Acute coronary syndrome classification prediction method and related equipment

Abstract

The invention relates to an acute coronary syndrome classification prediction method and related equipment. The method comprises the following steps: acquiring original information which is input by a user and comprises an acute coronary syndrome classification task about a target patient, and acquiring a hospital admission record text and target case information of the target patient; interacting with a user to determine the task content of the classification task, and constructing a cue word based on the task content; searching a plurality of historical admission record texts most similar to the admission record text of the target patient in a retrieval library as associated admission records; and inputting the classification task, the cue word and the associated admission record into a large language model, and processing the classification task by the model based on the cue word and the associated admission record to obtain the acute coronary syndrome category of the target patient. According to the method, the big language model is adopted to process the acute coronary syndrome classification task, resources consumed when the scheme is implemented are reduced, and the classification prediction performance and accuracy of the model are improved in combination with multiple prompt engineering technologies.

Description

Technical Field

[0001] The present application relates to the intersection of cardiovascular medicine and artificial intelligence technology, and in particular to an acute coronary syndrome classification and prediction method and related equipment. Background Art

[0002] Acute coronary syndrome refers to an acute ischemic syndrome of the heart caused by thrombosis due to rupture or erosion of unstable atherosclerotic plaques in the coronary arteries. It is a common and serious cardiovascular disease and a severe type of coronary heart disease. Patients often present with symptoms such as paroxysmal chest pain and chest tightness, which can lead to arrhythmias, heart failure, and even sudden death, seriously affecting the patient's quality of life and life span. Acute coronary syndrome includes ST-segment elevation myocardial infarction (STEMI), non-ST-segment elevation myocardial infarction (NSTEMI), and unstable angina (UA). Different types of acute coronary syndrome require different treatments and prognostic strategies. Therefore, accurately judging the type of acute coronary syndrome and taking appropriate treatments in a timely manner can greatly reduce the mortality rate, reduce complications, and improve the patient's prognosis.

[0003] In the cardiovascular field, there are currently some auxiliary diagnosis solutions based on deep learning algorithms such as BERT (Bidirectional Encoder Representations from Transformers), which doctors can use to complete clinical diagnosis more quickly and accurately. These solutions include predicting whether a patient has ST-segment elevation myocardial infarction, predicting whether a patient has non-ST-segment elevation myocardial infarction, and predicting whether a patient has unstable angina. However, these BERT-based solutions require more resources for model training during implementation. For example, training a prediction model for related diseases requires collecting a large number of case samples of related diseases, screening out disease characteristics related to related diseases and performing feature engineering, as well as subsequent iterative training and model evaluation. Summary of the invention

[0004] In view of the above-mentioned deficiencies or shortcomings, the present application provides an acute coronary syndrome classification prediction method and related equipment. The embodiment of the present application adopts a large language model to process the acute coronary syndrome classification task, which can accurately identify which type of acute coronary syndrome the patient suffers from. At the same time, the large language model does not need to be pre-trained and fine-tuned before use, which can reduce the resources required for the implementation of the solution. The embodiment of the present application also combines a variety of prompt engineering technologies to improve the classification prediction performance and accuracy of the model.

[0005] The present application provides a method for classification and prediction of acute coronary syndrome according to a first aspect. In some embodiments, the method comprises: Obtaining original information input by a user; the original information includes an acute coronary syndrome classification task about a target patient; Acquire the admission record text of the target patient and the target case information specified by the user, wherein the target case information includes typical case information of each category of acute coronary syndrome; Interact with the user to determine the task content of the acute coronary syndrome classification task, and construct prompt words based on the task content; the prompt words include the admission record text of the target patient and the target case information; Searching the search database for multiple historical admission record texts that are most similar to the admission record texts of the target patient as associated admission records; The acute coronary syndrome classification task, prompt words and associated admission records are input into the large language model, so that the large language model processes the acute coronary syndrome classification task based on the prompt words and the associated admission records to obtain the acute coronary syndrome category of the target patient.

[0006] In some embodiments, the task content includes a classification target and a precondition corresponding to the classification target; interacting with a user to determine the task content of the acute coronary syndrome classification task includes: A first prompt word is input into the large language model, where the first prompt word is used to instruct the large language model to interact with the user multiple times, and in each interaction, task content to be confirmed is generated according to task information and context information of the acute coronary syndrome classification task, and the user is prompted to confirm the task content to be confirmed; the context information includes the content to be confirmed generated by the large language model in each previous interaction and the confirmation information input by the user.

[0007] In some embodiments, the prompt word also includes a plurality of thinking points; the prompt word is constructed based on the task content, including: Fill the task content into the preset prompt word template to obtain the second prompt word; A second prompt word is input into the large language model, where the second prompt word is used to instruct the large language model to decompose the acute coronary syndrome classification task based on the user's judgment ideas to obtain the multiple thinking points.

[0008] In some embodiments, the target patient's admission record text includes element information of multiple clinical elements; the multiple clinical elements include multiple of the chief complaint, current medical history, past medical history, infectious disease history, personal life history, family history, physical examination, and auxiliary examinations; the multiple thinking points include the thinking points corresponding to each clinical element.

[0009] In some embodiments, the search library includes a first database for storing vector data and a second database for storing keywords; searching the search library for multiple historical admission record texts that are most similar to the admission record text of the target patient as associated admission records includes: Encode the target patient's admission record text into a vector, search the first database based on the encoded vector, and obtain n historical admission record texts based on the search results; Extract keywords from the admission record text of the target patient, search the second database based on the extracted keywords, and obtain m historical admission record texts based on the search results; Combining n historical admission record texts, m historical admission record texts and the ranking of each historical admission record text, a search result table is obtained; The historical admission record texts in the search result table are reordered, and the associated admission records are obtained based on the reordered results.

[0010] In some embodiments, the original information includes a patient identifier of a target patient; obtaining the admission record text of the target patient includes: The admission record text of the target patient is obtained from the medical database according to the patient identifier of the target patient.

