Method and system for selecting standard fetal section in early pregnancy

Abstract

The application relates to an early early pregnancy fetus standard section selection method and system, which indexes and filters invalid data of early early pregnancy fetus standard section data, generates a data set index report, performs TOP K iteration selection according to the data set index report, extracts ultrasound images for paired comparison and completes multi-dimensional scoring, filters in combination with multi-dimensional scoring and overall confidence, and determines optimal standard sections of each valid object. The optimal standard sections of each valid object and the labeled data are stored in association row by row, the stored data is synchronized to a disk, and a backup file is automatically generated and stored in an independent directory. Single labeling is replaced by paired comparison labeling, and subjective deviation is reduced in combination with structured multi-dimensional scoring, labeling consistency is improved, and section selection reliability is improved.

Description

Technical Field

[0001] This application relates to the field of ultrasound image detection technology, and in particular to a method and system for selecting standard sections of the fetus in early pregnancy. Background Technology

[0002] Prenatal screening is an important means of assessing fetal development. It involves using ultrasound equipment to acquire images and select standard sections to evaluate the fetus's development. In the early stages of pregnancy (4-10 weeks), the selection of standard fetal sections often relies on the physician's experience for manual screening, which is subject to significant subjective bias and has the disadvantage of low reliability in section selection. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, system, computer equipment, and storage medium for selecting standard fetal sections in early pregnancy that can improve the reliability of section selection, in order to address the above problems.

[0004] The first aspect of this application provides a method for selecting standard sections of a fetus in early pregnancy, including: Obtain standard cross-sectional data of the fetus in early pregnancy; The standard cross-sectional data of the fetus in early pregnancy are indexed and invalidated, and a dataset index report is generated; the dataset index report includes the ID, ultrasound image and image information of different valid objects; Based on the dataset index report, TOP K iterative selection is performed, ultrasound images are extracted for pairwise comparison and multi-dimensional scoring is completed, and the multi-dimensional scores are combined with the overall confidence to filter and determine the optimal standard section of each valid object. The optimal standard cross section and annotation data of each valid object are stored in a row-by-row association. The stored data is synchronized to the disk and a backup file is automatically generated and stored in an independent directory. The annotation data includes multi-dimensional scores, scoring criteria, overall confidence, annotation results, and image information.

[0005] In one embodiment, the early pregnancy fetal standard section data is indexed and invalid filtered to generate a dataset index report. This includes: automatically identifying all object ID subfolders and performing filtering operations to remove invalid objects containing only one or fewer images, images that cannot be parsed normally, and images without fetal core structures; counting the number of images and theoretical number of paired comparison groups for each valid object; and generating a dataset index report.

[0006] In one embodiment, ultrasound images are extracted for pairwise comparison and multi-dimensional scoring is performed. The multi-dimensional scores are then combined with the overall confidence level for screening to determine the optimal standard section for each valid object, including: Process the valid objects one by one according to their ID order, perform preliminary screening on all ultrasound images of each valid object, remove invalid images that are obviously blurry or lack core structure, and obtain the valid images of each valid object. Two images are randomly selected from the valid images of the current valid objects, and they are compared in pairs and scored in multiple dimensions. The one with the higher score is selected as the temporary optimal cross section. Randomly select the next image from the remaining valid images of the current valid object, compare it with the temporary best slice and score it, retain the one with the higher score as the new temporary best slice, until the initial comparison of all valid images of the current valid object is completed, and select the K candidate slices with the highest scores to form the TOP K candidate set; The images in the TOP K candidate set are compared in pairs, and the optimal standard cross section of the current valid object is determined by combining multi-dimensional scores and overall confidence.

[0007] In one embodiment, images within the TOP K candidate set are compared pairwise, and the optimal standard cross-section of the current valid object is determined by combining multi-dimensional scoring and overall confidence, including: Images are merged based on the multi-dimensional scores of each image in the TOP K candidate set to obtain equivalent clusters; each equivalent cluster represents a group of ultrasound sections with similar quality. Based on the win / loss results of pairwise comparison annotations, a directed tree is constructed between equivalent clusters; where the weight of the directed edge corresponds to the multi-dimensional score difference between the clusters, and the larger the score difference, the higher the weight of the directed edge. The root cluster with the highest score in the directed tree and no superior cluster is taken as the optimal standard section cluster of the current valid object. From the optimal standard section cluster, the image with the highest overall confidence and the most balanced multi-dimensional scores is selected as the optimal standard section of the current valid object.

[0008] In one embodiment, after storing the optimal standard cross-sections and annotation data of each valid object row by row, synchronizing the stored data to the disk, and automatically generating backup files stored in a separate directory, the method further includes: The optimal standard section, scores for each dimension, scoring criteria, and cluster structure relationships for each valid object are displayed through a visual interface.

