Cancer medical image based physical examination population experience map generation method and system

Abstract

The application discloses a medical examination crowd experience map generation method and system based on cancer medical images, and relates to the technical field of image recognition; its technical points are: obtaining the target organ medical image file, artificial intelligence reading output and patient end image marking disclosure log of the same examinee, forming image evidence fragments based on the target organ anatomical positioning benchmark; then combining the examination report conclusion and auxiliary image examination result to generate report receiving fragments, and determining the candidate area receiving state; when the patient end image marking is effectively disclosed, the disclosure conflict strength is calculated and the disclosure conflict event is generated; finally, the disclosure conflict event is input into the contact permission constraint graph to determine the main experience interpretation contact and the interpretation priority, and a medical examination crowd experience map that can be traced back is generated; the application can distinguish image marking disclosure advance, report description gap and cross-modal description gap, and reduce experience contact false generation.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to a method and system for generating experience maps of people undergoing physical examinations based on cancer medical images. Background Technology

[0002] In smart health checkup scenarios, medical imaging examination results are gradually shifting from simply receiving reports to the simultaneous presentation of intelligent image reading results, image labeling results, and results displayed on the examinee's end. For medical imaging examinations of target organs, intelligent image reading systems typically delineate abnormal candidate regions in the image. These abnormal candidate regions, at the image processing level, are merely candidate results that need to be further interpreted by combining image quality, regional boundary stability, examination report conclusions, and auxiliary imaging examination results. They do not necessarily correspond to the final medical conclusion. Examination reports usually need to combine the spatial location of the candidate region, boundary stability, adjacent image manifestations within the same target organ, auxiliary imaging examination results, and follow-up recommendations to form an interpretation. However, some health checkup platforms will display the image labels generated by intelligent image reading in a prominent manner to the examinee before the full interpretation of the report is displayed. This causes the examinee to see suspicious labels first, and then see the low-risk description or follow-up recommendations, thus creating a misunderstanding of the relationship between image labels and examination report conclusions.

[0003] Current intelligent medical image reading systems primarily target the detection of abnormal candidate regions. Experience maps for examinees typically generate report interpretation touchpoints based on report viewing records, consultation feedback, and satisfaction information. When a moderately confident abnormal candidate region exists in the medical image of a target organ, and the examination report has already interpreted this candidate region as low-risk for follow-up based on auxiliary imaging results, if the examinee's end is initially shown a prominent image marker, subsequent consultations, follow-up requests, and dissatisfaction feedback are easily categorized as ordinary report interpretation abnormalities by the experience map. This attribution method fails to identify the true relationship between experience abnormalities and image recognition evidence, the continuity of the examination report, the support of auxiliary imaging, and the intensity of image marker disclosure. It also fails to distinguish between physical examinations and other medical procedures. Anomalies in medical examinations can be caused by insufficient confidence in candidate regions, limited image quality, lack of supporting auxiliary images, insufficient acceptance of the examination report, or an intensity of image marker display exceeding the report's interpretive capacity. Therefore, a method for generating experience maps for physical examination populations based on medical image recognition results of target organs is needed. This method would enable the system to unify abnormal candidate regions, examination report conclusions, auxiliary image examination results, and image marker disclosure logs under the same candidate region primary key. Based on the acceptance relationship between image evidence fragments and report acceptance fragments, disclosure conflicts would be identified and converted into corresponding experience interpretation touchpoints. This would avoid generating experience map nodes solely based on examinee feedback or ordinary report viewing behavior. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for generating experience maps of physical examination populations based on cancer medical imaging includes:

[0006] Obtain medical image files of the target organ of the same subject, read the AI ​​image reading output and patient-side image labeling disclosure log associated with the same examination task, perform coordinate normalization on the target abnormal candidate region based on the target organ anatomical positioning benchmark determined by the effective region of the target organ, and perform corresponding verification within the same standardized organ coordinate domain with the actual labeled region displayed in the patient-side image labeling disclosure log, and generate image evidence fragments indexed by the primary key of the same candidate region and used to constrain the subsequent report acceptance verification.

[0007] Based on the primary key of the same candidate region, the inspection report conclusion and auxiliary image inspection results are retrieved. The description of the candidate region in the inspection report and the cross-modal support relationship of the auxiliary image inspection results for the candidate region are written together under the primary key of the same candidate region to generate the report receiving fragment and determine the receiving status of the candidate region.

[0008] When the patient-side image label disclosure log and candidate region pass the corresponding verification, and the image evidence fragments and report acceptance fragments do not have source-restricted status, pending review status, or insufficient boundary maintenance markings, the disclosure conflict intensity is calculated based on the image evidence fragments and report acceptance fragments. The source of conflict is determined based on the pressure contribution ratio in the disclosure conflict intensity calculation, the candidate region acceptance status, and the source distinction boundary, and a disclosure conflict event is generated.

[0009] The disclosure conflict event is input into the touchpoint permission constraint graph formed by historical review samples. The main experience interpretation touchpoint is selected and the interpretation priority is determined. The main experience interpretation touchpoint is written into the corresponding physical examination process node, and a physical examination population experience map that can be traced back to the image evidence fragment, report receiving fragment, and patient-end image tag disclosure log is generated.

[0010] Furthermore, the specific operation of coordinate normalization is as follows: taking the effective area of ​​the target organ in the target organ medical image file as the processing object, the first anatomical baseline and the second anatomical pointing axis are determined; based on the imaging position, the target abnormality candidate area in the AI ​​image reading output is converted to the standardized organ coordinate domain, so that the target abnormality candidate area uses the same candidate area primary key in report acceptance verification and disclosure conflict judgment.

[0011] Furthermore, the image evidence fragment generation operation is as follows: within the standardized organ coordinate domain, the normalized candidate region is overlap-checked with the actual marked region displayed in the patient-side image mark disclosure log, and the corresponding boundary of the overlap check is determined by the historical display review samples.

[0012] After the overlap verification is passed, the candidate credible information output by the artificial intelligence image reading is read. Based on the image perturbation copy whose perturbation range is determined by stable image samples, the consistency of regional overlap of the candidate region before and after the perturbation is verified. The display intensity of the patient end is normalized and converted according to the display level of the patient end and the display ratio of the candidate region. The candidate credible information, the regional overlap consistency verification result and the display intensity of the patient end are written into the same candidate region primary key to generate image evidence fragments that enter the report acceptance verification.

[0013] Furthermore, the specific operation describing the succession relationship is as follows: based on the primary key of the same candidate region, find the report statement corresponding to the candidate region in the inspection report conclusion; convert the target organ location expression in the report statement into a report location range, and perform same-side proximity verification with the candidate region. The proximity range of the same-side proximity verification is formed by the historical report succession review sample.

[0014] When the lesion terminology in the report statement covers the candidate image type output by the artificial intelligence image reading, the report statement is written into the report receiving element; when the location range fails the ipsilateral proximity check, a location receiving gap is written; when the lesion terminology does not cover the candidate image type, a terminology receiving gap is written, and the location receiving gap and terminology receiving gap are written into the report receiving element as report description gaps.

[0015] Furthermore, the specific operation of cross-modal support relationship is as follows: based on the primary key of the same candidate region, the auxiliary image examination results are converted to the target organ location range corresponding to the candidate region; within the target organ location range, auxiliary image abnormalities are first searched, and if no auxiliary image abnormalities are found, no abnormalities are searched.

[0016] When an abnormal description of an auxiliary image is found, it is written into the report receiving segment as cross-modal support information; when no abnormal description is found, and the examination report conclusion has already provided a receiving explanation for the difference between the no abnormal description and the candidate region, it is written into the report receiving segment as explained cross-modal difference; when no abnormal description is found but the examination report conclusion has not provided a receiving explanation, it is written into the report receiving segment as the basis for judging the unreceived cross-modal support gap; when the auxiliary image examination results cannot be converted to the target organ location range, the source-limited receiving status is written into the report receiving segment.

[0017] Furthermore, the calculation of disclosure conflict intensity is as follows: First, check whether the patient-side image tag disclosure log contains candidate region display records corresponding to the same candidate region primary key; if there are candidate region display records, and the image evidence fragments do not have source-restricted status, pending review status, or insufficient boundary maintenance markers, and the report acceptance fragments do not have source-restricted acceptance status, pending review acceptance status, or cross-modal source-restricted markers, then compare and convert the patient-side display intensity with the report acceptance gap converted from the report acceptance fragments under the same candidate region primary key, and output the disclosure advance pressure;

[0018] The cross-modal support gap obtained from the report acceptance fragment is compared and converted with the description missing result in the report acceptance fragment under the same candidate region primary key, and the cross-modal acceptance gap pressure is output.

[0019] When the candidate credibility information and the regional overlap consistency verification results meet the high credibility candidate conditions determined by the historical image review samples, the high credibility candidate is compared and converted with the results described in the inspection report, and the report omission pressure is output.

[0020] The intensity of disclosure conflict is calculated based on the pressure of premature disclosure, the pressure of cross-modal acceptance gaps, and the pressure of report omissions. When image evidence fragments are marked with a source-restricted status, a status pending review, or a boundary-insufficient mark, or when report acceptance fragments are marked with a source-restricted acceptance status, a status pending review, or a cross-modal source-restricted mark, a disclosure status pending review is output, and a high-credibility disclosure conflict event is not generated.

[0021] Furthermore, the process for determining the source of conflict is as follows: candidate regions whose disclosure conflict intensity reaches the disclosure conflict triggering conditions determined by historical review samples are taken as the objects to be attributed, and the acceptance status of candidate regions in the report acceptance fragments is read; according to the normalized contribution ratio corresponding to each pressure in the disclosure conflict intensity calculation, each pressure source is ranked.

[0022] If the pressure source at the top of the ranking and the candidate region acceptance status point to the same candidate region primary key, and the difference between the contribution ratio of the top-ranked source and the contribution ratio of the second-ranked source reaches the source distinction boundary determined by historical review samples, the pressure source at the top of the ranking will be written into the disclosure conflict event; if the difference does not reach the source distinction boundary, the disclosure conflict event will be written into the composite source status; if the composite source status does not meet the writing conditions, the disclosure conflict event will be written into the pending review status.

[0023] Furthermore, the selection process for the primary experience interpretation touchpoint is as follows: Based on historical review samples, an authorization relationship is established between the source of conflict and the nodes in the physical examination process. The source of conflict in the disclosed conflict event is entered into the authorization relationship for matching. After successful matching, among the physical examination process nodes allowed to accept the source of conflict, the acceptance conversion is performed on the nodes allowed to accept the source of conflict based on the intensity of the disclosed conflict, touchpoint authorization consistency, the contribution ratio of the primary source, and historical touchpoint review samples. Candidate touchpoints are ranked according to the acceptance conversion results. When the candidate touchpoint ranked first reaches the touchpoint acceptance boundary formed by historical touchpoint review samples, and the acceptance conversion difference between the first and second ranked touchpoints reaches the touchpoint differentiation boundary formed by historical touchpoint review samples, the node ranked first is designated as the primary experience interpretation touchpoint. If the matching fails, the touchpoint acceptance boundary is not reached, or the touchpoint differentiation boundary is not reached, the disclosed conflict event is written into the touchpoints to be reviewed.

