Lung CT report risk assessment method based on large language model

Abstract

The invention discloses a lung CT report risk assessment method based on a large language model, and the method is characterized in that the method comprises the following steps: obtaining the text content of a lung CT report of a current patient, and inputting the text content into a trained large language model; outputting a risk report by the large language model; the clinician evaluates the risk report output by the large language model, and obtains a clinical man-made risk scoring result based on an evaluation conclusion; and taking a clinical artificial risk scoring result and a subsequent CT progress result of the patient as training data of the large language model, and performing fine-tuning training on the large language model so as to improve the effect of lung CT risk assessment. According to the method, a risk level evaluation and judgment basis can be generated according to text content of a lung CT report in combination with factors such as knowledge and reasoning ability of a large language model and the age of a patient, and the risk level evaluation and judgment basis is used for prompting the risk of the disease course progress of the patient and measures possibly needing to be taken.

Description

Technical Field

[0001] The present invention relates to a lung CT report risk assessment method based on a large language model, and belongs to the technical field of artificial intelligence. Background Art

[0002] Lung CT is a commonly used imaging method for examining lung diseases. It can show the structure and function of the lungs, as well as abnormalities of the lungs, such as lung nodules, pneumonia, lung cancer, etc. A lung CT report is a diagnostic text made by a doctor based on lung CT images, usually including the anatomical location of the lungs, the size, location, morphology, density, margins and other characteristics of the lesions, as well as possible pathological diagnoses and recommendations.

[0003] Lung CT reports are of great reference value for doctors to formulate treatment plans and assess prognosis. However, reading and understanding lung CT reports requires certain professional knowledge and experience. For ordinary patients, it may be difficult to accurately grasp their own condition and risks. In addition, lung CT reports also have a certain degree of subjectivity and uncertainty. Different doctors may have different interpretations and judgments on the same CT image, which affects the consistency and accuracy of the diagnosis, and even misses the best time for early treatment due to subjective and conservative judgments. Summary of the invention

[0004] The purpose of the present invention is to use artificial intelligence technology to provide an objective, reliable and easy-to-understand method for assessing the risk level of lung abnormalities in lung CT reports.

[0005] In order to achieve the above object, the technical solution of the present invention is to provide a lung CT report risk assessment method based on a large language model, which is characterized by comprising the following steps:

[0006] Step 1: Obtain the text content of the current patient's lung CT report and input it into the trained large language model;

[0007] Step 2: In the large language model:

[0008] Using the natural language understanding, information extraction and generation capabilities of the large language model, the text content of the lung CT report is parsed and extracted to obtain lung-related features;

[0009] Using the reasoning ability of the large language model, based on lung-related features and combined with the current patient's age and gender factors and current environmental factors, a risk level score representing the patient's risk of disease progression is calculated;

[0010] Using the natural language generation capabilities of the large language model, a risk report is generated based on the risk level score and related evidence;

[0011] Step 3: The clinical physician evaluates the risk report output by the large language model and obtains the clinical human risk score result based on the evaluation conclusion;

[0012] Step 4: Use the clinical manual risk scoring results and the patient's subsequent CT progress results as training data for the large language model, and fine-tune the large language model to improve the effect of lung CT risk assessment.

[0013] Preferably, in step 2, when the large language model acquires the lung-related features, the input prompt and expected output format of the large language model are designed through prompt engineering to guide the large language model to focus on parsing and extracting the text content of the lung CT report.

[0014] Preferably, in step 2, the lung-related characteristics include the size, location and type of the lesion.

[0015] Preferably, in step 2, the risk level score is 1 to 10 points, and the higher the risk level score, the greater the risk of disease progression of the patient.

[0016] Preferably, in step 2, the risk report includes a specific value of the risk level score, and an explanation and suggestion of the risk level score.

[0017] The present invention discloses a lung CT report risk level assessment method based on LLM (Large Language Model). The method can generate a risk level assessment and judgment basis based on the text content of the lung CT report, combined with the knowledge and reasoning ability of the large language model, as well as factors such as the patient's age, to indicate the risk of disease progression of the patient and the measures that may need to be taken.

[0018] Compared with the prior art solutions, the present invention has the following beneficial effects:

[0019] 1. The present invention provides a lung CT report risk assessment method based on the LLM large language model. The method can generate a risk score and judgment basis based on the text content of the lung CT report, combined with the knowledge and reasoning ability of the large language model and related environmental factors, to indicate the risk of the patient's disease progression and the measures that may need to be taken. The method disclosed in the present invention can help patients better understand their own conditions and risks, improve their self-care awareness and ability, and can also help doctors better formulate treatment plans and evaluate prognosis, and improve medical quality and efficiency.