[0009] In one embodiment, after storing the optimal standard cross-sections and annotation data of each valid object row by row, synchronizing the stored data to the disk, and automatically generating backup files stored in a separate directory, the method further includes: Export the annotation results, optimal standard cross section, and multi-dimensional scoring report to Excel and JSONL formats. The exported files contain complete annotation trajectories and structured information.

[0010] In one embodiment, after extracting ultrasound images for pairwise comparison and performing multi-dimensional scoring, and combining the multi-dimensional scores with the overall confidence level to determine the optimal standard section for each valid object, the method further includes: Automatically and randomly select labeled image pairs for verification according to a preset ratio, and / or, upon receiving a manual command to start conflict review, identify all conflicting image pairs and push them out, and synchronously update the TOP K iteration selection results based on the received review and correction information.

[0011] In one embodiment, after storing the optimal standard cross-sections and annotation data of each valid object row by row, synchronizing the stored data to the disk, and automatically generating backup files stored in a separate directory, the method further includes: Based on the received instructions, the stored optimal standard cross section and annotation data are queried or corrected.

[0012] In one embodiment, the multi-dimensional rating includes three dimensions: centering degree, area ratio, and clarity.

[0013] A second aspect of this application provides a system for selecting standard sections of a fetus in early pregnancy, comprising: The data acquisition module is used to acquire standard cross-sectional data of the fetus in early pregnancy; The data processing module is used to index and filter invalid data of the early pregnancy fetal standard section data, and generate a dataset index report; the dataset index report includes the ID, ultrasound image and image information of different valid objects; The section analysis module is used to perform TOP K iterative selection based on the dataset index report, extract ultrasound images for pairwise comparison and complete multi-dimensional scoring, and combine the multi-dimensional scores with the overall confidence to filter and determine the optimal standard section for each valid object. The data storage module is used to associate and store the optimal standard cross-sections and annotation data of each valid object line by line, synchronize the stored data to the disk, and automatically generate backup files stored in an independent directory; the annotation data includes multi-dimensional scores, scoring criteria, overall confidence, annotation results, and image information.

[0014] A third aspect of this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0015] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0016] The aforementioned method, system, computer equipment, and storage medium for selecting standard fetal sections in early pregnancy involve indexing and filtering the dataset of standard fetal sections in early pregnancy, generating a dataset index report, performing TOP K iterative selection based on the report, extracting ultrasound images for pairwise comparison and completing multi-dimensional scoring, and combining the multi-dimensional scores with overall confidence to determine the optimal standard section for each valid object. The optimal standard section and labeled data for each valid object are stored row-by-row in association, and the stored data is synchronized to disk, with backup files automatically generated and stored in a separate directory. By replacing single labels with pairwise comparison labels and combining structured multi-dimensional scoring, subjective bias is reduced, label consistency is improved, and the reliability of section selection is enhanced. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for selecting standard cross-sections of the fetus in early pregnancy in one embodiment; Figure 2 This is a schematic diagram of the clusters selected by TOP K iteration in one embodiment; Figure 3 A block diagram of a system for selecting standard cross-sections of the fetus in early pregnancy in one embodiment; Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. When used herein, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising / including” or “having,” etc., specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used herein includes any and all combinations of the associated listed items.

[0020] Addressing the core pain points of traditional early pregnancy fetal standard section selection methods—namely, "only distinguishing between standard and non-standard, lacking interpretability" and "consuming significant human and material resources for annotation"—this application provides a method for early pregnancy fetal standard section selection. This method indexes and filters invalid data from early pregnancy fetal standard section data, generating a dataset index report. Based on the dataset index report, it performs TOP K iterative selection, extracting ultrasound images for pairwise comparison and completing multi-dimensional scoring. The optimal standard section for each valid object is determined by combining the multi-dimensional scores with the overall confidence level. The optimal standard section for each valid object and the labeled data are then stored row-by-row, synchronized to disk, and automatically generated as a backup file stored in a separate directory. By integrating a customized multi-dimensional scoring system based on industry standards for early pregnancy fetal ultrasound examinations and employing a TOP K selection model to replace traditional full-scale comparison, it achieves efficient, interpretable, and standardized section selection.

[0021] In one embodiment, such as Figure 1 As shown, a method for selecting standard sections of the fetus in early pregnancy is provided, including: Step S110: Obtain standard cross-sectional data of the fetus in early pregnancy.

[0022] Specifically, early pregnancy fetal standard section data includes identification IDs and ultrasound images for different subjects. Taking different cases as examples, a standardized directory structure is first constructed, using a hierarchical organization method of "root directory - case ID subfolder - ultrasound images". All ultrasound images of the same case (such as NT section, cranial section, trunk section, etc.) are uniformly stored in the subfolder corresponding to the case ID. The image format is uniformly DICOM format (compliant with medical ultrasound image industry standards), and the file name is named according to "case ID_image sequence number_section type" (such as "Case001_001_NT") to ensure the standardization and traceability of data management.