[0024] Furthermore, the process of generating the physical examination population experience map is as follows: taking the main experience explanation touchpoints as the writing objects, sorting the main experience explanation touchpoints according to the intensity of disclosure conflict and the acceptance conversion result; writing the main experience explanation touchpoints into the corresponding physical examination process nodes according to the sorting result; saving the traceability relationship between image evidence fragments, report acceptance fragments, disclosure conflict events and main experience explanation touchpoints under the same candidate region primary key, and generating the physical examination population experience map;

[0025] If any traceability index in the image evidence fragment, report receiving fragment, disclosure conflict event, or patient-side image tag disclosure log cannot be closed, the high-confidence main experience interpretation touchpoint will not be written, and the corresponding disclosure conflict event will be written to the pending review touchpoint.

[0026] A system for generating experience maps of physical examination populations based on cancer medical imaging, including:

[0027] The image evidence fragment construction module acquires the target organ medical image file of the same subject, reads the AI ​​image reading output and patient-side image labeling disclosure log associated with the same examination task, performs coordinate normalization on the target abnormal candidate region based on the target organ anatomical positioning benchmark determined by the effective region of the target organ, and performs corresponding verification with the actual labeled region displayed in the patient-side image labeling disclosure log within the same standardized organ coordinate domain, generating image evidence fragments indexed by the primary key of the same candidate region and used to constrain the subsequent report verification.

[0028] The report receiving fragment construction module retrieves the inspection report conclusion and auxiliary image inspection results based on the primary key of the same candidate region. It writes the description of the candidate region in the inspection report and the cross-modal support relationship of the candidate region in the auxiliary image inspection results into the primary key of the same candidate region, generates the report receiving fragment, and determines the receiving status of the candidate region.

[0029] The disclosure conflict event generation module calculates the disclosure conflict intensity based on the image evidence fragments and report acceptance fragments when the patient-side image label disclosure log and candidate regions pass the corresponding verification, and the image evidence fragments and report acceptance fragments do not have source-restricted status, pending review status, or insufficient boundary maintenance markings. Based on the pressure contribution ratio in the disclosure conflict intensity calculation, the candidate region acceptance status, and the source distinction boundary, the conflict source is determined, and a disclosure conflict event is generated.

[0030] The Experience Map Touchpoint Writing Module inputs disclosure conflict events into a touchpoint permission constraint graph formed by historical review samples, selects the main experience interpretation touchpoint and determines the interpretation priority, writes the main experience interpretation touchpoint into the corresponding physical examination process node, and generates a physical examination population experience map that can be traced back to image evidence fragments, report receiving fragments and patient-side image tag disclosure logs.

[0031] This invention provides a method and system for generating experience maps of physical examination populations based on cancer medical images, which has the following beneficial effects:

[0032] 1. This invention first encapsulates the target organ medical image file, the AI ​​image reading output, and the actual display record on the patient's end using the same candidate region as the primary key. It then confirms whether the markers seen by the patient do indeed correspond to the same target abnormal candidate region through standardized organ coordinate domain, display region correspondence verification, and candidate boundary stability verification. As a result, the generation of the experience map no longer depends on the surface display results on the patient's end page, but is based on traceable image evidence boundaries, which can effectively eliminate the accidental triggering of experience touchpoints caused by factors such as coordinate mismatch, incorrect association of page markers, and unstable candidate regions.

[0033] 2. This invention writes the examination report conclusions and auxiliary imaging examination results into the same report acceptance fragment, and judges the description acceptance, cross-modal support, cross-modal difference explanation and report acceptance gap under the primary key of the same candidate region, and then forms the disclosure conflict intensity and conflict source. In this way, the system can distinguish the premature disclosure of image tags, insufficient report description, cross-modal explanation gap and explained low-risk differences, and avoid directly misjudging the situation of medium confidence artificial intelligence candidates, no abnormalities found in auxiliary images or examination reports that have been followed up and explained as lack of medical explanation.

[0034] 3. When medical examination institutions need to configure service touchpoints such as report viewing, post-discovery consultation, and doctor review, this invention will disclose a conflict event input touchpoint permission constraint diagram. Based on the source of the conflict, the acceptance conversion result, and the interpretation priority, the main experience interpretation touchpoint will be determined and written into the specific medical examination process node. The resulting medical examination population experience map can trace back from the process touchpoint to the image evidence fragment, report acceptance fragment, and patient-side disclosure log. This allows the institution to identify which medical interpretation acceptance link the problem occurred in and which process node should handle it first. For events with limited sources, insufficient samples, multiple sources, or incomplete tracing, the system will not directly write them into the high-confidence main interpretation node, thereby improving the reliability of medical examination interpretation resource allocation, review triage, and patient-side communication management. Attached Figure Description

[0035] Figure 1 This is a flowchart of the method of the present invention.

[0036] Figure 2 This is a system framework diagram of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0038] Example 1:

[0039] Please see Figure 1 This embodiment provides a method for generating an experience map of a physical examination population based on cancer medical imaging, including the following specific steps:

[0040] In this embodiment, the target organ medical image refers to a medical image formed around the same target organ during the physical examination process. The type of target organ is determined by the examination item code, imaging protocol, and report template version. The target organ should be an organ or anatomical region that can form an effective area of ​​the target organ in the medical image file and can establish an anatomical positioning benchmark for the target organ based on the imaging protocol, organ outline, lumen centerline, organ long axis, fixed anatomical side, or fixed anatomical reference point specified by the examination item code. The target organ may correspond to the lung, liver, thyroid, gastrointestinal tract, colorectal, prostate, uterus and adnexa, kidney, pancreas, bladder, or other cancer screening sites with a fixed image acquisition protocol and structured report template in the physical examination project. When the target organ cannot form an effective area of ​​the target organ, a first anatomical benchmark, a second anatomical pointing axis, or a positioning reference point, the system will not use the target organ medical image for the generation of high-confidence image evidence pieces, but will write it into a source-restricted state or a pending review state. The target abnormality candidate area refers to the candidate abnormal image area corresponding to the target organ in the intelligent image reading output. The auxiliary imaging examination result refers to the imaging examination result used to form a cross-modal description of the target organ candidate area within the same physical examination cycle.

[0041] This implementation method sequentially forms a physical examination population experience map according to S1 to S4. Specifically, S1 unifies the medical image files of the target organs of the same examinee and the same examination task, the AI ​​image reading output, and the patient-side image tag disclosure log under the same candidate region primary key to form an image evidence fragment; S2 retrieves the examination report conclusion and auxiliary imaging examination results based on the same candidate region primary key to form a report receiving fragment and determine the candidate region receiving status; S3, under the condition of effective disclosure of patient-side image tags, forms the disclosure conflict intensity, conflict source, and disclosure conflict event based on the image evidence fragment and the report receiving fragment; S4 inputs the disclosure conflict event into the touchpoint permission constraint diagram, selects the main experience interpretation touchpoint and writes it into the physical examination process node to generate a physical examination population experience map that can be traced back to the image evidence fragment, the report receiving fragment, and the patient-side image tag disclosure log.

[0042] The conventional image reading, page log recording, and data indexing in the above steps can be implemented using existing methods in the field. In this embodiment, they are only used as input sources for each fragment and event. For the extraction of candidate regions, text structure parsing, location range conversion, pressure item conversion, and touch point permission matching that affect image evidence fragments, report receiving fragments, disclosure conflict events, and main experience interpretation touch points, this embodiment provides at least one executable implementation method and makes the corresponding output enter the subsequent access, competition, gating, or tracing chain. The above processing does not change the examination report conclusion, nor does it output the lesion diagnosis conclusion.

[0043] Using the same candidate region primary key as the data boundary spanning S1 to S4, the same candidate region primary key is formed by at least the inspection task number, the examinee's physical examination number, the imaging file identifier, the organ position identifier, the imaging position, the artificial intelligence candidate region identifier, and the standardized candidate region. Image evidence fragments, report acceptance fragments, disclosure conflict events, and main experience interpretation touchpoints are all traced back using this primary key. Records from different examinees, different inspection tasks, different organ position identifiers, or different candidate regions must not be mixed under the same primary key for calculation. When the same primary key closure cannot be formed, the system only writes the source restricted status or the pending review status, and does not output the highly reliable acceptance result, disclosure conflict event, or main experience interpretation touchpoint.

[0044] In this implementation, the boundaries, weights, windows, number of layered samples, and triggering conditions are not written as preset values ​​without a source. The display corresponding boundary, candidate region extraction boundary, perturbation range, stability boundary, and display intensity weight in S1 are formed by historical image review samples, organ image display review sample library, and stable image sample library. The location proximity boundary, terminology mapping boundary, cross-modal support boundary, and report acceptance intensity weight in S2 are formed by report acceptance review sample library. The pressure conversion weight, disclosure conflict trigger boundary, source differentiation boundary, and minimum contribution boundary in S3 are formed by historical disclosure review samples. The touch point permission relationship, touch point acceptance boundary, touch point differentiation boundary, display acceptance boundary, report interpretation acceptance boundary, interpretation priority boundary, and conversion weight in S4 are formed by historical touch point review samples. Each sample library is layered according to dimensions related to its function, such as inspection items, display template version, report template version, candidate image type, location expression type, or physical examination process node.

[0045] When historical samples are insufficient, the system can use neighboring template records from the same institution, records of similar equipment or processes, or manually confirmed samples to form initial boundaries or initial weights. However, the corresponding results are written to the pending review status and no highly reliable results are output. After the corresponding stratified samples reach the minimum sample size requirement, the system updates the boundaries and weights based on the newly added review samples. The updated boundaries are used for subsequent processing and do not retroactively rewrite the conclusions of inspection reports that have already been generated and archived.

[0046] In this implementation, the high-reliability output refers to the result that can be entered into subsequent automatic acceptance, conflict event generation, main experience interpretation touchpoint determination, or experience map main node writing; the output to be reviewed refers to the result that retains the object index and source record, but is not used as the basis for automatic licensing. Image evidence fragments, report acceptance fragments, disclosure conflict events, and main experience interpretation touchpoints all save the sample layer version, boundary version, and weight version at the time of formation, so that subsequent reviews can trace the calibration basis used at that time. When the boundary or weight is updated later, the system only applies the updated parameters to the candidate areas newly entered into the processing chain after the update; the disclosure conflict events and experience map nodes that have already been formed are not automatically rewritten due to parameter updates, and can only be reviewed through the manual review entry.