[0011] In some embodiments, the training process of the large language model includes: Acquire acute coronary syndrome case data; the acute coronary syndrome case data includes the admission record texts of multiple patients with ST-segment elevation myocardial infarction, the admission record texts of multiple patients with non-ST-segment elevation myocardial infarction, the admission record texts of patients with unstable angina, and the acute coronary syndrome category corresponding to each admission record text; The pre-trained general large language model is incrementally trained using acute coronary syndrome case data to obtain a large language model.

[0012] According to a second aspect, the present application provides an acute coronary syndrome classification and prediction device. In some embodiments, the device includes: The original information acquisition module is used to acquire the original information input by the user; the original information includes the acute coronary syndrome classification task about the target patient; A case information acquisition module is used to acquire the admission record text of the target patient and the target case information specified by the user, and the target case information includes typical case information of each category of acute coronary syndrome; A prompt word construction module is used to interact with the user to determine the task content of the acute coronary syndrome classification task and to construct prompt words based on the task content; the prompt words include the admission record text of the target patient and the target case information; A related record retrieval module is used to search the retrieval library for multiple historical admission record texts that are most similar to the admission record text of the target patient as related admission records; The disease classification module is used to input the acute coronary syndrome classification task, prompt words and related admission records into the large language model, so that the large language model processes the acute coronary syndrome classification task based on the prompt words and related admission records to obtain the acute coronary syndrome category of the target patient.

[0013] According to a third aspect, the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the acute coronary syndrome classification prediction method provided in any of the above embodiments are implemented.

[0014] According to a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the acute coronary syndrome classification and prediction method provided in any of the above embodiments are implemented.

[0015] The above-mentioned embodiment of the present application uses a large language model to process the acute coronary syndrome classification task, which can reduce the resources consumed when implementing the scheme. The embodiment of the present application automatically constructs a prompt word (prompt) including the admission record text of the target patient based on the original information input by the user, reducing the difficulty of the user to get started. The embodiment of the present application also combines few-sample learning and retrieval enhancement generation to improve the classification prediction performance and accuracy of the large language model for acute coronary syndrome, specifically adding target case information containing typical case information of each category of acute coronary syndrome to the prompt word, and obtaining multiple historical admission record texts that are most similar to the admission record text of the target patient as associated admission records. The associated admission record will be input into the large language model together with the prompt word and the acute coronary syndrome classification task about the target patient included in the original information as input data. Under the instruction of the prompt word, the large language model will process the acute coronary syndrome classification task based on the prompt word and the associated admission record, and identify the acute coronary syndrome category of the target patient. In addition, the embodiment of the present application also introduces adaptive thinking chain technology based on few-sample learning and retrieval enhanced generation, so as to further improve the classification prediction performance and accuracy of the large language model. Among them, the embodiment of the present application adaptively breaks down the acute coronary syndrome classification task into multiple thinking points and acts on the prompt words. These thinking points include thinking points corresponding to specified clinical elements, which can guide the large language model to think step by step when processing the acute coronary syndrome classification task, and finally summarize the thinking results of each step and derive the prediction results of the acute coronary syndrome category of the target patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a method for classification and prediction of acute coronary syndrome provided by the present application according to one or more embodiments; Figure 2 A schematic diagram of an acute coronary syndrome classification process combining multiple prompt engineering technologies based on a large language model according to one or more embodiments of the present application; Figure 3 This is a structural block diagram of an acute coronary syndrome classification and prediction device provided by the present application according to one or more embodiments; Figure 4 This is a diagram of the internal structure of a computer device provided by the present application according to one or more embodiments. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below in conjunction with the accompanying drawings. It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0018] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims.

[0019] In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances. In addition, in the description of the present application, unless otherwise specified, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship.

[0020] In response to the deficiencies or defects of related technologies, the present application provides an acute coronary syndrome classification and prediction method, which adopts a large language model to process the acute coronary syndrome classification task, can reduce the resources required for the implementation of the scheme, and also improves the classification prediction performance and accuracy of the large language model for the acute coronary syndrome classification task by introducing a variety of prompt engineering technologies.

[0021] In some embodiments, the method can be applied to a disease classification system, which can be a cloud platform (can be an independent server, server cluster, etc.). Users (usually doctors, but can also be other roles, such as researchers) can interact with the system through user terminals (such as smart phones, PCs, tablet computers and other computer devices), provide the system with acute coronary syndrome classification tasks (referred to as classification tasks), and the system processes the received classification tasks and provides feedback to the user.

[0022] The acute coronary syndrome classification prediction method provided in this embodiment includes the following steps: Figure 1 The steps shown are described below using the method applied to the above-mentioned disease classification system as an example.

[0023] S110: Acquire original information input by the user, where the original information includes an acute coronary syndrome classification task about a target patient.

[0024] The original information is the information input by the user in the disease classification system to instruct the system to identify which type of acute coronary syndrome the target patient suffers from. This embodiment does not specifically limit the format of the original information, which can be a text format, a voice format, or other formats. If the format of the original information is not a text format, it is necessary to convert the format into a text format before subsequent processing. For example, assuming that the user inputs a voice, the system can use speech recognition technology (Automatic Speech Recognition, ASR) to convert the voice into text.

[0025] The acute coronary syndrome classification task for the target patient is a disease classification problem in the original information. The target patient can be any patient who suffers from or is suspected of suffering from acute coronary syndrome, and the system can perform semantic analysis on the original information to determine the target patient. For example, if the original information is "determine which acute coronary syndrome the patient in bed 8 suffers from", then the acute coronary syndrome classification task can be "determine which acute coronary syndrome the patient in bed 8 suffers from", and the target patient is "the patient in bed 8"; for another example, if the original information is "the patient's admission record text is xxx, determine whether the patient has ST-segment elevation myocardial infarction or non-ST-segment elevation myocardial infarction", then the acute coronary syndrome classification task can be "determine whether the patient has ST-segment elevation myocardial infarction or non-ST-segment elevation myocardial infarction", and the target patient is "the patient", or "the patient whose admission record text is xxx".

