AI-fused auxiliary lymphangioleiomyomatosis LAM diagnosis system and reading medium thereof

Abstract

The invention discloses an AI-fused auxiliary LAM diagnosis system for lymphangioleiomyomatosis and a reading medium of the AI-fused auxiliary LAM diagnosis system. The above LAM diagnosis system is installed on a server, the LAM diagnosis system comprises a multi-modal data acquisition and preprocessing module responsible for collecting various types of data, and an AI diagnosis model construction module which is in data transmission connection with the multi-modal data acquisition and preprocessing module and is used for identifying relevant features of auxiliary lymphangioleiomyomatosis LAM, the dynamic risk assessment decision support module is in data transmission connection with the AI diagnosis model building module; and the intelligent referral system module is in data transmission connection with the AI diagnosis model building module and recommends an optimal referral path. The method has the advantages that the accuracy and efficiency of LAM diagnosis are improved, and uneven medical resource distribution and lack of personalized diagnosis are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of auxiliary medical image processing, and in particular to an AI-integrated auxiliary lymphangioleiomyomatosis (LAM) diagnosis system and a reading medium thereof. Background Art

[0002] Lymphangioleiomyomatosis (LAM) is a rare and aggressive lung disease that primarily affects young women. It is a multi-organ lesion caused by congenital smooth muscle cell proliferation, involving internal organs such as the lungs, pleura, heart, and kidneys. There is no cure for the disease, and the prognosis is poor, seriously affecting the patient's quality of life and having a high mortality rate. Currently, the diagnosis of LAM mainly relies on traditional medical methods, including imaging examinations, pathological diagnosis, and biomarker detection. (1) Imaging diagnosis: Doctors manually analyze lung cystic lesions in CT, HRCT, and other images. This method is time-consuming, and the diagnostic results are greatly affected by the doctor's subjective experience. Different doctors may have different judgments, making it difficult to ensure diagnostic accuracy. At the same time, LAM has similar cystic lesion characteristics to other lung diseases, making differential diagnosis difficult and the misdiagnosis rate high. (2) Pathological diagnosis: Pathologists rely on identifying HMB-45 positive cells in lung biopsy sections. However, grassroots hospitals lack experienced pathologists, which affects the accuracy and reliability of diagnosis. Moreover, lung biopsy is an invasive examination with risks and complications, which brings physical pain and potential health risks to patients. (3) Biomarker detection: VEGF-D is a commonly used biomarker, but detection relies on imported reagents, which are costly and time-consuming, increasing the financial burden on patients and potentially delaying diagnosis and treatment. (4) Multidisciplinary collaboration: The diagnosis and treatment of LAM require multidisciplinary collaboration, but the existing collaborative model lacks intelligent support, resulting in low consultation efficiency and requiring patients to visit multiple hospitals, wasting time and medical resources. As a result, existing treatment methods have the following defects: (1) High misdiagnosis rate: Due to the similarity of disease characteristics and the limitations of diagnostic methods, the misdiagnosis rate of LAM remains high, resulting in patients not receiving timely and correct treatment and delaying their condition. (2) Inefficiency: Traditional diagnostic methods are cumbersome and take a long time from examination to results, especially the long cycle of biomarker detection, which affects the early diagnosis and treatment of the disease. (3) Uneven distribution of medical resources: Grassroots hospitals lack professional diagnostic equipment and experienced doctors, resulting in patients often having to go to large hospitals in big cities for treatment, increasing the cost and burden of medical treatment for patients and causing uneven use of medical resources. (4) Lack of personalized diagnosis: Existing diagnostic methods are mainly based on general standards and experience, making it difficult to make personalized diagnosis and treatment plans based on individual differences of patients. Summary of the Invention

[0003] In view of the above problems, the purpose of the present invention is to provide an AI-assisted lymphangioleiomyomatosis LAM diagnostic system and its reading medium that improves the accuracy and efficiency of LAM diagnosis and avoids the uneven distribution of medical resources and lack of personalized diagnosis.

[0004] To achieve the above-mentioned objectives, the present invention provides an AI-integrated assisted lymphangiomyomatosis LAM diagnostic system, wherein the above-mentioned LAM diagnostic system is installed on a server, wherein the LAM diagnostic system includes a multimodal data acquisition and preprocessing module responsible for collecting various types of data, an AI diagnostic model construction module connected to the multimodal data acquisition and preprocessing module for data transmission and identifying assisted lymphangiomyomatosis LAM-related features, a dynamic risk assessment decision support module connected to the AI diagnostic model construction module for data transmission, and an intelligent referral system module connected to the AI diagnostic model construction module for data transmission and recommending the optimal referral path.

