Prescription image structured information extraction and correction method, system and program product
By combining image preprocessing and medical multimodal large models with medical knowledge graphs, the problems of recognition errors and privacy security of paper prescriptions are solved. This approach achieves high-accuracy extraction and error correction of structured prescription information, adapts to different layout methods of medical institutions, and ensures data compliance and privacy security.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU ZIJIELIU TECH CO LTD
- Filing Date
- 2026-06-09
- Publication Date
- 2026-07-14
AI Technical Summary
Existing technologies for processing paper prescriptions suffer from defects in the visual perception layer (such as errors caused by cursive writing, uneven lighting, or stamp obstruction), a lack of medical knowledge in the logical cognition layer leading to difficulties in error correction, poor structural generalization ability, and high risks to data compliance and privacy security.
Image preprocessing, dynamic layout recognition, and target semantic polygon region segmentation are employed, combined with a medical multimodal large model and knowledge graph for text extraction and error correction. Entity node matching and attribute correction are performed through a medical prescription knowledge graph to ensure the anonymization of privacy data.
It achieves high-accuracy prescription recognition and error correction, adapts to different layout methods, ensures data compliance and privacy security, and eliminates the potential medical accidents caused by recognition errors.
Smart Images
Figure CN122392079A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer vision technology, specifically relating to a method, system, and program product for extracting and correcting structured information from prescription images. Background Technology
[0002] Currently, medical institutions and third-party platforms primarily rely on optical character recognition (OCR) technology and layout analysis based on predefined templates for the digitization of paper prescriptions, along with simple post-processing corrections of typos using general-purpose language models. These traditional methods have the following shortcomings: 1. Deficiencies in the visual perception layer (inaccuracy): Some doctors' handwritten prescriptions may contain cursive writing, mixed Chinese and English characters, and the collected prescription images are often accompanied by creases, uneven lighting, and overlapping or obscuring stamps. Traditional optical character recognition (OCR) technology is prone to producing a large number of garbled characters or refusing to recognize them when dealing with these situations.
[0003] 2. Deficiencies in logical cognition (lack of medical knowledge): Traditional technologies lack prior knowledge in the medical field, only extracting the literal meaning and failing to identify and correct errors in medical values or names based on prior knowledge. For example, traditional OCR is very likely to misidentify "amoxicillin 0.25g" as "amoxicillin 0.25q", or "potassium chloride 10ml" as "10ml".
[0004] 3. Deficiencies in structural generalization ability (difficult to adapt): Prescription layouts may vary between different hospitals and platforms. Traditional layout analysis methods based on fixed coordinate templates or simple rules will completely fail when dealing with different prescription layouts, resulting in extremely poor generalization ability.
[0005] 4. Data Compliance and Privacy Security Risks (Easily Leaked): Prescription images may contain highly sensitive information such as patient names, ages, and diagnoses. Existing identification and extraction methods typically transmit the raw image information directly to the backend, which can easily lead to privacy leaks. Summary of the Invention
[0006] The purpose of this invention is to provide a method, system, and program product for extracting and correcting structured information from prescription images, in order to solve the aforementioned problems existing in the prior art.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, it provides methods for extracting and correcting structured information from prescription images, including: Acquire a clean prescription image transmitted by an image acquisition device, wherein the clean prescription image is obtained by preprocessing an initial prescription image acquired by the image acquisition device; Layout recognition is performed on the clean prescription image to identify the target semantic polygon region in the clean prescription image, and the target semantic polygon region is segmented from the clean prescription image; The target semantic polygon region is input into a pre-set medical multimodal large model for text extraction to obtain the initial structured text; The entity names and corresponding prescription attribute information in the initial structured text are determined, and the entity names are vector-encoded to obtain entity feature vectors. The entity name and its entity feature vector are substituted into a pre-set medical prescription knowledge graph to perform entity node matching, and the matching entity node and the standard entity name and standard prescription attribute information corresponding to the matching entity node are determined. Replace the corresponding entity names in the initial structured text with standard entity names, and correct the prescription attribute information of the corresponding entity names in the initial structured text using standard prescription attribute information to obtain the corrected structured text. The corrected structured text is converted into a standard format and transmitted to the hospital information system.
