-up, indicating that precancerous lesions of the stomach cannot be reliably evaluated based on a single examination result alone, but require comprehensive judgment based on multiple observations. Current techniques commonly use endoscopic findings and pathological results from each follow-up visit directly as input for risk scoring, or arrange re-examinations at fixed intervals. However, there is a lack of a unified quantitative processing mechanism for situations where, during continuous follow-up at the same fixed gastric site, "endoscopic abnormalities persist while pathological abnormalities remain low," "previous abnormal records are followed by subsequent endoscopic findings showing lower levels," or "insufficient previous sampling or positioning errors lead to decreased reliability of current results." Especially at fixed gastric sites such as the lesser curvature of the antrum, greater curvature of the antrum, angular notch, lesser curvature of the body, and greater curvature of the body, if a single biopsy results in insufficient sample quantity, localization deviation, or insufficient observation of key areas, the results of the current site may not fully reflect the true state of that site, thereby further affecting the robustness of the system-level risk scoring results.

[0003] Current research on the management of precancerous gastric lesions indicates that lesions such as chronic atrophic gastritis, intestinal metaplasia, and dysplasia exhibit long-term evolutionary characteristics, and the results of a single examination are insufficient to consistently reflect the true state of various fixed sites in the gastric mucosa. Existing guidelines and reviews generally emphasize that high-quality endoscopic observation, sampling at fixed gastric sites, standardized biopsy pathways, and stratified follow-up management are fundamental conditions for improving the identification ability and risk stratification stability of precancerous gastric lesions. In particular, at fixed gastric sites such as the lesser curvature of the antrum, greater curvature of the antrum, angular notch, lesser curvature of the body, and greater curvature of the body, the use of standardized sampling pathways combined with endoscopic grading information can provide a more stable data foundation for subsequent pathological analysis and risk stratification.

[0004] Related studies have shown that techniques such as the Sydney Protocol, targeted biopsy, and virtual chromoendoscopy all demonstrate that the observation and sampling of precancerous lesions of the stomach cannot be accomplished simply by "sampling at any location once," but requires standardized processing based on fixed gastric sites, lesion distribution, and endoscopic findings. Therefore, site-level observation and site-level sampling have a clear clinical and technical basis, providing a practical foundation for this invention to construct a historical sequence analysis framework at the fixed gastric site level.

[0005] (References: 1. MAPS III guidelines: Emphasizing the importance of high-quality endoscopy, standardized gastric biopsy sampling, OLGA / OLGIM stratification and stratified follow-up in the management of precancerous lesions of the stomach.)

[0006] 2. The review article "Endoscopic grading and sampling of gastric precancerous lesions" emphasizes the fundamental role of endoscopic grading, fixed-site sampling, and standardized sampling pathways in the identification and stratification of gastric precancerous lesions.

[0007] 3. Related research on the Updated Sydney System: This demonstrates that fixed gastric site sampling and standardized biopsy pathways have practical value in identifying precancerous lesions of the stomach.

[0008] 4. A comparative study of targeted biopsy and the Sydney protocol: demonstrating that different sampling routes affect the detection rate and stratification results of lesions such as GIM. Furthermore, existing risk scoring methods typically assume that all input results at each site have the same reliability, failing to control for reliability differences at the site level. When long-term endoscopic-pathological inconsistencies exist at certain sites, directly incorporating their current low-abnormality results into the systemic risk score with conventional weights can easily lead to an underestimation of overall risk. On the other hand, existing review sampling or focused review often relies on experience-triggered methods, lacking an information processing mechanism that can simultaneously combine long-term inconsistency information at the site level with systemic risk status, making it difficult to prioritize limited review resources. Therefore, it is necessary to provide an information processing method for follow-up data at fixed gastric sites. This method quantifies the degree of endoscopic-pathological mismatch at the same site across time series, generates a site-level detection result reliability level, and uses this level as an input control parameter for the systemic risk score. Simultaneously, it combines the systemic risk status to generate review sampling trigger information. After obtaining new results, the site-level mismatch and systemic risk score are updated in reverse, thereby improving the robustness of the risk scoring results and the relevance of the review process.

[0009] Although existing technologies have established management pathways for precancerous gastric lesions based on endoscopic observation, histological results, and stratified management, inconsistencies may still occur between endoscopic findings and pathological results at fixed gastric sites during long-term follow-up. These inconsistencies may stem from the spatial heterogeneity of the lesions themselves, or be related to sampling location bias, insufficient sample quantity, inadequate focused observation, or fluctuations in observation quality. Existing research has indicated that while there is a correlation between endoscopic findings and pathological results, they are not always completely consistent, and a single site result may not fully reflect the true degree of abnormality at that site.

[0010] On the other hand, existing studies have used endoscopic findings and histological results to predict the risk of gastric cancer, indicating that incorporating "endoscopic information + pathological information" into systematic risk scoring has a realistic basis. However, current risk models typically assume that the input results at each point have similar reliability, and lack a mechanism for quantitatively controlling "long-term inconsistencies at the site level." In other words, while current technology can use endoscopic and pathological information for risk scoring, a unified information processing method is still lacking for questions such as "whether the current results are sufficiently reliable, whether they should be downgraded, and whether they should be prioritized in the review and sampling process" for certain fixed gastric sites.

[0011] Therefore, based on existing technologies, it is necessary to further establish a quantitative mechanism for fixed gastric site historical sequences, which can transform the long-term inconsistency between endoscopic findings and pathological results into a calculable mismatch index. This index can then be used for site-level input confidence control, system-level risk recalibration, and review sampling gating, thereby reducing the interference of a single low-confidence input on the overall risk score.

[0012] (Reference: 1. Arai 2022 study: It has been demonstrated that initial endoscopic findings and histological results can be used for individualized prediction of gastric cancer risk.)

[0013] 2. Endoscopic findings and pathology correlation studies: These studies demonstrate a correlation between endoscopic findings and pathological results, but still leave room for inconsistency.