[0005] The data collected by the multimodal data acquisition and preprocessing module include imaging data, pathological data, biomarker data, and clinical data;

[0006] The preprocessing of image data uses the U-Net model to remove artifacts from the image and extract key features of lung cystic lesions, such as the size, number, and distribution density of the lesions;

[0007] The preprocessing of pathological data was to use the Mask R-CNN model to identify HMB-45 positive cell areas in the sections;

[0008] The preprocessing of biomarker data is to combine time series analysis methods to observe the dynamic trend of VEGF-D detection values over time;

[0009] AI diagnostic model building modules include image analysis, pathology analysis, and multi-task learning;

[0010] Image analysis uses a 3D ResNet model to automatically label and classify cystic lesions in the lungs, accurately identifying LAM-related features and providing high-quality data input for subsequent diagnostic models.

[0011] Pathological analysis is a Transformer model used to learn the morphological characteristics of smooth muscle-like cells to assist in pathological diagnosis;

[0012] Multi-task learning uses a multi-task learning approach, where the model simultaneously predicts the subtype of LAM (sporadic, TSC-associated) and the patient's risk of complications;

[0013] Model training uses a large amount of clinical data (including patient data from different regions and hospitals and public medical data sets) to ensure the accuracy and generalization ability of the model;

[0014] The dynamic risk assessment decision support module calculates and analyzes data input based on trained parameters and outputs LAM diagnosis results, subtype predictions, and complication risk assessments;

[0015] The dynamic risk assessment decision support module includes early warning, disease monitoring and consultation recommendations;

[0016] Early warning is based on the patient's genetic data, symptoms and various examination results. By building a risk assessment model, the patient's disease risk is given an early warning.

[0017] After a patient is diagnosed, disease monitoring involves continuously tracking changes in lung function indicators through time series models, predicting the risk of disease worsening, and adjusting treatment plans in a timely manner.

[0018] Consultation recommendations are generated for multidisciplinary consultations to promote collaboration and communication between different departments and improve diagnostic and treatment efficiency.

[0019] The intelligent referral system module refers to the use of optimization algorithms to recommend the optimal referral route based on the patient's geographical location, severity of the disease and the diagnosis and treatment capabilities of surrounding hospitals based on risk assessment results.

[0020] In some embodiments, the imaging data is chest CT or HRCT. The pathological data is lung biopsy slices. The biomarker data is VEGF-D detection value. The clinical data is patient symptom records and pulmonary function test results.

[0021] In some embodiments, the server is a cloud server or a local server of a medical institution.

[0022] In some embodiments, the dynamic risk assessment decision support module includes early warning, disease monitoring, and consultation recommendations. Early warning is based on the patient's genetic data, symptoms, and various test results. By constructing a risk assessment model, an early warning of the patient's risk of disease is provided. Disease monitoring is to continuously track changes in the patient's lung function indicators through a time series model after the patient is diagnosed, predict the risk of disease worsening, and adjust the treatment plan in a timely manner. Consultation recommendations are to generate multidisciplinary consultation recommendations, promote collaboration and communication between different departments, and improve the efficiency of diagnosis and treatment.

[0023] In some implementations, the reading medium includes a storage module and a processor execution module.

[0024] The processor execution module is arranged on the storage module.

[0025] The processor executor module performs the following steps: (1) receiving a first set of medical images. (2) determining whether the first set of medical images satisfies a first condition. (3) in response to determining that the first set of medical images does not satisfy the first condition, after acquiring a second set of medical images, using the first set of medical images and the second set of medical images as input to obtain a first diagnosis result for lymphangioleiomyomatosis (LAM). The first set of medical images includes positron emission tomography (PET) / computed tomography (CT) images.

[0026] Another object is to provide a reading medium of an AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system, wherein a processor executor module performs the following steps: (1) receiving a third set of medical images. (2) determining whether the third set of medical images satisfies a second condition. (3) in response to determining that the third set of medical images satisfies the second condition, using the third set of medical images as input to obtain a second diagnostic result for LAM. The third set of medical images is magnetic resonance imaging (MRI) images.

[0027] In some embodiments, a processor-executor module is associated with a database. The processor-executor module queries the database for multiple drugs corresponding to the first diagnostic result, and sorts the multiple drugs according to the ranking score of each drug. The sorted multiple drugs are output. The processor-executor module queries the database for a series of treatment plans corresponding to the first diagnostic result, and sorts the series of treatment plans according to the ranking score of each treatment plan. The sorted series of treatment plans are output. The processor-executor module queries the database for multiple drug names corresponding to the first diagnostic result, and sorts the multiple drug names according to the ranking score of each drug name. The sorted multiple drug names are output. The processor-executor module queries the database for multiple manufacturer information corresponding to the first diagnostic result, and sorts the multiple manufacturer information according to the ranking score of each manufacturer information. The sorted multiple manufacturer information is output. The processor-executor module queries the database for multiple generic drug names corresponding to the first diagnostic result, and sorts the multiple generic drug names according to the ranking score of each generic drug name. The sorted multiple generic drug names are output. The processor-executor module queries the database for multiple product information corresponding to the first diagnostic result, and sorts the multiple product information according to the ranking score of each product information. The sorted multiple product information is output. The processor-executor module queries a database for multiple chemical drugs corresponding to the first diagnostic result. The multiple chemical drugs are sorted according to the ranking score of each chemical drug. The sorted multiple chemical drugs are output. The processor-executor module queries a database for multiple molecular structural formulas corresponding to the first diagnostic result. The multiple molecular structural formulas are sorted according to the ranking score of each molecular structural formula. The sorted multiple molecular structural formulas are output.

