An ophthalmic biometry report automated analysis method, system, device, medium, and product

Abstract

This application discloses an automated parsing method, system, device, medium, and product for ophthalmic biometric reports, relating to the field of ophthalmic artificial intelligence research. The method includes: converting the ophthalmic biometric report to be processed into a report image; partitioning the report image into a table grid and locating cells to obtain a set of cells; sorting all cells by rows and columns in a grid, and cropping each cell to generate a cell image; recognizing text using OCR and filling it into the corresponding positions to obtain cell text; subsequently performing structured preprocessing and semantic expansion on the cell text, completing field parsing and standardized mapping based on the preprocessed two-dimensional grid structure to generate an initial structured record; extracting values ​​from the intraocular lens calculation parameter area and writing them into the initial record to obtain the final structured record, and then writing it into an output file according to the target database template. This application can effectively improve data entry efficiency and reduce field mismatch and data cleaning costs.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence research in ophthalmology, and in particular to a method, system, device, medium, and product for automated analysis of ophthalmic biometric reports. Background Technology

[0002] In the process of ophthalmic artificial intelligence research and clinical data mining, various ophthalmic examination reports / biometric reports (including OD / OS data for both eyes, corneal curvature K value, axial length AL, anterior chamber depth ACD and other parameters) need to be transformed into structured data to build a standardized database that can be used for model training, regression analysis and multi-center data fusion.

[0003] The current data processing solutions mainly suffer from the following technical limitations and shortcomings: (1) Manual input / proofreading: low processing efficiency, high labor cost, easy to make copying errors, and cannot meet the needs of batch data processing; (2) General OCR full-page recognition combined with keyword extraction: Ophthalmology reports generally have problems such as complex table format, bilateral partitioning, relatively fixed field positions but different formats of different devices and export methods, which can easily lead to field mismatch (especially OD / OS cross-column), unit and special symbol (△ / Δ, °, D, mm, etc.) recognition errors, and multiple expressions of the same field leading to extraction failure. (3) Regional OCR cropping based on fixed coordinate template: This method is less robust and has higher maintenance costs when the report version changes, the screenshot is cropped, the resolution changes, or there are gray bars or watermarks.

[0004] Therefore, there is an urgent need for an automated analysis method in clinical and research settings that can adapt to multiple templates and noise interference scenarios in ophthalmology reports, automatically complete table structure reconstruction, field error correction and fusion, and directly output a standardized database. Summary of the Invention

[0005] The purpose of this application is to provide an automated parsing method, system, device, medium, and product for ophthalmic biometric reports, which can effectively improve data entry efficiency and reduce field mismatch and data cleaning costs.

[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides an automated method for parsing ophthalmic biometric reports, including: Acquire the ophthalmic biometric report to be processed and convert it into a report image; The report image is divided into table grids and positioned by cells to obtain a set of cells; Sort all cells in the cell set by row and column to create a two-dimensional grid structure, and then crop each cell in the two-dimensional grid structure to generate a cell image; Perform OCR recognition on each cell image, and fill the recognized text into the corresponding cell in the two-dimensional grid structure to obtain the cell text; The text in each cell is preprocessed in a structured manner and semantically expanded to obtain a preprocessed two-dimensional grid structure. Based on the preprocessed two-dimensional grid structure, field parsing and normalization mapping are performed to generate initial structured records; Numerical extraction is performed on the artificial crystal computational parameter region in the preprocessed two-dimensional mesh structure, and the extraction results are written into the initial structured record to obtain the final structured record. The final structured records are written to the output file according to the target database template.

