Medical image report critical value early warning method and device and storage medium

Abstract

The invention discloses a medical image report critical value early warning method and device and a storage medium. The method comprises the steps that a first positive sentence corresponding to a medical image report text is determined, and the first positive sentence is used for describing human body abnormal information contained in the medical image report text; matching the first positive sentence with a preset critical value standard library, and determining an instruction corresponding to the first positive sentence; the instruction is input into an intelligent agent based on a large language model, a second positive sentence corresponding to the medical image report text is generated, and the second positive sentence is used for describing human body abnormal information contained in the medical image report text; and outputting critical value early warning information corresponding to the first positive sentence under the condition that the second positive sentence is matched with the first positive sentence.

Description

Technical Field

[0001] The present application relates to the field of medical care information technology, and in particular to a critical value warning method, device and storage medium for medical imaging reports. Background Art

[0002] In the daily management of hospitals, critical values ​​in medical imaging reports play an important role in clinical diagnosis and treatment. The establishment of a critical value system usually involves serious or potentially life-threatening diseases, such as aortic dissection, intracranial hemorrhage, acute pulmonary embolism, etc. Timely identification and reporting of these critical values ​​can help doctors quickly make correct clinical decisions and take necessary emergency measures. The hospital imaging department can improve medical quality and patient safety by formulating critical value standards and reporting processes for medical imaging reports. For example: Aortic dissection: The medical imaging report found that the patient had aortic dissection, which is an extremely dangerous disease that requires immediate surgical intervention; Intracranial hemorrhage: If a CT or MRI examination reveals intracranial hemorrhage, especially a large amount of hemorrhage or brain herniation, immediate surgery or drug treatment is required. Critical value reports can ensure that doctors take prompt measures to prevent the condition from worsening; or Acute pulmonary embolism: Imaging tests revealed that the patient had acute pulmonary embolism, a disease that can cause sudden death.

[0003] However, in actual operations, due to the sharp increase in the amount of imaging examination tasks, radiologists are often faced with the challenge of ensuring that critical values ​​in each imaging diagnosis report can be identified and reported in a timely and accurate manner under high-intensity work pressure.

[0004] Specifically, the amount of imaging data that physicians need to process every day is huge, and the tense working environment makes it easy for them to suffer from occupational fatigue and distraction, which increases the risk of missing critical values. In addition, the production cycle of medical imaging reports is usually tight, requiring physicians to complete heavy tasks within a limited time, which undoubtedly further increases the difficulty of paying attention to details. At the same time, there are obstacles to cross-departmental information exchange on critical values, and the imperfect communication mechanism leads to the inability to quickly and effectively transmit key information to relevant clinical departments, affecting the timeliness and safety of patient treatment. Finally, the existing medical information platform is still insufficient in the management of critical values, lacking intelligent reminder and tracking tools, making it difficult for physicians to fully cover all potential risk points even when they are highly focused.

[0005] The existing medical information platforms mentioned above are still insufficient in critical value management and lack intelligent reminder and tracking tools, making it difficult for doctors to fully cover all potential risk points even when they are highly focused. No effective solution has been proposed yet. Summary of the invention

[0006] The embodiments of the present disclosure provide a critical value warning method, device and storage medium for medical imaging reports, so as to at least solve the technical problems that the existing medical information platform is still insufficient in critical value management and lacks intelligent reminder and tracking tools, making it difficult for doctors to fully cover all potential risk points even in a highly focused state.

[0007] According to one aspect of an embodiment of the present disclosure, a critical value warning method for a medical imaging report is provided, comprising: determining a first positive sentence corresponding to a medical imaging report text, wherein the first positive sentence is used to describe abnormal human body information contained in the medical imaging report text; matching the first positive sentence with a preset critical value standard library to determine an instruction corresponding to the first positive sentence; inputting the instruction into an intelligent agent based on a large language model to generate a second positive sentence corresponding to the medical imaging report text, wherein the second positive sentence is used to describe abnormal human body information contained in the medical imaging report text; and when the second positive sentence matches the first positive sentence, outputting critical value warning information corresponding to the first positive sentence.

[0008] According to another aspect of the embodiments of the present disclosure, a storage medium is further provided. The storage medium includes a stored program, wherein the above method is executed by a processor when the program is running.

[0009] According to another aspect of the embodiments of the present disclosure, a critical value warning device for a medical imaging report is also provided, including: a first positive sentence determination module, used to determine a first positive sentence corresponding to a medical imaging report text, wherein the first positive sentence is used to describe abnormal human body information contained in the medical imaging report text; an instruction determination module, used to match the first positive sentence with a preset critical value standard library, and determine the instruction corresponding to the first positive sentence; a second positive sentence generation module, used to input the instruction into an intelligent agent based on a large language model, and generate a second positive sentence corresponding to the medical imaging report text, wherein the second positive sentence is used to describe abnormal human body information contained in the medical imaging report text; and a critical value warning module, used to output critical value warning information corresponding to the first positive sentence when the second positive sentence matches the first positive sentence.

[0010] According to another aspect of the embodiments of the present disclosure, a critical value warning device for a medical imaging report is also provided, comprising: a processor; and a memory, connected to the processor, for providing the processor with instructions for processing the following processing steps: determining a first positive sentence corresponding to a medical imaging report text, wherein the first positive sentence is used to describe abnormal human body information contained in the medical imaging report text; matching the first positive sentence with a preset critical value standard library to determine an instruction corresponding to the first positive sentence; inputting the instruction into an intelligent agent based on a large language model to generate a second positive sentence corresponding to the medical imaging report text, wherein the second positive sentence is used to describe abnormal human body information contained in the medical imaging report text; and when the second positive sentence matches the first positive sentence, outputting critical value warning information corresponding to the first positive sentence.