[0047] The through sample in this embodiment is only used to illustrate how the primary key of the same candidate region flows from S1 to S4. The sample does not limit the actual value, nor does it represent the actual judgment of the nature of the lesion. Each calculation quantity has been converted to a dimensionless result of 0 to 1 before entering the weighting, sorting or triggering judgment. Millimeter distance, standardized position difference, text terminology coverage results and page display records are not directly added. All records that cannot complete the dimension unification or the normalization boundary cannot be formed are written to the pending review status.

[0048] The status triage rules apply uniformly throughout the text. The "Source-Restricted" status indicates that the input object cannot be closed to the primary key of the same candidate region, or that key inputs are missing, cannot be parsed, or cannot be converted to the same coordinate domain. The "Pending Review" status indicates that the input object can be identified, but there are insufficient historical samples, the candidate competition cannot be distinguished, and the boundary verification does not meet the high-confidence output conditions. The "Not Displayed" status indicates that the corresponding candidate region is not displayed in the patient-side image label disclosure log. The "Boundary Insufficient Maintenance" mark indicates that the candidate region cannot maintain regional consistency under stable perturbation conditions. Records with the "Source-Restricted," "Pending Review," "Not Displayed," or "Boundary Insufficient Maintenance" marks must not be converted into high-confidence report acceptance gaps, high-confidence disclosure conflict events, or single main experience interpretation touchpoints in subsequent steps. They can only enter the "Source-Restricted Acceptance," "Pending Review Acceptance," "Pending Review Disclosure," composite interpretation touchpoint, or manual review entry point.

[0049] S1 is used to unify the target abnormality candidate regions in the target organ medical image file, the AI ​​image reading output, and the patient-side image labeling disclosure log under the same candidate region primary key, forming image evidence fragments that enter the S2 report for verification. The target organ medical image file serves as the source of image evidence objects and the anatomical positioning benchmark for the target organ; the AI ​​image reading output serves as the source of target abnormality candidate regions and candidate credible information; the patient-side image labeling disclosure log serves as the source of the actual display area, display status, and display intensity; the historical image review sample, the organ image display review sample library, and the stable image sample library serve as the calibration function for the candidate region extraction boundary, the corresponding display boundary, the perturbation stability boundary, and the display intensity weight.

[0050] First, based on the examination task number, the examinee's physical examination number, the imaging file identifier, the organ location identifier, and the imaging position, the system performs task attribution verification on the target organ's medical image file, the AI-generated image reading output, and the patient's image labeling disclosure log. After the attribution verification is passed, the system reads the effective area of ​​the target organ in the target organ's medical image file, determines the first anatomical baseline and the second anatomical pointing axis, and establishes a standardized organ coordinate domain with the first anatomical baseline as the starting reference and the second anatomical pointing axis as the unified direction.

[0051] The effective region of the target organ can be obtained using existing target organ contour segmentation methods in the field. During segmentation, the image background, ruler characters, and non-target organ tissue display areas are excluded. In a preferred implementation, the system first removes the image background, ruler character areas, and non-target organ tissue display areas based on the grayscale distribution of the target organ medical image. Then, it performs connected component filtering on the remaining target organ tissue areas and takes the largest connected region of the target organ tissue as the effective region of the target organ. Subsequently, it extracts the continuous boundary near the fixed anatomical side corresponding to the examination item code in the effective region of the target organ as the first anatomical baseline. The second anatomical pointing axis is determined based on the protruding direction away from the first anatomical baseline, the direction of the lumen centerline, or the direction of the organ's long axis in the target organ contour. When the largest connected region, the first anatomical baseline, or the second anatomical pointing axis cannot be stably formed, the system does not use the segmentation result to form an unrestricted image evidence fragment, but writes it into a source-restricted state or a state pending review.

[0052] The system converts the target anomaly candidate regions in the AI ​​image reading output into standardized candidate regions. A common primary key for each candidate region is formed by the examination task number, imaging file identifier, organ location identifier, imaging position, AI candidate region identifier, and standardized candidate region. When the AI ​​image reading output includes candidate region outlines, bounding boxes, or pixel masks, the system directly reads the corresponding region representation. When the AI ​​image reading output only includes heatmaps or response maps, the system first normalizes the heatmap or response map to a range of 0 to 1, and then extracts the response connected regions based on the candidate response boundaries formed by the AI ​​image reading system output instructions or historical image review samples. Connected regions below the minimum area boundary, crossing the outer boundary of the effective target organ region, or not corresponding to the anomaly candidate type are excluded. Pixel masks or bounding boxes are formed for the retained connected regions, and these are used as target anomaly candidate regions in the standardized organ coordinate domain. When multiple connected regions cannot be distinguished by response intensity, area boundaries, or spatial separation relationships, the system writes a pending review status. When a candidate region representation cannot be formed, the system writes a source-restricted status and records the reason for the missing candidate region.

[0053] S1 also sets up candidate object admission and evidence display admission. Candidate object admission is used to confirm whether the AI ​​image reading output can form a clear target abnormal candidate region expression. Evidence display admission is used to confirm whether the actual display mark on the patient end is located in the same standardized organ coordinate domain as the candidate region and reaches the corresponding display boundary. The candidate region expression, display mark region, and stability perturbation result do not determine the diagnostic conclusion independently, but play the roles of candidate object formation, patient end disclosure proof, and candidate boundary stability proof, respectively. If there are multiple display marks on the same patient end page, the system does not directly select the marks according to the page display order, but converts them to the standardized organ coordinate domain and performs overlap verification. When multiple marks are close to the same candidate region and cannot be distinguished by the corresponding display boundary, the system writes a pending review display status.

[0054] The system reads the actual displayed marked area from the patient-side image marking disclosure log. This marked area is the image marking actually displayed to the examinee on the patient-side page, and is not directly replaced by the backend AI candidate area. The system converts the actual displayed marked area to the same standardized organ coordinate domain and performs overlap verification with the standardized candidate area. The overlap ratio is a dimensionless result of 0 to 1, which is used to determine whether the patient-side marked area can serve as valid disclosure evidence of the same candidate area. The corresponding boundary is formed by the organ image display review sample library. The sample library comes from the same institution and the same display template, and the review confirms that the patient-side marked area can correspond to the backend candidate area. The sample library records include at least the backend candidate area, the actual displayed marked area on the patient-side, the display template version, the page display position, and the review correspondence result. Preferably, there are no fewer than 20 manual confirmation records used to form the corresponding boundary for each display template version and page display position layer.

[0055] The system writes candidate confidence information around the primary key of the same candidate region. The candidate confidence information comes from the candidate confidence results given by the AI ​​image reading output for the target abnormal candidate region. The result is a dimensionless result with a value of 0 to 1. It is used to determine whether the image markers displayed on the patient end have a stable image recognition basis and is not used as a conclusion on lesion risk. When candidate confidence information is missing, the system writes a source-restricted state. When there are insufficient historical image review samples, the system writes a pending review confidence state. For the examinee, when the candidate confidence result given by the AI ​​image reading output is in the medium confidence interval, the candidate region can be included in the image evidence fragment, but the candidate confidence result must not trigger the report omission pressure in S3 on its own.

[0056] The system further verifies the boundary preservation of the candidate regions. Based on a stable image sample library, the system determines the perturbation range of the image perturbation copy. The stable image sample library consists of medical images of target organs that have been verified and confirmed to be stable under the same equipment, imaging position, and compression conditions. The perturbation copy can be processed by slightly adjusting the window width and window level, noise suppression, or local contrast normalization. The perturbation amplitude is limited by the allowable change boundary formed by similar processing records in the stable image sample library. The system compares the overlap consistency of the candidate regions before and after the perturbation and uses this consistency as the basis for the stability of the candidate boundaries. The overlap consistency is a dimensionless result of 0 to 1. The stable boundary is used to control whether the candidate region can be used as valid image evidence for the report acceptance verification. Preferably, for each equipment protocol, imaging position, and compression thickness layer, there are no fewer than 30 verified stable images used to form the perturbation range and stable boundary. If the minimum sample size requirement is not met, the system writes to the pending verification state and does not output a high-confidence boundary preservation result.

[0057] The system normalizes and converts the display intensity on the patient side. The display intensity is determined by the patient side display level and the candidate area display ratio, both of which are dimensionless normalized results ranging from 0 to 1. The patient side display level is determined by the patient side page embedding records, the display template version, and the page position of the image markers. The page position includes the first screen image preview area, the collapsed area, the pop-up prompt area, or the local magnified area. The candidate area display ratio is used to reflect the display scale of the marked area relative to the original image or the local magnified image. The conversion weights of the patient side display intensity are formed by the organ image display review sample library. Each weight is a non-negative dimensionless value and the sum of the weights is 1. The patient side display intensity serves as the input for the S3 disclosure advance pressure and is used to determine whether the patient side image marker display exceeds the interpretability of the examination report.

[0058] After pixel pitch conversion, the effective length of the target organ along the second anatomical pointing axis is 120mm. The center distance of the candidate region of the target abnormality output by the artificial intelligence image reading is 54mm from the first anatomical reference side. Therefore, the normalized position of the candidate region along the second anatomical pointing axis is 54 / 120=0.45. The lateral offset of the candidate region relative to the second anatomical pointing axis is 18mm. The effective width of the target organ at the same position is 90mm. Therefore, the absolute value of the lateral normalized offset is 18 / 90=0.2. The direction label relative to the second anatomical pointing axis is recorded.

[0059] After completing the aforementioned processing, the system generates image evidence fragments. Each image evidence fragment is indexed by the primary key of the same candidate region and contains standardized candidate regions, candidate credibility information, region overlap consistency verification results, boundary preservation status, patient-side display intensity, and status fields. The image evidence fragments do not output diagnostic conclusions but provide a unified object boundary for S2 to retrieve examination report conclusions and auxiliary imaging examination results. When the source of medical images of the target organ is missing, the AI ​​image reading output cannot be attributed to the same examination task, the target abnormal candidate region cannot complete standardized coordinate normalization, the patient-side image label disclosure log cannot be converted to the same standardized organ coordinate domain, the historical samples do not meet the minimum sample size requirement, or the perturbation copy cannot form a valid corresponding candidate region, the system writes a source-restricted status, a pending review status, or a boundary preservation insufficiency mark, and restricts its subsequent high-credibility output according to the shared abnormal diversion rules.