[0023] Step S120: Index and filter invalid data from standard fetal cross-sections in early pregnancy, generating a dataset index report. The dataset index report includes the IDs of different valid objects, ultrasound images, and image information. Image information may include ultrasound image paths, basic information, etc., while basic information may include the number of images for each valid object and the theoretical number of paired comparison groups.

[0024] Using different cases as examples, we performed dataset indexing and invalid case filtering: We scanned the root directory, automatically identified all object ID subfolders, established a mapping relationship of "case ID - ultrasound image path - basic information", and simultaneously performed invalid case filtering to remove invalid cases containing only one or fewer images, images that could not be properly resolved, or those lacking fetal core structures (such as no nuchal translucency in the NT section or no skull structure in the cranial section). We counted the number of images and the theoretical number of paired comparison groups for each valid case, and generated a dataset index report to provide a basis for subsequent annotation work.

[0025] Step S130: Perform TOP K iterative selection based on the dataset index report, extract ultrasound images for pairwise comparison and complete multi-dimensional scoring, combine multi-dimensional scores with overall confidence to filter and determine the optimal standard section for each valid object.

[0026] The relevant parameters for TOP K iteration selection can be preset. Combined with clinical industry needs and annotation efficiency goals, the K value can be flexibly adjusted (the conventional value is 3-5, and the 3 best sections are selected first). The core rules for TOP K selection are clearly defined - taking "multi-dimensional comprehensive score" as the core basis, taking into account the balance of scores of each dimension, avoiding the influence of single-dimensional extreme values ​​(such as extremely high clarity but insufficient centering) on ​​the selection results. At the same time, the threshold for the number of comparison iterations is preset to prevent invalid iterations.

[0027] The specific types of multi-dimensional scoring are not unique. In this embodiment, multi-dimensional scoring includes three dimensions: centering degree, area proportion, and clarity. By strictly referring to the industry standards for early pregnancy fetal ultrasound examinations, three core evaluation dimensions—centering degree, area proportion, and clarity—were customized. Each dimension corresponds to specific requirements in the industry standards, ensuring the standardization and interpretability of the annotations. The specific dimension definitions are as follows: Centering degree: Refer to the industry standard for the positioning requirements of the fetal core structure (such as skull, NT, trunk) in the cross-section, evaluate whether the core structure is located in the central area of ​​the cross-section. The offset between the center point of the core structure and the center point of the cross-section shall not exceed 10% of the side length of the cross-section to meet the standard. Area ratio: Based on industry standards for the display range of the fetal core structure, evaluate whether the proportion of the core structure in the section meets the standards. Specifically, the proportion of the nuchal translucency in the NT section should not be less than 30%, and the proportion of the skull contour in the cranial section should not be less than 40%. Clarity: In accordance with industry standards for ultrasound image resolution and detail display, the clarity of the fetal core structures (such as skull outline, NT boundary, and trunk outline) is evaluated. The core structure boundaries are clear and unbroken, and detailed features can be clearly identified to meet the standards and satisfy the needs of clinical diagnosis and AI model training.

[0028] By setting annotation rules and scoring criteria, each dimension adopts a five-level scoring system from -2 to 2. The scoring criteria strictly correspond to industry standards to ensure the interpretability of the annotation results. The specific scoring rules are as follows: -2 points: The left image is significantly better than the industry standard minimum requirement in the corresponding dimension, while the right image is significantly substandard (e.g., the core structure of the left image is centered, while the core structure of the right image is offset by more than 20% of the cross-sectional side length). -1 point: The left image is slightly better than the industry standard minimum requirement in the corresponding dimension, and the right image is close to meeting the standard (e.g., the core structure accounts for 35% in the left image and 28% in the right image). 0 points: The corresponding dimensions of the left and right images meet industry standards and the differences are negligible (e.g., the centering offset of the core structure of the left and right images is within 5%). 1 point: The right image is slightly better than the industry standard minimum requirement in the corresponding dimension, and the left image is close to meeting the standard (e.g., the core structure of the right image is clear enough to identify details, while the core structure of the left image is slightly blurred). 2 points: The right image is significantly better than the industry standard minimum requirement in the corresponding dimension, while the left image is significantly substandard (e.g., the core structure accounts for 45% in the right image and 20% in the left image).

[0029] In one embodiment, step S130 includes steps 131 to 134.

[0030] Step 131: Process the valid objects one by one according to their ID order. Perform preliminary screening on all ultrasound images of each valid object, and remove invalid images that are obviously blurry or lack core structures to obtain the valid images of each valid object.

[0031] First, an initial screening is performed, processing each valid object sequentially by its ID. Each valid object is then designated as the current valid object, and all ultrasound images of that object undergo preliminary screening. Image clarity thresholds and core structure recognition algorithms are used to eliminate obviously blurry images (clarity score below 1) and images lacking core structures, thus obtaining valid images for each valid object. This reduces unnecessary comparisons, lowers manpower and time consumption, and improves annotation efficiency. After the initial screening, a TOP K iterative selection is performed using the valid images of each valid object, employing a "winner-takes-all + gradient comparison" mechanism, eliminating the need for full pairwise comparisons of all images.

