A method, apparatus, system, and readable storage medium for parsing medical examination reports.

Abstract

A physical examination report analysis method, device, system and readable storage medium are applied to the technical field of information detection. The physical examination report is disassembled and processed to obtain multiple physical examination report pages. Text recognition is performed on each physical examination report page to obtain a text recognition result and extract medical text information. According to a pre-set connection relationship, medical text groups corresponding to the medical text information are determined. The application reduces the difficulty of medical long text analysis, can be directly applied to a review system of underwriting personnel and a customer health portrait system, can obtain high-accuracy and high-robustness information structured output for physical examination reports of different institutions and different formats, and improves underwriting efficiency.

Description

Technical Field

[0001] This application relates to the field of information detection technology, and in particular to a method, apparatus, system and readable storage medium for parsing physical examination reports. Background Technology

[0002] A medical examination report is a formatted document generated based on data from a physical examination. In current medical examination processes, various examination data need to be compiled into a single report. To enable insurance companies to thoroughly understand a policyholder's health condition before they apply for insurance, and to ultimately decide whether to accept the application or design and select a suitable insurance policy, the information from the policyholder's medical examination report needs to be extracted and utilized.

[0003] Early medical examination reports largely required manual extraction and analysis of information, which was highly inefficient and caused eye strain due to the lengthy recognition and extraction process. With advancements in science and technology, the medical and insurance sectors are gradually moving towards digitalization, enabling the automatic extraction of information from medical examination reports based on intelligent vision technology. However, current technology can only represent one-to-one relationships within groups, not one-to-many relationships. For example, an ophthalmology examination in a medical examination report might include two items: left eye vision and right eye vision. Current technology can only display the result of one item, failing to establish a one-to-many relationship. Summary of the Invention

[0004] This application provides a method, apparatus, system, and readable storage medium for parsing medical examination reports, which can obtain highly accurate and robust structured information output for medical examination reports from different institutions and in different formats.

[0005] This application discloses the following technical solution:

[0006] Firstly, this application provides a method for parsing a medical examination report, the method comprising:

[0007] The medical examination report is split into multiple pages to obtain the report.

[0008] Perform text recognition on each page of the medical examination report to obtain the text recognition results for each page of the medical examination report;

[0009] Based on the text recognition results of each page of the physical examination report, extract medical text information;

[0010] Based on the pre-defined connection relationships, the medical text groups corresponding to each medical text information are determined.

[0011] In some possible implementations, the pagination process of the medical examination report includes:

[0012] Medical examination reports stored in TIFF format are split into pages through file parsing;

[0013] For medical examination reports stored in PDF format, page splitting is performed using automated or open-source tools.

[0014] In some possible implementations, the text recognition for each page of the medical examination report includes:

[0015] The text on each page of the medical examination report was recognized using the OCR method.

[0016] In some possible implementations, the method for learning the connection relationship includes:

[0017] Based on the positional relationships of the medical text information, deep learning is used to learn the connection relationships of each piece of medical text information, and the positional relationships are represented by relative position encoding.

[0018] In some possible implementations, the medical text information includes: the name of the examination item, the result of the examination item, the unit of measurement for the examination item, and the normal range value.

[0019] Secondly, this application provides a medical examination report parsing device, which includes: a page splitting module, a character recognition module, an extraction module, and a determination module;

[0020] The page splitting module is used to split the physical examination report into multiple pages.

[0021] The text recognition module is used to perform text recognition on each page of the physical examination report and obtain the text recognition result of each page of the physical examination report.

[0022] The extraction module is used to extract medical text information based on the text recognition results of each page of the physical examination report.

[0023] The determining module is used to determine the medical text group corresponding to each medical text information according to the pre-set connection relationship.

[0024] In some possible implementations, the page splitting module is specifically used for:

[0025] Medical examination reports stored in TIFF format are split into pages through file parsing;

[0026] For medical examination reports stored in PDF format, page splitting is performed using automated or open-source tools.