[0060] Step S2 uses the image evidence fragments output by S1 as the processing object. Based on the primary key of the same candidate region, it retrieves the examination report conclusion and the auxiliary imaging examination results. It writes the description and inheritance relationship of the examination report to the standardized candidate region and the cross-modal support relationship of the auxiliary imaging examination results to the standardized candidate region under the same primary key to generate report inheritance fragments and determine the candidate region inheritance status. S2 does not re-identify the target abnormal candidate region in the medical image of the target organ, nor does it modify the standardized candidate region formed by S1. The image evidence fragments serve as the boundary of the candidate region object and the admission of image evidence; the examination report conclusion serves as the source of description and inheritance; the auxiliary imaging examination results serve as the source of cross-modal support or cross-modal difference explanation; the report inheritance review sample library serves as the calibration function of location boundary, terminology mapping boundary, cross-modal support boundary, sentence distinction boundary and inheritance strength weight.

[0061] The system performs high-confidence report acceptance verification only on image evidence fragments that do not have source-restricted status, pending review status, or insufficient boundary markers and have formed standardized candidate regions. The system retrieves the examination report conclusions for the same subject and the same examination task based on the primary key of the same candidate region. The examination report conclusions can be structured report fields or examination report text. For structured report fields, the system reads the lateral location, location, lesion type, imaging findings, conclusion level, and follow-up recommendations. For natural language report text, the system performs controlled structured parsing based on the report template field, lesion location vocabulary, lesion terminology vocabulary, negative word rules, and follow-up recommendation field. This converts the existing location descriptions, lesion terms, negative descriptions, follow-up recommendations, and explanatory notes in the report text into report description fragments. In one preferred implementation, the system first segments the examination report text into candidate report statements according to the report paragraph title, punctuation marks, and report template field.

[0062] Subsequently, the system retrieves organ location identifiers, location expression terms, lesion terms, negation terms, and follow-up suggestion terms from each candidate report statement. Organ location identifiers are used to determine the left organ location, right organ location, upper organ location, lower organ location, or other organ locations defined by the examination item code. Location expression terms are used to extract location partitions, directional location expressions, distance from the positioning reference point, depth level, or inner / outer location. Lesion terms are used to determine whether they cover abnormal candidate types. Negation terms are used to determine whether negative descriptions such as "not seen," "not found," or "no clear indication" modify the same lesion term. Follow-up suggestion terms are used to determine whether the report forms a low-risk follow-up interpretation for the candidate region. If there are multiple lateral or location expressions in the same report statement, the system forms a candidate report description fragment based on the expression that is on the same side as the standardized candidate region and has the smallest location difference. If multiple expressions cannot be distinguished by the boundary of the same-side proximity checksum statement, the system writes a pending review status. This parsing does not add diagnostic conclusions not recorded in the examination report conclusion, nor does it use patient feedback or AI image reading output as a substitute for the examination report conclusion.

[0063] The report description unit includes at least the report statement index, organ location identifier, location expression, lesion terminology, conclusion level, follow-up recommendation, negative description, and explanatory notes. The system converts the target organ location expression in the examination report statement into a report location range. For quadrant expressions, the system determines the location reference point of the target organ based on the examination item code and report template version. Using the second anatomical pointing axis and a horizontal reference line that passes through the location reference point and intersects the second anatomical pointing axis, the standardized organ coordinate domain is divided into multiple location partitions, and the corresponding location partitions are used as the report location range. The location reference point can be determined by the geometric center of the target organ contour, the intersection of the lumen centerline, the endpoint of the organ's long axis, or a fixed anatomical reference point specified by the examination item code. When the location reference point cannot be stably formed, the system does not output a high-confidence location acceptance result, but writes it to the pending review acceptance status.

[0064] For directional location expressions, the system converts the directional expression into an angular range relative to the second anatomical pointing axis and combines it with the distance from the positioning reference point to form a corresponding sector range. When the distance from the positioning reference point is not recorded, the system only forms a directional range and does not directly serve as a high-confidence location basis for local candidate regions. For distance expressions of approximately 30 mm from the positioning reference point, the system converts the millimeter distance into a radial range in the standardized organ coordinate domain based on the pixel spacing in the imaging file and the effective length of the target organ formed by S1. Statements that only record scattered abnormal manifestations within the target organ without a clear local location are marked as generalized location expressions. The converted report location range is compared with the standardized candidate region using the same coordinate domain. Generalized location expressions can be written into the report receiving fragment as report background descriptions, but are not directly used as a high-confidence basis for local standardized candidate regions.

[0065] The report acceptance review sample library uses a common stratification but is used in a differentiated manner. The location proximity boundary is used to determine whether the report location expression can accept the standardized candidate region; the terminology mapping boundary is used to determine whether the lesion terminology in the examination report covers the artificial intelligence candidate image type; the statement differentiation boundary is used to select a unique accepting statement among multiple candidate report statements; the cross-modal support boundary is used to determine whether the description of auxiliary image abnormalities can serve as cross-modal support for the same candidate region; and the acceptance strength weight is used to form the report acceptance strength. All boundaries share the same stratified sample basis but do not replace each other, avoiding the misuse of the cross-modal support admission boundary for the report acceptance strength admission or the misuse of the location proximity boundary for the terminology coverage judgment.

[0066] The system performs contiguous proximity verification between the reported location range and the standardized candidate region, and verifies whether the lesion terminology in the report statement covers the candidate image type output by the artificial intelligence image reading. If the candidate image type of S1 is a target abnormality candidate type, the system checks whether the report statement contains standard terms, synonyms, or higher-level terms that can cover the target abnormality candidate type as confirmed by the report template terminology table. If the report statement only mentions local dense shadows or structural distortions without including terms that can cover the abnormality candidate type, the system writes a terminology transition gap. The location proximity boundary and terminology coverage relationship are formed by the report template terminology table, the report transition verification sample library, and the manually confirmed terminology mapping record. Before comparison, the location differences are uniformly converted to the standardized organ coordinate domain to avoid directly adding millimeter distances, quadrant ranges, and normalization ratios.

[0067] When multiple candidate report statements exist under the same candidate region primary key, the system calculates and sorts the report region acceptance score for each candidate report statement. The report region acceptance score is a dimensionless result ranging from 0 to 1, formed by the proximity of the ipsilateral location, the coverage of lesion terminology, and the consistency of the lateral orientation. When the examination report conclusion includes imaging position information, imaging position consistency can be used as an additional sub-item. When imaging position information is unavailable, this sub-item is not used as a positive bonus item, and the weights of the remaining available sub-items are normalized and adjusted. If the report region acceptance score of the first ranked statement reaches the report acceptance threshold, and the score difference between the first and second ranked statements reaches the statement distinction threshold, the system will use the first ranked report statement as a valid report description acceptance statement. If the difference does not reach the statement distinction threshold, it will be written to the pending review acceptance status to avoid multiple similar report statements matching the same standardized candidate region at the same time.

[0068] The system retrieves auxiliary imaging examination results based on the primary key of the same candidate region. The auxiliary imaging examination results should belong to the same examinee, the same physical examination cycle, or the same target organ imaging examination task. The system converts the organ location identifier, location partition, directional location expression, distance from the positioning reference point, depth level, abnormal description, and no abnormality description into the target organ location range corresponding to the standardized candidate region. If the auxiliary imaging examination results are missing, cannot be attributed to the same physical examination cycle or the same target organ imaging examination task, or only contain generalized descriptions and cannot form the location range corresponding to the candidate region, the system writes a cross-modal source restricted flag. This flag does not negate the inheritance of the already formed examination report description, but prohibits the formation of high-confidence cross-modal abnormality support and high-confidence cross-modal support gap.

[0069] During the formation of cross-modal support relationships, the system first searches for auxiliary image anomaly descriptions within the location range corresponding to the standardized candidate region. When anomalies such as abnormal echoes, density anomalies, signal anomalies, structural anomalies, or other anomaly descriptions consistent with the candidate region location are found, the system generates cross-modal anomaly support. Cross-modal anomaly support is a dimensionless result ranging from 0 to 1, formed jointly by the degree of location consistency, the degree of coverage of anomaly terms, and the clarity of auxiliary image descriptions. When multiple auxiliary image anomaly descriptions can correspond to the same candidate region, the system sorts them according to cross-modal anomaly support. When the top-ranked description reaches the cross-modal support admission boundary and the difference between it and the second-ranked description reaches the cross-modal distinction boundary, the system uses the top-ranked auxiliary image anomaly description as cross-modal support information; otherwise, it is written to the pending review status. If no auxiliary image anomaly description is found, the system continues to search for unseen anomaly descriptions corresponding to that location. If the inspection report has already stated that no corresponding anomaly was found in the auxiliary image, the system writes it as an explained cross-modal difference into the report acceptance fragment. If the inspection report does not state this difference, the system uses it as one of the criteria for judging the unaccepted status of cross-modal support gaps.

[0070] The system generates report acceptance strength and report acceptance gap. Report acceptance strength is a dimensionless result ranging from 0 to 1, used to characterize whether the inspection report has provided an interpretation and acceptance for the same candidate region. Its inputs include the report region acceptance score, whether the conclusion level or follow-up recommendation is clear, whether the limitation of image evidence is explained, and whether cross-modal differences are explained. Each item comes from image evidence fragments, inspection report conclusions, auxiliary imaging examination results, and the report acceptance review sample library. The weight of each item is a non-negative dimensionless value and the sum of the weights is 1. When the report acceptance strength is a valid value, the report acceptance gap can be converted to 1 minus the report acceptance strength. When the report acceptance fragment is in a limited acceptance state or a pending acceptance state, a high-confidence report acceptance gap is not output.

[0071] The standardized candidate region for the left organ formed by S1 is located at a normalized position of 0.45 and a horizontal normalized offset absolute value of 0.2. If the examination report only contains a generalized follow-up description and cannot pass the local location acceptance and abnormal candidate type term acceptance verification, the system will convert the report region acceptance score to 0.42. If this score does not reach the report acceptance threshold of 0.7, the report acceptance fragment will be written into the report description gap unaccepted state. If the report conclusion and follow-up recommendations are analyzable but fail to provide a sufficient explanation for the local candidate region, the report acceptance strength will be converted to 0.3, and the report acceptance gap will be 1-0.3=0.7. If the auxiliary imaging examination results at the same location show no abnormality, but the examination report does not explain this cross-modal difference, the system will use it as the basis for judging the cross-modal support gap unaccepted state. The above boundaries and scores are all from the corresponding stratified report acceptance review sample library and are not used as the basis for disease risk judgment.