[0008] In one possible design, the method further includes, before converting the corrected structured text into standard format data: Search for the compatibility and incompatibilities between entity nodes corresponding to each standard entity name in the corrected structured text of the medical prescription knowledge graph; When a mismatch taboo relationship is found between the entity nodes corresponding to the corresponding standard entity name, the corresponding standard entity name is marked with a mismatch risk label in the corrected structured text.
[0009] In one possible design, the image preprocessing includes: A lightweight convolutional neural network is used to locate privacy regions in an initial prescription image, and the privacy regions include personal identification information regions. The privacy region of the location is obfuscated with pixels to obtain the prescription image after pixel obfuscation. Image correction processing is performed on the prescription image after pixel obfuscation to obtain a clean prescription image.
[0010] In one possible design, the layout recognition of the clean prescription image to determine the target semantic polygon region in the clean prescription image includes: A pre-trained LayoutLM model is used to perform layout recognition on a clean prescription image to determine each semantic polygon region in the clean prescription image. The semantic polygon regions include the header area, diagnosis area, medication details area, or signature area. The medication details area is used as the target semantic polygon region.
[0011] In one possible design, the step of inputting the target semantic polygon region into a pre-set medical multimodal large model for text extraction to obtain initial structured text includes: The target semantic polygon region and the preset Prompt words are input into a pre-set medical multimodal large model. The Prompt words guide the medical multimodal large model to perform text recognition and extraction on the target semantic polygon region to obtain initial structured text. The initial structured text contains structured entity names and their corresponding prescription attribute information.
[0012] In one possible design, the medical prescription knowledge graph contains several entity nodes and prescription attribute nodes that are topologically connected to each entity node. Each entity node is associated with a corresponding standard entity name and a standard entity feature vector, and the prescription attribute nodes are associated with corresponding standard prescription attribute information. The step of substituting the entity name and its entity feature vector into the pre-set medical prescription knowledge graph for entity node matching to determine the matching entity node and the corresponding standard entity name and standard prescription attribute information includes: The entity feature vector of the entity name is compared with the standard entity feature vector of each entity node. The entity node with the highest similarity between the corresponding standard entity feature vector and the entity feature vector is taken as the matching entity node. Determine the standard entity name corresponding to the matching entity node and the standard prescription attribute information corresponding to the prescription attribute node connected to the topology of the matching entity node.
[0013] In one possible design, the step of correcting the prescription attribute information corresponding to the entity name in the initial structured text using standard prescription attribute information includes: The standard prescription attribute information is used to correct errors and / or complete missing content in the prescription attribute information corresponding to the entity names in the initial structured text.
[0014] Secondly, a system for extracting and correcting structured information from prescription images is provided, comprising an image acquisition unit, a layout segmentation unit, a text extraction unit, a feature encoding unit, a map matching unit, an information correction unit, and a data transmission unit, wherein: The image acquisition unit is used to acquire a clean prescription image transmitted by the image acquisition device. The clean prescription image is obtained by preprocessing the initial prescription image acquired by the image acquisition device. The layout segmentation unit is used to perform layout recognition on the clean prescription image, determine the target semantic polygon region in the clean prescription image, and segment the target semantic polygon region from the clean prescription image; The text extraction unit is used to input the target semantic polygon region into a pre-set medical multimodal large model for text extraction to obtain the initial structured text; The feature encoding unit is used to determine the entity name and the prescription attribute information corresponding to the entity name in the initial structured text, and to perform vector encoding on the entity name to obtain the entity feature vector; The graph matching unit is used to substitute the entity name and its entity feature vector into the preset medical prescription knowledge graph to perform entity node matching, and determine the matching entity node as well as the standard entity name and standard prescription attribute information corresponding to the matching entity node. The information correction unit is used to replace the corresponding entity name in the initial structured text with the standard entity name, and to correct the prescription attribute information of the corresponding entity name in the initial structured text with the standard prescription attribute information, so as to obtain the corrected structured text. The data transmission unit is used to convert the corrected structured text into a standard format and transmit it to the hospital information system.
[0015] Thirdly, a system for extracting and correcting structured information from prescription images is provided, including: Memory, used to store instructions; A processor is configured to read instructions stored in the memory and execute the method described in any one of the first aspects above, according to the instructions.