[0011] In the disclosed embodiment, after the doctor has completed the text of the medical image report. The key positive sentence corresponding to the text of the medical image report can be determined by a computing device such as a server. Then the key positive sentence is matched with the critical value standard library to determine the instruction corresponding to the key positive sentence. Then the instruction is input into an intelligent agent based on a large language model to generate a verification positive sentence corresponding to the instruction. Finally, when the verification positive sentence matches the key positive sentence successfully, the critical value warning information corresponding to the key positive sentence is output. Thus, in this way, the text content in the medical image report can be automatically parsed with the help of artificial intelligence and natural language processing technology, and then based on the pre-established preset critical value standard, the technical solution of this embodiment can automatically determine whether there is a critical value in the report and issue an early warning. Thus, the existing medical information platform is still insufficient in critical value management, lacks intelligent reminder and tracking tools, and makes it difficult for doctors to fully cover all potential risk points even in a highly focused state. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present application. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings: Figure 1 is a hardware structure block diagram of a computing device for implementing the method according to Embodiment 1 of the present disclosure; Figure 2 is a schematic diagram of a critical value warning system according to Embodiment 1 of the present disclosure; Figure 3 is a schematic diagram of the architecture of a server of the critical value warning system according to Embodiment 1 of the present disclosure; Figure 4 is a flowchart of a critical value warning method for a medical imaging report according to the first aspect of Embodiment 1 of the present disclosure; Figure 5 is a schematic diagram of an interface containing warning information displayed on a terminal device according to Embodiment 1 of the present disclosure; Figure 6 is a schematic diagram of fine-tuning a large language model according to Embodiment 1 of the present disclosure; Figure 7 is a schematic diagram of a critical value warning device for a medical imaging report according to Embodiment 2 of the present disclosure; and Figure 8 It is a schematic diagram of a critical value warning device for medical imaging reports according to Embodiment 3 of the present disclosure. DETAILED DESCRIPTION

[0013] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only embodiments of a part of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0014] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0015] Example 1 According to this embodiment, a method embodiment of a critical value warning method for a medical imaging report is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0016] The method embodiment provided in this embodiment can be executed in a computer terminal, a server or a similar computing device. Figure 1 The hardware structure block diagram of a computing device for implementing a critical value warning method for medical image reports is shown. Figure 1As shown, the computing device may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, the transmission device, and the input / output interface are connected to the processor via a bus. In addition, it may also include: a display, a keyboard, and a cursor control device connected to the input / output interface. A person skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0017] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computing device. As involved in the embodiments of the present disclosure, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0018] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the critical value warning method of the medical image report in the embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the critical value warning method of the medical image report of the above-mentioned application is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computing device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0019] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computing device. In one example, the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0020] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of the computing device.

[0021] It should be noted that, in some optional embodiments, the above Figure 1 The computing device shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computing devices described above.

[0022] Figure 2 Schematic diagram of the critical value warning system according to this embodiment. Figure 2 As shown, the system includes: a terminal device 100 and a server 200. The terminal device 100 is connected to the server 200 through a network. The doctor of the imaging department can complete the preparation and processing of the medical imaging report on the terminal device 100, and the doctor uploads the completed medical imaging report to the server 200 through the terminal device 100. After receiving the medical imaging report, the server 200 can perform a warning process of a critical value for the medical imaging report.

[0023] It should be noted that the terminal device 100 and the server 200 in the system can both be applicable to the hardware structure described above.

[0024] also, Figure 3 The schematic diagram of the architecture of the server 200 is further shown. Figure 3 As shown, the server 200 is provided with a preprocessing module 210, a named entity recognition module 220, a relationship extraction module 230, a key positive sentence generation module 240, an instruction generation module 250, a critical value standard library 260, a large language model intelligent agent 270, a positive sentence matching module 280 and an early warning module 290.

[0025] Under the above operating environment, according to the first aspect of this embodiment, a method for warning critical values ​​of medical image reports is provided. The method comprises: Figure 2 The server 200 shown in FIG. Figure 4 A schematic diagram showing the process of the method is shown in FIG. Figure 4 As shown, the method includes: S402: Determine a first positive sentence corresponding to the medical imaging report text, wherein the first positive sentence is used to describe human body abnormality information contained in the medical imaging report text; S404: Matching the first positive sentence with a preset critical value standard library to determine an instruction corresponding to the first positive sentence; S406: inputting the instruction into the agent based on the large language model to generate a second positive sentence corresponding to the medical imaging report text, wherein the second positive sentence is used to describe the human body abnormality information contained in the medical imaging report text; and S408: When the second positive sentence matches the first positive sentence, output critical value warning information corresponding to the first positive sentence.

[0026] Specifically, in this embodiment, the physician prepares and completes the medical imaging report through the terminal device 100. The text of the medical imaging report (in this embodiment, the "examination findings" part of the medical imaging report is taken as an example for explanation) is as follows: "The thorax is symmetrical on both sides, with the trachea and mediastinum in the middle. Lamellar translucent shadows can be seen in the bilateral chest cavities, and the edges of the compressed lungs can be seen. The right lung is compressed by about 75%, and the left lung is compressed by less than 10%. The pleura is thickened on both sides. The translucency of both lungs is reduced." Then, the doctor sends the medical imaging report text to the server 200 via the terminal device 100 , so that the server 200 obtains the medical imaging report text from the terminal device 100 .