[0026] S120: Acquire the admission record text of the target patient and the target case information specified by the user. The target case information includes typical case information of each category of acute coronary syndrome.

[0027] In some embodiments, if the user also inputs the target patient's admission record text when providing the acute coronary syndrome classification task, the target patient's admission record text is directly extracted from the original information.

[0028] In other embodiments, the original information includes the patient identification of the target patient; obtaining the admission record text of the target patient includes: obtaining the admission record text of the target patient from a medical database according to the patient identification of the target patient. If the user does not enter the admission record text of the target patient when providing the acute coronary syndrome classification task, but enters the patient identification of the target patient, the admission record text corresponding to the patient identification of the target patient can be queried and obtained from the medical database according to the patient identification of the target patient. A medical database refers to a database used to store the medical data of each patient (including the patient's admission record text). For example, when the system is a cloud platform of one or more hospitals, the medical database can be used to store the medical data of each patient in each hospital.

[0029] The target case information specified by the user refers to the typical case information of each category of acute coronary syndrome specified by the user, which may include the admission record texts of multiple typical patients with ST-segment elevation myocardial infarction, the admission record texts of multiple typical patients with non-ST-segment elevation myocardial infarction, and the admission record texts of multiple typical patients with unstable angina pectoris.

[0030] In prompt engineering, few-shot learning usually refers to including a small number of contextual examples (usually 1 to 5) in the prompt to help the model understand the task requirements and generate expected outputs. In this embodiment, the system can input some typical cases of acute coronary syndrome specified by the user into the large language model, allowing the model to learn case samples that are strongly related to the classification task before processing the classification task, thereby improving the performance of the model.

[0031] In each embodiment of the present application, the admission record text of each patient includes element information of multiple clinical elements; the multiple clinical elements include multiple of the chief complaint, current medical history, past medical history, infectious disease history, personal life history, family history, physical examination, and auxiliary examination. It should be noted that the admission record text of the target patient does not contain a classification label, while the admission record text of other patients for reference by the large language model contains a classification label, which is used to indicate the patient's acute coronary syndrome category.

[0032] For example, the admission record text of a patient with unstable angina pectoris includes the following element information of each clinical element: Chief complaint: chest tightness and chest pain for 24 hours, aggravated for 16 hours Current medical history: The patient had chest tightness without obvious cause 24 hours ago, accompanied by chest and back pain, which was tolerable, without obvious radiation, no sweating, nausea and vomiting, no headache and dizziness, no cough, sputum and fever, no syncope, and improved after resting for 2-3 minutes. 16 hours ago, the patient had worsening chest tightness after activity, accompanied by a feeling of throat tightening and sweating. 120 was sent to the emergency department of our hospital for treatment. The electrocardiogram showed: Q wave formation in the inferior wall lead. Myocardial infarction three items + BNP showed: * creatine kinase isoenzyme 1.4 IU / L, serum myoglobin 23.2 ug / L, troponin I 0.018; BNP 13.2 pg / mL. Blood biochemistry showed: hyperlipidemia, hyperuricemia. After the patient arrived at the emergency department, the chest tightness and other symptoms had improved, and he was treated with oral aspirin + clopidogrel anticoagulant, enoxaparin anticoagulant, etc. The patient is in good condition now and was admitted to our department today with "unstable angina pectoris". Since the onset of the disease, the patient's mental state, sleep and physical strength are normal, and there is no special condition in bowel movements and diet, and there is no significant change in weight within half a year.

[0033] Past medical history: denied history of hypertension, diabetes, liver disease, kidney disease, hyperuricemia and hyperlipidemia for 3 years, not controlled by regular medication, denied history of blood transfusion, no history of food allergy, no history of drug allergy, vaccination history was as planned, history of surgical trauma, post-surgery for varicose veins of the lower limbs, no metal implants.

[0034] History of infectious diseases: Denying history of hepatitis, tuberculosis, SARS, avian influenza, and close contact.

[0035] Personal life history: born in the native place, no long-term residence in other places, no history of contact with schistosomiasis water, no history of living in endemic or infectious areas, no history of contact with toxic substances, dust and radioactive substances. He lives a relatively regular life, has a history of smoking for 22 years, 10 cigarettes / day, has a history of drinking for 30 years, is married, his spouse and children are healthy, and he has no history of prostitution.

[0036] Family history: Denies family history of genetic diseases and similar diseases.

[0037] Physical examination: General condition: T: 36.4℃ P: 68 times / min R: 18 times / min BP: 123 / 84mmHg, normal development, good nutrition, height 178cm, weight 87kg. Clear consciousness, autonomous signs, normal facial features, and cooperative with doctors.

[0038] Skin and mucous membranes: normal color, no edema, normal elasticity, no rash, no subcutaneous nodules or lumps, no spider nevi, no bleeding spots, and no ecchymosis.

[0039] Superficial lymph nodes: No enlargement was felt in the preauricular, postauricular, occipital, posterior cervical triangle, anterior cervical triangle, submandibular, submental, supraclavicular, axillary, supratrochlear, inguinal, and popliteal fossa. No redness, swelling, or ulceration was found on the surface of the skin.