[0014] 3. Studies on the relationship between sampling and biopsy of precancerous lesions of the stomach: These studies indicate that the results of a single sampling attempt may be affected by the adequacy of the sample, localization bias, and observation quality. Summary of the Invention

[0015] Based on existing research confirming the practical basis of fixed gastric site observation, standardized sampling, and combined endoscopic-histological analysis, this invention does not establish a new risk scoring model that deviates from existing management pathways. Instead, it further addresses the insufficiently addressed technical problem of "site-level input reliability discrepancies" on top of existing pathways. Specifically, this invention first collects endoscopic abnormality scores, pathological results, sampling adequacy, localization accuracy, and time-series information from continuous follow-up at fixed gastric sites. Then, it constructs an endoscopic-pathological sequence mismatch degree to address persistent inconsistencies at the same site across time series. This mismatch degree is then used as a site-level input reliability control parameter to weight the site inputs in the system-level risk scoring model. In cases of high mismatch, it generates a review sampling trigger information. After obtaining new results, these are then used to update the site-level mismatch degree and the system-level risk score.

[0016] Therefore, the technical problem this invention aims to solve is not only "how to perform risk scoring," but also "how to control input reliability under long-term inconsistency at a fixed gastric site, and thereby achieve risk recalibration and review resource allocation." This technical approach can transform the inconsistencies originally scattered in endoscopic observation, pathological results, and sampling quality into a computable, updatable, and gated input control mechanism.

[0017] (Reference 1. Existing guidelines and reviews have demonstrated that site-level observation and standardized sampling are the practical basis for the management of precancerous lesions of the stomach.)

[0018] 2. Arai 2022 has demonstrated that endoscopic and histological information can be incorporated into risk prediction models.

[0019] 3. Relevant correlation studies have demonstrated that endoscopic findings and pathological results are not consistently consistent, thus leaving room for further technical quantification of "site-level mismatches." The purpose of this invention is to provide a risk recalibration method and system based on site-level endoscopic pathology mismatch, in order to solve the technical problems in the prior art such as insufficient reliability assessment of site-level pathology test results, insufficient robustness of system-level risk assessment, lack of quantitative basis for verification sampling gating, and lack of reverse update closed-loop mechanism.

[0020] To achieve the above objectives, the present invention adopts the following technical solution: A risk recalibration method based on site-level endoscopic pathology mismatch includes the following steps: S1: Collect site-level data from multiple follow-up visits for a fixed gastric site in the patient. The site-level data includes endoscopic abnormality scores, pathological test results, information on adequacy of tissue sampling, information on accuracy of localization, and follow-up time series information. S2: Calculate the endoscopic-pathological sequence mismatch at the same site based on multiple follow-up data. S3: Determine the confidence level of the current detection result for the site based on the endoscopic-pathological sequence mismatch. S4: In system-level risk assessment, when the current pathological test result of the site is used as the risk assessment input, the weight of the pathological test result of the site is adjusted according to the reliability level; S5: Based on the endoscopy-pathology sequence mismatch and systemic risk status of the site, trigger the re-sampling or key re-examination gating of the site; S6: Based on the new results obtained from the review sampling or key review, update the endoscopic-pathological sequence mismatch, reliability level, and systemic risk score of the site in reverse.

[0021] Furthermore, the fixed gastric sites include at least three sites selected from the following: lesser antral curvature, greater antral curvature, angular notch, lesser body curvature, and greater body curvature.

[0022] Further, in step S2, the calculation of the endoscopy-pathology sequence mismatch considers at least three of the following factors: (1) the cumulative number of times the endoscopy abnormality score and the pathology detection level are inconsistent; (2) the duration or number of times the endoscopy continuously indicates abnormality while the pathology remains negative; (3) the number of times the historical high abnormality level is recorded and the subsequent endoscopy score is continuously underestimated; (4) the average or weighted value of the sufficiency score of the site in all previous samplings; (5) the average or weighted value of the localization accuracy score of the site in all previous samplings; and (6) the time decay factor or the time weighting factor.

[0023] Furthermore, in step S3, the reliability level is divided into at least three levels: high reliability, medium reliability, and low reliability. When the mismatch between the endoscopy and pathology sequence exceeds a preset threshold, the current pathology test result at that site is marked as low reliability.

[0024] Furthermore, in step S4, for sites with low reliability levels, the weight of their pathological test results in the system-level risk assessment is reduced; for sites with high reliability levels, the weight of their pathological test results in the system-level risk assessment is increased or maintained at the original weight.

[0025] Furthermore, step S4 also includes: when the reliability level of the site is low reliability, when the site is at a low confidence level and there are historical abnormality records or a persistently high endoscopic abnormality score, an alternative risk contribution value is introduced, which consists of the current endoscopic abnormality score, the number of historical abnormalities and the insufficient sampling penalty. The alternative risk contribution value consists of at least two of the following factors: (1) the current endoscopic abnormality score of the site; (2) the number of historical pathological positive detections of the site; (3) the insufficient sampling penalty of the site.

[0026] Furthermore, step S4 also includes a judgment of the overall risk score low confidence state: when at least one of the following conditions is met, the system enters the overall risk score low confidence state, and it is prohibited to directly lower the system-level risk based solely on the current low anomaly level detection result: (1) multiple sites are simultaneously in a high mismatch state, and the number exceeds a preset threshold; (2) key sites are in a high mismatch state and the site has been detected as positive in the past; (3) the overall sampling sufficiency score is lower than a preset threshold.

[0027] Further, in step S5, the triggering conditions for review sampling or key review include: (1) the endoscopic-pathological sequence mismatch of the site exceeds the first preset threshold; (2) the endoscopic-pathological sequence mismatch of the site exceeds the second preset threshold, and the system-level risk score is higher than the risk threshold; (3) the current endoscopic abnormality score of the site is high, but the sufficiency score of the sampling is lower than the sufficiency threshold.

[0028] Furthermore, in step S5, the priority of re-sampling for multiple sites is determined according to the following rules: the priority is positively correlated with the degree of mismatch between the endoscopy and pathology sequences of the site, positively correlated with the systemic risk status, and positively correlated with the number of historical pathological positive detections of the site.