[0027] Then, the server 200 determines the key positive sentence (i.e., the first positive sentence) corresponding to the medical image report text through the preprocessing module 210, the named entity recognition module 220, the relationship extraction module 230, and the key positive sentence generation module 240. The key positive sentence is used to describe the abnormal human body information contained in the medical image report text (S402). Specifically, in the medical image report text, only the description of "the bilateral chest is symmetrical, and the trachea and mediastinum are centered" is normal, so the key positive sentences corresponding to the text include: Bilateral chest lucencies; Compression of lung margins; Right lung compression 75%; Left lung compression less than 10%; The translucency of both lungs decreased.

[0028] It should be noted that the key positive sentence determined based on the text of the medical imaging report can be one or more. If the body of the detected object is normal, the key positive sentence is zero, and no critical value warning is issued in this case. The specific method for determining the key positive sentence will be described in detail below.

[0029] Then, the server 200 matches the key positive sentences with the preset critical value standard library 260 one by one, and determines the instructions corresponding to the key positive sentences (S404). Specifically, the critical value standard library 260 is pre-deployed in the server 200 for matching, wherein the following table shows part of the emergency value information of the critical value standard library: Table 1 Critical value standard library

[0030] Therefore, the server 200 matches each key positive sentence in the critical value standard library through the instruction generation module 250 .

[0031] For example, the server 200 matches the key positive sentence "chest translucency" in the critical value standard library, but cannot match the corresponding critical value. This indicates that the key positive sentence does not involve a critical value. Therefore, no further processing is performed.

[0032] For another example, the server 200 matches the key positive sentence "lung edge compression" in the critical value standard library, but cannot match the corresponding critical value. This indicates that the key positive sentence does not involve a critical value. Therefore, no further processing is performed.

[0033] For example, the server 200 matches the key positive sentence "right lung compression is about 75%" in the critical value standard library, thereby matching the corresponding critical value "lung compression".

[0034] For another example, the server 200 matches the key positive sentence "left lung compression is less than 10%" in the critical value standard library, but cannot match the corresponding critical value. This indicates that the key positive sentence does not involve a critical value. Therefore, no further processing is performed.

[0035] By analogy, among the key positive sentences corresponding to the above medical imaging report text, only the key positive sentence "right lung compression is about 75%" matches the corresponding critical value "lung compression".

[0036] Then further, the server 200 determines the corresponding instruction according to the matched critical value through the instruction generation module 250. The specific technical details of determining the instruction will be described in detail below.

[0037] Then, the server 200 inputs the instruction to the large language model agent 270, thereby generating a verification positive sentence (i.e., the second positive sentence) corresponding to the above-mentioned medical imaging report text (S406). For example, in this embodiment, the verification positive sentence generated by the large language model agent 270 is also "right lung compression is approximately 75%." Among them, the large language model agent 270 can be, for example, an agent fine-tuned based on the llama large language model, which is used to generate corresponding positive sentences according to the input instructions. The fine-tuning method of the agent will be described in detail below.

[0038] Then, the server 200 matches the verification positive sentence with the key positive sentence through the positive sentence matching module 280. If the verification positive sentence matches the key positive sentence successfully, it indicates that the key positive sentence is a critical value. Otherwise, if the verification positive sentence matches the key positive sentence unsuccessfully, it indicates that the key positive sentence is not a critical value.

[0039] For example, in this embodiment, the verification positive sentence "right lung compression is about 75%" is successfully matched with the key positive sentence "right lung compression is about 75%", so that the key positive sentence "right lung compression is about 75%" is a critical value.

[0040] Therefore, when the server 200 successfully verifies that the positive sentence matches the key positive sentence, the server 200 outputs critical value warning information corresponding to the key positive sentence through the warning module 290 ( S408 ).

[0041] Specifically, the server 200 may send the warning information to the terminal device 100 so that the warning information is displayed on the terminal device 100. Figure 5 FIG. 1 is a schematic diagram showing an interface containing warning information displayed on the terminal device 100. Figure 5 As shown, the terminal device 100 displays a box at the position corresponding to the text "right lung compression is about 75%" in the medical imaging report text for warning, and displays the following warning information: Compression: Please note the critical value "compression"! Thus, through the above method, doctors can be warned of critical values.

[0042] Preferably, the server 200 further monitors whether the physician fills in the report in accordance with the critical value management requirements on the terminal device 100. If the server 200 determines that the physician does not fill in the report in accordance with the critical value management requirements, it further issues an intervention warning message.

[0043] Specifically, the server 200 will automatically check whether the physician has reported relevant information in accordance with the critical value management requirements. If it is found that the physician has not reported as required, the system will immediately issue a reminder to ensure the timely communication of critical value information. This function helps to reduce human omissions and improve the accuracy and timeliness of reports. Finally, when the server 200 identifies a critical value and the physician has not reported as required, the server 200 will automatically send a reminder notification to ensure that the physician can pay attention and handle it in a timely manner. Reminder methods can include pop-up prompts, SMS notifications, and other forms to ensure the efficiency of information communication.

[0044] As described in the background technology, doctors need to deal with a huge amount of imaging data every day, and the tense working environment makes it easy to produce occupational fatigue and distraction, thereby increasing the risk of missing critical values. In addition, the production cycle of medical imaging reports is usually relatively tight, requiring doctors to complete heavy tasks within a limited time, which undoubtedly further increases the difficulty of paying attention to details. At the same time, there are obstacles to cross-departmental information exchange on critical values, and the imperfect communication mechanism leads to the inability to quickly and effectively transmit key information to relevant clinical departments, affecting the timeliness and safety of patient treatment. Finally, the existing medical information platform still has shortcomings in critical value management, lacks intelligent reminder and tracking tools, making it difficult for doctors to fully cover all potential risk points even in a highly focused state.