[0020] 2. The present invention utilizes the natural language understanding and generation capabilities of a large language model to parse and extract the text content of lung CT reports and obtain relevant features of the lungs, such as the size, location, and type of lesions. This eliminates the need for complex image processing and analysis of lung CT images, simplifies the process and difficulty of risk assessment, and improves the speed and accuracy of risk assessment.

[0021] 3. The present invention utilizes the knowledge base and reasoning ability of a large language model, combined with relevant characteristics of the lungs, and other environmental factors, such as the air quality where the patient is located, to calculate a risk score that indicates the risk level of the patient's disease progression. This does not require reliance on professional medical knowledge and experience, thus expanding the scope of application and population of risk assessment and enhancing the universality and reliability of risk assessment.

[0022] 4. The present invention utilizes the natural language generation capability of the large language model to generate a risk report based on the risk score and related evidence, including the specific value of the risk score, as well as explanations and suggestions for the risk score. No manual writing and editing is required, which improves the quality and efficiency of the risk report, and also improves the readability and understandability of the risk report.

[0023] 5. The present invention receives doctors' feedback on risk reports, such as scores or comments on risk scores and the degree and speed of subsequent CT progression of patients, converted into risk scores, and uses them as training data for the large language model to fine-tune the large language model and optimize its risk assessment performance, thereby achieving dynamic updating and continuous improvement of risk assessment and enhancing the adaptability and flexibility of risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0026] like Figure 1 As shown, a lung CT report risk assessment method based on LLM disclosed in an embodiment of the present invention specifically includes the following steps:

[0027] Step 1: Input the text content of the lung CT report into LLM. In the embodiment of the present invention, LLM adopts the open source large model - Tongyi Qianwen-72B.

[0028] Step 2: Utilize LLM's natural language understanding, information extraction and generation capabilities, mainly through prompt engineering, to design LLM's input prompts and expected output formats, and guide LLM to focus on parsing and extracting the text content of the lung CT report to obtain relevant features of the lungs, including the size, location, and type of lesions.

[0029] Step 3: Utilize the reasoning ability of LLM, combine the lung features extracted in step 2, and add the patient's age and gender factors to calculate the risk level score. The risk level indicates the degree of risk of the patient's disease progression. In the embodiment of the present invention, the risk level score is 1 to 10 points, and the higher the risk level score, the greater the risk.

[0030] Step 4: Use the natural language generation capabilities of LLM to generate a risk report based on the risk level score and related evidence, including the specific value of the risk score, as well as the explanation and suggestions for the risk score;

[0031] Step 5: The doctor provides feedback on the risk report output by LMM to obtain the clinical human risk score result.

[0032] Step 6: Use the clinical risk score results and the patient's subsequent CT progression results as the training data for LMM, and then fine-tune the LMM to improve the effect of lung CT risk assessment.

Claims

1. A lung CT report risk assessment method based on a large language model, characterized in that: The following steps are involved: Step 1: Obtain the text content of the current patient's lung CT report and input it into the trained large language model; Step 2: In the large language model: Using the natural language understanding, information extraction and generation capabilities of the large language model, the text content of the lung CT report is parsed and extracted to obtain lung-related features; Using the reasoning ability of the large language model, based on lung-related features and combined with the current patient's age and gender factors and current environmental factors, a risk level score representing the patient's risk of disease progression is calculated; Using the natural language generation capabilities of the large language model, a risk report is generated based on the risk level score and related evidence; Step 3: The clinical physician evaluates the risk report output by the large language model and obtains the clinical human risk score result based on the evaluation conclusion; Step 4: Use the clinical manual risk scoring results and the patient's subsequent CT progress results as training data for the large language model, and fine-tune the large language model to improve the effect of lung CT risk assessment.

2. A lung CT report risk assessment method based on a large language model as claimed in claim 1, characterized in that: In step 2, when the large language model acquires the lung-related features, the input prompt and expected output format of the large language model are designed through prompt engineering to guide the large language model to focus on parsing and extracting the text content of the lung CT report.

3. The lung CT report risk assessment method based on a large language model as claimed in claim 1, characterized in that: In step 2, the lung-related characteristics include the size, location and type of the lesion.

4. The lung CT report risk assessment method based on a large language model as claimed in claim 1, characterized in that: In step 2, the risk level score is 1 to 10 points, and the higher the risk level score, the greater the risk of disease progression of the patient.

5. The lung CT report risk assessment method based on a large language model as claimed in claim 1, characterized in that: In step 2, the risk report includes specific values ​​of the risk level scores, as well as explanations and suggestions for the risk level scores.

Images