[0040] Head and facial features: Skull: No deformity, no mass, evenly distributed hair, no tenderness in the skull, no scars. Eyes: No edema of the eyelids, no protrusion of the eyeballs, no tremor of the eyeballs, no restricted eye movement, no conjunctival congestion or edema, no yellowing of the sclera, transparent cornea, corneal reflex, equal size and roundness of the pupils on both sides, diameter: 3mm on the left and 3mm on the right. Sensitive to light reflex, normal convergence reflex. Ears: Normal auricle, no secretions in the external auditory canal, no tenderness in the mastoid, and normal hearing. Nose: Normal nose, centered nasal septum, no flaring of the nasal wings, no secretions, no bleeding, unobstructed nasal cavity, symmetrical nasolabial grooves, no tenderness in the maxillary sinus, ethmoid sinus, and frontal sinus. Oral cavity: No pale lips, no cyanosis, no swelling of the gums, no bleeding, intact oral mucosa, centered tongue, no tremor, no bad smell in the mouth, no missing teeth. Pharynx: No congestion in the pharynx, uvula in the middle, no swelling of the tonsils on both sides, no purulent secretions, and clear pronunciation.

[0041] Neck: The neck moves freely without neck resistance, no abnormal carotid pulsation, no jugular vein distention, no vascular murmur, the trachea is in the middle, the thyroid gland is not palpably enlarged, but is soft and non-tender, without tremor and no vascular murmur.

[0042] Thorax: The thoracic cage is without deformity, with symmetrical movement, no sternal tenderness, and no subcutaneous emphysema. Lungs: The respiratory movements are consistent, vocal fremitus is bilaterally symmetrical, percussion of both lungs is resonant, vocal resonance is normal, breath sounds of both lungs are clear, and no dry or wet rales are heard. There is no pleural friction rub. Heart: There is no prominence in the precordial area. The point of apical impulse is located 0.5 cm medial to the midclavicular line at the 5th left intercostal space. No abnormal impulse is seen, no thrill is palpable, no pericardial friction sensation is palpable. Heart rate is 68 beats per minute, rhythm is regular, heart sounds are normal, P2 > A2, no murmur is heard, and no pericardial friction rub is heard. The point of apical impulse is located 0.5 cm medial to the midclavicular line at the 5th left intercostal space. In the measurement of the cardiac silhouette, on the right side, it is 2 cm at the 2nd intercostal space (II), 3 cm at the 3rd intercostal space (III), and 4 cm at the 4th intercostal space (IV); on the left side, it is 2 cm at the 2nd intercostal space (II), 4 cm at the 3rd intercostal space (III), 6 cm at the 4th intercostal space (IV), and 8 cm at the 5th intercostal space (V). The size of the cardiac silhouette is 8.5 cm at maximum systole (MCL).

[0043] Peripheral vascular signs: No abnormal vascular signs are seen.

[0044] Abdomen: The abdomen is flat, abdominal respiration is normal, no gastric or intestinal peristaltic waves are seen, no abdominal wall varices are seen, the abdomen is soft, no muscle tension, no tenderness or rebound tenderness, no abdominal mass, percussion of the abdomen is tympanic, shifting dullness is negative, bowel sounds are 3 times per minute, and no vascular murmur is heard. Liver (gallbladder): The liver is not palpable below the costal margin. Murphy's sign is negative, hepatojugular reflux sign is negative. The hepatic dullness boundary exists, and the upper border of the liver is at the 5th right intercostal space along the right midclavicular line. Spleen: The spleen is not palpable below the costal margin, and there is no percussion pain in the splenic area. Kidneys: Both kidneys are not palpable, there is no percussion pain in the renal areas, no tenderness at the costal points, no tenderness at the upper ureteral points, and the middle ureteral orifice is left for further examination.

[0045] Spine: The physiological curvature of the spine exists, movement is not restricted, no tenderness, no percussion pain.

[0046] Extremities (joints): The extremities are without deformity, the joints are not swollen or tender, active movement is normal, no clubbing of fingers or toes, no muscle atrophy, no varicose veins, no edema.

[0047] Nervous system: The abdominal reflex is normal. The bilateral biceps brachii, bilateral triceps brachii, bilateral patellar tendons, and bilateral Achilles tendon reflexes are symmetrical. Babinski sign is negative, Kernig sign is negative, Brudzinski sign is negative, and Brunnstrom sign is positive.

[0048] Specialist examination: The patient is in good spirits, respiratory movements are consistent, vocal fremitus is bilaterally symmetrical, percussion of both lungs is resonant, vocal resonance is normal, breath sounds of both lungs are clear, and no dry or wet rales are heard. There is no pleural friction rub. There is no prominence in the precordial area, no abnormal impulse is seen, no thrill is palpable, no pericardial friction sensation is palpable. Heart rate is 68 beats per minute, rhythm is regular, heart sounds are normal, P2 < A2, no murmur is heard, and no pericardial friction rub is heard. The abdomen is soft, no muscle tension, and both lower extremities are not swollen.

[0049] Auxiliary examination: 2024-8-24 ECG: Q wave formation in inferior wall leads 2024-08-25 Myocardial infarction three items: *Creatine kinase isoenzyme 14 IU / L, serum myoglobin 23.2 ug / L, troponin I0.018; BNP13.2 pg / ml 2024-08-25 Triglycerides 3.30 mmol / L; Total cholesterol 5.88 mmol / L; Uric acid 567 umol / L Classification label: unstable angina In this example, the admission record texts of other types of acute coronary syndrome patients also include the element information of each clinical element in the above example, but the specific element information is different. Therefore, this example no longer provides admission record text samples for patients with ST-segment elevation myocardial infarction and non-ST-segment elevation myocardial infarction.

[0050] S130: Interact with the user to determine the task content of the acute coronary syndrome classification task, and construct prompt words based on the task content; the prompt words include the admission record text of the target patient and the target case information.

[0051] In order to improve the accuracy of classification prediction, after obtaining the acute coronary syndrome classification task, the system also interacts with the user to clarify the task content of the acute coronary syndrome classification task, and to make the large language model's understanding of the classification task consistent with the user's understanding as much as possible. Among them, the task content can include the classification target and the preconditions corresponding to the classification target. The classification target can refer to the user's diagnostic target.

[0052] S140: Searching the search library for multiple historical admission record texts that are most similar to the admission record text of the target patient as associated admission records.