[0072] The system ultimately generates report acceptance fragments, indexed by the primary key of the same candidate region. These fragments include at least the report region acceptance score, cross-modal anomaly support, report acceptance strength, report acceptance gap, cross-modal source-restricted marker, and candidate region acceptance status. Candidate region acceptance status includes: low-risk conclusion accepted, limited image evidence not accepted, cross-modal support gap not accepted, report description gap not accepted, source-restricted acceptance status, and pending review acceptance status. Report acceptance strength is used in S3 to suppress disclosure conflicts arising from candidate regions already explained and accepted by the reviewed report; report acceptance gap is used in S3 to construct disclosure advance pressure and report omission pressure; cross-modal anomaly support is used to distinguish between genuine cross-modal support and explained cross-modal differences; cross-modal source-restricted marker is used to limit the direct formation of high-confidence gaps from cross-modal support relationships; and candidate region acceptance status is used in S3 to determine the source of the disclosure conflict event.

[0073] When an inspection report lacks a conclusion, cannot be matched with the same inspection task, the report acceptance and review sample library does not meet the corresponding minimum sample size requirement, or the inspection report text cannot form a valid parsing result, the system writes a source-restricted acceptance status or a pending review acceptance status, and does not output a high-confidence report acceptance gap. When auxiliary imaging examination results are missing, cannot be attributed to the same physical examination cycle or the same target organ imaging examination task, or cannot be converted to the target organ location range corresponding to the candidate region, the system writes a cross-modal source-restricted flag. After entering S3, the above statuses cannot trigger a high-confidence disclosure conflict event on their own; after entering S4, they cannot directly form a main experience interpretation touchpoint.

[0074] Step S3 uses the image evidence fragments output by S1 and the report acceptance fragments output by S2 as inputs to perform patient-side image tag disclosure access verification, forming disclosure advance pressure, cross-modal acceptance gap pressure, and report omission pressure, calculating disclosure conflict intensity, and generating disclosure conflict events when conflict triggering conditions are met. S3 does not re-identify target abnormality candidate regions in the medical images of the target organ, nor does it re-parse the examination report conclusions. Instead, it judges whether the patient-side image tag disclosure exceeds the medical interpretation acceptance capacity within the same candidate region primary key range already formed by S1 and S2. The image evidence fragments serve as inputs for patient-side display source, candidate credibility information, and candidate boundary stability; the report acceptance fragments serve as inputs for examination report interpretation acceptance, cross-modal support or difference explanation, report acceptance gap, and candidate region acceptance status; and the historical disclosure review samples serve as calibrations for pressure conversion weights, disclosure conflict triggering boundaries, source differentiation boundaries, minimum contribution boundaries, and review status rollback conditions.

[0075] The system reads image evidence fragments, report receiving fragments, and patient-side image tag disclosure logs using the same candidate region primary key. If the patient-side image tag disclosure log does not contain a display record corresponding to the same candidate region primary key, or if S1 is written to a non-disclosure state, the system outputs a non-disclosure state and does not proceed to disclosure conflict intensity calculation. If the patient-side image tag disclosure log contains a valid display record, and neither the image evidence fragments nor the report receiving fragments are in a source-restricted state, a pending review state, or a boundary-insufficient mark, the system proceeds to stress formation and disclosure conflict intensity calculation; otherwise, it outputs a pending review disclosure state.

[0076] The three stressors play different roles in S3: disclosure advance stress is used to determine whether the image markers seen by the patient exceed the report interpretation acceptance level; cross-modal acceptance gap stress is used to determine whether the support or difference between the auxiliary imaging examination results and the target abnormality candidate region has been explained; and report omission stress is used to determine whether the high-confidence image candidate has not been accepted by the examination report description.

[0077] Disclosure overrun pressure is calculated by combining the intensity of patient-side presentation and the report acceptance gap. The system first treats both as dimensionless inputs from 0 to 1, weights them according to the non-negative weights formed by historical disclosure review samples, and then obtains P_i through monotonically normalized mapping. Cross-modal acceptance gap pressure is calculated by combining the degree of unexplained cross-modal differences and the degree of support for unstated cross-modal anomalies. When either of these sources is limited, a high-confidence Ki is not formed. Report omission pressure is formed by combining the high-confidence candidate admission results and the results of insufficient acceptance of examination report descriptions. When the high-confidence candidate conditions are not met but can be effectively judged, O_i is treated as a valid but not enabled result. When the high-confidence candidate conditions cannot be judged, it is written to the disclosure status pending review. The weights and mapping boundaries of each pressure item are formed by the same stratified historical disclosure review samples.

[0078] All three are positive pressure inputs for the intensity of disclosure conflicts, but they only enter high-confidence calculations when the corresponding input sources are valid, the sample boundaries are valid, and the state is unrestricted. The report acceptance strength and the effective cross-modal anomaly support are used as suppression terms to suppress the formation of disclosure conflicts in candidate regions that have been reported or cross-modal explanations accepted.

[0079] Under the same candidate region primary key, the system generates disclosure advance pressure P_i, cross-modal acceptance gap pressure K_i, and report omission pressure O_i. P_i originates from the patient-side display intensity of S1 and the report acceptance gap of S2, characterizing the degree of advance of the patient-side image marker display intensity relative to the degree of interpretation and acceptance in the examination report. It is a dimensionless result with a value between 0 and 1. K_i originates from the unaccepted state of the cross-modal support gap and the missing cross-modal difference description result of S2, characterizing whether the support or difference between the target abnormality candidate region and the auxiliary imaging examination results has been effectively described by the examination report. It has a value between 0 and... When a dimensionless result of 1 exists and there is a cross-modal source-restricted marker, K_i is not used as a high-confidence positive pressure input, but instead puts the candidate region into a pending review and disclosure state. O_i comes from the candidate confidence information of S1, the regional overlap consistency verification result, and the inspection report description acceptance result of S2. It is only enabled when the high-confidence candidate condition is met and the inspection report description acceptance is insufficient, and is a dimensionless result with a value of 0 to 1. When the high-confidence candidate condition can be effectively judged but is not met, O_i is treated as a valid but not enabled result. When the high-confidence candidate condition cannot be judged due to source restrictions or insufficient samples, the system writes it to the pending review and disclosure state.

[0080] The system simultaneously reads the report acceptance strength A_i and the effective cross-modal anomaly support U_i. A_i is derived from the report acceptance fragments in S2 and is a dimensionless result ranging from 0 to 1. It is used to suppress the formation of disclosure conflicts in candidate regions that have been interpreted and accepted by the inspection report. If the inspection report can be parsed and the report acceptance fragments are confirmed not to have formed effective acceptance, then A_i can be treated as a valid 0 value. If the inspection report is missing, cannot match the same inspection task, cannot form a report acceptance fragment, or the report acceptance fragment is in a source-restricted acceptance state, then A_i must not be substituted into the formula as a 0 value. U_i is derived from the report acceptance fragments in S2 and is a dimensionless result ranging from 0 to 1. It is only used as a suppression item when cross-modal anomaly support has been formed by the inspection report or auxiliary image report. When the cross-modal inspection is effective and it is confirmed that there is no anomaly support, U_i can be treated as a valid 0 value. When the cross-modal source is restricted, the sample is insufficient, or it is pending review, U_i must not be substituted into the formula as 0.

[0081] Before calculating the disclosure conflict intensity, the system defines the i-th standardized candidate region as the i-th target abnormality candidate region determined by the same candidate region primary key in S1. It belongs to the same examinee, the same examination task, the same organ position identifier, and the same patient-side disclosure evaluation window. The patient-side disclosure evaluation window is the current processing window from the first display of the candidate region mark in the patient-side image mark disclosure log to the time when the disclosure conflict intensity calculation is started in S3. Consultation records or review records formed after the window is exceeded are not used as real-time input of the current candidate region, but are only used for subsequent updates of historical disclosure review samples. All quantities with subscript i in the formula come from the image evidence fragments, report receiving fragments, and patient-side image mark disclosure logs under the same candidate region primary key, and must not be mixed across candidate regions, examinees, or examination tasks.

[0082] The disclosure conflict intensity is calculated using the following function: F_i = sigma(b0 + betaP×P_i + betaK×K_i + betaO×O_i - betaA×A_i - betaU×U_i); where F_i represents the disclosure conflict intensity of the i-th standardized candidate region, a dimensionless result ranging from 0 to 1; sigma represents the monotonic normalization function, used to map the dimensionless linear combination result to the range of 0 to 1, preferably the Sigmoid function; b0 is the dimensionless intercept term obtained from the calibration of historical disclosure review samples, which is not a dimensionless intercept term. For the observation input of the current candidate region, the preferred value range is -3 to 3; betaP, betaK, betaO, betaA and betaU represent the dimensionless weight parameters of disclosure advance pressure, cross-modal acceptance gap pressure, report omission pressure, report acceptance strength and effective cross-modal anomaly support, respectively. They are all obtained by training or calibration from historical disclosure review samples, and are preferably non-negative dimensionless parameters of 0 to 3. If a piecewise normalization function is used and the upper and lower limits of normalization are equal, the system will not perform the normalization calculation, but will write the corresponding candidate region into the disclosure pending review status.

[0083] Historical disclosure review samples include at least patient-side image-marked disclosure records, image evidence fragments, report receiving fragments, subject follow-up consultation records, manually confirmed disclosure conflict types, and manually confirmed review results regarding the existence of disclosure conflicts. Subject follow-up consultation records in the historical disclosure review samples are only used for historical sample annotation and boundary calibration, and are not used for real-time disclosure conflict intensity calculation in the current candidate region. Historical disclosure review samples under the same stratum are also used to calibrate pressure conversion weights, disclosure conflict trigger boundaries, source differentiation boundaries, and minimum contribution boundaries. Pressure conversion weights control the contribution of each pressure and inhibition term to the disclosure conflict intensity; disclosure conflict trigger boundaries control whether a disclosure conflict event is generated; source differentiation boundaries control whether a single primary conflict source is output; and minimum contribution boundaries exclude situations where the contribution of each pressure term is insufficient to support source judgment. Preferably, each stratum contains at least 30 manually confirmed samples used to form the above boundaries and function parameters. If the minimum sample size requirement is not met, the system writes a pending disclosure status and does not output high-confidence disclosure conflict events.

[0084] In the above-described implementation, the patient-side display strength is 0.72, the report acceptance gap is 0.7, and the system, after calibration and conversion using historical disclosure review samples, obtains the following: disclosure advance pressure P_i = 0.62; cross-modal acceptance gap pressure K_i = 0.4; report omission pressure O_i = 0.2; report acceptance strength A_i = 0.3; effective cross-modal anomaly support U_i = 0.1. The parameters formed by the historical disclosure review samples are b0 = -0.5, betaP = 1.2, betaK = 0.8, betaO = 0.7, betaA = 1, betaU = 0.6. The linear combination result is: -0.5 + 1.2 × 0.62 + 0.8 × 0.4 + 0.7 × 0.2 - 1 × 0.3 - 0.6 × 0.1 = 0.344.