[0045] In view of this, according to the technical solution of this embodiment, after the doctor has completed the text of the medical image report. The key positive sentence corresponding to the text of the medical image report can be determined by a computing device such as a server. Then the key positive sentence is matched with the critical value standard library to determine the instruction corresponding to the key positive sentence. Then the instruction is input into an intelligent agent based on a large language model to generate a verification positive sentence corresponding to the instruction. Finally, when the verification positive sentence matches the key positive sentence successfully, the critical value warning information corresponding to the key positive sentence is output. Thus, in this way, the text content in the medical image report can be automatically parsed with the help of artificial intelligence and natural language processing technology, and then based on the pre-established preset critical value standard, the technical solution of this embodiment can automatically determine whether there is a critical value in the report and issue an early warning. Thus, the existing medical information platform in the prior art is still insufficient in critical value management, lacks intelligent reminder and tracking tools, and makes it difficult for doctors to fully cover all potential risk points even in a highly focused state.

[0046] Optionally, before determining the first positive sentence corresponding to the medical imaging report text, the method further includes preprocessing the medical imaging report text, wherein the preprocessing operation includes at least one of the following: text cleaning; word segmentation; and stemming and lemmatization.

[0047] Specifically, after acquiring the medical image text, the server 200 performs at least one of the following preprocessing operations through the preprocessing module 210: Text cleaning: remove irrelevant characters and punctuation marks, convert to lowercase, etc.

[0048] Word segmentation: Split the text into words or phrases, using Chinese word segmentation tools such as Jieba.

[0049] Stemming and lemmatization: reducing words to their base form, reducing lexical variations.

[0050] Optionally, the operation of determining the first positive sentence corresponding to the medical imaging report text includes: determining key information of the medical imaging report text through named entity recognition, wherein the key information includes medical terms and diagnostic conclusions; determining relationship information of the medical imaging report text through relationship extraction, wherein the relationship information is used to indicate the relationship between the key information; and determining the first positive sentence based on the key information and the relationship information.

[0051] Specifically, in this embodiment, entity types corresponding to medical terms and diagnosis conclusions are predefined. Table 2 below shows some examples of entity types: Table 2 Entity Types

[0052] Thus, after preprocessing the medical imaging report text, the server 200 determines the key information of the medical imaging report text through the named entity recognition module 220, wherein the key information includes medical terms and diagnostic conclusions. For example, the server 200 can use traditional NLP technology (such as BERT, RoBERTa, combined with conditional random fields CRF, support vector machines SVM, etc.) to perform named entity recognition and extract medical terms and diagnostic conclusions in the report. Then preferably, the server 200 can also perform further entity recognition and context understanding in combination with the Tongyi language model, for example, to improve the accuracy and robustness of recognition.

[0053] Therefore, for the medical imaging report text mentioned above, the following key information can be identified: Bilateral, thorax, symmetrical; trachea, mediastinum, centered; bilateral, patchy, translucent shadows, visible in the thorax; visible, compressed, lung, edge; right, lung, compressed, about, 75%; left, lung, compressed, less than, 10%; bilateral, pleura, thickened; bilateral lung, translucency, reduced.

[0054] In addition, in this embodiment, relationship types for describing the relationship between key information are also predefined. Table 3 below shows some examples of relationship types: Table 3 Relationship types

[0055] Thus, the server 200 further determines the relationship between the key information as the corresponding relationship information through the relationship extraction module 230. For example, the server 200 can extract the relationship between the key information by using NLP technology, dependency syntax analysis and graph neural network (GNN).

[0056] Further preferably, the server 200 can also combine the general semantic model to perform context understanding, identify and extract more complex semantic relationships, and improve the accuracy of relationship extraction.

[0057] Thus, for the key information identified in the above medical imaging report text, the server 200 can extract the following relationships: The relationship between "bilateral" and "thorax" is one of orientation, and the relationship between "thorax" and "symmetry" is one of negative signs; The relationship between "trachea" and "mediastinum" is a parent position relationship, and the relationship between "mediastinum" and "center" is a negative sign relationship; There is a directional relationship between "bilateral" and "chest cavity", a state description relationship between "flaky" and "translucent shadow", a qualitative relationship between "visible" and "flaky translucent shadow", and a positive relationship between "chest cavity" and "visible flaky translucent shadow"; etc.

[0058] Then, the server 200 determines the key positive sentence according to the key information and the relationship information between the key information through the key positive sentence generation module 240. Specifically, the part entity and the sign entity of the positive relationship can be combined according to a predetermined rule to obtain the key positive sentence. Of course, in order to be more accurate, a correction rule can also be introduced, which will not be repeated here.

[0059] Therefore, the key positive sentence generation module 240 can generate the following positive sentences for the above medical imaging report text: Bilateral chest lucencies; Compression of lung margins; Right lung compression 75%; Left lung compression less than 10%; The translucency of both lungs decreased.

[0060] Therefore, this embodiment can more accurately identify positive information in medical imaging report texts by combining named entity recognition and relationship extraction technology.

[0061] For example, when encountering a specific scenario such as "lung compression exceeding 70%", the system can quickly and accurately perform risk classification and early warning, providing intelligent and precise technical support for medical risk management. From the following five common descriptions of lung compression, it can be seen that the descriptions are different. If traditional NLP (natural language processing) is used, it is very easy to miss or misidentify.

[0062] 1) The right lung lobe was compressed by about 30%, and there was pneumothorax on the left side, with lung tissue compressed by 80%.

[0063] 2) Textureless lucency can also be seen in the upper field of the left lung, which is compressed by about 70%.