[0053] The retrieval library stores the admission records of multiple patients with acute coronary syndrome who have been diagnosed. The system can retrieve multiple historical admission records that are most similar to the admission records of the target patient for reference by the large language model to improve the classification prediction performance and accuracy of the model.

[0054] In some embodiments, the search library includes a first database for storing vector data and a second database for storing keywords. Accordingly, searching the search library for multiple historical admission record texts that are most similar to the admission record text of the target patient as associated admission records includes: (1) Encode the target patient's admission record text into a vector, search the first database based on the encoded vector, and obtain n historical admission record texts based on the search results; (2) extracting keywords from the target patient's admission record text, searching the second database based on the extracted keywords, and obtaining m historical admission record texts based on the search results; n and m are both positive integers, and the two can be the same or different; (3) Combining the n historical admission record texts, the m historical admission record texts, and the ranking of each historical admission record text, a search result table is obtained; (4) Reorder the text of historical admission records in the search result table, and obtain related admission records based on the reordering results.

[0055] Retrieved Augmented Generation (RAG) technology is a method that combines information retrieval and natural language generation. It enhances the model's generation capabilities by retrieving relevant information from external knowledge sources, making the generated content more accurate and rich. Retrieved Augmented Generation solutions usually use vector-based or keyword-based methods to retrieve documents related to the question for reference by the large language model to enhance the model's generation capabilities.

[0056] In this embodiment, considering that the medical data that the large language model needs to process is mainly the patient's admission record text, which is relatively complex medical data, the current retrieval enhancement generation scheme is prone to missed detection. For example, when searching based on vectors, the retriever can only search for sentences with similar semantics, but the admission record text also includes information specific to the part of speech level, which is often not found when using vector retrieval, resulting in missed detection. To this end, this embodiment provides a retrieval enhancement generation scheme based on hybrid retrieval, which can improve the comprehensiveness of the retrieval results on the basis of ensuring the accuracy of the retrieval results and avoid missed detection.

[0057] Among them, this embodiment sets up two major databases, namely, a first database for storing vector data and a second database for storing keywords. Exemplarily, the first database can be a milvus database, and the second database can be an Elastic Search. During the search, the first database and the second database are searched based on the vector and the keyword respectively, and multiple cases of the admission record text closest to the target patient are found in terms of semantics and keywords. Then, the search results of the two databases are merged, and the case data in the merged results are re-sorted using a re-sorting algorithm. Finally, the top-ranked multiple case data are selected as the final search results.

[0058] Based on the admission record text of the target patient, vector search is performed in the first database, and keyword search is performed in the second database.

[0059] Regarding vector retrieval, the historical admission record text of each patient is first encoded into an embedded vector using an embedding model, and the flat index FLAT is used as the vector indexing method. When using the target patient's admission record text for retrieval, the target patient's admission record text is first encoded into an embedded vector, and then the vector similarity between the embedded vector and each embedded vector in the first database is calculated, and then the historical admission record text corresponding to the multiple embedded vectors with the highest vector similarity is output. For example, the historical admission record text can be encoded into a 1024-dimensional embedding vector (including 1023 dimensions and 1-dimensional classification labels), the target patient's admission record text can be encoded into a 1023-dimensional embedding vector, the L2 Euclidean distance between the 1023-dimensional embedding vector and the first 1023 dimensions of each 1024-dimensional embedding vector is calculated, and the Top-50 results are output based on the score.

[0060] Regarding keyword retrieval, a word segmenter, such as Jieba word segmenter, can be used to remove stop words (such as "almost", "almost", "!", "》" and other adverbs that may affect the generated keyword combination, or meaningless symbols) from the historical admission record text of each patient. Secondly, the word segmenter is used to automatically segment and extract keywords. The extracted keywords may include keyword information corresponding to each clinical element in the admission record text. Finally, the keywords extracted from each historical admission record text are stored in the second database. When the target patient's admission record text is used for retrieval, the admission record text is processed by the word segmenter to extract keywords, and then the similarity between the keywords extracted from the target patient's admission record text and the keywords in the second database is calculated (the BM25 algorithm can be used to calculate the similarity score). Finally, the historical admission record text corresponding to multiple keywords with the highest similarity score is output. The keyword retrieval can output the top 50 results like the vector retrieval. The ranking of the retrieval results of the keyword retrieval depends on the number of similar keywords. The number of similar keywords is large and the corresponding high score is obtained. In some embodiments, the similarity score of two keywords may be obtained based on the similarity score between the keyword information of each clinical element in the two keywords, for example, the average value of the similarity scores between the keyword information of all clinical elements.

[0061] Finally, the results of the two searches and their rankings are combined into a search result table, and RRF is used as a simple re-ranking algorithm to calculate the score of each historical admission record text in the search result table. Finally, the top multiple (e.g., 10) historical admission record texts are selected as associated record texts.

[0062] S150: Inputting the acute coronary syndrome classification task, prompt words and associated admission records into the large language model, so that the large language model processes the acute coronary syndrome classification task based on the prompt words and associated admission records to obtain the acute coronary syndrome category of the target patient.

[0063] In this embodiment, the basic framework of the large language model is zero-shot learning. The prompt words that need to be input into the large language model may be, for example, "You are a clinical cardiologist specializing in ST-segment elevation myocardial infarction (STEMI), non-ST-segment elevation myocardial infarction (NSTEMI) and unstable angina (UA). Your job is to judge whether the patient has unstable angina, ST-segment elevation myocardial infarction or non-ST-segment elevation myocardial infarction based on the patient's admission record text (including chief complaint, current medical history, past history, infectious disease history, personal life history, family history, physical examination report, auxiliary examination report, etc.) as a criterion."

[0064] Based on the above-mentioned prompt words, a variety of prompt engineering technologies are introduced, including few-shot learning and retrieval-enhanced generation (or retrieval-enhanced generation based on hybrid retrieval), to enhance the acute coronary syndrome classification prediction performance of the large language model.