[0032] Step 132: Randomly select two images from the valid images of the current valid object, perform pairwise comparison and complete multi-dimensional scoring, and select the one with the higher score as the temporary optimal cross-section.

[0033] Step 133: Randomly select the next image from the remaining valid images of the current valid object, compare it with the temporary optimal slice and score it, retain the one with the higher score as the new temporary optimal slice, until the preliminary comparison of all valid images of the current valid object is completed, and select the K candidate slices with the highest scores to form the TOP K candidate set.

[0034] Step 134: Compare the images in the TOP K candidate set in pairs, and determine the optimal standard cross-section of the current valid object by combining multi-dimensional scores and overall confidence.

[0035] Specifically, the TOP K iterative selection steps are as follows: ① Randomly select two images from the valid images of the current valid object (case), perform pairwise comparison and complete multi-dimensional scoring, and select the one with the higher score as the temporary optimal section; ② Randomly select the next image from the remaining valid images, compare it with the temporary optimal section and score it, and retain the one with the higher score as the new temporary optimal section; ③ Repeat step ② until the preliminary comparison of all valid images of the case is completed, and select K candidate sections with the highest scores to form the TOP K candidate set; ④ Perform pairwise comparison of the images in the TOP K candidate set, combine multi-dimensional scoring and overall confidence, and finally determine the TOP K optimal section of the valid object. Prioritize outputting the top-ranked optimal standard section for user confirmation, and the remaining K-1 sections are used as candidate sections.

[0036] Furthermore, the overall confidence level can be filled in by the annotator or by the computer device scoring the image according to set rules. In this embodiment, the annotator fills in the overall confidence level (1-5 points, 1 point being the lowest and 5 points being the highest) and remarks for each pair of comparative annotations. The remarks clearly explain the scoring basis (e.g., "The left image clarity is rated 1 point because the NT boundary is slightly blurry and does not fully meet the industry standard requirement for clear display"). This makes the annotation results traceable and interpretable, and completely solves the pain point of traditional annotations that "only distinguishes between standard and non-standard, but cannot explain the reasons for the superiority or inferiority".

[0037] In one embodiment, step 134 includes: merging images based on the multi-dimensional scores of each image in the TOP K candidate set to obtain equivalent clusters; each equivalent cluster represents a set of ultrasound sections with similar quality; constructing a directed tree between equivalent clusters based on the win / loss results of pairwise comparison annotations; wherein the weight of the directed edge corresponds to the difference in multi-dimensional scores between clusters, and the larger the score difference, the higher the weight of the directed edge; taking the root cluster with the highest score in the directed tree and no superior cluster as the optimal standard section cluster of the current valid object, and selecting the image with the highest overall confidence and the most balanced multi-dimensional scores from the optimal standard section cluster as the optimal standard section of the current valid object.

[0038] Specifically, firstly, based on the "about the same" label in the annotation results, images with negligible score differences (total difference in multi-dimensional scores ≤ 1) are merged into equivalent clusters, such as... Figure 2 As shown, each equivalent cluster represents a group of ultrasound sections of similar quality, reducing redundant data and facilitating rapid location of the optimal section. Then, based on the results of pairwise comparison annotations, a directed tree is constructed among the equivalent clusters, indicating that "cluster A is superior to cluster B." The weight of the directed edge corresponds to the difference in multi-dimensional comprehensive scores between the clusters; the larger the score difference, the higher the weight of the directed edge, intuitively reflecting the superiority or inferiority relationship between clusters. Finally, the root cluster with the highest score and no superior higher-level cluster is selected as the optimal standard section cluster for that case. From the root cluster, the image with the highest overall confidence and the most balanced multi-dimensional scores is selected as the optimal standard section for the current valid object. Simultaneously, the optimal standard section of the valid object, along with the scores for each dimension and the scoring criteria, are output for user final confirmation.

[0039] Step S140: The optimal standard section and annotation data of each valid object are associated and stored row by row, the stored data is synchronized to the disk, and backup files are automatically generated and stored in an independent directory.

[0040] The labeled data includes multi-dimensional scores, scoring criteria, overall confidence levels, annotation results, and image information. Specifically, by encapsulating structured labeled data, the optimal standard cross-sections of each valid object, along with corresponding multi-dimensional scores, scoring criteria (notes), overall confidence levels, annotation results (TOP K selection results, pairwise comparison results), image paths, and other labeled data, are stored together to ensure that the selection of each optimal cross-section is supported by a clear score. The labeled data can be directly used for subsequent clinical diagnostic reference and AI model training.