[0007] Secondly, this application provides an automated analysis system for ophthalmic biometric reports, comprising: The report processing module is used to acquire ophthalmic biometric reports to be processed and convert them into report images. The cell set determination module is used to perform table grid partitioning and cell positioning on the report image to obtain a cell set; The cell image generation module is used to sort all cells in the cell set by rows and columns in a grid, establish a two-dimensional grid structure, and crop each cell in the two-dimensional grid structure to generate a cell image. The OCR recognition module is used to perform OCR recognition on each cell image and fill the recognized text into the corresponding cell in the two-dimensional grid structure to obtain the cell text; The preprocessing module is used to perform structured preprocessing and semantic expansion on the cell text in each cell to obtain a preprocessed two-dimensional grid structure. The initial structure generation module is used to perform field parsing and normalization mapping based on the preprocessed two-dimensional grid structure to generate initial structured records; The final structured record determination module is used to extract numerical values ​​from the artificial crystal calculation parameter region in the preprocessed two-dimensional mesh structure and write the extraction results into the initial structured record to obtain the final structured record. The write module is used to write the final structured records to the output file according to the target database template.

[0008] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described automated analysis method for ophthalmic biometric reports.

[0009] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for automatically parsing ophthalmic biometric reports.

[0010] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for automatically parsing ophthalmic biometric reports.

[0011] According to the specific embodiments provided in this application, this application has the following technical effects: This application employs a technical solution involving report image conversion, table grid partitioning and cell positioning, 2D grid construction, cell OCR recognition, structured preprocessing and semantic expansion, field parsing and standardized mapping, and extraction of intraocular lens parameter values, outputting structured files according to templates. This solution enables automated parsing of ophthalmic biometric reports, effectively solving the problems of low efficiency, error-proneness, and inability to batch process manual data entry. Field positioning via table grid partitioning and 2D grid structure adapts to differences in layout across multiple devices, reducing the risk of field mismatch and cross-references. Adaptive table structure reconstruction eliminates the need for fixed coordinate templates, improving adaptability to report version changes, image cropping, and resolution variations. Furthermore, structured preprocessing and standardized mapping enable unified field output, directly generating structured data suitable for database storage, reducing data cleaning and processing costs. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating an automated parsing method for ophthalmic biometric reports, provided as an embodiment of this application. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] In one exemplary embodiment, such as Figure 1 As shown, an automated parsing method for ophthalmic biometric reports is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is described using a server as an example, and includes the following steps S1 to S8.

[0017] S1: Obtain the ophthalmic biometric report to be processed and convert it into a report image.

[0018] S2: Perform table grid partitioning and cell positioning on the report image to obtain a set of cells.

[0019] S3: Sort all cells in the cell set by rows and columns in a grid to create a two-dimensional grid structure, and then crop each cell in the two-dimensional grid structure to generate a cell image.

[0020] S4: Perform OCR recognition on each cell image and fill the recognized text into the corresponding cell in the two-dimensional grid structure to obtain the cell text.

[0021] S5: Perform structured preprocessing and semantic expansion on the cell text in each cell to obtain a preprocessed two-dimensional grid structure.

[0022] S6: Based on the preprocessed two-dimensional grid structure, perform field parsing and normalization mapping to generate initial structured records.

[0023] S7: Numerical extraction is performed on the artificial crystal calculation parameter region in the preprocessed two-dimensional mesh structure, and the extraction results are written into the initial structured record to obtain the final structured record.

[0024] S8: Write the final structured records to the output file according to the target database template.

[0025] By implementing steps S1 to S8 above, this application can replace manual data entry, improve processing efficiency, and avoid transcription errors; it can adapt to multiple version differences and reduce field mismatches; it does not require a fixed coordinate template and has stronger anti-interference capabilities; and it can directly output standardized data, reducing data cleaning and warehousing costs.

[0026] In one specific embodiment, step S1 includes: The process involves acquiring the ophthalmic biometric report (PDF format) to be processed and converting it into a page-by-page report image. Global OCR recognition is then performed on the entire report image to extract the patient's basic information fields for future reference. These basic information fields include at least name, patient ID, date of birth, and examination date. Further, extended fields such as calculation formulas, target refractive error, surgeon's information, and corneal refractive index parameter n can be extracted. To improve robustness to mixed Chinese and English text and OCR errors, a multi-modal regularization matching strategy is employed for field extraction. For example, different keyword and date format matching branches are set for name, ID, date of birth, and examination date, ensuring stable output even when text is missing or its order changes.