[0065] In some embodiments, the large language model can be a general large language model, such as a GPT-4o large language model. In other embodiments, the large language model can be obtained based on the training of the general large language model, wherein the training process of the large language model includes: obtaining acute coronary syndrome case data; the acute coronary syndrome case data includes the admission record texts of multiple ST-segment elevation myocardial infarction patients, the admission record texts of multiple non-ST-segment elevation myocardial infarction patients, the admission record texts of unstable angina patients, and the acute coronary syndrome category corresponding to each admission record text; using the acute coronary syndrome case data to incrementally train the pre-trained general large language model to obtain a large language model. This embodiment only needs to use a small amount of acute coronary syndrome-related data to fine-tune the general large language model, so that the model can learn professional knowledge, thereby improving the classification prediction performance of the model for the acute coronary syndrome classification task.

[0066] In some embodiments, interacting with a user to determine the task content of an acute coronary syndrome classification task includes: inputting a first prompt word into a large language model, the first prompt word being used to instruct the large language model to interact with the user multiple times, generating task content to be confirmed in each interaction based on task information and context information of the acute coronary syndrome classification task, and prompting the user to confirm the task content to be confirmed; the context information includes the content to be confirmed generated by the large language model in each previous interaction and the confirmation information input by the user.

[0067] A web front end can be constructed, and the large language model and the user interact through the front end page, and the user's classification target and preconditions are constantly clarified through interaction. For example, the first prompt word can be as follows: “You are an expert at expressing, exploring and clarifying the user’s classification task goals. Your task is to continuously clarify the classification tasks input by the user until you finally summarize it into a sentence when you think it is consistent with the user’s goal and very clear.

[0068] Please follow the following rules when clarifying user classification requirements: -1. Since users may use the service in any clinical scenario or scientific research environment, you need to adapt to the user's language style considering their different backgrounds.

[0069] -2. After clarifying the user’s background and research direction, you need to put yourself in the role of an expert in the field and say: “I am a (mask) expert” when outputting.

[0070] -3. When clarifying user goals, you need to think as a (mask) expert and explore the prerequisites required for classification.

[0071] -4. Always remind the user whether the text you provide perfectly describes his classification task.

[0072] -5. If the user is satisfied, you will output: Since you think my answer is detailed enough, your goal is actually: ________ -6. The output target should include all the conditions that have been clarified and output as a heuristic fill-in-the-blank text.

[0073] -7. If you have identified the user's goal, your answer must include: Since you think my answer is detailed enough, your goal is actually: ________ -8. If you think the goal is still not specific enough then the output must include: I don’t think I fully understand your task yet.

[0074] -9. When users determine the input and output of a task, they must provide input examples to help clarify the task. The (mask) in the prompt word in the above example is a prompt. During the interaction between the large language model and the user, it is used to guide the large language model to fill in specific information to ensure that the final output can accurately meet the user's needs.

[0075] In some embodiments, the prompt words constructed by the system also include multiple key thinking points; constructing prompt words based on task content includes: filling the task content into a preset prompt word template to obtain a second prompt word; inputting the second prompt word into the large language model, and the second prompt word is used to instruct the large language model to disassemble the acute coronary syndrome classification task based on the user's judgment ideas to obtain the multiple key thinking points.

[0076] After the system determines the task content, it can generate a second prompt word based on a preset prompt word template. The second prompt word is used to instruct the large language model to generate a thought chain, which includes multiple thinking points. After the large language model generates the thought chain, the system adds the thought chain to the prompt word, thereby guiding the large language model to think step by step when processing the acute coronary syndrome classification task, and finally summarizes the thinking results of each step and derives the prediction result of the acute coronary syndrome category of the target patient.

[0077] For example, the preset prompt word template may be as follows: "You are a rational decision maker. Now there is a classification task. The goal is (mask) I need you to design a series of important thinking chains to assist this judgment process.

[0078] **Be sure to follow the following rules when designing a thinking chain: -1. Accurate and rich description of thinking nodes.

[0079] -2. It is best to use thinking nodes based on inspirational filling-in-the-blanks to reduce ambiguity and enhance the accuracy of the answer. For example: Is there a glass shadow: __ yes / no __ -3. Ensure that the overall chain of thought conforms to the expert logic for solving the problem.

[0080] -4. The design thinking chain will be used for judgment after the big model, and the thinking chain should be based on the characteristics of the classification task.

[0081] -5. It is impossible to know whether the classification result is correct or not without comparing with the gold standard.

[0082] -6. Give an example of the use of thought chain.

[0083] -7. Since toolcalling cannot be performed, all statistical and mathematical evaluation features such as weights, probabilities, frequencies, etc. should be replaced directly with binary classification problems.

[0084] -8. Apply the chain of thought to the examples provided by the user, solve the identified chain of thought, and if too many are provided, then only one of them can be selected. " The system fills the task content determined by interacting with the user into the mask in the prompt word template to obtain the second prompt word.

[0085] See also Figure 2 This embodiment also introduces an adaptive thinking chain based on few-sample learning and hybrid RAG, which can further improve the classification prediction performance and accuracy of the large language model. Among them, the related operations of the adaptive thinking chain can break down the acute coronary syndrome classification task into multiple thinking points and act on the prompt words. These thinking points include the thinking points corresponding to the specified clinical elements, which can guide the large language model to think step by step when processing the acute coronary syndrome classification task, and finally summarize the thinking results of each step and derive the prediction results of the acute coronary syndrome category of the target patient.

[0086] In some embodiments, the multiple key thinking points included in the thinking chain include key thinking points corresponding to each clinical factor. For example, the thinking chain decomposed by the large language model combined with the doctor's judgment thinking for acute coronary syndrome may include the following ten key thinking points: “1. Chief complaint Unstable angina (UA): Patients usually describe chest pain or discomfort that is similar in nature and characteristics to stable angina, but the pain attacks are more frequent, more severe, or last longer.