[0064] 3) The lung tissue in the left lung was compressed (lung tissue compression was about 90%), and there was no lung texture inside.

[0065] 4) The lung volume is compressed by approximately 75%.

[0066] 5) The lingular segment of the left upper lung is compressed and atelectasis, with a compression of about 70%.

[0067] However, the technical solution of this embodiment can extract the "key information" of the positive report content through entity naming recognition and relationship extraction of the above five report description sentences, and generate corresponding key positive sentences based on the determined relationships, which correspond to the following (the first sentence can generate two key positive sentences): 1) The right lung lobe is compressed by 30%.

[0068] 2) The left lung tissue is compressed by 80%.

[0069] 3) The left upper lung field is compressed by 70%.

[0070] 4) The left lung tissue is compressed by 90%.

[0071] 5) Lung volume is compressed by 55%.

[0072] 6) The tongue of the left upper lung is compressed by 70%.

[0073] Then, in subsequent operations, the technical solution of this embodiment matches the above key positive sentences with the rules in the critical value standard library. Then, with the help of the intelligent agent based on the large language model, verification matching is performed to identify the corresponding rule: "lung compression exceeds 70%".

[0074] Then, these five key information are verified by the intelligent body using matching critical rules, and the output of the 1st, 2nd, 3rd, and 5th descriptions indicates that critical warnings are required. The first right lung does not require a warning, but the left lung does.

[0075] Thus, through the above method, this embodiment further improves the accuracy of extracting positive content from the medical imaging report text.

[0076] Optionally, the first positive sentence is matched with a preset critical value standard library to determine the operation of the instruction corresponding to the first positive sentence, including: matching the critical value in the critical value standard library with the first positive sentence to determine the target critical value corresponding to the first positive sentence; and determining the instruction based on the keywords corresponding to the target critical value and the medical imaging report text.

[0077] Specifically, the instruction generation module 250 of the server 200 matches the key positive sentence "right lung compression 75%" with the critical value standard library shown in Table 1, thereby determining the corresponding critical value "lung compression" (ie, the target critical value).

[0078] Further, referring to Table 1, the critical value standard library is also pre-set with keywords corresponding to each critical value. Thus, the instruction generation module 250 of the server 200 further determines the keywords corresponding to the critical value "lung compression": "lung", "compression", "number>70%".

[0079] Then, the instruction generation module 250 obtains the preprocessed medical imaging report text "The bilateral thorax is symmetrical, and the trachea and mediastinum are centered. Flake-like translucent shadows can be seen in the bilateral chest cavities, and the edges of the compressed lungs can be seen. The right lung is compressed by about 75%, and the left lung is compressed by less than 10%. The bilateral pleura is thickened. The translucency of the bilateral lungs is reduced" from the preprocessing module 210, and generates instructions in combination with the above keywords. Specifically, for example, the instruction generation module 250 can fill the medical imaging report text and the above keywords into a pre-set instruction template to generate instructions.

[0080] For example, the instruction template is: "Keywords:... Medical imaging report text:... Task: Extract positive sentences corresponding to keywords from medical imaging report texts".

[0081] Thus, the instruction generation module 250 may generate the following instructions: “Keywords: lungs, compression, numbers > 70%.

[0082] Medical imaging report text: The chest is symmetrical on both sides, with the trachea and mediastinum in the middle. Lamellar translucent shadows can be seen in the chest cavity on both sides, and the edge of the compressed lung can be seen. The right lung is compressed by about 75%, and the left lung is compressed by less than 10%. The pleura is thickened on both sides. The translucency of both lungs is reduced.

[0083] Task: Extract positive sentences corresponding to keywords from medical imaging report texts".

[0084] Optionally, when the second positive sentence matches the first positive sentence, the operation of outputting critical value warning information corresponding to the first positive sentence includes: when the matching degree between the second positive sentence and the first positive sentence is greater than a predetermined threshold, outputting critical value warning information.

[0085] Specifically, the server 200 performs similarity matching between the verified positive sentence and the key positive sentence through the positive sentence matching module 280. When the similarity matching is greater than 90%, it indicates that the key positive sentence "right lung compression 75%" is a critical value. Thus, the server 200 outputs critical value warning information through the warning module 290. Among them, the method for similarity matching in this embodiment can refer to the commonly used methods in the prior art, which will not be repeated here. Of course, in addition to 90%, other values ​​close to 1 can also be used according to actual conditions, which will not be repeated here.

[0086] Optionally, the method further includes fine-tuning the intelligent agent through the following steps: connecting the large language model to a positive sentence matching module for matching positive sentences, connecting the positive sentence matching module to a binary classification model, and freezing the parameters of the positive sentence matching module and the binary classification model, wherein the binary classification model is used to map the similarity matching value output by the positive sentence matching module to a probability value for determining whether it is a critical value; constructing a training sample set, wherein each training sample in the training sample set includes a sample instruction, a sample positive sentence, and a true critical state value, wherein the true critical state value is used to indicate whether the sample positive sentence is a critical value; inputting each sample instruction into the large language model to generate a corresponding predicted positive sentence, and inputting the predicted positive sentence into the positive sentence matching module to match the corresponding sample positive sentence, and outputting the corresponding similarity matching degree; inputting the similarity matching degree into the binary classification model, thereby outputting a predicted critical probability that the predicted sample positive sentence is a critical value; and fine-tuning the large language model using a pre-set loss function according to the predicted critical probability corresponding to each sample instruction and the corresponding true critical state value.

[0087] Specifically, in order to obtain the large language model agent 270, the large language model needs to be fine-tuned.