[0027] In one specific embodiment, step S2 includes: S21: After converting the report image to grayscale, perform adaptive thresholding binarization to obtain a binary image.

[0028] First, the report image is converted to grayscale and then subjected to adaptive threshold binarization (color inversion) to make the table lines and text appear as the foreground.

[0029] S22: Construct horizontal and vertical structuring elements, and perform erosion and dilation processing on the binary image respectively to obtain horizontal and vertical lines.

[0030] The horizontal structural element uses a rectangular structure with a width greater than its height, and its size is set to ( ). ,1); Vertical structural elements adopt a rectangular structure with a height greater than its width, and the size is set to (1, ).in, Width of the horizontal structural element ， Based on the width W of the binary image and The ratio is determined. Height of the vertical structural element ， Based on the height H of the binary image and The ratios are determined and can be expressed as follows: , ; This indicates rounding down to the nearest integer.

[0031] in, These are the line structure scale parameters, used to control the lengths of horizontal and vertical structural elements; The smaller the value, the longer the structural element is constructed, which is more conducive to extracting long table lines. The settings can be adaptively adjusted based on the input image resolution, table density, and page width and height; for report images with a more regular layout, preset empirical values ​​can be used. A value between 15 and 50 is preferred.

[0032] Subsequently, horizontal and vertical structuring elements are used to erode and dilate the binary image to enhance horizontal and vertical line segments and suppress irrelevant noise.

[0033] S23: Overlay the horizontal lines and vertical lines to obtain a table grid mask.

[0034] The extracted horizontal and vertical lines are superimposed to obtain a table grid mask; to connect broken line segments, a closing operation can be performed on the table grid mask.

[0035] S24: Perform contour detection on the table grid mask to obtain multiple candidate cells.

[0036] Contour detection is performed on the table grid mask, each closed rectangular region is treated as a candidate cell, and its bounding rectangle is calculated.

[0037] S25: Filter and deduplicate multiple candidate cells based on area threshold and intersection-union comparison to obtain a set of cells.

[0038] Noisy candidate boxes are filtered based on the minimum area threshold min_cell_area and minimum width and height thresholds min_w and min_h. To avoid interference from large boxes such as "table outlines", candidate boxes with areas exceeding a preset proportion of the maximum area (e.g., 70%) are further removed based on the area distribution of the candidate boxes. Then, lightweight deduplication is performed based on the Intersection over Union (IoU). If the IoU of two candidate boxes exceeds a threshold (e.g., 0.85), they are identified as duplicate boxes and removed.

[0039] In one specific embodiment, step S3 includes: The cells obtained in step S2 are sorted into a grid by rows and columns, and each cell area is cropped to generate the corresponding cell image. At the same time, an intermediate result file JSON containing cell row and column indices and coordinate size information is generated, and a debugging annotation image is generated: the bounding rectangle of each cell and its row and column numbers are drawn on the original image for manual review and regression testing.

[0040] In one specific embodiment, step S4 includes: Cell-level OCR recognition is performed on each cell image generated in step S3, and the recognized text is written back to the corresponding cell. The original OCR result for each cell can also be saved for later debugging. By employing a dual-channel recognition strategy of "global OCR extraction of key fields + cell-level OCR extraction of table fields," the risk of missed recognition and crosstalk caused by relying solely on a single OCR region can be reduced.

[0041] In one specific embodiment, step S5 includes: S51: Traverse each cell in the two-dimensional grid structure and obtain the corresponding cell text.

[0042] S52: If the cell text contains a predefined delimiter, then the cell text is split into a label substring and a numeric substring, with the first matched delimiter as the boundary, and the corresponding cell is marked as a cell to be expanded.

[0043] Within a two-dimensional grid structure, detect cells containing colons and split them into "label substrings" and "numeric substrings" with the first colon as the boundary, marking the corresponding cells as cells to be expanded.