[0029] Further, in step S6, the reverse update includes: (1) recalculating the endoscopic-pathological sequence mismatch degree of the site based on the new pathological test results of the re-examination sampling or key re-examination; (2) redetermining the reliability level of the site based on the new endoscopic-pathological sequence mismatch degree; (3) substituting the new reliability level and the new pathological test results into the system-level risk scoring model to update the system-level risk scoring results; (4) adjusting the subsequent re-examination time window and re-examination priority based on the updated system-level risk scoring results.

[0030] This invention also provides a risk recalibration method system based on site-level endoscopic pathological mismatch, including: a data acquisition module, a mismatch calculation module, a reliability assessment module, a risk recalibration module, a verification sampling gating module, and a reverse update module. Beneficial effects

[0031] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a site-level endoscopy-pathology sequence mismatch index, which can quantify the consistency between endoscopic observation and pathological detection results at the same site in multiple follow-ups, providing a quantitative basis for the reliability assessment of site-level detection results and avoiding treating all pathological results indiscriminately.

[0032] 2. This invention uses the site-level reliability level as a credibility control parameter for system-level risk assessment, downweights pathological results of low-reliability sites, and introduces alternative risk contribution values, which can improve the robustness of system-level risk assessment and reduce risk underestimation caused by insufficient sampling or location deviation in a single sampling.

[0033] 3. This invention establishes a quantitative gating mechanism for review sampling or key review based on mismatch degree and system-level risk status. Compared with the traditional model that relies on doctors' subjective experience, it can more accurately identify high-risk sites that need to be reviewed and sampled, thereby improving the targeting and efficiency of review sampling.

[0034] 4. This invention establishes a reverse update closed-loop mechanism, which uses the new results obtained from the review sampling or key review to update the site mismatch, reliability level and system-level risk assessment, so that the system can be continuously optimized during long-term follow-up and avoid repeated and inefficient sampling patterns.

[0035] 5. The technical solution of this invention focuses on data processing, reliability assessment, weight adjustment and resource allocation optimization. This invention focuses on site-level detection result credibility control, system-level risk score recalibration and review resource allocation, and is not intended to obtain disease diagnosis conclusions or treatment plans, thus meeting the requirements for patent protection subject matter.

[0036] 6. The coupling mechanism of this invention organically combines the four links of mismatch degree, confidence weighting, verification sampling gating, and reverse update to form an inseparable closed-loop system, which has a significant synergistic effect compared with using a single link.

[0037] 7. "Mismatch" is not only a measure of historical consistency, but also a risk amplifier for "occult lesions." This design transforms inconsistencies, originally considered "interference data" or "physician's experience judgment," into quantifiable "secondary verification gating," thereby significantly improving the detection efficiency of precancerous lesions of the stomach through algorithmic optimization without increasing medical resource investment. Comparison: Compared to conventional schemes that rely solely on a single OLGA / OLGIM staging, this scheme adds a time-series consistency dimension, enabling earlier identification of high-risk sites that are characteristic under endoscopy but "off-target" during biopsy. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the overall process of the method of the present invention; Figure 2 A schematic diagram of the site-level mismatch calculation process; Figure 3 This is a schematic diagram illustrating the linkage between risk recalibration and verification sampling gating. Figure 4 This is a schematic diagram of the reverse update closed loop; Figure 5 This is a schematic diagram of the module structure of the system of the present invention. Detailed Implementation

[0039] The present invention will be further described in detail below with reference to specific embodiments.

[0040] Example 1: System Flow Example This embodiment provides a complete workflow for a risk recalibration method based on site-level endoscopic pathology mismatch.

[0041] Step S1: Data Acquisition For patients with precancerous gastric lesions undergoing follow-up, fixed gastric sites were selected for long-term observation. In this embodiment, the selected fixed gastric sites include: lesser curvature of the antrum, greater curvature of the antrum, angular notch, lesser curvature of the body, and greater curvature of the body, totaling five sites.

[0042] For each site, the following data were collected at each follow-up visit: (1) Endoscopic abnormality scoring: The degree of abnormality at this site is scored based on the observation results of white light endoscopy, magnifying endoscopy, or NBI. The scoring can use a numerical scale, such as 0-10, where 0 indicates completely normal and 10 indicates highly abnormal. The scoring criteria may include endoscopically observable features such as mucosal color, vascular morphology, and glandular opening morphology.

[0043] (2) Pathological examination results: Pathological diagnosis results after biopsy of the site, including normal, chronic inflammation, atrophy, intestinal metaplasia, low-grade dysplasia, high-grade dysplasia, etc.

[0044] (3) Information on adequacy of biopsy sample: Record information such as the quantity and depth of biopsy sample taken at this site, and score it. For example, adequacy of sample is scored as 3 points, partial adequacy as 2 points, inadequacy as 1 point, and no sample is scored as 0 points.

[0045] (4) Location accuracy information: Record the accuracy of endoscopic location of the site. It can be evaluated by image comparison, consistency of location description, etc., and is rated as high, medium and low.

[0046] (5) Follow-up time series information: Record the date of each follow-up visit to calculate the time interval and time decay factor.

[0047] Step S2: Calculate site-level endoscopic-pathological sequence mismatch. For each site, the endoscopy-pathology sequence mismatch was calculated based on historical data from multiple follow-ups.

[0048] The degree of mismatch is determined by a combination of the following factors: Factor 1: The cumulative number of discrepancies between endoscopic abnormality scores and pathological examination grades The endoscopic abnormality score is mapped to the pathological examination level. For example, an endoscopic score of 0-2 corresponds to normal pathology or mild inflammation, 3-5 corresponds to atrophy or intestinal metaplasia, 6-8 corresponds to low-grade dysplasia, and 9-10 corresponds to high-grade dysplasia or higher.

[0049] The number of times N_high the endoscopic score indicated a grade higher than the pathological examination grade, and the number of times N_low the endoscopic score indicated a grade lower than the pathological examination grade, were statistically analyzed during each follow-up.

[0050] Factor 2: The number of times endoscopy consistently showed abnormalities while pathology remained negative. The number of consecutive follow-ups with endoscopic abnormality scores consistently above the threshold (e.g., ≥6 points) but pathological test results consistently negative or low-grade (e.g., ≤intestinal metaplasia) is counted as N_persist.