[0016] Fourthly, a computer-readable storage medium is provided, on which instructions are stored, which, when executed on a computer, cause the computer to perform any of the methods described in the first aspect. A computer program product is also provided, which, when executed on a computer, performs any of the methods described in the first aspect.
[0017] Beneficial effects: This invention effectively ensures the confidentiality of sensitive medical data and guarantees compliance and privacy security across the entire data chain through a collaborative processing mechanism of desensitized and parsed prescription images; dynamic layout recognition and target semantic polygon region segmentation eliminate the need for manual template annotation, adapting to the prescription layout methods of various medical institutions; the use of a multimodal large model for structured text extraction abandons the traditional "character segmentation + sequence recognition" path, significantly improving the accuracy of prescription recognition; deep integration with the medical prescription knowledge graph solves the problem of existing technologies "only recognizing characters, not understanding medicine, and outputting incorrectly"; and it possesses powerful error correction capabilities for extracted prescription content, achieving logical self-correction similar to a medical brain, eliminating the potential for major medical accidents caused by prescription recognition errors. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of the steps in the method of Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the system configuration in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the system configuration in Embodiment 3 of the present invention. Detailed Implementation
[0020] It should be noted that the descriptions of these embodiments are intended to aid in understanding the invention and do not constitute a limitation thereof. The specific structural and functional details disclosed herein are merely for describing exemplary embodiments of the invention. However, the invention may be embodied in many alternative forms and should not be construed as being limited to the embodiments described herein.
[0021] It should be understood that, unless otherwise explicitly specified and limited, the corresponding terms should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments according to the specific circumstances.
[0022] Specific details are provided in the following description to provide a complete understanding of the exemplary embodiments. However, those skilled in the art will understand that the exemplary embodiments can be implemented without these specific details. For example, apparatus may be shown in block diagrams to avoid obscuring the examples with unnecessary details. In other embodiments, well-known processes, structures, and techniques may be omitted with non-essential details to avoid obscuring the embodiments.
[0023] Example 1: This embodiment provides a method for extracting and correcting structured information from prescription images, which can be applied to corresponding image processing systems, such as... Figure 1 As shown, the method includes the following steps: S1. Acquire a clean prescription image transmitted by an image acquisition device, wherein the clean prescription image is obtained by preprocessing an initial prescription image acquired by the image acquisition device.
[0024] In practice, an image acquisition device can be used to acquire images of the corresponding prescription (such as a paper prescription) to obtain an initial prescription image. Then, a pre-deployed lightweight convolutional neural network (such as a YOLO-Lite network) is used to locate privacy regions (i.e., target region detection) in the initial prescription image, identifying privacy regions in the initial prescription image. These privacy regions include personal identification information regions, such as "name" and "ID number." The located privacy regions are then subjected to pixel obfuscation processing to obtain a pixel-obfuscated prescription image. Finally, a generative adversarial network combined with a spatial transformation network is used to perform image correction processing on the pixel-obfuscated prescription image, mapping deformed, distorted, or stamp-obscured prescription images to standard orthogonal clean prescription images. The image acquisition device uploads the pre-processed clean prescription image to an image processing system for subsequent processing.
[0025] S2. Perform layout recognition on the clean prescription image, determine the target semantic polygon region in the clean prescription image, and segment the target semantic polygon region from the clean prescription image.
[0026] In practical implementation, the image processing system can use a pre-trained LayoutLM model to perform layout recognition on the clean prescription image, identifying various semantic polygon regions in the clean prescription image. These semantic polygon regions include the header area, diagnosis area, medication details area, or signature area. Then, the medication details area is used as the target semantic polygon region. The LayoutLM model is based on the core idea of ViT (Vision Transformer) and does not rely on any fixed template. It can divide the clean prescription image into semantic polygon regions such as the header area, diagnosis area, medication details area, and signature area, and extract the high-value medication details area.
[0027] S3. Input the target semantic polygon region into a pre-set medical multimodal large model for text extraction to obtain the initial structured text.
[0028] In practice, the image processing system can input the target semantic polygon region and the preset Prompt words into a pre-set medical multimodal large model. The Prompt words guide the medical multimodal large model (based on the cross-attention mechanism) to perform text recognition and extraction on the target semantic polygon region to obtain initial structured text. The initial structured text contains structured entity names and their corresponding prescription attribute information.