[0085] After mapping using the Sigmoid function, F_i = 1 / (1 + exp(-0.344)), which is approximately 0.585. If the disclosure conflict trigger boundary formed by the historical disclosure review samples under this layer is 0.58, then the candidate region meets the disclosure conflict trigger condition, and the system generates a disclosure conflict event. If the trigger boundary is not reached, the system outputs a disclosed but not conflicted state, and the disclosure conflict trigger boundary is not used as a basis for lesion risk judgment.

[0086] After a disclosure conflict event is generated, the system determines the source of the conflict based on positive pressure terms, including betaP×P_i, betaK×K_i, and betaO×O_i. Inhibition terms A_i and U_i do not participate in the ranking of conflict source contribution ratios. In the above embodiment, the positive pressure terms are 0.744, 0.32, and 0.14, respectively, and the sum of the positive pressure terms is 1.204, corresponding to contribution ratios of approximately 0.618, 0.266, and 0.116. If the source distinction boundary is 0.12, the difference between the first and second contribution ratios is approximately 0.352, reaching the source distinction boundary. If the image evidence fragment confirms effective display on the patient's end, and the report receiving fragment confirms insufficient report receiving, the system writes the image marker corresponding to the disclosure advance pressure into the main conflict source field. If the difference between the first and second positions does not reach the source distinction boundary, the system writes a composite source status. If the contribution ratios of each source are all below the minimum contribution boundary, the system outputs a disclosure status pending review.

[0087] The system ultimately generates disclosure conflict events, which are indexed by the same candidate region primary key. These events include at least the disclosure conflict intensity, conflict triggering status, primary conflict source, subordinate conflict source, contribution ratio of each stress item, source differentiation result, source restricted flag, pending review flag, and composite source status. The disclosure conflict intensity is used by S4 to determine the order of experience interpretation touchpoints. The primary conflict source is used to determine the allowed write nodes in the physical examination process. Subordinate conflict sources are used to form composite interpretation touchpoints or pending review touchpoints. The source restricted flag and pending review flag are used to restrict the event from directly forming the primary experience interpretation touchpoint. The system does not generate high-confidence disclosure conflict events when patient-side display records are missing, two types of fragments cannot correspond to the same candidate region primary key, historical disclosure review samples are insufficient, positive stress items cannot form an effective contribution order, or the conflict source cannot form a consistency verification with image evidence fragments or report receiving fragments under the same candidate region primary key.

[0088] Step S4 uses the disclosure conflict event generated in S3 as the sole event input, inputs the touchpoint permission constraint diagram formed by historical review samples, selects the main experience interpretation touchpoint and determines the interpretation priority, writes the main experience interpretation touchpoint into the corresponding physical examination process node, and generates a physical examination population experience map that can be traced back to image evidence fragments, report receiving fragments and patient-end image tag disclosure logs. S4 does not recalculate the disclosure conflict intensity or reassess the lesion risk, but determines which physical examination process touchpoint should receive the disclosure conflict event within the scope of the disclosure conflict event, conflict source, source differentiation result and status mark already formed in S3.

[0089] The system reads the disclosure conflict events output by S3, as well as the image evidence fragment index, report acceptance fragment index, and patient-side image tag disclosure log index corresponding to the primary key of the same candidate region. The disclosure conflict events serve as the event objects for touchpoint selection; the image evidence fragments serve as the image source tracing and patient-side display verification; the report acceptance fragments serve as the report interpretation acceptance tracing and acceptance gap verification; the patient-side image tag disclosure log serves as the patient-side display behavior tracing; the historical touchpoint review samples serve as the calibration function for touchpoint permission relationships, touchpoint acceptance boundaries, touchpoint differentiation boundaries, display acceptance boundaries, report interpretation acceptance boundaries, interpretation priority boundaries, and conversion weights; the physical examination process node configuration serves as the experience map writing boundary. When a disclosure conflict event does not contain the primary key of the same candidate region, or cannot be traced back to the fragment index formed by S1 and S2, the system does not generate a main experience interpretation touchpoint, but writes the event to the touchpoint to be reviewed.

[0090] The touchpoint permission constraint graph does not directly determine the map writing result based on the intensity of disclosure conflict. The intensity of disclosure conflict only indicates the intensity of the event that needs to be explained and accepted. The main conflict source is used to determine the type of touchpoint that can be accepted. Touchpoint permission consistency is used to determine the degree of matching between candidate touchpoints and conflict sources. The touchpoint acceptance boundary and the touchpoint differentiation boundary jointly control whether a single main experience explanation touchpoint can be output. Therefore, the same disclosure conflict event must go through source permission, touchpoint access, acceptance competition, differentiation gating and tracing closure in sequence before it can be written into the main explanation node of the physical examination population experience map.

[0091] The system constructs a contact permission constraint graph, which is formed from historical review samples. Among them, the historical review samples used for calibration of contact permission relationships, contact acceptance boundaries, contact differentiation boundaries, and interpretation priority boundaries are denoted as historical contact review samples. In a preferred implementation, the system uses the conflict source type as the starting point of the permission relationship and the physical examination process node as the ending point of the permission relationship, and performs group statistics on the historical contact review samples.

[0092] For the same conflict source type and the same physical examination process node, the system counts the number of samples that can be manually confirmed to accept the conflict source, the number of samples that pass the review after acceptance, and the number of samples that fail to accept the conflict source. When the review pass rate of the node reaches the permission boundary formed by the corresponding layer, a permission edge from the conflict source to the physical examination process node is formed in the touchpoint permission constraint graph. Nodes that do not reach the permission boundary do not enter the candidate touchpoint set of the corresponding conflict source. When multiple nodes reach the permission boundary, they enter the acceptance conversion competition. When there are insufficient historical touchpoint review samples, the system can form an initial permission edge based on the manual review records, but the corresponding event only enters the touchpoint to be reviewed and does not output a highly reliable main experience explanation touchpoint.

[0093] The historical touchpoint review samples are derived from the historical disclosure conflict event database, the manual review system, patient-side display logs, physical examination process node configurations, and touchpoint handling records. They include at least the historical disclosure conflict events, manually confirmed conflict sources, manually confirmed adapted physical examination process nodes, touchpoint explanation results, review handling records, patient-side follow-up consultation records, and patient-side display template versions. Patient-side follow-up consultation records are only used for historical touchpoint compatibility, permission relationships, and priority boundary calibration, and are not used as real-time input for current disclosure conflict events.

[0094] The nodes in the touchpoint permission constraint graph are the physical examination process nodes that are allowed to accept disclosure conflict events. Preferred nodes include image quality interpretation touchpoints, report conclusion interpretation touchpoints, patient-side display control touchpoints, post-discovery consultation touchpoints, doctor review touchpoints, and manual review entry points. The edges in the touchpoint permission constraint graph define the permission acceptance relationship between the conflict source and the touchpoint. The system first performs admission screening of touchpoints based on the permission edges. Touchpoints that do not pass the permission edge matching do not enter the candidate touchpoint set, nor do they participate in the main experience interpretation touchpoint ranking. Touchpoints that pass the permission edge matching and meet the corresponding acceptance boundary enter the candidate touchpoint set and participate in the competition in the subsequent acceptance conversion.

[0095] The image quality interpretation touchpoint is used to address situations where image evidence is limited, boundaries are insufficiently maintained, or candidate credible information requires verification; the report conclusion interpretation touchpoint is used to address situations where there are gaps in report description, insufficient interpretation of low-risk conclusions, or report acceptance gaps reach the report interpretation acceptance boundary; the patient-side display control touchpoint is used to address situations where the patient-side display intensity reaches the display acceptance boundary and the examination report acceptance is insufficient, resulting in premature disclosure; the post-discovery consultation touchpoint is used to address situations where the patient-side report viewing node has occurred and the process configuration includes a post-discovery consultation node; the physician review touchpoint and manual review entry point are used to address situations where the source is limited, awaiting verification, multiple sources, insufficient historical samples, or medical interpretations are not suitable for automatic acceptance.

[0096] The acceptance boundary, report interpretation acceptance boundary, touch point acceptance boundary, touch point differentiation boundary, and interpretation priority boundary are all formed from historical touch point review samples under the same stratum. They are all dimensionless results from 0 to 1. The acceptance boundary is used to determine whether the display intensity of the patient-side image marker is sufficient to allow the patient-side display control touch point to enter the candidate touch point set. The report interpretation acceptance boundary is used to determine whether the report acceptance gap is sufficient to allow the report conclusion interpretation touch point to enter the candidate touch point set. The touch point acceptance boundary is used to determine whether a candidate touch point can become a primary experience interpretation touch point. The touch point differentiation boundary is used to determine whether the first-ranked candidate touch point and the second-ranked candidate touch point can be distinguished. The interpretation priority boundary is used to divide the primary experience interpretation touch point into a priority processing queue, a routine processing queue, or a low-priority review queue. The above boundaries are not used as a basis for lesion risk assessment, nor do they change the examination report conclusion.

[0097] During the touchpoint permission matching process, if the main source of conflict is the premature disclosure of image tags, the image evidence fragment confirms the effective display on the patient's end, the report acceptance fragment confirms insufficient report acceptance, and the display intensity on the patient's end reaches the display acceptance boundary, then the system allows the patient's end display control touchpoint to enter the candidate touchpoint set; if the report acceptance gap reaches the report interpretation acceptance boundary, then the report conclusion interpretation touchpoint is allowed to enter the candidate touchpoint set; if the patient's end image tag disclosure log shows that the patient's end report viewing node has occurred, or the physical examination process node configuration includes a post-discovery consultation node, then the post-discovery consultation touchpoint is allowed to enter the candidate touchpoint set.

[0098] If the primary source of conflict is a cross-modal support gap, and the candidate region's acceptance status is "cross-modal support gap not accepted," the system allows the report conclusion interpretation touchpoint, post-discovery consultation touchpoint, and doctor review touchpoint to enter the candidate touchpoint set. If the primary source of conflict is a report acceptance gap, and the candidate region's acceptance status is "report description gap not accepted," the system allows the report conclusion interpretation touchpoint and doctor review touchpoint to enter the candidate touchpoint set. When the disclosed conflict event has a source-restricted mark or a pending review mark, the system does not allow it to enter the high-confidence subjective experience interpretation touchpoint sorting, and only matches the doctor review touchpoint or manual review entry. When the disclosed conflict event has a composite source status, the system does not perform single subjective experience interpretation touchpoint sorting, but matches permitted touchpoints according to the primary conflict source and subordinate conflict sources to form composite interpretation touchpoints, and does not split them into multiple independent high-confidence subjective experience interpretation touchpoints.