[0044] If the cell text does not contain a predefined separator, the cell content remains unchanged.

[0045] S53: After completing the traversal and marking of all cells, perform a logical column expansion operation on the two-dimensional grid structure: For each row, if there is a cell to be expanded, insert a new logical column after the logical column index of the cell to be expanded, keep the label substring in the cell corresponding to the original logical column index, and fill the numeric substring into the cell corresponding to the newly inserted logical column.

[0046] The "insert new logical column" step in this process involves logically expanding the two-dimensional grid structure, rather than physically cropping the cell images. Therefore, for marked cells, the coordinates of their bounding rectangle in the cell image do not need to be changed; only the logical column indexes in the two-dimensional grid structure need to be updated. Specifically, after inserting a new column at the nth column position, the logical column indices of the original (n+1)th column and all subsequent cells are incremented by 1 sequentially, while their cell image coordinates remain unchanged.

[0047] S54: For other cells in the current row that have not undergone cell text splitting, if the original logical column index is greater than or equal to the original logical column index of the cell to be expanded, then increment the original logical column index by 1 in sequence to obtain the preprocessed two-dimensional grid structure.

[0048] In one specific embodiment, step S6 includes: S61: Normalize the cell text in the preprocessed two-dimensional grid structure, and map the right-eye markers of various expressions to a unified right-eye anchor point, and map the left-eye markers of various expressions to a unified left-eye anchor point.

[0049] First, normalize the cell text by mapping OD, Right, right, right, and right eye to the right eye anchor point, and mapping OS, Left, left, and left eye to the left eye anchor point.

[0050] S62: In the preprocessed 2D grid structure, retrieve the cells containing the right eye anchor point and the left eye anchor point to obtain the eye-specific anchor point cells; the eye-specific anchor point cells include the right eye anchor point cells and the left eye anchor point cells.

[0051] In the preprocessed two-dimensional grid structure, cells containing the above anchor points are retrieved, and cells located in the header row or near the header area are preferentially selected as eye anchor point cells.

[0052] S63: Determine the eye data range based on the eye anchor point cell.

[0053] When both right-eye and left-eye anchor cells are detected simultaneously, the midline of their horizontal center coordinates is used as the dividing line between the two eyes. The column range on one side of the dividing line is defined as the right-eye data region, and the column range on the other side is defined as the left-eye data region.

[0054] When only one side of the anchor cell is detected, the other eye region can be inferred by combining the center line of the page, the overall left and right layout of the table, or the symmetry of the other side area; when there are multiple anchor cells of the same type, the optimal anchor cell can be selected to complete the eye region division by combining factors such as OCR confidence, row position and distance from the table header.

[0055] S65: Within the eye-specific data area, extract the corresponding conventional biometric values ​​based on the predefined mapping relationship between field labels and column offsets. Conventional biometrics include at least axial length, corneal curvature, anterior chamber depth, and white-to-white distance.

[0056] S66: Fill the values ​​of routine biometric indicators into the corresponding key-value pairs in the data structure template according to the preset eye identifier and field name to generate the initial structured record.

[0057] In one specific embodiment, step S7 specifically includes: S71: Obtain the cell text in the cell corresponding to the artificial crystal calculation parameter area and construct a text sequence.

[0058] S72: Detect adjacent text segments in the same line of a text sequence with a horizontal spacing of less than a preset threshold. If the preceding text segment ends with a non-numeric character and the following text segment begins with a number or a decimal point, then concatenate the two text segments to obtain the repaired numerical sequence.

[0059] The text in the cells corresponding to the artificial lens calculation parameter area is directly scanned to locate the A-Const constant (e.g., 117.x to 120.x), and the numerical sequence is extracted only from the text fragment containing the IOL header. Token merging is performed when the OCR splits a number into multiple fragments (e.g., "-0" is separated from ".11", or "-1.47" is broken).

[0060] S73: Arrange the repaired numerical sequence in order, with each pair of consecutive numerical values ​​forming a numerical pair.