[0088] See also Figure 6 As shown, in order to fine-tune the large language model, the large language model is first connected to the positive sentence matching module. The positive sentence matching module is, for example, connected to Figure 3 The masculine sentence matching modules shown in are preferably the same modules.

[0089] Construct a training sample set, where the training sample set includes multiple training samples {S 1 , S 2 ,..., S n}, each training sample includes sample instructions, sample positive sentences and real critical state values, where Si represents the i-th training sample. And S i ={I i , C i , Y i}. i represents the sample instruction of the i-th training sample, C i represents the sample positive sentence of the i-th training sample, Y i represents the true critical state value of the i-th training sample, which is used to indicate the corresponding sample positive sentence C i Whether it is a critical value. Among them, Y i The value is 0 or 1. i =0, indicating the corresponding sample positive sentence C i is a non-critical value; when Y i =1, indicating that the corresponding sample positive sentence C i It is a critical value.

[0090] refer to Figure 6 As shown, during the training process, the training sample S i Sample instruction I i Input the large language model to generate the corresponding predicted positive sentence E i Then, the positive sentence matching module will predict the positive sentence E i and the corresponding sample positive sentence C i Perform similarity matching to obtain the corresponding similarity matching value H i .

[0091] In this embodiment, the similarity matching value H i If the percentage is greater than 90%, the corresponding sample positive sentence C will be judged i is a critical value. Or the threshold must be close to 1. Therefore, during training, the threshold needs to be input into a pre-trained binary classification model (such as logistic regression, etc.) to obtain the value of the sample positive sentence C. i The corresponding predicted critical probability X i When the predicted probability of emergency is X i When it is greater than 50%, the corresponding sample positive sentence C can be predicted. i It is a critical value.

[0092] Thus, this embodiment predicts the critical probability X i and the corresponding real critical state value Y i The pre-set loss function is input to fine-tune the large language model. During the fine-tuning process, the parameters of the binary classification model and the positive sentence matching module are fixed (i.e. frozen), and only the large language model is fine-tuned.

[0093] Thus, in this way, the task of the large language model agent (i.e., generating positive sentences) can be converted into the task of predicting critical values. Since the converted task can be implemented by binary classification, it is easier to set the loss function, and the positive sentences generated by the large language model agent fine-tuned in this way are more conducive to the prediction of critical values ​​in this embodiment. Thus, in this way, the large language model can be fine-tuned more accurately with a simpler loss function.

[0094] Further optionally, the cost value corresponding to each predicted critical probability in the loss function is determined in the following manner: based on each predicted critical probability and the corresponding actual critical state value, the cost value corresponding to each predicted critical probability is determined in a pre-set asymmetric cost matrix, wherein in the asymmetric cost matrix, the cost value of predicting a critical value as a non-critical value is greater than the cost value of predicting a non-critical value as a critical value.

[0095] Furthermore, in this embodiment, an asymmetric cost matrix M is set, whose elements M jk Represents the cost of classifying the true category j as category k. The asymmetric matrix M is as follows:

[0096] Where c_TN = 0 represents the cost of predicting a non-critical value as a non-critical value; c_FP = 0.5 represents the cost of predicting a non-critical value as a critical value; c_FN = 1 represents the cost of predicting a critical value as a non-critical value; c_TP = 0 represents the cost of predicting a critical value as a critical value.

[0097] Therefore, in this embodiment, each predicted emergency probability X i The corresponding predicted critical state value X' i (X i Greater than 50% is a critical value, X' i =1;X i Less than or equal to 50% is a non-critical value, X' i =0) and the corresponding real critical state value Y i Perform a comparison. Determine the comparison type based on the comparison results. The comparison types include the following 4 types: The sample positive sentence C of non-critical value i The prediction is non-critical value; The sample positive sentence C of non-critical value i The prediction is critical value; The sample positive sentence C of the critical value i Predicted to be non-critical; and The sample positive sentence C of the critical value i The prediction is critical.

[0098] Then, according to the comparison type of each training sample, the corresponding cost value F is determined in the cost matrix i . Then substitute the loss function shown in the following formula: .

[0099] in, X i To predict the probability of criticality; Max() means taking the maximum value; Y i is the real critical status value; X' i For X i The corresponding predicted critical state value, when X i ≤50% when X' i =0, when X i >50% when X' i =1; F i For X i and Y i The corresponding cost value.

[0100] Thus, the large language model is fine-tuned through the above loss function to obtain the large language model agent 270.

[0101] Misclassifying a non-critical situation as critical (false positive) and misclassifying a critical situation as non-critical (false negative), the latter is usually more dangerous. Therefore, this embodiment further customizes the loss function by introducing an asymmetric cost matrix to reflect the cost difference of different types of errors. Thus, the verified positive sentences generated by the fine-tuned large language model agent can be more accurately used for early warning of critical values.

[0102] In addition, reference Figure 1 As shown, according to the second aspect of this embodiment, a storage medium is provided, wherein the storage medium includes a stored program, wherein when the program is run, a processor executes any one of the above methods.

[0103] Thus, according to the technical solution of this embodiment, after the doctor has completed the text of the medical image report. The key positive sentence corresponding to the text of the medical image report can be determined by a computing device such as a server. Then the key positive sentence is matched with the critical value standard library to determine the instruction corresponding to the key positive sentence. Then the instruction is input into an intelligent agent based on a large language model to generate a verification positive sentence corresponding to the instruction. Finally, when the verification positive sentence matches the key positive sentence successfully, the critical value warning information corresponding to the key positive sentence is output. Thus, in this way, the text content in the medical image report can be automatically parsed with the help of artificial intelligence and natural language processing technology, and then based on the pre-established preset critical value standard, the technical solution of this embodiment can automatically determine whether there is a critical value in the report and issue an early warning. Thus, the existing medical information platform in the prior art is still insufficient in critical value management, lacks intelligent reminder and tracking tools, and makes it difficult for doctors to fully cover all potential risk points even in a highly focused state.