[0099] The system performs acceptance conversion within the candidate touchpoint set. This conversion determines the suitability of each candidate touchpoint for accepting the current disclosure conflict event. Its inputs include disclosure conflict intensity, touchpoint permission consistency, main source contribution ratio, patient-side display intensity, report acceptance gap, historical touchpoint suitability, source-restricted marker, pending review marker, and historical touchpoint review samples. Touchpoint permission consistency is a dimensionless result ranging from 0 to 1, formed by the permission edge matching results in the touchpoint permission constraint graph; direct permission edges can be mapped to 1, conditionally... The permission edge can be mapped to 0.85, and the auxiliary acceptance permission edge can be mapped to 0.65. The actual value is formed by calibration of the corresponding layer's historical touch point review sample. The historical touch point adaptability is formed by conversion of the manually confirmed touch point acceptance results, touch point processing completion records and subsequent review results in the corresponding layer's historical touch point review sample. The value is a dimensionless result from 0 to 1. When the touch point permission consistency, historical touch point adaptability or acceptance conversion weight cannot form a valid value, the system will not substitute 0, but will write the disclosure conflict event into the touch point to be reviewed.

[0100] Historical touchpoint review samples are stratified according to screening items, display template versions, report template versions, conflict source types, and physical examination process nodes. Historical touchpoint review samples under the same stratum are used to form touchpoint permission relationships, touchpoint acceptance boundaries, touchpoint differentiation boundaries, interpretation priority boundaries, and acceptance conversion weights. Preferably, there are no fewer than 30 manually confirmed samples used to form the above relationships, boundaries, and weights under each stratum. If the minimum sample size requirement is not met, the system uses neighboring display template records or manually confirmed records from the same institution to form initial permission relationships, initial boundaries, and initial conversion weights. However, the corresponding events are written to the touchpoint to be reviewed, and no highly reliable main experience interpretation touchpoints are output. After the manual confirmed samples of the corresponding stratum reach the minimum sample size requirement, the system updates the above relationships, boundaries, and weights based on the newly added historical touchpoint review samples.

[0101] The disclosure conflict intensity is 0.585, with the primary source of conflict being the premature disclosure of image-marked sources. The contribution ratio of premature disclosure pressure to the positive pressure term is 0.618. The patient-side display intensity is 0.72, the report acceptance gap is 0.7, the display acceptance boundary formed by the corresponding stratified historical touchpoint review samples is 0.6, and the report interpretation acceptance boundary is 0.5. Therefore, the patient-side display control touchpoint, the report conclusion interpretation touchpoint, and the post-discovery consultation touchpoint are included in the candidate touchpoint set. In the acceptance conversion weight, the weights of touchpoint permission consistency, disclosure conflict intensity, primary source contribution ratio, and historical touchpoint fit are 0.35, 0, and 0, respectively. 25, 0.25, and 0.15 are all non-negative dimensionless values ​​with a weighted sum of 1. For the patient-side display control touchpoint, the touchpoint compliance is 1, and the historical touchpoint fit is 0.9. The conversion result is 1×0.35+0.585×0.25+0.618×0.25+0.9×0.15=0.786. For the post-discovery consultation touchpoint, the touchpoint compliance is 0.85, and the historical touchpoint fit is 0.68. The conversion result is 0.85×0.35+0.585×0.25+0.618×0.25+0.68×0.15=0.7.

[0102] For the report conclusion interpretation touchpoint, the touchpoint compliance is 0.65, the historical touchpoint fit is 0.55, and the conversion result is 0.65×0.35+0.585×0.25+0.618×0.25+0.55×0.15=0.611. If the touchpoint acceptance boundary is 0.68 and the touchpoint differentiation boundary is 0.08, then both the patient-side display control touchpoint and the post-discovery consultation touchpoint have reached the touchpoint acceptance boundary, and the difference between the two is approximately 0.086, reaching the touchpoint differentiation boundary. The system will then determine the patient-side display control touchpoint as the primary experience interpretation touchpoint.

[0103] The system further determines the interpretation priority, which is used to determine the display or processing order of the main experience interpretation touchpoint in the physical examination population experience map. Its inputs include disclosure conflict intensity, main experience interpretation touchpoint acceptance conversion result, patient-side display intensity, and report acceptance gap, all of which are dimensionless results from 0 to 1. The source restricted mark, pending review mark, and composite source status are used to restrict the event from entering the high-confidence single main experience interpretation touchpoint sorting.

[0104] If the priority conversion weights are 0.35, 0.3, 0.2, and 0.15 respectively, then the priority conversion result of the above-mentioned explanation throughout the embodiment is 0.585×0.35+0.786×0.3+0.72×0.2+0.7×0.15=0.69. If the priority processing boundary formed by the corresponding hierarchical historical touch point review sample is 0.65 and the regular processing boundary is 0.45, then the main experience explanation touch point is written into the priority processing queue. The explanation priority conversion weights and priority boundaries are only used for touch point sorting and explanation resource allocation in the physical examination population experience map, do not control touch point access, and are not used as a basis for disease risk judgment.

[0105] The system writes the main experience interpretation touchpoints into the corresponding physical examination process nodes. The physical examination process nodes are formed according to the examinee's physical examination process timeline, and include at least the imaging examination node, report generation node, report viewing node, patient-side image labeling display node, post-discovery consultation node, auxiliary image collaborative interpretation node, and doctor review node. The image quality interpretation touchpoint is written into the imaging examination node or report generation node; the report conclusion interpretation touchpoint is written into the report generation node or report viewing node; the patient-side display control touchpoint is written into the patient-side image labeling display node; the post-discovery consultation touchpoint is written into the post-discovery consultation node; the doctor review touchpoint is written into the doctor review node; the manual review entry is written into the manual review queue; and the composite interpretation touchpoint is written into the composite relationship field between multiple allowed nodes, without being split into multiple independent high-confidence main experience interpretation touchpoints.

[0106] The system synchronously saves traceability relationships, which include at least the primary key of the same candidate region, the index of disclosure conflict events, the index of image evidence fragments, the index of report acceptance fragments, the index of patient-side image tag disclosure logs, the main experience interpretation touchpoint, the corresponding physical examination process node, the interpretation priority, the source of the permission relationship, the touchpoint review status, and the writing time. The traceability relationship is used to reverse locate the source of image evidence, the source of report acceptance, and the source of patient-side display of disclosure conflict events during subsequent reviews. It is not used to rewrite the examination report conclusion. If any index is missing, the system will not write the high-confidence main experience interpretation touchpoint, but will write the event to the touchpoint to be reviewed.

[0107] The system ultimately generates a physical examination population experience map. This map uses the examinee's physical examination process as a timeline and examination process nodes as units. It writes the main experience interpretation touchpoints, interpretation priorities, and traceability relationships into the corresponding nodes. Each touchpoint in the physical examination population experience map can be traced back to image evidence fragments, report acceptance fragments, and patient-side image marker disclosure logs. The physical examination population experience map does not output diagnostic conclusions, does not change the content of the examination report, and does not directly replace medical interpretation acceptance results with patient satisfaction. Instead, it is used to display the mismatch between image marker disclosure and medical interpretation acceptance in the cancer medical imaging physical examination process, and its priority interpretation or review nodes. Undisclosed states do not enter the touchpoint permission constraint graph; restricted source disclosure states or pending review disclosure states are only written to pending review touchpoints or manual review entry points; composite source states are only written to composite interpretation touchpoints or pending review touchpoints. High-credibility physical examination population experience map touchpoints are not generated when touchpoint permission relationships do not match, historical touchpoint review samples are insufficient, touchpoint acceptance results do not reach boundaries, touchpoint differentiation results do not reach boundaries, or the traceability index cannot be closed.

[0108] In the engineering deployment at each step, the system can set the image evidence meta table, report acceptance meta table, disclosure conflict event table, and experience map touchpoint table as traceable record tables. Each record table stores the primary key of the same candidate region, formation time, input index, status field, sample stratification version used, and calculation result version. This record structure does not require a specific database format and can be implemented using a relational database, document database, or structured records within the inspection task. Its function is to ensure that during subsequent review, the image evidence, report acceptance, and patient-side display logs can be traced back from the experience map touchpoints, preventing the generation of experience interpretation conclusions solely based on patient feedback or AI image reading output.

[0109] This implementation, through the sequential processing of S1 to S4 described above, ensures that patient-side image tag disclosure, examination report interpretation acceptance, cross-modal description of auxiliary images, and physical examination process touchpoint writing are all constrained by the same candidate region primary key. Image evidence fragments address the question of whether the candidate region and the patient-side display are the same object; report acceptance fragments address the question of whether the examination report and auxiliary image examination results have already accepted the candidate region; disclosure conflict events address the question of whether a conflict has occurred between the disclosed image tag and the medical interpretation acceptance, and the source of the conflict; the physical examination population experience map addresses the question of which physical examination process touchpoint should accept and review the conflict. Each step retains abnormal states and traceability indexes, thus preventing events with limited sources, insufficient samples, mixed objects, or composite sources from being directly written into the high-confidence master interpretation node.

[0110] During system implementation, all the above status fields are saved as access control fields for subsequent steps, rather than just as log notes. When subsequent steps read a restricted or pending review status, they should perform corresponding routing. This saving method can ensure that object boundaries, parameter versions, and output permission conditions are continuously transmitted between S1 and S4.

[0111] Example 2:

[0112] Please see Figure 2 Based on Example 1, this embodiment also provides a physical examination population experience map generation system based on cancer medical images, including: an image evidence fragment construction module, which acquires the target organ medical image file of the same examinee, reads the artificial intelligence image reading output and patient-end image marking disclosure log associated with the same examination task, performs coordinate normalization on the target abnormal candidate area according to the target organ anatomical positioning benchmark determined by the effective area of ​​the target organ, and performs corresponding verification with the marked area actually displayed in the patient-end image marking disclosure log in the same standardized organ coordinate domain, and generates image evidence fragments indexed by the primary key of the same candidate area and used to constrain the subsequent report acceptance verification;

[0113] The report receiving fragment construction module retrieves the inspection report conclusion and auxiliary image inspection results based on the primary key of the same candidate region. It writes the description of the candidate region in the inspection report and the cross-modal support relationship of the candidate region in the auxiliary image inspection results into the primary key of the same candidate region, generates the report receiving fragment, and determines the receiving status of the candidate region.

[0114] The disclosure conflict event generation module calculates the disclosure conflict intensity based on the image evidence fragments and report acceptance fragments when the patient-side image label disclosure log and candidate regions pass the corresponding verification, and the image evidence fragments and report acceptance fragments do not have source-restricted status, pending review status, or insufficient boundary maintenance markings. Based on the pressure contribution ratio in the disclosure conflict intensity calculation, the candidate region acceptance status, and the source distinction boundary, the conflict source is determined, and a disclosure conflict event is generated.