[0061] S74: Based on the preset numerical range constraints, filter the numerical pairs to obtain a set of valid numerical pairs.

[0062] S75: Generate a list of artificial lens parameters based on the set of valid numerical values.

[0063] Based on the relative position of the horizontal coordinate of each cell containing a valid value in the table to the center line, the valid value pairs are classified as right-eye data or left-eye data, and a list of intraocular lens parameters is generated by sequentially numbering them.

[0064] S76: Add the list of artificial lens parameters as a new field and merge it into the initial structured record to obtain the final structured record.

[0065] In one specific embodiment, step S8 specifically includes: The final structured records are written to the output file according to the target database template. For templates with a fixed column order, a dictionary to be written can be generated based on the template column names, and unit conversion and field placement can be completed. When the template is missing some fields, they can be written to the extended field set and the new columns can be appended to the end of the output table, thereby completing field expansion without breaking the compatibility of the existing template. At the same time, append writing in batch processing mode (append_mode) is supported to avoid repeatedly writing to the table header.

[0066] Compared with the prior art, the method provided in this application has the following significant technical effects: 1. Data processing efficiency is significantly improved, supporting batch automated production of structured databases.

[0067] Description of effects: This application enables end-to-end fully automated processing of ophthalmic biometric reports, transforming unstructured data into structured data, completely replacing traditional manual data entry, proofreading, copying and pasting, and verification processes. This not only eliminates the bottleneck of manual operation but also significantly reduces labor and time costs. It can efficiently handle the needs of large-scale historical data migration and batch data entry of new data, significantly improving the productivity of clinical data mining and scientific research data construction.

[0068] Technology source: Step S1: Quickly extract key metadata such as basic patient information through global OCR, avoiding manual page-by-page review.

[0069] Steps S2-S3: Through automated table grid partitioning and cell positioning, the entire report is intelligently broken down into a set of independently processable cells, forming standardized "batch-processable" input units.

[0070] Steps S4, S6, and S8: Through cell-level OCR recognition, automatic field parsing, and templated output, structured records that meet database requirements are directly generated without the need for manual secondary processing.

[0071] Reason for this: This application automatically transforms complex "full-page unstructured images" into a "two-dimensional table grid + field dictionary" data stream that can be continuously processed by a computer, realizing a closed-loop automation of the entire process.

[0072] 2. Field extraction accuracy and stability are higher, effectively overcoming layout differences and OCR noise interference.

[0073] Results Description: This application maintains an extremely high success rate and consistency in field extraction even when faced with ophthalmology reports exported from different devices, with different versions, interference from gray bars / watermarks, or poor scan quality. It effectively solves the problems of string interference and missed recognition caused by complex backgrounds in general full-page OCR, as well as the problem of large-scale extraction failures caused by minor changes in the layout of fixed coordinate templates.

[0074] Technology source: Step S2: The table lines are extracted using morphological operations based on adaptive structuring elements, and the broken line segments are connected by closing operations to enhance the robustness of cell positioning; the interference of the entire table's outer frame is eliminated through a large frame filtering mechanism; and duplicate candidate boxes are eliminated through an IoU threshold deduplication mechanism, ensuring the accuracy of cell positioning from the source.

[0075] Steps S1+S4: Employ a dual-channel recognition strategy of "global OCR + cell OCR". Global OCR is responsible for extracting scattered patient information, while cell OCR focuses on fields within the table. The two complement each other, avoiding missed recognition or contextual interference under a single path.

[0076] Cause: By employing the strategy of "first structured positioning (grid / cell), then local recognition (cell OCR)," the OCR recognition task is decoupled from the complex full-page background, isolating visual interference from adjacent fields, thereby significantly reducing systematic errors caused by misalignment, omission, and repeated recognition.

[0077] 3. It has adaptive parsing capabilities for scenarios where "labels and values ​​are linked", improving the accuracy of field placement.