[0041] Continuing with examples from different cases, the labeled data is stored line by line in JSONL format. Each record includes core information such as timestamp, session ID, annotator ID, case ID, image path, pairwise comparison results, multi-dimensional scores, overall confidence level, and remarks, ensuring the uniqueness and integrity of each labeled record for easy subsequent querying and tracing. Immediately after each labeled data entry is written, `flush()` and `os.fsync()` operations are executed to force data synchronization to disk, preventing data loss due to abnormal program exit. For every 100 original labeled events accumulated, the system automatically generates a `.bak` backup file, stored in a separate directory. Manual backup is also supported, forming a triple security guarantee of "real-time storage + automatic backup + manual backup".

[0042] In one embodiment, after step S130, the method further includes: automatically randomly selecting labeled image pairs for verification according to a preset ratio, and / or, after receiving a manual conflict review instruction, identifying all conflicting image pairs and pushing them, and synchronously updating the TOP K iteration selection results according to the received review and correction information.

[0043] This embodiment adds mode switching and verification: The TOP K selection mode is the default core annotation mode. After the TOP K selection of all valid objects is completed, the system automatically switches to the full pair random mode and randomly selects labeled image pairs according to a preset ratio (e.g., 30%) (prioritizing image pairs with conflicting annotation results) for verification to ensure the accuracy of the TOP K selection results. At the same time, it supports manually enabling the conflict review mode. The system automatically identifies all conflicting image pairs (the same image pair has annotation results that are mutually superior and inferior) for annotators to review and correct. After correction, the TOP K optimal cross section result is updated synchronously, which avoids the waste of resources caused by full comparison and ensures the annotation quality.

[0044] Furthermore, after step S140, the method may include: querying or correcting the stored optimal standard cross section and annotation data according to the received instructions. Further, after step S140, the method may also include: displaying the optimal standard cross section, scores for each dimension, scoring criteria, and cluster structure relationships for each valid object through a visual interface; and / or exporting the annotation results, optimal standard cross section, and multi-dimensional scoring report to Excel or JSONL format, with the exported file containing complete annotation trajectories and structured information.

[0045] Specifically, the system supports historical query, correction, result display, and result export functions, as detailed below: Historical query: Supports filtering and viewing by object ID, annotation time, and annotator ID. It also allows viewing the complete structured annotation information and paired comparison trajectories for each record, making it convenient for annotators to review and verify.

[0046] Correction function: Supports loading historically labeled image pairs, modifying labeling results, multi-dimensional scores, overall confidence and remarks. Correction records are stored incrementally, retaining the original labeling trajectory, correction time and corrector information, ensuring that the labeling correction process is traceable and avoiding data chaos caused by arbitrary modifications.

[0047] Results Display: The system displays the TOP K optimal profiles, scores for each dimension, scoring criteria, and cluster structure relationships for each object through a visual interface. Users can intuitively view the advantages of the optimal profiles, understand the scoring logic, and improve the efficiency of clinical judgment.

[0048] Results Export: Supports exporting annotation results, TOP K optimal cross section information, and multi-dimensional scoring reports to Excel and JSONL formats. The exported files contain complete annotation trajectories and structured information, which can be used for scenarios such as clinical case archiving, AI model training data input, and annotation quality review.

[0049] The above-mentioned method for selecting standard fetal sections in early pregnancy addresses the existing technical pain points such as the difficulty in selecting standard fetal sections in early pregnancy (4-10 weeks), non-standard labeling, and low efficiency. The following four key invention points are summarized: The three-level paired annotation system is the first to combine the Top K mode (winner retention mechanism), the full pair random mode, and the conflict review mode. It prioritizes and quickly locks the optimal aspect ratio, and then ensures the accuracy of annotation through random verification and conflict correction, solving the practical problems of low efficiency and poor consistency of judgment among different annotators in traditional annotation.

[0050] Multi-dimensional structured and detailed annotation: In response to the core evaluation needs of standard fetal cross-sections in early pregnancy, the three core evaluation dimensions of "centering degree, area ratio, and clarity" are strictly customized in accordance with industry standards. A five-level scoring system of -2 to 2 is adopted, combined with overall confidence and remarks, to achieve quantifiable and interpretable annotation of cross-section quality. This solves the practical problems of traditional annotations that can only distinguish between "qualified / unqualified", cannot explain the reasons for the quality, and lack interpretability.

[0051] Intra-case image cluster structure construction and visualization: Based on the annotation results (merging equivalent clusters of “nearly” annotations and constructing directed trees from the win and lose results), the optimal cross-section cluster (root cluster) is automatically identified, and the cluster relationship and image location are visualized to help users quickly confirm the optimal standard cross-section, solving the practical problem that users need to manually select the optimal cross-section after traditional annotation, which is cumbersome.

[0052] A secure and traceable annotation management mechanism: Annotation records are stored in real time using a standardized format. Each record is forcibly synchronized to the storage device after being written. Automatic backup is performed after a certain number of annotation records are accumulated. Historical annotation queries, loading, and corrections are also supported, preserving the original annotations and correction trajectories. This solves the practical problems of traditional annotation data being easily lost, the annotation process being untraceable, and not complying with medical data management standards.