[0027] In some possible implementations, the method for learning the connection relationship includes:

[0028] Based on the positional relationships of the medical text information, deep learning is used to learn the connection relationships of each piece of medical text information, and the positional relationships are represented by relative position encoding.

[0029] Thirdly, this application provides a medical examination report parsing device, including: a memory and a processor;

[0030] The memory is used to store programs;

[0031] The processor is configured to execute the program to implement the steps of the method as described in the first aspect.

[0032] Fourthly, this application provides a computer storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the method described in the first aspect.

[0033] Compared with the prior art, this application has the following beneficial effects:

[0034] This application provides a method, device, system, and readable storage medium for parsing medical examination reports, which reduces the difficulty of parsing long medical documents. It can be directly applied to the review system of underwriting personnel and the customer health profiling system. It can obtain highly accurate and robust structured information output for medical examination reports of different institutions and different formats, thereby improving the efficiency of underwriting operations. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 A flowchart of a method for parsing a medical examination report provided in this application embodiment;

[0037] Figure 2 A schematic diagram of ophthalmology-based medical text information provided for an embodiment of this application;

[0038] Figure 3 This is a schematic diagram of a medical examination report parsing device provided in an embodiment of this application. Detailed Implementation

[0039] Early medical examination reports largely required manual extraction and analysis of information, which was highly inefficient and caused eye strain due to the lengthy recognition and extraction process. With advancements in science and technology, the medical and insurance sectors are gradually moving towards digitalization, enabling the automatic extraction of information from medical examination reports based on intelligent vision technology. However, current technology can only represent one-to-one relationships within groups, not one-to-many relationships. For example, an ophthalmology examination in a medical examination report might include two items: left eye vision and right eye vision. Current technology can only display the result of one item, failing to establish a one-to-many relationship.

[0040] In view of this, this application provides a method, device, system and readable storage medium for parsing medical examination reports, which reduces the difficulty of parsing long medical documents and can be directly applied to the review system of underwriting personnel and the customer health profiling system. It can obtain highly accurate and robust information structured output for medical examination reports of different institutions and different formats, thereby improving the efficiency of underwriting operations.

[0041] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0042] See Figure 1 The figure is a flowchart of a physical examination report parsing method provided in an embodiment of this application.

[0043] S11: Split the medical examination report into multiple pages.

[0044] After obtaining the user's physical examination report, the first step is to split the report into multiple pages.

[0045] In some possible implementations, medical examination reports are typically multi-page documents, primarily stored in TIFF and PDF formats. To facilitate OCR text recognition, the file package needs to be split into individual JPEG image formats for each page. Specifically, TIFF format can be split primarily through file parsing, while PDF format can be split primarily through automated or open-source tools. It should be noted that this application does not limit the method for splitting pages or the format of the medical examination report.

[0046] S12: Perform text recognition on each page of the medical examination report to obtain the text recognition results for each page of the medical examination report.

[0047] After splitting the medical examination report into multiple pages, text recognition can be performed on each page to obtain the text recognition result for each page.

[0048] In some possible implementations, text recognition can be performed on each page of a multi-page medical examination report using OCR (Optical Character Recognition). OCR refers to the process by which an electronic device (such as a scanner or digital camera) examines characters printed on paper and then translates the shapes into computer text using character recognition methods; that is, scanning text documents and then analyzing and processing the image files to obtain text and layout information. OCR can provide multi-scene, multi-language, and high-precision text detection and recognition services. It should be noted that this application does not limit the specific methods used for text recognition.

[0049] S13: Based on the text recognition results of each page of the physical examination report, extract the basic information of the examinee and various medical text information, and assign corresponding tags.

[0050] The basic information of the examinee refers to their name, gender, ID number, and other basic information. The various medical text information refers to the name of the examination item, the result of the examination item, the unit of measurement for the result, and the normal range values, etc. It should be noted that other information can also be used as the basic information of the examinee and the various medical text information; this application does not limit the specific content of the information.