[0051] Factor 3: Number of times a historical high-abnormality score was followed by a persistently low endoscopic score. If a site has been pathologically positive in the past (e.g., dysplasia), but the endoscopic abnormality score continues to be below the threshold during subsequent follow-ups, then the cumulative number of such occurrences is N_underestimate.

[0052] Factor 4: Weighted average of the sufficiency score Calculate the weighted average S_adequacy of the adequacy scores of each sampling site. The weights can be applied using time decay, meaning that the weight of recent follow-ups is higher than that of earlier follow-ups.

[0053] Factor 5: Weighted average of positioning accuracy scores Calculate the weighted average of the location accuracy scores for each location.

[0054] Factor 6: Time decay factor Historical data is assigned a time decay weight, with data closer to the current time having a higher weight.

[0055] Taking all the above factors into account, the mismatch can be calculated using a weighted summation or a comprehensive scoring method. An example calculation method is as follows: Mismatch_Score = w1 (N_high + N_low) + w2 N_persist + w3 N_underestimate + w4 (1 - S_adequacy) + w5 (1 - S_location) Among them, w1, w2, w3, w4, and w5 are weight coefficients, which can be determined based on clinical experience or data training. Currently, they are set as placeholders [w1], [w2], [w3], [w4], and [w5].

[0056] Step S3: Determine the site-level reliability level Based on the mismatch degree calculated in step S2, the reliability of the current pathological test results for this site is graded.

[0057] In this embodiment, the reliability level is divided into three levels: High reliability: Mismatch_Score < T1 Medium reliability: T1 ≤ Mismatch_Score < T2 Low reliability: Mismatch_Score ≥ T2 T1 and T2 are preset thresholds, currently set as placeholders [T1] and [T2].

[0058] Step S4: Risk credibility weighting and system-level risk recalibration In system-level risk assessment, pathological test results at each site are used as input. Traditional methods assign equal weight to pathological results at all sites. This invention assigns different weights to pathological results at different sites based on their reliability level.

[0059] Weighting adjustment rules: High-reliability sites: weighting coefficient of 1.0 or higher, e.g., [w_high] Medium-reliable sites: weighted by moderate values, e.g., [w_mid] Low reliability sites: weighting coefficients are reduced, for example, [w_low] For low-reliability sites, in addition to reducing the weight of their pathological results, a surrogate risk contribution value is introduced. The surrogate risk contribution value consists of the following components: (1) Current endoscopic abnormality score at this site; (2) The number of historical positive pathological findings at this site; (3) Penalty for insufficient sampling at this site.

[0060] Example of calculating alternative risk contribution value: Alt_Risk = alpha Endoscopy_Score + beta Positive_History_Count + gamma (1 - Adequacy_Score) Where alpha, beta, and gamma are coefficients, currently set as placeholders [alpha], [beta], and [gamma].

[0061] System-level risk scoring models can employ multi-factor scoring models or machine learning models, with inputs including weighted pathological results at each point, surrogate risk contribution values, patient basic information, OLGA / OLGIM staging, etc.

[0062] Overall risk score low confidence level assessment: The system determines the overall risk score to be in a low-confidence state when one of the following conditions is met: (1) The number of low-reliability sites is ≥ [N_threshold]; (2) Key sites (e.g., angular notches) are in a low reliability state and have a history of positive detection; (3) The average adequacy score of all sites < [Adequacy_threshold].

[0063] In a low-confidence state of the overall risk score, even if all current pathological test results are negative or low-level, the system does not allow the overall risk score to be significantly lowered directly, but rather remains in a state of alert.

[0064] Step S5: Verify sampling or focus on gating Based on the site mismatch and system-level risk status, trigger a review sampling or key review.

[0065] Triggering rules: Rule 1: If the site mismatch is greater than or equal to [T_trigger1], then the site will be directly resampled.

[0066] Rule 2: If the site mismatch is ≥ [T_trigger2] (T_trigger2 < T_trigger1) and the system-level risk score is ≥ [Risk_threshold], then a resampling of that site is triggered.

[0067] Rule 3: If the current endoscopic abnormality score at the site is ≥ [Endo_threshold], but the adequacy score is < [Adequacy_min], then a review sampling is triggered.

[0068] When multiple sites simultaneously meet the triggering condition, they are ordered according to priority: Priority = k1 Mismatch_Score + k2 Overall_Risk + k3 Positive_History_Count Where k1, k2, and k3 are coefficients, currently set as placeholders [k1], [k2], and [k3].

[0069] Prioritize sampling the sites with the highest Priority values.

[0070] Step S6: Reverse update closed loop After re-sampling or focused re-examination, new pathological test results are obtained. The system updates accordingly based on these new results. (1) Update site mismatch: The new endoscopic score and new pathological results were incorporated into the historical sequence of this locus, and the mismatch was recalculated.

[0071] If the new results show improved consistency with the endoscopy score, the mismatch decreases; if they remain inconsistent, the mismatch increases further.

[0072] (2) Update reliability level: The reliability level of the site is reassessed based on the updated mismatch.

[0073] (3) Update system-level risk assessment: The new reliability level and new pathological results are substituted into the system-level risk scoring model to recalculate the overall risk score.

[0074] (4) Adjust the time window and priority of subsequent review: Based on the updated overall risk score, the follow-up intervals and observation priorities will be adjusted.

[0075] Example 2: Parametric Example This embodiment provides a parameterized configuration scheme for parameter adjustment and optimization in practical applications.