[0104] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0105] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0106] Example 2 Figure 7 FIG. 7 shows a critical value warning device 700 for a medical image report according to this embodiment, and the device 700 corresponds to the method according to the first aspect of embodiment 1. Figure 7As shown, the device 700 includes: a first positive sentence determination module 710, which is used to determine the first positive sentence corresponding to the medical imaging report text, wherein the first positive sentence is used to describe the abnormal human body information contained in the medical imaging report text; an instruction determination module 720, which is used to match the first positive sentence with a preset critical value standard library to determine the instruction corresponding to the first positive sentence; a second positive sentence generation module 730, which is used to input the instruction into an intelligent agent based on a large language model to generate a second positive sentence corresponding to the instruction; and a critical value warning module 740, which is used to output critical value warning information corresponding to the first positive sentence when the second positive sentence matches the first positive sentence.

[0107] Optionally, the device further comprises a preprocessing module, which is used to preprocess the medical imaging report text before determining the first positive sentence corresponding to the medical imaging report text, wherein the preprocessing operation comprises at least one of the following: text cleaning; word segmentation; and stemming and lemmatization.

[0108] Optionally, the first positive sentence determination module 710 includes: a key information determination submodule, which is used to determine the key information of the medical imaging report text through named entity recognition, wherein the key information includes medical terms and diagnostic conclusions; a relationship information determination submodule, which is used to determine the relationship information of the medical imaging report text through relationship extraction, wherein the relationship information is used to indicate the relationship between the key information; and a first positive sentence determination submodule, which is used to determine the first positive sentence based on the key information and the relationship information.

[0109] Optionally, the instruction determination module 720 includes: a target critical value determination submodule, which is used to match the critical value in the critical value standard library with the first positive sentence to determine the target critical value corresponding to the first positive sentence; and an instruction determination submodule, which is used to determine the instruction based on the keywords corresponding to the target critical value and the medical imaging report text.

[0110] Optionally, the critical value warning module 740 includes: a critical value warning information output submodule, which is used to output critical value warning information when the matching degree between the second positive sentence and the first positive sentence is greater than a predetermined threshold.

[0111] Thus, according to the technical solution of this embodiment, after the doctor has completed the text of the medical image report. The key positive sentence corresponding to the text of the medical image report can be determined by a computing device such as a server. Then the key positive sentence is matched with the critical value standard library to determine the instruction corresponding to the key positive sentence. Then the instruction is input into an intelligent agent based on a large language model to generate a verification positive sentence corresponding to the instruction. Finally, when the verification positive sentence matches the key positive sentence successfully, the critical value warning information corresponding to the key positive sentence is output. Thus, in this way, the text content in the medical image report can be automatically parsed with the help of artificial intelligence and natural language processing technology, and then based on the pre-established preset critical value standard, the technical solution of this embodiment can automatically determine whether there is a critical value in the report and issue an early warning. Thus, the existing medical information platform in the prior art is still insufficient in critical value management, lacks intelligent reminder and tracking tools, and makes it difficult for doctors to fully cover all potential risk points even in a highly focused state.

[0112] Example 3 Figure 8 FIG. 8 is a diagram showing a critical value warning device 800 for a medical image report according to this embodiment, and the device 800 corresponds to the method according to the first aspect of Embodiment 1. Figure 8 As shown, the device 800 includes: a processor 810; and a memory 820, which is connected to the processor 810 and is used to provide the processor 810 with instructions for processing the following processing steps: determining a first positive sentence corresponding to the medical imaging report text, wherein the first positive sentence is used to describe the abnormal human body information contained in the medical imaging report text; matching the first positive sentence with a preset critical value standard library to determine the instruction corresponding to the first positive sentence; inputting the instruction into an intelligent agent based on a large language model to generate a second positive sentence corresponding to the instruction; and when the second positive sentence matches the first positive sentence, outputting critical value warning information corresponding to the first positive sentence.

[0113] Optionally, the memory 820 is also used to provide instructions for the processor 810 to process the following processing steps: before determining the first positive sentence corresponding to the medical imaging report text, preprocessing the medical imaging report text, wherein the preprocessing operation includes at least one of the following: text cleaning; word segmentation; and stemming and lemma restoration.

[0114] Optionally, the operation of determining the first positive sentence corresponding to the medical imaging report text includes: determining key information of the medical imaging report text through named entity recognition, wherein the key information includes medical terms and diagnostic conclusions; determining relationship information of the medical imaging report text through relationship extraction, wherein the relationship information is used to indicate the relationship between the key information; and determining the first positive sentence based on the key information and the relationship information.

[0115] Optionally, the first positive sentence is matched with a preset critical value standard library to determine the operation of the instruction corresponding to the first positive sentence, including: matching the critical value in the critical value standard library with the first positive sentence to determine the target critical value corresponding to the first positive sentence; and determining the instruction based on the keywords corresponding to the target critical value and the medical imaging report text.

[0116] Optionally, when the second positive sentence matches the first positive sentence, the operation of outputting critical value warning information corresponding to the first positive sentence includes: when the matching degree between the second positive sentence and the first positive sentence is greater than a predetermined threshold, outputting critical value warning information.