[0051] In some possible implementations, see Table 1, which is a physical examination text recognition result table provided by an embodiment of this application. Here, "examiner" refers to the basic information of the examinee, and "skin color," "no abnormalities found," etc., are the various medical text information.

[0052] Table 1

[0053] Surgical examination Inspector Check item name result skin color No abnormalities were found. subcutaneous lump none Lymph nodes No abnormalities were found. thyroid No abnormalities were found. Limb joints No abnormalities were found. spine No abnormalities were found. Other checks none summary No obvious abnormalities were found during the surgical examination.

[0054] In some other possible implementations, see Figure 2 This figure is a schematic diagram of ophthalmological medical text information provided in an embodiment of this application. "Left eye vision", "0.8", "Right eye vision", and "0.9" represent the various medical text information.

[0055] After extracting the basic information of the examinees and various medical text information, corresponding tags are assigned based on the Named Entity Recognition (NER) method. Named entities generally refer to entities in text that have specific meaning or strong referentiality, typically including names of people, places, organizations, dates and times, proper nouns, etc. The NER system extracts these entities from unstructured input text and can identify more categories of entities according to business needs, such as product names, models, and prices. In some possible implementations, named entity recognition can be performed based on rule-based methods, machine learning methods, or deep learning methods, and then the identified medical text information is labeled using the BIO tag task. BIO tagging is a method of labeling each element as "BX", "IX", or "O". Specifically, "BX" indicates that the segment containing this element belongs to type X and this element is at the beginning of the segment; "IX" indicates that the segment containing this element belongs to type X and this element is in the middle of the segment; and "O" indicates that it does not belong to any type. It should be noted that this application does not limit the specific methods for named entity recognition and tagging.

[0056] S14: Determine the medical topic category based on the acquired medical text information.

[0057] Based on the various medical text information obtained in S13, the medical subject category is determined. In some possible implementations, referring to Table 1, "surgical examination" is a medical subject category. In other implementations, ophthalmological examination, pediatric examination, ultrasound examination, gynecological examination, etc., can all be medical subject categories. It should be noted that this application does not limit the specific content of the medical subject category.

[0058] S15: Based on the acquired medical text information, learn the connection relationships between the various medical text information.

[0059] Based on the medical text information obtained in S13, the connection relationships between the various medical text information are learned through deep learning.

[0060] In some possible implementations, referring to Table 1, there is a connection between "Inspection Item Name" and "Result," and a connection between "Skin Color" and "No Abnormalities Found." These connections can be learned using deep learning methods. In other possible implementations, see... Figure 2 The phrase "left eye visual acuity" is connected to "0.8", and the phrase "right eye visual acuity" is connected to "0.9". It should be noted that this application does not limit the specific content of these connections.

[0061] In some possible implementations, deep learning methods can be used to learn the connections between various medical text information. Deep learning is a new research direction in the field of machine learning. It can leverage the characteristics of artificial neural networks to train and adjust parameters, obtaining the weights and representations, i.e., features, in each layer. In some possible implementations, graph neural networks (GNNs) can be used for deep learning. GNNs are neural networks that operate directly on graph structures. GNNs can be used to learn the connections between various medical text information. It should be noted that other methods can also be used for deep learning, and this application is not limited to any particular method.

[0062] In some possible implementations, the connection relationships between various pieces of medical text information can be learned through deep learning based on their positional relationships. These positional relationships are represented using relative positional encoding.

[0063] In some possible implementations, the connections between various medical text information can be represented by a matrix, and deep learning methods learn precisely the matrix relationships between these medical text information.

[0064] For details, see Figure 2 This diagram can represent the matrix relationship between "left eye vision" and "0.8" and "right eye vision" and "0.9". Among them, "trigger" is a binary category, where 0 indicates no trigger and 1 indicates trigger. In some possible implementations, it can also be that 1 indicates no trigger and 0 indicates trigger, which is not limited in this application.

[0065] S16: Determine the medical text group to which each piece of medical text information belongs based on the connection relationship between the acquired medical text information.