[0076] Mismatch calculation parameters: w1 (weight of cumulative inconsistencies) = [w1_value] w2 (weight of consecutive mismatches) = [w2_value] w3 (underestimation of frequency weight) = [w3_value] w4 (sufficiency weight) = [w4_value] w5 (location accuracy weight) = [w5_value] Reliability level threshold: T1 (the boundary between high reliability and medium reliability) = [T1_value] T2 (the boundary between medium and low reliability) = [T2_value] Weighting coefficients: w_high (weight of highly reliable sites) = [w_high_value] w_mid (weight of reliable site) = [w_mid_value] w_low (weight of low-reliability sites) = [w_low_value] Substitution risk contribution coefficient: alpha (endoscopy rating coefficient) = [alpha_value] beta (historical positive coefficient) = [beta_value] gamma (insufficient material penalty coefficient) = [gamma_value] Overall low-confidence triggering parameters: N_threshold (threshold for the number of low-reliability sites) = [N_threshold_value] Adequacy_threshold (Overall sufficiency threshold) = [Adequacy_threshold_value] Verify the sampling gate threshold: T_trigger1 (high-priority mismatch threshold) = [T_trigger1_value] T_trigger2 (medium priority mismatch threshold) = [T_trigger2_value] Risk_threshold (system-level risk threshold) = [Risk_threshold_value] Endo_threshold (Endoscopic Abnormality Threshold) = [Endo_threshold_value] Adequacy_min (Minimum Sufficiency of Material) = [Adequacy_min_value] Priority calculation coefficient: k1 (mismatch coefficient) = [k1_value] k2 (Overall Risk Coefficient) = [k2_value] k3 (historical positive coefficient) = [k3_value] The above parameters can be optimized and adjusted based on actual clinical data. It is recommended to fine-tune the parameters through retrospective data validation or prospective cohort studies to achieve optimal review sampling efficiency and risk assessment accuracy.

[0077] Example 3: Implementation of the Effect Verification Framework This embodiment provides an effectiveness verification design framework for evaluating the effectiveness of the present invention. This framework does not contain fictitious experimental results; all result positions are reserved as placeholders for subsequent supplementation with actual data.

[0078] Validate the design: Validation Objective 1: Prove that the mismatch degree is predictive. We retrospectively collected fixed-site data from patients with precancerous lesions of the stomach. The sample size was [N_patients] patients, and a total of [N_sites] sites were observed. The follow-up period was ≥ [Follow-up months] months.

[0079] All loci were divided into a high mismatch group (mismatch ≥ [T_high]) and a low mismatch group (mismatch < [T_low]) based on the degree of mismatch.

[0080] The incidence of pathological escalation or abnormal detection in the two groups during subsequent follow-up was statistically analyzed: High mismatch group anomaly detection rate: [Detection_rate_high]% Low mismatch group anomaly detection rate: [Detection_rate_low]% Statistical difference: p = [p_value_1] If the subsequent anomaly detection rate in the high mismatch group is significantly higher than that in the low mismatch group, it proves that the degree of mismatch is predictive of subsequent risks.

[0081] Validation Objective 2: Demonstrate the effectiveness of the dual-gated verification sampling strategy. Patients were randomly divided into three groups: Group A: Dual-gated verification sampling (considering both mismatch and system-level risk) Group B: Single-gated verification sampling (considering only mismatch, not system-level risk) Group C: Random verification sampling (randomly selecting sites without considering mismatch) Compare the positive rates of the re-examination sampling in the three groups (i.e., the proportion of pathological upgrades or new abnormalities detected after re-examination sampling): Positive rate of group A retesting: [Positive_rate_A]% Positive rate of group B retesting: [Positive_rate_B]% Positive rate of retesting in Group C: [Positive_rate_C]% Statistical differences: p(A vs C) = [p_value_2], p(A vs B) = [p_value_3] If the positive rate of the retested samples in group A is significantly higher than that in groups B and C, it proves that the dual-gating strategy is superior to the single-gating and randomization strategies.

[0082] Validation Objective 3: Prove credibility-weighted reduction of false negative rate The risk assessment model is divided into two versions: Version 1: Risk assessment using credibility weighting (invention solution) Version 2: Risk assessment without credibility weighting (traditional approach) Within the same patient cohort, the false negative rates (i.e., the proportion of cases assessed as low risk by the system but subsequently progressing to pathological escalation) were compared between the two versions: Version 1 False Negative Rate: [FN_rate_weighted]% Version 2 False Negative Rate: [FN_rate_unweighted]% False negative rate reduction: [FN_reduction]% Statistical difference: p = [p_value_4] Validation objective 4: The results show that it outperforms the conventional strategy. This approach (which includes a complete closed loop of mismatch, confidence weighting, review sampling gating, and reverse update) is compared with the conventional review time window and review priority (which only follows up at fixed intervals recommended by the guidelines, without site-level mismatch analysis).

[0083] Comparison indicators: (1) Accuracy of risk prediction: Risk prediction AUC for this proposal: [AUC_proposed] Risk prediction AUC for conventional strategies: [AUC_conventional] (2) False negative compensation rate (the proportion of false negative cases recovered through resampling out of the total false negatives): The false negative compensation rate for this scheme is: [FN_compensation_proposed]% False negative compensation rate using conventional strategies: [FN_compensation_conventional]% (3) Risk underestimation rate in scenarios with insufficient material availability: The risk level of this plan is underestimated: [Underestimation_proposed]% Risk underestimation rate of traditional solutions: [Underestimation_conventional]% Validation Objective 5: Ablation experiments to verify the indivisibility of the coupling mechanism. Design ablation experiments, removing either the confidence weighting module or the verification sampling gating module, and observe the changes in system performance: Complete solution (baseline): False negative rate: [FN_baseline]% Verification sampling effectiveness: [Biopsy_efficiency_baseline]% Remove credibility weighting: False negative rate: [FN_no_weighting]% (expected to increase) Verification sampling efficiency: [Biopsy_efficiency_no_weighting]% Remove the verification sampling gating: False negative rate: [FN_no_gating]% (expected to increase) Verification sampling effectiveness rate: Cannot be calculated (because verification sampling was not triggered). Remove reverse updates: False negative rate in follow-up: [FN_no_update]% (expected to increase) If removing any module significantly reduces system performance (p < 0.05), it proves that the coupling mechanism between the modules is inseparable, and the ingenuity of this invention lies in the overall closed-loop design.

[0084] Validation Objective 6: Small-sample validation of mismatch prediction (preliminary example provided) This validation objective is a reserved implementation example, intended for subsequent supplementary small-sample retrospective validation data.

[0085] Validate the design: Collect fixed-site follow-up data from [N_small] patients, with a follow-up period of ≥ [T_followup] months.