[0053] This invention differs fundamentally from conventional image sorting, general pairwise comparison, and simple clustering methods. Its core lies in constructing a hierarchical decision-making chain of "highly efficient initial selection of TOP K → controllable pairwise verification → label-driven final selection of cluster structure." It also combines standardized dataset construction, a multi-dimensional labeling system, and a secure storage mechanism to form a complete technical closed loop, rather than simply processing a single link. Therefore, it has achieved improvements in efficiency, accuracy, interpretability, and clinical consistency.

[0054] Compared to conventional image sorting and simple scoring sorting, conventional methods only use a single indicator to globally rank images, directly outputting the top-ranked images. They only focus on image scores and do not pay attention to their clinical representativeness. They are easily affected by noise, artifacts, and angular deviations, lack interpretability, and are not specifically optimized for clinical annotation scenarios. In contrast, this invention adopts a two-level screening mechanism that combines TOP K initial selection and cluster structure final selection. First, the effective images of the cases are initially screened, iteratively compared, and scored in multiple dimensions to select K candidate sections with the highest scores, effectively compressing the number of images to be processed and ensuring processing efficiency. Then, through cluster structure construction, equivalent clusters are merged and a directed hierarchical tree is built to accurately determine the unique standard section. At the same time, the structured annotation results are associated, including three-dimensional scores, scoring basis, and overall confidence. This not only preserves high-value images but also achieves clinical interpretability, structural traceability, and unique and stable results. It is not a simple image sorting but a hierarchical screening mechanism that meets clinical needs.

[0055] Compared to general pairwise comparison annotation, traditional pairwise comparison uses a full random comparison method, which lacks clear priorities and focus, resulting in a large amount of repetitive work, low efficiency, and no distinction between "initial selection" and "verification" stages, leading to high annotation costs. Furthermore, it lacks a conflict correction mechanism. This invention, however, strictly limits pairwise comparison to the verification stage. It first uses TOPK initial selection to significantly reduce the number of images to be annotated, then employs a controllable random strategy of "30% repeated verification of already annotated pairs + 70% new image pairs" for verification. Simultaneously, a manually triggered conflict review branch is set up. Conflicting image pairs are identified, extracted, and displayed for annotation by the annotator for verification and correction, while the optimal TOP K cross-section is adjusted concurrently. This forms an annotation mode that prioritizes efficiency, uses verification as a secondary measure, and allows for retrospective correction, rather than blindly performing a full comparison. This effectively reduces annotation costs and improves annotation accuracy.

[0056] Compared to simple clustering and traditional clustering algorithms, conventional clustering only groups based on feature distance, without considering clinical semantics, lacking hierarchical structure, and failing to output a unique standard cross-section, resulting in unstable and uninterpretable results. Furthermore, it does not correlate with labeled data for semantic alignment. The cluster structure construction of this invention is not simply feature clustering, but rather establishes cluster relationships based on labeled multi-dimensional scores, merges clinically equivalent cross-section clusters based on clinical needs, constructs a directed hierarchical tree to determine the root cluster and optimal representative, and ultimately outputs a unique, standard, and interpretable optimal cross-section. This belongs to "label-driven semantic clustering," and the clustering results can retain modification trajectories through the labeling traceability function in step six. This is fundamentally different from traditional data-driven clustering and better meets the clinical practical needs of medical image labeling.

[0057] This application addresses four core technical problems existing in the current process of selecting standard cross-sections of fetuses in early pregnancy, and provides precise solutions through code functionality, as detailed below: The problems of high subjectivity and poor consistency: Existing technologies rely on manual screening based on physician experience, and the judgments of different physicians vary greatly; This application replaces single annotation with paired comparison annotation and combines structured multi-dimensional scoring (strictly referring to industry standards) to reduce subjective bias. At the same time, it corrects annotation contradictions through a conflict review mode to improve annotation consistency.

[0058] The problem of low selection efficiency: Existing technologies require physicians to sift through a large number of images one by one, which is cumbersome and consumes a lot of manpower and time; This application adopts the TOP K selection mode (winner retention + gradient comparison), which eliminates the need for full comparison and quickly focuses on the optimal section. Compared with traditional full combination annotation, the efficiency is improved by more than 50%, which meets the needs of clinicians for efficient operation.

[0059] The problems of incomplete annotation system and poor data quality: Existing general annotation tools are not customized for fetal cross-sections in early pregnancy, and can only achieve simple "standard / non-standard" distinction. They lack structured dimensional scoring and cannot meet the needs of AI training for refined and interpretable data. This application customizes a multi-dimensional scoring system that conforms to industry standards, realizes structured and interpretable annotation, and ensures the quality of annotated data through duplication verification and conflict correction, which can be directly used for AI model training.

[0060] The problems of unsafe and untraceable labeled data: Existing labeling tools lack a sound storage and backup mechanism, which makes data loss easy and makes it impossible to trace the labeling correction trajectory; This application ensures data security through real-time storage, forced synchronization and automatic backup, while retaining the original labeling and correction records, realizing full-process traceability and meeting medical data management standards.