[0066] Based on the connection relationships of the various medical text information obtained in S15, the main group category of each medical text is determined. A pair of connected medical text information constitutes a medical text group.

[0067] In some possible implementations, referring to Table 1, "skin color" and "no abnormalities" form one medical text group. In other implementations, see... Figure 2 "Left eye visual acuity" and "0.8" form one medical text group, and "Right eye visual acuity" and "0.9" form another medical text group. It should be noted that this application does not limit the specific content of the medical text groups.

[0068] S17: Determine the medical topic category to which each medical text group belongs based on the obtained medical topic category, and obtain the structured output of each medical text.

[0069] Based on the medical topic categories obtained in S14, the medical topic category to which each medical text group belongs is determined in S16. In some possible implementations, referring to Table 1, the medical topic category to which the medical text groups "skin color" and "no abnormalities found" belong is surgical examination. In other implementations, see... Figure 2 The medical text groups for "left eye vision" and "0.8" and "right eye vision" and "0.9" both belong to the medical subject category of ophthalmology.

[0070] At this point, the medical topic category includes medical text groups, and the medical text groups further include medical texts, thereby obtaining highly accurate and robust structured information outputs from each medical text.

[0071] In some possible implementations, since the existing technology uses the Layout LM model for structured output, which has a weak representation of page position, this application introduces additional relative spatial position encoding.

[0072] Specifically, there are many types of relative spatial location encoding, and the following formula is one implementation method.

[0073]

[0074] in, To ensure stable scaling factors during training, Let x be a parameter matrix, x = (x1, ..., x2) n ) represents the input text sequence, a ij This represents the absolute distance between token_i and token_j.

[0075] It should be noted that the medical examination report parsing method provided in this application can not only parse medical examination reports, but also other documents and reports with similar formats. This application does not limit the specific content of the reports.

[0076] This application provides a method for parsing medical examination reports, which addresses the problem of weak positional representation in medical examination reports from different institutions and in different formats. Specifically, when obtaining medical examination reports, by selecting a fixed medical subject category, such as surgical examination, the location of the corresponding medical text information within that category can be predicted. Furthermore, it solves the one-to-many problem within groupings in medical examination reports. Specifically, when obtaining ophthalmology medical text information, the medical subject category "ophthalmology" can include two groups of medical text: left eye visual acuity, right eye visual acuity, and their examination results.

[0077] Although this application describes the operations in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous.

[0078] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0079] See Figure 3 The figure is a schematic diagram of a physical examination report parsing device provided in an embodiment of this application. The device 300 includes: a page splitting module 301, a text recognition module 302, an extraction module 303, and a determination module 304.

[0080] The page splitting module 301 is used to split the physical examination report into multiple pages.

[0081] The text recognition module 302 is used to perform text recognition on each page of the physical examination report and obtain the text recognition result of each page of the physical examination report;

[0082] The extraction module 303 is used to extract medical text information based on the text recognition results of each page of the physical examination report.

[0083] The determination module 304 is used to determine the medical text group corresponding to each medical text information according to the pre-set connection relationship.

[0084] In some possible implementations, the pagination module is specifically used for: paginating medical examination reports stored in TIFF format through file parsing; and paginating medical examination reports stored in PDF format using automated or open-source tools.

[0085] In some possible implementations, the method for learning connectivity relationships includes: deep learning the connectivity relationships between individual pieces of medical text information based on their positional relationships. These positional relationships are represented using relative positional encoding.

[0086] This application provides a medical examination report parsing device that addresses the problem of weak positional representation in medical examination reports from different institutions and in different formats. Specifically, when retrieving a medical examination report, by selecting a fixed medical subject category, such as surgical examination, the device can predict the position of the corresponding medical text information within that category. Furthermore, it resolves the one-to-many problem within groupings in the medical examination report. Specifically, when retrieving ophthalmology medical text information, the medical subject category "ophthalmology" can include two groups of medical text: left eye visual acuity, right eye visual acuity, and their examination results.