[0078] Effect Description: In response to the common problem of "label:value" text being copied and pasted in the same field in ophthalmology reports (such as "AL:23.70"), this application can automatically split it and accurately place it in the standard field, avoiding regular expression matching failures or field mismatches caused by missing colons, abnormal spaces, or character copying, and significantly improving the compatibility of cross-version and cross-format reports.

[0079] Technology source: Step S5: A semantic expansion mechanism was designed to detect cells containing delimiters, split the "label substring" and "numerical substring", and insert a new column in the two-dimensional grid structure to build a stable "cell to be expanded - adjacent numerical cell" structure.

[0080] Step S8: Map the split stable structure to a unified standard template field.

[0081] Reason for this: By converting unstructured concatenated strings into structured key-value adjacency relationships, subsequent field parsing no longer relies on unstable regular expression guessing, but is extracted based on definite grid position relationships, thereby greatly improving the accuracy of placement.

[0082] 4. The distinction between left and right eye (OD / OS) data is more reliable, completely solving the problems of cross-column and confusion.

[0083] Effect description: This application can reliably identify and distinguish the data areas of the left eye (OS) and the right eye (OD). Even with different report layouts (such as left and right interchange, increase or decrease in the number of columns), it can ensure that biometric indicators (such as AL, K1, ACD, etc.) are accurately classified under the corresponding eye, which greatly reduces serious errors caused by left and right eye confusion in clinical data analysis.

[0084] Technology source: Step S6: Support OD / OS multi-anchor point compatible parsing (such as OD / Right / Right normalization), and based on the grid spatial position, use the midline of the horizontal center of the anchor point to divide the left and right eye data regions, and extract data by column range.

[0085] Steps S3 and S6: Field location is performed based on the reconstructed two-dimensional grid row and column structure, rather than relying on the reading order of the global text.

[0086] Cause: The key to eye-detection lies in the "consistency of spatial location". This application uses the physical coordinates of cells and logical row and column structure to rigidly separate the eye-detection area, which is less affected by changes in layout than methods that rely solely on the global OCR text order.

[0087] 5. Special values ​​such as IOL / REF are more robust for extraction, effectively repairing broken numbers and pairing anomalies.

[0088] Effect Description: Addressing common OCR numerical breakage issues (such as separation of the minus sign and digit, and decimal point breakage) and multi-value pairing problems in intraocular lens (IOL) calculation parameters, this application can correctly recover numerical values ​​and form stable (IOL constant, REF refractive power) numerical pairs, improving the integrity and usability of key clinical data.

[0089] Technology source: Step S7: Adopt a three-level processing strategy: First, repair the split digital fragments through the token merging mechanism; second, pair them in order and perform numerical range constraint filtering to remove outliers; finally, classify the data to the corresponding eye based on the grid centerline.

[0090] Step S7: Use key anchor points such as A-Const to limit the search area and avoid accidentally selecting irrelevant numbers.

[0091] Cause: By using a dedicated algorithm that first corrects and restores the numbers, then performs structured pairing, and then performs constraint filtering, the random noise generated by OCR is transformed into a controllable processing flow, which significantly improves the success rate of extracting complex numerical fields.

[0092] 6. Enhanced traceability and quality control capabilities facilitate the long-term maintenance and review of research datasets.

[0093] Effect description: This application not only outputs the final structured data, but also generates traceable intermediate products in sync, which makes it easier for researchers to find the source of errors, conduct sampling verification, and manage the version of model training data, thereby improving data governance capabilities and the reproducibility of research results.

[0094] Technology source: Step S3: Output cell images to provide visual evidence; generate debugging annotation images with cell boxes and numbers drawn on the cell images to facilitate intuitive location of partitioning issues.

[0095] Steps S3 and S4: Save the intermediate JSON file containing the cell coordinates, row and column relationships, and parsing results.

[0096] Cause: Traditional solutions often only produce the final table, making it difficult to troubleshoot errors. This application saves end-to-end evidence of "original coordinates - cell image - OCR text - parsing result," enabling each structured record to be traced back to a specific area of ​​the original image, thus achieving transparent management of data quality.