[0029] S4. Determine the entity names and corresponding prescription attribute information in the initial structured text, and perform vector encoding on the entity names to obtain entity feature vectors.
[0030] In practice, the image processing system extracts entity names and corresponding prescription attribute information from the initial structured text, and then performs vector encoding on the entity names (such as using one-hot encoding or word embedding encoding) to obtain entity feature vectors. The entity names include drug names, and the prescription attribute information includes drug specifications, standard single dose, route of administration, and frequency of administration.
[0031] S5. Substitute the entity name and its entity feature vector into the preset medical prescription knowledge graph to perform entity node matching, and determine the matching entity node as well as the standard entity name and standard prescription attribute information corresponding to the matching entity node.
[0032] In practice, the image processing system can pre-construct a medical prescription knowledge graph based on existing standard entity names and standard prescription attribute information. This graph includes multi-level topological relationships such as drug name, drug specifications, routine single dose, route of administration, and frequency of administration. In other words, the medical prescription knowledge graph contains several entity nodes and prescription attribute nodes that are topologically connected to each entity node. Each entity node is associated with a corresponding standard entity name and standard entity feature vector, and the prescription attribute nodes are associated with corresponding standard prescription attribute information.
[0033] In matching applications, the image processing system calculates the vector similarity between the entity feature vector of the entity name and the standard entity feature vector of each entity node. The entity node with the highest similarity between its corresponding standard entity feature vector and the entity feature vector is selected as the matching entity node. The system then determines the standard entity name corresponding to the matching entity node and the standard prescription attribute information corresponding to the prescription attribute nodes connected to the topological link of the matching entity node.
[0034] S6. Replace the corresponding entity names in the initial structured text with standard entity names, and correct the prescription attribute information of the corresponding entity names in the initial structured text using standard prescription attribute information to obtain the corrected structured text.
[0035] In practice, the system can use standard prescription attribute information to correct errors and / or complete missing content in the prescription attribute information corresponding to the entity names in the initial structured text. For example, if the system recognizes the entity name "Aspirin" and the prescription attribute information "100g", and traverses the standard dose information corresponding to the "legitimate dose" attribute node of the matching "Aspirin" entity node in the medical prescription knowledge graph, finding that "100g" far exceeds the standard dose of "Aspirin" and does not conform to the specification, the system will automatically correct the dose attribute information to "100mg". Similarly, if a doctor omits a unit (e.g., only writing "cephalosporin 250"), the system can automatically complete the unit based on the default specification of the corresponding drug in the medical prescription knowledge graph (e.g., complete it to "250mg"), and so on. Finally, the corrected structured text is obtained.
[0036] S7. Convert the corrected structured text into a standard format and transmit it to the hospital information system.
[0037] In practice, the system can query the medical prescription knowledge graph to find the incompatibilities between the entity nodes corresponding to each standard entity name in the corrected structured text (the corresponding entity nodes in the medical prescription knowledge graph are also connected by edges representing incompatibilities between the corresponding drugs). When an incompatibility relationship is found between the entity nodes corresponding to the standard entity name, the system can label the corresponding standard entity name with a compatibility risk label in the corrected structured text. Finally, the system converts the corrected and labeled structured text into standard format data (such as standard JSON format data) and transmits the standard format data to the Hospital Information System (HIS) for display.
[0038] This method effectively ensures the confidentiality of sensitive medical data and guarantees compliance and privacy security across the entire data chain through a collaborative processing mechanism of desensitized and parsed prescription images. Dynamic layout recognition and target semantic polygon region segmentation eliminate the need for manual template annotation, adapting to various prescription layout methods across different medical institutions. Utilizing a multimodal large-scale model for structured text extraction abandons the traditional "character segmentation + sequence recognition" approach, significantly improving prescription recognition accuracy. Deep integration with a medical prescription knowledge graph solves the problem of existing technologies "only recognizing characters, not understanding medical principles, and outputting incorrect information." Furthermore, it possesses powerful error correction capabilities for extracted prescription content, achieving logical self-correction similar to a medical brain, eliminating the potential for major medical accidents caused by prescription recognition errors.