[0087] STEMI and NSTEMI: Patients usually present with acute, severe chest pain, which may be accompanied by sweating, nausea, vomiting, or difficulty breathing.

[0088] What disease do you think the patient has now? [mask] 2. History of present illness UA: Ask whether there are any triggers such as fatigue, emotional excitement or cold, as well as the frequency, duration and relief of chest pain.

[0089] STEMI and NSTEMI: Pay attention to the onset time, nature, radiation site, accompanying symptoms and history of angina pectoris of chest pain.

[0090] What disease do you think the patient has now? [mask] 3. Past history Find out whether the patient has a history of coronary heart disease, hypertension, diabetes, hyperlipidemia, etc.

[0091] What disease do you think the patient has now? [mask] 4. History of infectious diseases Check for infectious diseases related to heart disease, such as rheumatic fever, syphilis, etc.

[0092] What disease do you think the patient has now? [mask] 5. Personal life history Consider the patient's smoking, drinking, eating habits, exercise frequency, etc.

[0093] What disease do you think the patient has now? [mask] 6. Family history Whether there is a history of premature coronary heart disease in the family.

[0094] What disease do you think the patient has now? [mask] 7. Physical examination Pay attention to heart rate, blood pressure, heart sounds, lung rales, etc.

[0095] What disease do you think the patient has now? [mask] 8. Auxiliary examination Electrocardiogram (ECG): STEMI: There is usually ST-segment elevation and / or new left bundle branch block on the ECG.

[0096] NSTEMI: There may be ST depression or T wave inversion on the ECG, but no ST elevation.

[0097] UA: ECG may be normal or show nonspecific ST segment and T wave changes.

[0098] Myocardial injury markers: Troponin (cTn): Usually elevated in STEMI and NSTEMI, but usually normal or only slightly elevated in UA.

[0099] Creatine kinase isoenzyme (CK-MB): elevated in STEMI, possibly elevated in NSTEMI, usually normal in UA.

[0100] Coronary angiography: For patients with suspected ACS, coronary angiography can confirm the diagnosis and guide treatment.

[0101] What disease do you think the patient has now? [mask] 9. The disease with the most occurrences is: [mask] 10. Based on all your diagnoses, the disease you think is most likely right now is: [mask]” The (mask) in the thought chain of the above example is used to guide the large language model to fill in specific information when processing classification tasks.

[0102] It should be noted that in the embodiments of the present application, patient information or patient data, such as the patient's admission record text and other related data, need to be obtained and processed with the patient's authorization and consent. When the embodiments of the present application are applied to specific products or technologies, the patient's permission or consent must be obtained, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0103] In addition, with respect to the various steps included in the classification prediction method for acute coronary syndrome provided in any of the above embodiments, unless clearly stated in this document, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of these steps can include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0104] Based on the same inventive concept, the present application also provides an acute coronary syndrome classification prediction device. In some embodiments, Figure 3 As shown, the device includes the following modules: The original information acquisition module 110 is used to acquire the original information input by the user; the original information includes the acute coronary syndrome classification task about the target patient; The case information acquisition module 120 is used to acquire the admission record text of the target patient and the target case information specified by the user, wherein the target case information includes typical case information of each category of acute coronary syndrome; The prompt word construction module 130 is used to interact with the user to determine the task content of the acute coronary syndrome classification task and construct prompt words based on the task content; the prompt words include the admission record text of the target patient and the target case information; The associated record retrieval module 140 is used to search the retrieval library for multiple historical admission record texts that are most similar to the admission record text of the target patient as associated admission records; The disease classification module 150 is used to input the acute coronary syndrome classification task, prompt words and related admission records into the large language model, so that the large language model processes the acute coronary syndrome classification task based on the prompt words and related admission records to obtain the acute coronary syndrome category of the target patient.

[0105] In some embodiments, the task content includes a classification target and a precondition corresponding to the classification target; the prompt word construction module includes: The task content confirmation submodule is used to input a first prompt word into the large language model, where the first prompt word is used to instruct the large language model to interact with the user multiple times, generate task content to be confirmed based on the task information and context information of the acute coronary syndrome classification task in each interaction, and prompt the user to confirm the task content to be confirmed; the context information includes the content to be confirmed generated by the large language model in each previous interaction and the confirmation information input by the user.

[0106] In some embodiments, the prompt word also includes a plurality of thinking points; the prompt word construction module includes: A filling submodule is used to fill the task content into the preset prompt word template to obtain a second prompt word; The thinking key points generation submodule is used to input a second prompt word into the large language model, and the second prompt word is used to instruct the large language model to disassemble the acute coronary syndrome classification task based on the user's judgment ideas to obtain the multiple thinking key points.

[0107] In some embodiments, the target patient's admission record text includes element information of multiple clinical elements; the multiple clinical elements include multiple of the chief complaint, current medical history, past medical history, infectious disease history, personal life history, family history, physical examination, and auxiliary examinations; the multiple thinking points include the thinking points corresponding to each clinical element.

[0108] In some embodiments, the search library includes a first database for storing vector data and a second database for storing keywords; the associated record search module includes: A vector retrieval submodule, used to encode the target patient's admission record text into a vector, search the first database based on the encoded vector, and obtain n historical admission record texts based on the search results; A keyword search submodule is used to extract keywords from the admission record text of the target patient, search the second database based on the extracted keywords, and obtain m historical admission record texts based on the search results; The search result combination submodule is used to combine n historical admission record texts, m historical admission record texts and the ranking of each historical admission record text to obtain a search result table; The reordering submodule is used to reorder the text of historical admission records in the search result table and obtain related admission records based on the reordering results.

[0109] In some embodiments, the original information includes a patient identification of a target patient; and the case information acquisition module includes: The admission record acquisition submodule is used to acquire the admission record text of the target patient from the medical database according to the patient identification of the target patient.