[0087] The modules described in the embodiments of this disclosure can be implemented in software or hardware. The names of the modules are not necessarily limiting in certain circumstances; for example, a first building module can also be described as a "module for building a pre-trained dataset". The functions described above can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field-Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), etc.

[0088] This application also provides corresponding generation equipment and computer storage media for implementing the solutions provided in this application.

[0089] The device includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code to cause the device to perform a lighting control method according to any embodiment of this application.

[0090] The computer storage medium stores code, and when the code is run, the device running the code implements the method described in any embodiment of this application.

[0091] In the embodiments of this application, the terms "first" and "second" (if they exist) are used only as name identifiers and do not represent the order of first and second.

[0092] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus a general-purpose hardware platform. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0093] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0094] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for parsing a medical examination report, characterized in that, include: The medical examination report is split into multiple pages to obtain the report. Perform text recognition on each page of the medical examination report to obtain the text recognition results for each page of the medical examination report; Based on the text recognition results of each page of the physical examination report, medical text information is extracted, including the name of the examination item, the result of the examination item, the unit of examination item, and the normal range value; Based on the medical text information, medical topic categories are determined, including surgical examination, ophthalmological examination, pediatric examination, ultrasound examination, and gynecological examination. Based on the positional relationship of each medical text information, the spatial positional representation between the medical text information is enhanced by relative positional encoding, and the connection relationship between each medical text information is learned by deep learning methods to determine the medical text group corresponding to each medical text information, wherein multiple parallel medical text groups are included under the same medical topic category; Based on the connection relationship of each medical text information, the medical text group corresponding to each medical text information is determined. Each medical text group consists of medical text information with connection relationship. Multiple parallel medical text groups are included under the same medical topic category. By determining the medical topic category corresponding to each medical text group, the structured output of the physical examination report is obtained. The structured output includes the medical topic category, the medical text group, and the medical text information.

2. The method according to claim 1, characterized in that, The process of splitting the medical examination report into pages includes: Medical examination reports stored in TIFF format are split into pages through file parsing; For medical examination reports stored in PDF format, page splitting is performed using automated or open-source tools.

3. The method according to claim 1, characterized in that, The text recognition process for each page of the medical examination report includes: The text on each page of the medical examination report was recognized using the OCR method.

4. A medical examination report parsing device, characterized in that, include: Page splitting module, text recognition module, extraction module, and confirmation module; The page splitting module is used to split the physical examination report into multiple pages. The text recognition module is used to perform text recognition on each page of the physical examination report and obtain the text recognition result of each page of the physical examination report. The extraction module is used to extract medical text information based on the text recognition results of each page of the physical examination report. The medical text information includes the name of the examination item, the result of the examination item, the unit of examination item, and the normal range value. The determining module is used to determine medical topic categories based on the medical text information, including surgical examination, ophthalmological examination, pediatric examination, ultrasound examination, and gynecological examination; enhance the spatial representation between the medical text information by using relative position encoding based on the positional relationship of each medical text information, and use deep learning methods to learn the connection relationships of each medical text information to determine the medical text groups corresponding to each medical text information, wherein multiple parallel medical text groups are included under the same medical topic category; determine the medical text groups corresponding to each medical text information based on the connection relationships of each medical text information, each medical text group is composed of medical text information with connection relationships, and multiple parallel medical text groups are included under the same medical topic category; and obtain the structured output of the physical examination report by determining the medical topic category corresponding to each medical text group, wherein the structured output includes the medical topic category, the medical text group, and the medical text information.

5. The apparatus according to claim 4, characterized in that, The page splitting module is specifically used for: Medical examination reports stored in TIFF format are split into pages through file parsing; For medical examination reports stored in PDF format, page splitting is performed using automated or open-source tools.

6. A medical examination report parsing device, characterized in that, include: Memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the steps of the method as described in any one of claims 1 to 3.

7. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 3.

Images