[0086] The mismatch degree of each site was calculated and divided into three groups: high, medium and low, based on the mismatch degree.

[0087] At the end of the follow-up, the pathological progression or abnormality detection of each site was statistically analyzed.

[0088] Expected results: Subsequent anomaly detection rate for the high mismatch group: [Rate_high_mismatch]% Subsequent anomaly detection rate for the mid-mismatch group: [Rate_mid_mismatch]% Subsequent anomaly detection rate for the low mismatch group: [Rate_low_mismatch]% Trend test p-value: [p_trend] ROC curve analysis: The AUC for predicting subsequent anomalies based on mismatch: [AUC_mismatch] Sensitivity (with specificity set at 80%): [Sensitivity_mismatch]% Specificity (when sensitivity is set to 80%): [Specificity_mismatch]% This validation data will be supplemented and improved in subsequent studies.

[0089] Example 4: System Example This embodiment provides a risk recalibration method system based on site-level endoscopic pathology mismatch, including the following modules: Data acquisition module 41: This module is used to collect multiple follow-up data from fixed gastric sites in patients, including endoscopic abnormality scores, pathological examination results, adequacy of tissue sampling, accuracy of localization, and follow-up time. It can interface with hospital information systems, endoscopy reporting systems, and pathology reporting systems to automatically extract relevant data.

[0090] Mismatch Calculation Module 42: This module is used to calculate the mismatch between endoscopic and pathological sequences at the same locus. It implements the mismatch calculation algorithm described in step S2, taking historical data of the locus as input and outputting a mismatch score for that locus.

[0091] Reliability assessment module 43: This module is used to determine the reliability level of site pathology test results based on the degree of mismatch. It categorizes sites into three levels: high reliability, medium reliability, and low reliability, based on preset thresholds.

[0092] Risk recalibration module 44: This module consists of three sub-units: (1) Weighting adjustment unit: Adjust the weight of the pathological results of the site in the system-level risk assessment according to the site reliability level.

[0093] (2) Alternative Risk Calculation Unit: Calculates the alternative risk contribution value for low reliability sites.

[0094] (3) Low confidence state judgment unit: Determine whether the system has entered the low confidence state of the overall risk score, and limit the risk downgrade in the low confidence state.

[0095] Verification sampling gating module 45: Based on the site mismatch and system-level risk status, a review sampling or focused review is triggered. This module implements the triggering rules and priority ranking algorithm described in step S5, and outputs a list of sites requiring review sampling and their priorities.

[0096] Reverse update module 46: Based on the new results of the review sampling or key review, the site mismatch, reliability level, and system-level risk assessment are updated. This module implements the reverse update algorithm described in step S6, ensuring continuous optimization of the system during long-term follow-up.

[0097] System output module 299: The system displays information such as the mismatch rate, reliability level, system-level risk score, recommended sampling sites and priorities for review, and subsequent review time windows and priorities for each site. The user interface can use tables, charts, and other formats to intuitively display site-level and system-level information.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0099] Example 5: Predictive Validation Example of Site-Level Endoscopic-Pathological Sequence Mismatch This embodiment aims to verify the correlation between the site-level endoscopic-pathological sequence mismatch index and subsequent abnormality detection or pathological escalation, thereby demonstrating the technical rationale for using this index as a trigger parameter for review sampling. Existing guidelines and related studies have shown that fixed gastric site observation, standardized sampling pathways, and high-quality endoscopic observation are of great significance during the follow-up of precancerous gastric lesions. At the same time, inconsistencies may exist between endoscopic observation results and pathological results due to insufficient sampling adequacy, positioning deviations, or fluctuations in observation quality. Therefore, this embodiment introduces the endoscopic-pathological sequence mismatch index at the fixed gastric site level and verifies its relationship with subsequent abnormality detection or pathological escalation.

[0100] 1. Scoring Rules and Statistical Methods (1) Scoring Rules: Based on the model described in step S2, weighting coefficients w1-w3=1.5 and w4-w5=1.0 are set. Higher weights are assigned to factors directly reflecting long-term inconsistencies between endoscopy and pathology, while lower weights are assigned to factors related to sampling quality and localization quality, highlighting the contribution of historical inconsistency patterns to the mismatch. Statistical loci are based on endoscopy-pathology sequence data from the past 24 months, including the cumulative number of inconsistencies, historical abnormality level records, and sampling adequacy scores.

[0101] (2) Statistical methods: A retrospective cohort study was conducted to collect historical data from 425 sites in 85 follow-up patients. The chi-square test was used to compare the subsequent abnormality detection rate or pathological escalation rate of different mismatch groups; the receiver operating characteristic (ROC) curve was used to analyze the ability of the score to distinguish subsequent abnormality detection or pathological escalation.

[0102] 2. Threshold Determination Method Based on the comprehensive balance of subsequent anomaly detection rate, sensitivity, and specificity under different thresholds in the historical sample database, the grading threshold is determined with the principle of maximizing the Youden index.

[0103] (1) T1 = 4.0: This serves as the dividing point between high and medium confidence.

[0104] (2) T2 = 7.5: This is the dividing point between medium and low confidence. A value higher than this will trigger the risk recalibration process.

[0105] In addition to balancing the sensitivity and specificity of this sample library, the threshold can also refer to publicly available research findings on the stratified management of precancerous lesions of the stomach, standardized sampling, and differences in abnormal detection rates, so that high mismatch sites can more sensitively correspond to the risk of subsequent abnormal detection.

[0106] 3. Verification Results and Technical Effects (1) Group comparison results: Statistical analysis showed that the pathological escalation rate or abnormal detection rate in the subsequent review sampling of the high mismatch group (Mismatch_Score ≥ 7.5) was 32.4%, which was significantly higher than that of the low mismatch group (Mismatch_Score < 4.0) at 4.0% (p < 0.001).

[0107] (2) ROC analysis results: The AUC of mismatch in predicting subsequent abnormal detection or pathological escalation was 0.82, and the sensitivity reached 74.2% when the specificity was set at 85%.