[0110] In some embodiments, the device further comprises a training module. The training module comprises: The sample data acquisition submodule is used to acquire the case data of acute coronary syndrome; the case data of acute coronary syndrome includes the admission record texts of multiple patients with ST-segment elevation myocardial infarction, the admission record texts of multiple patients with non-ST-segment elevation myocardial infarction, the admission record texts of patients with unstable angina pectoris, and the acute coronary syndrome category corresponding to each admission record text; The fine-tuning submodule is used to perform incremental training on the pre-trained general large language model using the acute coronary syndrome case data to obtain a large language model.

[0111] The specific definition of the acute coronary syndrome classification prediction device can be found in the definition of the acute coronary syndrome classification prediction method above, which will not be repeated here. Each module in the above-mentioned acute coronary syndrome classification prediction device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0112] The present application also provides a computer device. In some embodiments, the computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the acute coronary syndrome classification and prediction method provided in any of the above embodiments can be implemented.

[0113] Furthermore, in some embodiments, the internal structure diagram of the computer device may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as the admission record text of each patient with acute coronary syndrome. The specific stored data can also refer to the definition in the above method embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a classification prediction method for acute coronary syndrome is implemented.

[0114] Those skilled in the art will understand that Figure 4The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0115] The present application also provides a computer-readable storage medium. In some embodiments, a computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of the acute coronary syndrome classification prediction method provided in any of the above embodiments are implemented.

[0116] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0117] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink), DRAM (SLDRAM), memory bus (Rambus), direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0118] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A classification and prediction method for acute coronary syndrome, characterized in that: The method comprises: Acquiring original information input by a user; the original information includes an acute coronary syndrome classification task for a target patient; Acquire the admission record text of the target patient and target case information specified by a user, wherein the target case information includes typical case information of each category of acute coronary syndrome; Interacting with the user to determine the task content of the acute coronary syndrome classification task, and constructing prompt words based on the task content; the prompt words include the admission record text of the target patient and the target case information; Searching the search library for multiple historical admission record texts that are most similar to the admission record text of the target patient as associated admission records; The acute coronary syndrome classification task, the prompt words and the associated admission records are input into a large language model, so that the large language model processes the acute coronary syndrome classification task based on the prompt words and the associated admission records to obtain the acute coronary syndrome category of the target patient.

2. The method according to claim 1, characterized in that The task content includes a classification target and a precondition corresponding to the classification target; interacting with a user to determine the task content of the acute coronary syndrome classification task includes: A first prompt word is input into the large language model, where the first prompt word is used to instruct the large language model to interact with the user multiple times, generate task content to be confirmed according to task information and context information of the acute coronary syndrome classification task in each interaction, and prompt the user to confirm the task content to be confirmed; the context information includes the content to be confirmed generated by the large language model in each previous interaction and the confirmation information input by the user.

3. The method according to claim 2, characterized in that The prompt words also include a plurality of thinking points; the prompt words are constructed based on the task content, including: Filling the task content into a preset prompt word template to obtain a second prompt word; The second prompt word is input into the large language model, and the second prompt word is used to instruct the large language model to disassemble the acute coronary syndrome classification task based on the user's judgment ideas to obtain the multiple thinking points.

4. The method according to claim 3, characterized in that The admission record text of the target patient includes element information of multiple clinical elements; the multiple clinical elements include multiple of the chief complaint, current medical history, past medical history, infectious disease history, personal life history, family history, physical examination, and auxiliary examinations; the multiple thinking points include the thinking points corresponding to each of the clinical elements.

5. The method according to claim 1, characterized in that The search library includes a first database for storing vector data and a second database for storing keywords; Search the search library for multiple historical admission record texts that are most similar to the admission record text of the target patient as associated admission records, including: Encoding the target patient's admission record text into a vector, searching the first database based on the encoded vector, and obtaining n historical admission record texts based on the search results; Extracting keywords from the admission record text of the target patient, searching the second database based on the extracted keywords, and obtaining m historical admission record texts based on the search results; Combining the n historical admission record texts, the m historical admission record texts and the ranking of each historical admission record text, obtaining a search result table; The historical admission record texts in the search result table are reordered, and the associated admission record is obtained based on the reordered result.

6. The method according to claim 1, characterized in that The original information includes the patient identification of the target patient; obtaining the admission record text of the target patient includes: The admission record text of the target patient is obtained from a medical database according to the patient identification of the target patient.

7. The method according to claim 1, characterized in that The training process of the large language model includes: Acquiring acute coronary syndrome case data; the acute coronary syndrome case data includes admission record texts of multiple ST-segment elevation myocardial infarction patients, admission record texts of multiple non-ST-segment elevation myocardial infarction patients, admission record texts of unstable angina patients, and the acute coronary syndrome category corresponding to each admission record text; The pre-trained general large language model is incrementally trained using the acute coronary syndrome case data to obtain the large language model.

8. An acute coronary syndrome classification and prediction device, characterized in that: The device comprises: An original information acquisition module is used to acquire original information input by a user; the original information includes an acute coronary syndrome classification task about a target patient; A case information acquisition module, used to acquire the admission record text of the target patient and the target case information specified by the user, wherein the target case information includes typical case information of each category of acute coronary syndrome; A prompt word construction module is used to interact with the user to determine the task content of the acute coronary syndrome classification task, and to construct prompt words based on the task content; the prompt words include the admission record text of the target patient and the target case information; An associated record retrieval module is used to search in a retrieval library for a plurality of historical admission record texts that are most similar to the admission record text of the target patient as associated admission records; A disease classification module is used to input the acute coronary syndrome classification task, the prompt word and the associated admission record into a large language model, so that the large language model processes the acute coronary syndrome classification task based on the prompt word and the associated admission record to obtain the acute coronary syndrome category of the target patient.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images