[0115] The Experience Map Touchpoint Writing Module inputs disclosure conflict events into a touchpoint permission constraint graph formed by historical review samples, selects the main experience interpretation touchpoint and determines the interpretation priority, writes the main experience interpretation touchpoint into the corresponding physical examination process node, and generates a physical examination population experience map that can be traced back to image evidence fragments, report receiving fragments and patient-side image tag disclosure logs.

[0116] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0117] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0118] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for generating a physical examination population experience map based on cancer medical imaging, characterized in that, include: Obtain medical image files of the target organ of the same subject, read the AI ​​image reading output and patient-side image labeling disclosure log associated with the same examination task, perform coordinate normalization on the target abnormal candidate region based on the target organ anatomical positioning benchmark determined by the effective region of the target organ, and perform corresponding verification within the same standardized organ coordinate domain with the actual labeled region displayed in the patient-side image labeling disclosure log, and generate image evidence fragments indexed by the primary key of the same candidate region and used to constrain the subsequent report acceptance verification. Based on the primary key of the same candidate region, the inspection report conclusion and auxiliary image inspection results are retrieved. The description of the candidate region in the inspection report and the cross-modal support relationship of the auxiliary image inspection results for the candidate region are written together under the primary key of the same candidate region to generate the report receiving fragment and determine the receiving status of the candidate region. When the patient-side image label disclosure log and candidate region pass the corresponding verification, and the image evidence fragments and report acceptance fragments do not have source-restricted status, pending review status, or insufficient boundary maintenance markings, the disclosure conflict intensity is calculated based on the image evidence fragments and report acceptance fragments. The source of conflict is determined based on the pressure contribution ratio in the disclosure conflict intensity calculation, the candidate region acceptance status, and the source distinction boundary, and a disclosure conflict event is generated. The disclosure conflict event is input into the touchpoint permission constraint graph formed by historical review samples. The main experience interpretation touchpoint is selected and the interpretation priority is determined. The main experience interpretation touchpoint is written into the corresponding physical examination process node, and a physical examination population experience map that can be traced back to the image evidence fragment, report receiving fragment, and patient-end image tag disclosure log is generated.

2. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 1, characterized in that, The specific operation of coordinate normalization is as follows: taking the effective area of ​​the target organ in the target organ medical image file as the processing object, determining the first anatomical baseline and the second anatomical pointing axis; based on the imaging position, converting the target abnormality candidate area in the AI ​​image reading output to the standardized organ coordinate domain, so that the target abnormality candidate area uses the same candidate area primary key in report acceptance verification and disclosure conflict judgment.

3. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 1, characterized in that, The image evidence fragment generation operation is as follows: within the standardized organ coordinate domain, the normalized candidate region is overlap-checked with the actual marked region displayed in the patient-side image mark disclosure log, and the corresponding boundary of the overlap check is determined by the historical display review samples. After the overlap verification is passed, the candidate credible information output by the artificial intelligence image reading is read. Based on the image perturbation copy whose perturbation range is determined by stable image samples, the consistency of regional overlap of the candidate region before and after the perturbation is verified. The display intensity of the patient end is normalized and converted according to the display level of the patient end and the display ratio of the candidate region. The candidate credible information, the regional overlap consistency verification result and the display intensity of the patient end are written into the same candidate region primary key to generate image evidence fragments that enter the report acceptance verification.

4. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 1, characterized in that, The specific operation to describe the succession relationship is as follows: based on the primary key of the same candidate region, find the report statement corresponding to the candidate region in the inspection report conclusion; convert the target organ location expression in the report statement into the report location range, and perform same-side proximity verification with the candidate region. The proximity range of the same-side proximity verification is formed by the historical report succession review sample. When the lesion terminology in the report statement covers the candidate image type output by the artificial intelligence image reading, the report statement is written into the report receiving element; when the location range fails the ipsilateral proximity check, a location receiving gap is written; when the lesion terminology does not cover the candidate image type, a terminology receiving gap is written, and the location receiving gap and terminology receiving gap are written into the report receiving element as report description gaps.

5. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 1, characterized in that, The specific operation of cross-modal support relationship is as follows: based on the primary key of the same candidate region, the auxiliary image examination results are converted to the target organ location range corresponding to the candidate region; within the target organ location range, auxiliary image abnormalities are first searched, and if no auxiliary image abnormalities are found, no abnormalities are searched. When an abnormal description of an auxiliary image is found, it is written into the report receiving segment as cross-modal support information; when no abnormal description is found, and the examination report conclusion has already provided a receiving explanation for the difference between the no abnormal description and the candidate region, it is written into the report receiving segment as explained cross-modal difference; when no abnormal description is found but the examination report conclusion has not provided a receiving explanation, it is written into the report receiving segment as the basis for judging the unreceived cross-modal support gap; when the auxiliary image examination results cannot be converted to the target organ location range, the source-limited receiving status is written into the report receiving segment.

6. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 1, characterized in that, The calculation of disclosure conflict intensity is as follows: First, check whether the patient-side image tag disclosure log contains candidate region display records corresponding to the same candidate region primary key; if there are candidate region display records, and the image evidence fragments do not have source-restricted status, pending review status, or insufficient boundary maintenance marks, and the report acceptance fragments do not have source-restricted acceptance status, pending review acceptance status, or cross-modal source-restricted marks, then compare and convert the patient-side display intensity with the report acceptance gap converted from the report acceptance fragments under the same candidate region primary key, and output the disclosure advance pressure; The cross-modal support gap obtained from the report acceptance fragment is compared and converted with the description missing result in the report acceptance fragment under the same candidate region primary key, and the cross-modal acceptance gap pressure is output. When the candidate credibility information and the regional overlap consistency verification results meet the high credibility candidate conditions determined by the historical image review samples, the high credibility candidate is compared and converted with the results described in the inspection report, and the report omission pressure is output. The intensity of disclosure conflict is calculated based on the pressure of premature disclosure, the pressure of cross-modal acceptance gaps, and the pressure of report omissions. When image evidence fragments are marked with a source-restricted status, pending review status, or insufficient boundary status, or when report receiving fragments are marked with a source-restricted receiving status, pending review receiving status, or cross-modal source-restricted status, a pending review disclosure status is output, and no high-credibility disclosure conflict event is generated.

7. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 6, characterized in that, The procedure for determining the source of conflict is as follows: the candidate regions whose disclosure conflict intensity reaches the disclosure conflict triggering conditions determined by historical review samples are taken as the objects to be attributed, and the acceptance status of the candidate regions in the report acceptance fragments is read. Based on the normalized contribution ratio of each pressure in the calculation of the disclosed conflict intensity, the sources of pressure are ranked. If the pressure source at the top of the ranking and the candidate region acceptance status point to the same candidate region primary key, and the difference between the contribution ratio of the top-ranked source and the contribution ratio of the second-ranked source reaches the source distinction boundary determined by historical review samples, the pressure source at the top of the ranking will be written into the disclosure conflict event; if the difference does not reach the source distinction boundary, the disclosure conflict event will be written into the composite source status; if the composite source status does not meet the writing conditions, the disclosure conflict event will be written into the pending review status.

8. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 1, characterized in that, The selection process for the main experience interpretation touchpoint is as follows: Based on historical review samples, establish a permission relationship between the source of conflict and the physical examination process nodes. Input the source of conflict in the disclosed conflict event into the permission relationship for matching. After successful matching, among the physical examination process nodes allowed to accept the conflict source, perform acceptance conversion based on the intensity of the disclosed conflict, touchpoint permission consistency, the contribution ratio of the main source, and historical touchpoint review samples. Sort the candidate touchpoints according to the acceptance conversion results. When the candidate touchpoint ranked first reaches the touchpoint acceptance boundary formed by historical touchpoint review samples, and the acceptance conversion difference between the first and second ranked touchpoints reaches the touchpoint differentiation boundary formed by historical touchpoint review samples, the node ranked first is designated as the main experience interpretation touchpoint. If the match fails, the touchpoint acceptance boundary is not reached, or the touchpoint differentiation boundary is not reached, the disclosed conflict event is written into the touchpoints to be reviewed.

9. The method for generating a physical examination population experience map based on cancer medical imaging according to claim 8, characterized in that, The process of generating the experience map for the physical examination population is as follows: taking the main experience interpretation touchpoints as the writing objects, and sorting the interpretation priority of the main experience interpretation touchpoints according to the intensity of disclosure conflict and the conversion result of acceptance; Based on the sorting results, the main experience explanation touchpoints are written into the corresponding physical examination process nodes; Under the same candidate region primary key, save the traceability relationship between image evidence fragments, report receiving fragments, disclosed conflict events and main experience explanation touchpoints to generate a physical examination population experience map; If any traceability index in the image evidence fragment, report receiving fragment, disclosure conflict event, or patient-side image tag disclosure log cannot be closed, the high-confidence main experience interpretation touchpoint will not be written, and the corresponding disclosure conflict event will be written to the pending review touchpoint.

10. A system for generating a physical examination population experience map based on cancer medical imaging, using the method for generating a physical examination population experience map based on cancer medical imaging as described in any one of claims 1-9, characterized in that... include: The image evidence fragment construction module acquires the target organ medical image file of the same subject, reads the AI ​​image reading output and patient-side image labeling disclosure log associated with the same examination task, performs coordinate normalization on the target abnormal candidate region based on the target organ anatomical positioning benchmark determined by the effective region of the target organ, and performs corresponding verification with the actual labeled region displayed in the patient-side image labeling disclosure log within the same standardized organ coordinate domain, generating image evidence fragments indexed by the primary key of the same candidate region and used to constrain the subsequent report verification. The report receiving fragment construction module retrieves the inspection report conclusion and auxiliary image inspection results based on the primary key of the same candidate region. It writes the description of the candidate region in the inspection report and the cross-modal support relationship of the candidate region in the auxiliary image inspection results into the primary key of the same candidate region, generates the report receiving fragment, and determines the receiving status of the candidate region. The disclosure conflict event generation module calculates the disclosure conflict intensity based on the image evidence fragments and report acceptance fragments when the patient-side image label disclosure log and candidate regions pass the corresponding verification, and the image evidence fragments and report acceptance fragments do not have source-restricted status, pending review status, or insufficient boundary maintenance markings. Based on the pressure contribution ratio in the disclosure conflict intensity calculation, the candidate region acceptance status, and the source distinction boundary, the conflict source is determined, and a disclosure conflict event is generated. The Experience Map Touchpoint Writing Module inputs disclosure conflict events into a touchpoint permission constraint graph formed by historical review samples, selects the main experience interpretation touchpoint and determines the interpretation priority, writes the main experience interpretation touchpoint into the corresponding physical examination process node, and generates a physical examination population experience map that can be traced back to image evidence fragments, report receiving fragments and patient-side image tag disclosure logs.

Images