[0097] Based on the same inventive concept, this application also provides a system for implementing the above-described automated parsing method for ophthalmic biometric reports. The solution provided by this system is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the automated parsing system for ophthalmic biometric reports provided below can be found in the limitations of the automated parsing method for ophthalmic biometric reports described above, and will not be repeated here.

[0098] In one exemplary embodiment, an automated analysis system for ophthalmic biometric reports is provided, comprising the following modules.

[0099] The report processing module is used to acquire ophthalmic biometric reports to be processed and convert them into report images.

[0100] The cell set determination module is used to partition the report image into a table grid and locate the cells to obtain a cell set.

[0101] The cell image generation module is used to sort all cells in the cell set by rows and columns in a grid, establish a two-dimensional grid structure, and crop each cell in the two-dimensional grid structure to generate a cell image.

[0102] The OCR recognition module is used to perform OCR recognition on each cell image and fill the recognized text into the corresponding cell in the two-dimensional grid structure to obtain the cell text.

[0103] The preprocessing module is used to perform structured preprocessing and semantic expansion on the cell text in each cell to obtain a preprocessed two-dimensional grid structure.

[0104] The initial structure generation module is used to perform field parsing and normalization mapping based on the preprocessed two-dimensional grid structure to generate initial structured records.

[0105] The final structured record determination module is used to extract numerical values ​​from the artificial crystal calculation parameter region in the preprocessed two-dimensional mesh structure and write the extraction results into the initial structured record to obtain the final structured record.

[0106] The write module is used to write the final structured records to the output file according to the target database template.

[0107] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments. The computer device may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device stores data to be processed. The I / O interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with an external terminal via a network connection. When the computer program is executed by the processor, it implements the steps in the above-described method embodiments.

[0108] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0109] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0110] 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 relevant regulations.

[0111] 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. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, 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).

[0112] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may 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 to these.

[0113] 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.

[0114] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An ophthalmic biometry report automated parsing method, characterized in that, include: Acquire the ophthalmic biometric report to be processed and convert it into a report image; The report image is divided into table grids and positioned by cells to obtain a set of cells; Sort all cells in the cell set by row and column to create a two-dimensional grid structure, and then crop each cell in the two-dimensional grid structure to generate a cell image; Perform OCR recognition on each cell image, and fill the recognized text into the corresponding cell in the two-dimensional grid structure to obtain the cell text; The text in each cell is preprocessed in a structured manner and semantically expanded to obtain a preprocessed two-dimensional grid structure. Based on the preprocessed two-dimensional grid structure, field parsing and normalization mapping are performed to generate initial structured records; Numerical extraction is performed on the artificial crystal computational parameter region in the preprocessed two-dimensional mesh structure, and the extraction results are written into the initial structured record to obtain the final structured record. The final structured records are written to the output file according to the target database template.

2. The ophthalmic biometry report automated parsing method of claim 1, wherein, Also includes: Perform global OCR recognition on the report image to extract the patient's basic information fields.

3. The ophthalmic biometry report automated parsing method of claim 1, wherein, The report image is divided into table grids and cells are positioned to obtain a set of cells, specifically including: The report image is converted to grayscale and then subjected to adaptive threshold binarization to obtain a binary image; Construct horizontal and vertical structuring elements, and perform erosion and dilation processing on the binary image respectively to obtain horizontal and vertical lines. By overlaying the horizontal and vertical lines, a table grid mask is obtained; Perform contour detection on the table grid mask to obtain multiple candidate cells; The cell set is obtained by filtering and deduplicating multiple candidate cells based on area thresholds and intersection-union comparison.