[0039] Example 2: This embodiment provides a system for extracting and correcting structured information from prescription images, such as... Figure 2 As shown, it includes an image acquisition unit, a page segmentation unit, a text extraction unit, a feature encoding unit, a graph matching unit, an information error correction unit, and a data transmission unit, wherein: The image acquisition unit is used to acquire a clean prescription image transmitted by the image acquisition device. The clean prescription image is obtained by preprocessing the initial prescription image acquired by the image acquisition device. The layout segmentation unit is used to perform layout recognition on the clean prescription image, determine the target semantic polygon region in the clean prescription image, and segment the target semantic polygon region from the clean prescription image; The text extraction unit is used to input the target semantic polygon region into a pre-set medical multimodal large model for text extraction to obtain the initial structured text; The feature encoding unit is used to determine the entity name and the prescription attribute information corresponding to the entity name in the initial structured text, and to perform vector encoding on the entity name to obtain the entity feature vector; The graph matching unit is used to substitute the entity name and its entity feature vector into the preset medical prescription knowledge graph to perform entity node matching, and determine the matching entity node as well as the standard entity name and standard prescription attribute information corresponding to the matching entity node. The information correction unit is used to replace the corresponding entity name in the initial structured text with the standard entity name, and to correct the prescription attribute information of the corresponding entity name in the initial structured text with the standard prescription attribute information, so as to obtain the corrected structured text. The data transmission unit is used to convert the corrected structured text into a standard format and transmit it to the hospital information system.
[0040] Furthermore, the information correction unit is also used to query the incompatibilities between the entity nodes corresponding to each standard entity name in the corrected structured text in the medical prescription knowledge graph; when an incompatibility relationship is found between the entity nodes corresponding to the corresponding standard entity name, the corresponding standard entity name is labeled with a compatibility risk label in the corrected structured text.
[0041] Example 3: This embodiment provides a system for extracting and correcting structured information from prescription images, such as... Figure 3 As shown, at the hardware level, it includes: The data interface is used to establish data communication between the processor, image acquisition equipment, and the hospital information system (HIS). Memory, used to store instructions; The processor is used to read the instructions stored in the memory and execute the prescription image structured information extraction and error correction method in Embodiment 1 according to the instructions.
[0042] Optionally, the system also includes an internal bus, through which the processor, memory, and data interface can be interconnected. This internal bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.
[0043] The memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory. The processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0044] Example 4: This embodiment provides a computer-readable storage medium storing instructions. When these instructions are executed on a computer, the computer performs the prescription image structured information extraction and error correction method described in Embodiment 1. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
[0045] This embodiment also provides a computer program product that, when run on a computer, executes the prescription image structured information extraction and error correction method described in Embodiment 1. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.
[0046] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for extracting and correcting structured information from prescription images, characterized in that, include: Acquire a clean prescription image transmitted by an image acquisition device, wherein the clean prescription image is obtained by preprocessing an initial prescription image acquired by the image acquisition device; Layout recognition is performed on the clean prescription image to identify the target semantic polygon region in the clean prescription image, and the target semantic polygon region is segmented from the clean prescription image; The target semantic polygon region is input into a pre-set medical multimodal large model for text extraction to obtain the initial structured text; The entity names and corresponding prescription attribute information in the initial structured text are determined, and the entity names are vector-encoded to obtain entity feature vectors. The entity name and its entity feature vector are substituted into a pre-set medical prescription knowledge graph to perform entity node matching, and the matching entity node and the standard entity name and standard prescription attribute information corresponding to the matching entity node are determined. Replace the corresponding entity names in the initial structured text with standard entity names, and correct the prescription attribute information of the corresponding entity names in the initial structured text using standard prescription attribute information to obtain the corrected structured text. The corrected structured text is converted into a standard format and transmitted to the hospital information system.
2. The method for extracting and correcting structured information from prescription images according to claim 1, characterized in that, Before converting the corrected structured text into standard format data, the method further includes: Search for the compatibility and incompatibilities between entity nodes corresponding to each standard entity name in the corrected structured text of the medical prescription knowledge graph; When a mismatch taboo relationship is found between the entity nodes corresponding to the corresponding standard entity name, the corresponding standard entity name is marked with a mismatch risk label in the corrected structured text.