[0117] Thus, according to the technical solution of this embodiment, after the doctor has completed the text of the medical image report. The key positive sentence corresponding to the text of the medical image report can be determined by a computing device such as a server. Then the key positive sentence is matched with the critical value standard library to determine the instruction corresponding to the key positive sentence. Then the instruction is input into an intelligent agent based on a large language model to generate a verification positive sentence corresponding to the instruction. Finally, when the verification positive sentence matches the key positive sentence successfully, the critical value warning information corresponding to the key positive sentence is output. Thus, in this way, the text content in the medical image report can be automatically parsed with the help of artificial intelligence and natural language processing technology, and then based on the pre-established preset critical value standard, the technical solution of this embodiment can automatically determine whether there is a critical value in the report and issue an early warning. Thus, the existing medical information platform in the prior art is still insufficient in critical value management, lacks intelligent reminder and tracking tools, and makes it difficult for doctors to fully cover all potential risk points even in a highly focused state.

[0118] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0119] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0120] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0121] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0122] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0123] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0124] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A critical value warning method for medical imaging reports, characterized in that: include: Determining a first positive sentence corresponding to the medical imaging report text, wherein the first positive sentence is used to describe human body abnormality information contained in the medical imaging report text; Matching the first positive sentence with a preset critical value standard library to determine an instruction corresponding to the first positive sentence; Inputting the instruction into an agent based on a large language model to generate a second positive sentence corresponding to the medical imaging report text, wherein the second positive sentence is used to describe the human body abnormality information contained in the medical imaging report text; as well as When the second positive sentence matches the first positive sentence, critical value warning information corresponding to the first positive sentence is output.

2. The method according to claim 1, characterized in that Before determining the first positive sentence corresponding to the medical imaging report text, the method also includes preprocessing the medical imaging report text, wherein the preprocessing operation includes at least one of the following: text cleaning; word segmentation; and stemming and lemmatization.

3. The method according to claim 1, characterized in that The operation of determining the first positive sentence corresponding to the medical imaging report text includes: Determining key information of the medical imaging report text by named entity recognition, wherein the key information includes medical terms and diagnosis conclusions; Determining the relationship information of the medical imaging report text by relationship extraction, wherein the relationship information is used to indicate the relationship between the key information; and The first masculine sentence is determined according to the key information and the relationship information.

4. The method according to claim 1, characterized in that: Matching the first positive sentence with a preset critical value standard library to determine the operation of the instruction corresponding to the first positive sentence includes: Matching the critical value in the critical value standard library with the first positive sentence to determine a target critical value corresponding to the first positive sentence; and The instruction is determined according to a keyword corresponding to the target critical value and the medical imaging report text.

5. The method according to claim 1, characterized in that When the second positive sentence matches the first positive sentence, the operation of outputting the critical value warning information corresponding to the first positive sentence includes: When the matching degree between the second positive sentence and the first positive sentence is greater than a predetermined threshold, the critical value warning information is output.

6. The method according to claim 1, characterized in that The process also includes fine-tuning the agent by: Connecting the large language model to a positive sentence matching module for matching positive sentences, connecting the positive sentence matching module to a binary classification model, and freezing parameters of the positive sentence matching module and the binary classification model, wherein the binary classification model is used to map the similarity matching value output by the positive sentence matching module to a probability value for determining whether it is a critical value; Constructing a training sample set, wherein each training sample in the training sample set includes a sample instruction, a sample positive sentence, and a true critical state value, wherein the true critical state value is used to indicate whether the sample positive sentence is a critical value; Input each sample instruction into the large language model to generate a corresponding predicted positive sentence, input the predicted positive sentence into the positive sentence matching module to match it with the corresponding sample positive sentence, and output the corresponding similarity matching degree; Inputting the similarity matching degree into the binary classification model, thereby outputting the predicted critical probability that the sample positive sentence is a critical value; According to the predicted critical probabilities corresponding to each sample instruction and the corresponding actual critical state values, the large language model is fine-tuned using a preset loss function.

7. The method according to claim 6, characterized in that The cost value corresponding to each predicted critical probability in the loss function is determined by: According to each predicted critical probability and the corresponding actual critical state value, a cost value corresponding to each predicted critical probability is determined in a preset asymmetric cost matrix, wherein in the asymmetric cost matrix, the cost value of predicting a critical value as a non-critical value is greater than the cost value of predicting a non-critical value as a critical value.

8. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the processor executes the method according to any one of claims 1 to 7.

9. A critical value warning device for medical image reports, characterized in that: include: A first positive sentence determination module, used to determine a first positive sentence corresponding to the medical imaging report text, wherein the first positive sentence is used to describe the human body abnormality information contained in the medical imaging report text; An instruction determination module, used for matching the first positive sentence with a preset critical value standard library to determine an instruction corresponding to the first positive sentence; A second positive sentence generation module, used for inputting the instruction into an agent based on a large language model, and generating a second positive sentence corresponding to the medical imaging report text, wherein the second positive sentence is used for describing the human body abnormality information contained in the medical imaging report text; as well as The critical value warning module is used to output critical value warning information corresponding to the first positive sentence when the second positive sentence matches the first positive sentence.

10. A critical value warning device for medical image reports, characterized in that: include: processor; as well as A memory, connected to the processor, configured to provide the processor with instructions for processing the following processing steps: Determine a first positive sentence corresponding to the medical imaging report text, wherein the first positive sentence is used to describe human body abnormality information contained in the medical imaging report text; Matching the first positive sentence with a preset critical value standard library to determine an instruction corresponding to the first positive sentence; Inputting the instruction into an agent based on a large language model to generate a second positive sentence corresponding to the medical imaging report text, wherein the second positive sentence is used to describe the human body abnormality information contained in the medical imaging report text; as well as When the second positive sentence matches the first positive sentence, critical value warning information corresponding to the first positive sentence is output.

Images