4. The ophthalmic biometry report automated parsing method of claim 1, wherein, Perform OCR recognition on each cell image and fill the recognized text into the corresponding cell in the two-dimensional grid structure to obtain the cell text, specifically including: Iterate through each cell in the two-dimensional grid structure and obtain the corresponding cell text; If the cell text contains a predefined delimiter, the cell text is split into a tag substring and a numeric substring, with the first matched delimiter as the boundary, and the corresponding cell is marked as a cell to be expanded. After completing the traversal and marking of all cells, a logical column expansion operation is performed on the two-dimensional grid structure: for each row, if there are cells to be expanded, a new logical column is inserted after the logical column index of the cell to be expanded, the label substring is kept in the cell corresponding to the original logical column index, and the numerical substring is filled into the cell corresponding to the newly inserted logical column. For other cells in the current row that have not undergone cell text splitting, if the original logical column index is greater than or equal to the original logical column index of the cell to be expanded, then the original logical column index is incremented by 1 sequentially to obtain the preprocessed two-dimensional grid structure.

5. The ophthalmic biometry report automation parsing method of claim 1, wherein, Based on the preprocessed two-dimensional grid structure, field parsing and normalization mapping are performed to generate initial structured records, specifically including: Normalize the cell text in the preprocessed two-dimensional grid structure, and map the right-eye markers of various expressions to a unified right-eye anchor point, and map the left-eye markers of various expressions to a unified left-eye anchor point; In the preprocessed 2D mesh structure, cells containing right eye anchor points and left eye anchor points are retrieved to obtain eye-specific anchor point cells; eye-specific anchor point cells include right eye anchor point cells and left eye anchor point cells; Determine the eye-specific data range based on the eye-specific anchor point cells; Within the eye-specific data area, based on the mapping relationship between predefined field labels and column offsets, the corresponding conventional biometric values ​​are extracted. Conventional biometrics include at least axial length, corneal curvature, anterior chamber depth, and white-to-white distance. The values ​​of routine biometric indicators are filled into the corresponding key-value pairs in the data structure template according to the preset eye identifier and field name to generate the initial structured record.

6. The automated analysis method for ophthalmic biometric reports according to claim 1, characterized in that, Numerical extraction is performed on the artificial crystal computational parameter region within the preprocessed two-dimensional mesh structure, and the extraction results are written into the initial structured record to obtain the final structured record, which specifically includes: Obtain the cell text in the corresponding cell of the artificial crystal calculation parameter area and construct a text sequence; Detect adjacent text segments in the same line of a text sequence with a horizontal spacing of less than a preset threshold. If the preceding text segment ends with a non-numeric character and the following text segment begins with a number or a decimal point, then concatenate the two text segments to obtain the repaired numerical sequence. Arrange the repaired numerical sequence in order, with each pair of consecutive numerical values ​​forming a numerical pair; Based on preset numerical range constraints, the numerical pairs are filtered to obtain a set of valid numerical pairs; A list of artificial lens parameters is generated based on the set of valid values; The list of artificial lens parameters is added as a new field and merged into the initial structured record to obtain the final structured record.

7. An automated analysis system for ophthalmic biometric reports, characterized in that, include: The report processing module is used to acquire ophthalmic biometric reports to be processed and convert them into report images. The cell set determination module is used to perform table grid partitioning and cell positioning on the report image to obtain a cell set; The cell image generation module is used to sort all cells in the cell set by rows and columns in a grid, establish a two-dimensional grid structure, and crop each cell in the two-dimensional grid structure to generate a cell image. The OCR recognition module is used to perform OCR recognition on each cell image and fill the recognized text into the corresponding cell in the two-dimensional grid structure to obtain the cell text; The preprocessing module is used to perform structured preprocessing and semantic expansion on the cell text in each cell to obtain a preprocessed two-dimensional grid structure. The initial structure generation module is used to perform field parsing and normalization mapping based on the preprocessed two-dimensional grid structure to generate initial structured records; The final structured record determination module is used to extract numerical values ​​from the artificial crystal calculation parameter region in the preprocessed two-dimensional mesh structure and write the extraction results into the initial structured record to obtain the final structured record. The write module is used to write the final structured records to the output file according to the target database template.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for automated parsing of ophthalmic biometric reports according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for automated parsing of ophthalmic biometric reports as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for automated parsing of ophthalmic biometric reports as described in any one of claims 1-6.

Images