3. The method for extracting and correcting structured information from prescription images according to claim 1, characterized in that, The image preprocessing includes: A lightweight convolutional neural network is used to locate privacy regions in an initial prescription image, and the privacy regions include personal identification information regions. The privacy region of the location is obfuscated with pixels to obtain the prescription image after pixel obfuscation. Image correction processing is performed on the prescription image after pixel obfuscation to obtain a clean prescription image.
4. The method for extracting and correcting structured information from prescription images according to claim 1, characterized in that, The step of performing layout recognition on the clean prescription image to determine the target semantic polygon region in the clean prescription image includes: A pre-trained LayoutLM model is used to perform layout recognition on a clean prescription image to determine each semantic polygon region in the clean prescription image. The semantic polygon regions include the header area, diagnosis area, medication details area, or signature area. The medication details area is used as the target semantic polygon region.
5. The method for extracting and correcting structured information from prescription images according to claim 1, characterized in that, The process of inputting the target semantic polygon region into a pre-set medical multimodal large model for text extraction to obtain initial structured text includes: The target semantic polygon region and the preset Prompt words are input into a pre-set medical multimodal large model. The Prompt words guide the medical multimodal large model to perform text recognition and extraction on the target semantic polygon region to obtain initial structured text. The initial structured text contains structured entity names and their corresponding prescription attribute information.
6. The method for extracting and correcting structured information from prescription images according to claim 1, characterized in that, The medical prescription knowledge graph contains several entity nodes and prescription attribute nodes that are topologically connected to each entity node. Each entity node is associated with a corresponding standard entity name and a standard entity feature vector. The prescription attribute nodes are associated with corresponding standard prescription attribute information. The step of substituting the entity name and its entity feature vector into the pre-set medical prescription knowledge graph for entity node matching to determine the matching entity node and the corresponding standard entity name and standard prescription attribute information includes: The entity feature vector of the entity name is compared with the standard entity feature vector of each entity node. The entity node with the highest similarity between the corresponding standard entity feature vector and the entity feature vector is taken as the matching entity node. Determine the standard entity name corresponding to the matching entity node and the standard prescription attribute information corresponding to the prescription attribute node connected to the topology of the matching entity node.
7. The method for extracting and correcting structured information from prescription images according to claim 1, characterized in that, The step of correcting the prescription attribute information corresponding to the entity name in the initial structured text using standard prescription attribute information includes: The standard prescription attribute information is used to correct errors and / or complete missing content in the prescription attribute information corresponding to the entity names in the initial structured text.
8. A system for extracting and correcting structured information from prescription images, characterized in that, It includes an image acquisition unit, a page segmentation unit, a text extraction unit, a feature encoding unit, a graph matching unit, an information error correction unit, and a data transmission unit, wherein: The image acquisition unit is used to acquire a clean prescription image transmitted by the image acquisition device. The clean prescription image is obtained by preprocessing the initial prescription image acquired by the image acquisition device. The layout segmentation unit is used to perform layout recognition on the clean prescription image, determine the target semantic polygon region in the clean prescription image, and segment the target semantic polygon region from the clean prescription image; The text extraction unit is used to input the target semantic polygon region into a pre-set medical multimodal large model for text extraction to obtain the initial structured text; The feature encoding unit is used to determine the entity name and the prescription attribute information corresponding to the entity name in the initial structured text, and to perform vector encoding on the entity name to obtain the entity feature vector; The graph matching unit is used to substitute the entity name and its entity feature vector into the preset medical prescription knowledge graph to perform entity node matching, and determine the matching entity node as well as the standard entity name and standard prescription attribute information corresponding to the matching entity node. The information correction unit is used to replace the corresponding entity name in the initial structured text with the standard entity name, and to correct the prescription attribute information of the corresponding entity name in the initial structured text with the standard prescription attribute information, so as to obtain the corrected structured text. The data transmission unit is used to convert the corrected structured text into a standard format and transmit it to the hospital information system.
9. A system for extracting and correcting structured information from prescription images, characterized in that, include: Memory, used to store instructions; A processor is configured to read instructions stored in the memory and execute the prescription image structured information extraction and error correction method according to any one of claims 1-7.
10. A computer program product, characterized in that, When the computer program product is run on a computer, it executes the prescription image structured information extraction and error correction method according to any one of claims 1-7.