[0108] (3) Technical effect expression: The results show that the subsequent abnormal detection rate of the high mismatch group is significantly higher, which supports the effectiveness of the site-level endoscopy-pathology sequence mismatch as a system review sampling trigger parameter or key review gating parameter. It can provide a quantitative basis for site-level detection result credibility assessment, review sampling priority allocation and subsequent risk recalibration.

[0109] Example 6: Demonstration of Risk Recalibration and Reverse Update Processing Based on Low-Confidence Sites This embodiment illustrates the roles of low-confidence site input weighting, the introduction of surrogate risk contributions, and result feedback updates in system-level risk recalibration. Existing research suggests that a single sampling result does not necessarily equate to the true state of the site, especially in long-term follow-up scenarios at fixed gastric sites. If there are persistently high endoscopic abnormality scores, previous abnormality level records, or insufficient sampling adequacy, the current low-abnormality level result may not be sufficient to directly represent the true degree of abnormality at that site. Therefore, in this embodiment, under long-term inconsistency at the site level, the current low-confidence input is not directly included in the system-level risk score with conventional weights. Instead, it first undergoes confidence weighting and introduces surrogate risk contribution values ​​based on historical abnormality level records and the current endoscopic abnormality score. After re-sampling, the site-level mismatch and system-level risk score are then updated in reverse using the new results.

[0110] (References for experimental design: 1. Endoscopic / pathological inconsistency and sampling error study: supports the possibility of bias in single site results.)

[0111] 2. Risk prediction studies such as Arai 2022: support the inclusion of endoscopic and histological information in systemic risk scoring.

[0112] 3. Relevant biopsy and standardized sampling studies: Support the rationale for prioritizing verification sampling in cases of high mismatch. This embodiment illustrates the processing flow of low-confidence site input weighting, introduction of alternative risk contributions, and reverse updating, serving as an example of how system-level risk recalibration is implemented. Existing research indicates that during follow-up of precancerous gastric lesions, the results of a single sampling may be affected by sampling location, sampling adequacy, and observation quality, leading to discrepancies between site-level results and the actual degree of abnormality. Therefore, when a site consistently exhibits a high mismatch, this embodiment does not directly incorporate its current low-abnormality level result into the system-level risk score with conventional weights. Instead, it first performs confidence weighting and introduces alternative risk contribution values ​​in conjunction with historical abnormality level records.

[0113] 1. Risk recalibration processing Select an example of site data: The current detection result of a certain site (corner notch) is at a low anomaly level, but its historical mismatch score is 8.2 (greater than the threshold T2), so it is marked as low confidence.

[0114] (1) Weight adjustment: The input weight of this site is reduced from 1.0 to 0.3.

[0115] (2) Calculation of surrogate risk: When a site is marked as low confidence and has a history of abnormal endoscopic findings or a record of historical abnormality levels, a surrogate risk contribution value, Alt_Risk, is introduced. Alt_Risk is calculated based on the site's current endoscopic abnormality score (8 points), the number of historical abnormality level records, and the undersampling penalty.

[0116] (3) Technical effect: The system-level risk score result was not directly downgraded based solely on the low confidence and low anomaly level input, thereby reducing the possibility of overall risk underestimation caused by a single low confidence input.

[0117] 2. Verification sampling and reverse update closed loop The system triggers a review sampling logic based on a high mismatch. After the review sampling is completed, the new detection result for this site shows low-grade dysplasia. Since the new detection result is consistent with the consistently high endoscopic abnormality score for this site, and the sufficiency of this sampling meets the preset threshold, the system determines that the consistency of this input has improved.

[0118] (1) Data feedback: The reverse update module will substitute the new test results, the current sampling adequacy score and the current endoscopy abnormality score into the historical sequence and recalculate the mismatch.

[0119] (2) State transition: After recalculation, the mismatch of this site is reduced to 3.5. Since the updated score is lower than the threshold T1, the input confidence level of this site changes from low confidence to high confidence, and the system-level risk score result is updated synchronously.

[0120] (3) Technical effect: The results show that the closed-loop mechanism of this scheme can correct the risk bias caused by the initial low confidence input through targeted review sampling, and dynamically adjust the subsequent review time window and review priority, thereby improving the robustness of the system-level risk scoring results.

[0121] All references and sources in this article: 1. MAPS III 2025 Guide Chinese description: International guidelines for the management of precancerous lesions of the stomach emphasize high-quality endoscopy, standardized gastric biopsy sampling, OLGA / OLGIM stratification, and stratified follow-up management.

[0122] Purpose: Used in the first half of the background technology to support the practical basis of "fixed gastric site observation + standardized sampling + stratified follow-up".

[0123] 2. Endoscopic grading and sampling of gastric precancerous lesions (2024 review) This review summarizes endoscopic grading and sampling strategies for precancerous lesions of the stomach, discussing fixed-site sampling, endoscopic grading, and EGGIM / OLGA / OLGIM.

[0124] Purpose: Used in the first half of the background technology section and at the beginning of Example 5 to support the rationality of site-level observation and site-level sampling.

[0125] 3. Arai 2022: Endoscopy + Histology for Predicting Gastric Cancer Risk Chinese explanation: This demonstrates that initial endoscopic findings and histological results can be incorporated into a personalized gastric cancer risk prediction model.

[0126] Purpose: Used in the latter half of the background technology section and the section connecting the invention content, to support the existing foundation of the "system-level risk scoring backbone".

[0127] 4. Implementation / Analysis of the Updated Sydney System Biopsy Protocol This article, based on the implementation and analysis of the standardized sampling pathway of the Updated Sydney System, illustrates the importance of a fixed gastric biopsy site for lesion identification.

[0128] Purpose: Used in the first half of the background technology to support the technical basis of "fixed gastric site + standardized sampling".

[0129] 5. Targeted biopsy alone vs Sydney protocol in GIM diagnosis Chinese description: This study compares the differences between targeted biopsy and the Sydney protocol in the identification of gastrointestinal metaplasia, illustrating that the sampling path affects the detection rate and stratification results.

[0130] Purpose: Used in the first half of the background section and at the beginning of Example 5 to support the statement that "sampling method affects site-level results".