[0061] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0062] Based on the same inventive concept, this application also provides a system for selecting standard sections of the fetus in early pregnancy to implement the aforementioned method. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the system for selecting standard sections of the fetus in early pregnancy provided below can be found in the limitations of the method for selecting standard sections of the fetus in early pregnancy described above, and will not be repeated here.

[0063] In one embodiment, such as Figure 3 As shown, a standard section selection system for fetuses in early pregnancy is provided, comprising: The data acquisition module 110 is used to acquire standard cross-sectional data of the fetus in early pregnancy.

[0064] The data processing module 120 is used to index and filter invalid data of standard cross-sectional data of fetuses in early pregnancy and generate a dataset index report. The dataset index report includes the ID of different valid objects, ultrasound images and image information.

[0065] The section analysis module 130 is used to perform TOP K iterative selection based on the dataset index report, extract ultrasound images for pairwise comparison and complete multi-dimensional scoring, and combine the multi-dimensional scores with the overall confidence level to filter and determine the optimal standard section for each valid object.

[0066] The data storage module 140 is used to store the optimal standard cross section and annotation data of each valid object in a row-by-row association, synchronize the stored data to the disk, and automatically generate backup files stored in an independent directory; the annotation data includes multi-dimensional scores, scoring basis, overall confidence, annotation results, and image information.

[0067] In one embodiment, the data processing module 120 is used to automatically identify all object ID subfolders and perform filtering operations, remove invalid objects containing only one or fewer images, images that cannot be parsed normally, and images without fetal core structures, count the number of images and theoretical number of paired comparison groups for each valid object, and generate a dataset index report.

[0068] In one embodiment, the section analysis module 130 is used to process the valid objects one by one according to their ID order, perform preliminary screening of all ultrasound images of each valid object, remove invalid images that are obviously blurry or lack core structures, and obtain valid images of each valid object; randomly select two images from the valid images of the current valid object, perform pairwise comparison and complete multi-dimensional scoring, and select the one with the higher score as the temporary optimal section; randomly select the next image from the remaining valid images of the current valid object, compare and score it with the temporary optimal section, and retain the one with the higher score as the new temporary optimal section, until the preliminary comparison of all valid images of the current valid object is completed, and select the K candidate sections with the highest scores to form the TOPK candidate set; perform pairwise comparison of the images in the TOP K candidate set, and determine the optimal standard section of the current valid object by combining multi-dimensional scoring and overall confidence.

[0069] In one embodiment, the section analysis module 130 is used to merge images based on the multi-dimensional scores of each image in the TOP K candidate set to obtain equivalent clusters; each equivalent cluster represents a set of ultrasound sections with similar quality; based on the win and loss results of pairwise comparison annotations, a directed tree is constructed between the equivalent clusters; wherein, the weight of the directed edge corresponds to the difference in multi-dimensional scores between the clusters, and the larger the score difference, the higher the weight of the directed edge; the root cluster with the highest score in the directed tree and no superior cluster is taken as the optimal standard section cluster of the current valid object, and the image with the highest overall confidence and the most balanced multi-dimensional scores is selected from the optimal standard section cluster as the optimal standard section of the current valid object.

[0070] In one embodiment, the section analysis module 130 is also used to automatically randomly select labeled image pairs for verification according to a preset ratio, and / or, after receiving a manual conflict review instruction, identify all conflicting image pairs and push them, and synchronously update the TOP K iteration selection results according to the received review and correction information.

[0071] In one embodiment, the data storage module 140 is also used to display the optimal standard section, scores of each dimension, scoring basis and cluster structure relationship of each valid object through a visual interface; and to export the annotation results, optimal standard section and multi-dimensional scoring report to Excel and JSONL formats. The exported file contains complete annotation trajectory and structured information.

[0072] In one embodiment, the data storage module 140 is further configured to query or correct the stored optimal standard cross section and annotation data according to the received instructions.

[0073] The modules in the aforementioned early pregnancy fetal standard section selection system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0074] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for selecting standard cross-sections of the fetus in early pregnancy. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0075] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0076] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0077] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above method.

[0078] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0079] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0080] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The processors involved in the embodiments provided in this application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited thereto.

[0081] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0082] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for selecting standard cross-sections of the fetus in early pregnancy, characterized in that, include: Obtain standard cross-sectional data of the fetus in early pregnancy; The dataset indexing and invalid filtering of the early pregnancy fetal standard section data are performed to generate a dataset index report; The dataset index report includes the IDs, ultrasound images, and image information of different valid objects; Based on the dataset index report, TOP K iterative selection is performed, ultrasound images are extracted for pairwise comparison and multi-dimensional scoring is completed, and the multi-dimensional scores are combined with the overall confidence to filter and determine the optimal standard section of each valid object. The optimal standard cross section and annotation data of each valid object are stored in a row-by-row association. The stored data is synchronized to the disk and a backup file is automatically generated and stored in an independent directory. The annotation data includes multi-dimensional scores, scoring criteria, overall confidence, annotation results, and image information.