[0131] 6. Studies on the correlation between endoscopic findings and pathology Chinese description: Analyze the correspondence and inconsistencies between endoscopic findings and pathological results.

[0132] Purpose: Used in the latter half of the background technology section, at the beginning of Example 5 and Example 6, to support the source of the problem and the technical rationale of "site-level mismatch".

Claims

1. A risk recalibration method based on site-level endoscopic pathology mismatch, characterized in that, Includes the following steps: S1: Collect site-level data from multiple follow-up visits for a fixed gastric site in the patient. The site-level data includes endoscopic abnormality scores, pathological test results, information on adequacy of tissue sampling, information on accuracy of localization, and follow-up time series information. S2: Calculate the endoscopic-pathological sequence mismatch at the same site based on multiple follow-up data. S3: Determine the confidence level of the current detection result for the site based on the endoscopic-pathological sequence mismatch. S4: In system-level risk assessment, when the current pathological test result of the site is used as the risk assessment input, the weight of the pathological test result of the site is adjusted according to the reliability level; S5: Based on the endoscopy-pathology sequence mismatch and systemic risk status of the site, trigger the re-sampling or key re-examination gating of the site; S6: Based on the new results obtained from the review sampling or key review, update the endoscopic-pathological sequence mismatch, reliability level, and systemic risk score of the site in reverse.

2. The risk recalibration method based on site-level endoscopic pathology mismatch according to claim 1, characterized in that, The fixed gastric sites include at least three sites among the lesser antral curvature, greater antral curvature, angular notch, lesser body curvature, and greater body curvature.

3. The risk recalibration method based on site-level endoscopic pathology mismatch according to claim 1, characterized in that, In step S2, the calculation of the endoscopy-pathology sequence mismatch considers at least three of the following factors: (1) The cumulative number of times the endoscopic abnormality score and the pathological examination grade are inconsistent; (2) The duration or frequency of continuous abnormalities reported by endoscopy but continuous negative pathology results; (3) The number of times a historical record of high abnormality level was followed by a persistently low endoscopic score; (4) The average or weighted average of the sufficiency scores for this site in all previous sampling attempts; (5) The average or weighted average of the location accuracy scores from all previous positioning tests; (6) Time decay factor or time weighting factor.

4. The risk recalibration method based on site-level endoscopic pathology mismatch according to claim 1, characterized in that, In step S3, the reliability level is divided into at least three levels: high reliability, medium reliability, and low reliability. When the mismatch between the endoscopy and pathology sequence exceeds a preset threshold, the current pathology test result at that site is marked as low reliability.

5. The risk recalibration method based on site-level endoscopic pathology mismatch according to claim 1, characterized in that, In step S4, for sites with low reliability, the weight of their pathological test results in the system-level risk assessment is reduced; for sites with high reliability, the weight of their pathological test results in the system-level risk assessment is increased or maintained. When the reliability level of a site is low, and when the site is at a low confidence level and has historical abnormality records or persistently high endoscopic abnormality scores, an alternative risk contribution value is introduced, consisting of the current endoscopic abnormality score, the number of historical abnormalities, and an undersampling penalty. The alternative risk contribution value consists of at least two of the following factors: (1) Current endoscopic abnormality score at this site; (2) The number of historical positive pathological findings at this site; (3) Penalty for insufficient sampling at this site.

6. The risk recalibration method based on site-level endoscopic pathology mismatch according to claim 1, characterized in that, Step S4 also includes determining the low-confidence state of the overall risk score: The system enters a low-confidence state for the overall risk score when at least one of the following conditions is met, and it is prohibited to directly downgrade the system-level risk based solely on the current low-anomaly detection result: (1) Multiple sites are simultaneously in a high mismatch state, and the number exceeds the preset threshold; (2) The key site is in a state of high mismatch and the site has been detected as positive in the past; (3) The overall material sufficiency score is lower than the preset threshold.

7. The risk recalibration method based on site-level endoscopic pathology mismatch according to claim 1, characterized in that, In step S5, the triggering conditions for review sampling or key review include: (1) The mismatch between the endoscopic and pathological sequences at the site exceeds the first preset threshold; (2) The endoscopic-pathological sequence mismatch at the site exceeds the second preset threshold, and the systemic risk score is higher than the risk threshold; (3) The site currently has a high endoscopic abnormality score, but the adequacy score is below the adequacy threshold; (4) Priority is positively correlated with the degree of mismatch between endoscopic and pathological sequences at the site, positively correlated with systemic risk status, and positively correlated with the number of historical pathological positive detections at the site.

8. The risk recalibration method based on site-level endoscopic pathology mismatch according to claim 1, characterized in that, In step S6, the reverse update includes: (1) Based on the results of new pathological examinations from the re-examination sampling or key re-examination, recalculate the endoscopic-pathological sequence mismatch at this site; (2) The reliability level of the site is re-determined based on the new endoscopy-pathology sequence mismatch. (3) Substitute the new reliability level and new pathological test results into the system-level risk scoring model to update the system-level risk scoring results; (4) Adjust the subsequent review time window and review priority based on the updated system-level risk score results.

9. A risk recalibration system based on site-level endoscopic pathology mismatch, characterized in that, include: The data acquisition module is used to collect site-level data from multiple follow-up visits at fixed gastric sites in patients; The mismatch calculation module is used to calculate the mismatch between endoscopic and pathological sequences at the same site. A reliability assessment module is used to determine the reliability level of the site pathology test results based on the mismatch degree. The risk recalibration module is used to adjust the weight of site pathology test results in system-level risk assessment according to the reliability level. The review sampling gating module is used to trigger review sampling or key review based on the mismatch degree and system-level risk status. The reverse update module is used to update the mismatch degree, reliability level, and system-level risk score results based on the new results of the review sampling or key review.

10. A risk recalibration system based on site-level endoscopic pathology mismatch according to claim 9, characterized in that, The risk recalibration module also includes: The weighting adjustment unit is used to adjust the weights of site pathology test results according to the reliability level. Alternative risk calculation unit, used to calculate the alternative risk contribution value for low reliability sites; The low-confidence state judgment unit is used to determine whether the system has entered a low-confidence state in the overall risk score.

Images