2. The method according to claim 1, characterized in that, The dataset indexing and invalid filtering of the early pregnancy fetal standard section data are performed to generate a dataset index report, including: automatically identifying all object ID subfolders and performing filtering operations to remove invalid objects containing only 1 or fewer images, images that cannot be parsed normally, and images without fetal core structures; counting the number of images and theoretical number of paired comparison groups for each valid object; and generating a dataset index report.

3. The method according to claim 1, characterized in that, Ultrasound images were extracted, paired, and scored in multiple dimensions. The scores were then combined with the overall confidence level to determine the optimal standard section for each valid subject, including: Process the valid objects one by one according to their ID order, perform preliminary screening on all ultrasound images of each valid object, remove invalid images that are obviously blurry or lack core structure, and obtain the valid images of each valid object. Two images are randomly selected from the valid images of the current valid objects, and they are compared in pairs and scored in multiple dimensions. The one with the higher score is selected as the temporary optimal cross section. Randomly select the next image from the remaining valid images of the current valid object, compare it with the temporary best slice and score it, retain the one with the higher score as the new temporary best slice, until the initial comparison of all valid images of the current valid object is completed, and select the K candidate slices with the highest scores to form the TOP K candidate set; The images in the TOP K candidate set are compared in pairs, and the optimal standard cross section of the current valid object is determined by combining multi-dimensional scores and overall confidence.

4. The method according to claim 3, characterized in that, Images within the TOP K candidate set are compared pairwise, and the optimal standard cross-section of the current valid object is determined by combining multi-dimensional scores and overall confidence, including: Images are merged based on the multi-dimensional scores of each image in the TOP K candidate set to obtain equivalent clusters; each equivalent cluster represents a group of ultrasound sections with similar quality. Based on the win / loss results of pairwise comparison annotations, a directed tree is constructed between equivalent clusters; where the weight of the directed edge corresponds to the multi-dimensional score difference between the clusters, and the larger the score difference, the higher the weight of the directed edge. The root cluster with the highest score in the directed tree and no superior cluster is taken as the optimal standard section cluster of the current valid object. From the optimal standard section cluster, the image with the highest overall confidence and the most balanced multi-dimensional scores is selected as the optimal standard section of the current valid object.

5. The method according to claim 4, characterized in that, After storing the optimal standard cross-sections and annotation data of each valid object row by row, synchronizing the stored data to disk, and automatically generating backup files stored in a separate directory, the process also includes: The optimal standard section, scores for each dimension, scoring criteria, and cluster structure relationships for each valid object are displayed through a visual interface.

6. The method according to claim 5, characterized in that, After storing the optimal standard cross-sections and annotation data of each valid object row by row, synchronizing the stored data to disk, and automatically generating backup files stored in a separate directory, the process also includes: Export the annotation results, optimal standard cross section, and multi-dimensional scoring report to Excel and JSONL formats. The exported files contain complete annotation trajectories and structured information.

7. The method according to claim 1, characterized in that, After extracting ultrasound images for pairwise comparison and completing multi-dimensional scoring, and combining the multi-dimensional scores with the overall confidence level for screening, the optimal standard section for each valid subject is determined, and the process also includes: Automatically and randomly select labeled image pairs for verification according to a preset ratio, and / or, upon receiving a manual command to start conflict review, identify all conflicting image pairs and push them out, and synchronously update the TOP K iteration selection results based on the received review and correction information.

8. The method according to claim 1, characterized in that, After storing the optimal standard cross-sections and annotation data of each valid object row by row, synchronizing the stored data to disk, and automatically generating backup files stored in a separate directory, the process also includes: Based on the received instructions, the stored optimal standard cross section and annotation data are queried or corrected.

9. The method according to any one of claims 1 to 8, characterized in that, The multi-dimensional rating includes three dimensions: centering degree, area ratio, and clarity.

10. A system for selecting standard sections of fetuses in early pregnancy, characterized in that, include: The data acquisition module is used to acquire standard cross-sectional data of the fetus in early pregnancy; The data processing module is used to index and filter invalid data of the early pregnancy fetal standard section data, and generate a dataset index report; the dataset index report includes the ID, ultrasound image and image information of different valid objects; The section analysis module is used to perform TOP K iterative selection based on the dataset index report, extract ultrasound images for pairwise comparison and complete multi-dimensional scoring, and combine the multi-dimensional scores with the overall confidence to filter and determine the optimal standard section for each valid object. The data storage module is used to associate and store the optimal standard cross-sections and annotation data of each valid object line by line, synchronize the stored data to the disk, and automatically generate backup files stored in an independent directory; the annotation data includes multi-dimensional scores, scoring criteria, overall confidence, annotation results, and image information.

Images