[0028] The beneficial effects of the present invention are to improve the accuracy and efficiency of LAM diagnosis and avoid the uneven distribution of medical resources and the lack of personalized diagnosis. The details are as follows:

[0029] (1) Improve diagnostic accuracy: AI models can analyze data more objectively and accurately in imaging and pathological diagnosis, reduce the influence of doctors' subjective factors, and significantly reduce the misdiagnosis rate.

[0030] (2) Improve diagnostic efficiency: The system can complete the analysis and diagnosis of patients' multimodal data in a short time, shortening the diagnosis cycle. Grassroots hospitals can quickly complete preliminary screening and diagnosis and refer suspected patients in a timely manner.

[0031] (3) Reduce medical costs: Reduce unnecessary examination items, avoid waste of resources, accurately formulate treatment plans, avoid overtreatment and ineffective treatment, and reduce patients' medical expenses. Optimize referral pathways, reduce patients' travel, and reduce medical costs.

[0032] (4) Promote precision treatment: The dynamic risk assessment system monitors the changes in the patient's condition in real time, adjusts the treatment plan according to individual conditions, realizes personalized treatment, improves treatment effects, and reduces the occurrence of complications.

[0033] (5) Promote medical equity: The intelligent referral system rationally allocates high-quality medical resources, improves the diagnostic capabilities of primary hospitals, narrows the medical gap between different regions, and promotes the balanced distribution of medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagram of the architecture of the present invention. DETAILED DESCRIPTION

[0035] The invention will be further described in detail below with reference to the accompanying drawings.

[0036] A diagnostic system for assisted lymphangioleiomyomatosis (LAM) that integrates AI. The LAM diagnostic system is installed on a server. The LAM diagnostic system includes a multimodal data acquisition and preprocessing module responsible for collecting various types of data, an AI diagnostic model construction module that is connected to the multimodal data acquisition and preprocessing module for data transmission and identifies features related to assisted lymphangioleiomyomatosis (LAM), a dynamic risk assessment decision support module that is connected to the AI diagnostic model construction module for data transmission, and an intelligent referral system module that is connected to the AI diagnostic model construction module for data transmission and recommends the optimal referral path.

[0037] The multimodal data acquisition and preprocessing module collects data including imaging, pathology, biomarker, and clinical data. Imaging data preprocessing uses a U-Net model to remove artifacts and extract key features of lung cystic lesions, such as lesion size, number, and distribution density. Pathology data preprocessing uses a Mask R-CNN model to identify HMB-45-positive cell regions in sections. Biomarker data preprocessing combines time series analysis to observe the dynamic changes in VEGF-D test values over time.

[0038] The AI diagnostic model building modules include image analysis, pathological analysis, and multi-task learning. Image analysis uses the 3DResNet model to automatically label and classify cystic lesions in the lungs, accurately identify LAM-related features, and provide high-quality data input for subsequent diagnostic models. Pathological analysis uses the Transformer model to learn the morphological characteristics of smooth muscle-like cells to assist in pathological diagnosis. Multi-task learning uses a multi-task learning method, and the model simultaneously predicts the subtype of LAM (sporadic, TSC-related) and the risk of complications in patients. Model training uses a large amount of clinical data (including patient data from different regions and hospitals and public medical data sets) to ensure the accuracy and generalization ability of the model.

[0039] The dynamic risk assessment decision support module calculates and analyzes the data input based on the trained parameters and outputs the LAM diagnosis results, subtype predictions, and complication risk assessments.

[0040] The dynamic risk assessment decision support module includes early warning, disease monitoring, and consultation recommendations. Early warning provides early warning of a patient's risk of developing the disease by building a risk assessment model based on the patient's genetic data, symptoms, and various test results. Disease monitoring, after a patient's diagnosis, continuously tracks changes in lung function indicators using a time series model to predict the risk of disease progression and promptly adjust treatment plans. Consultation recommendations generate multidisciplinary consultation recommendations, promoting collaboration and communication between different departments and improving diagnostic and treatment efficiency.

[0041] The intelligent referral system module uses an optimization algorithm based on risk assessment results to recommend the optimal referral path based on the patient's location, severity of the condition, and the diagnostic and treatment capabilities of surrounding hospitals. The server can be a cloud server or a local server at the medical institution.

[0042] The dynamic risk assessment decision support module includes early warning, disease monitoring, and consultation recommendations. Early warning provides early warning of a patient's risk of developing the disease by building a risk assessment model based on the patient's genetic data, symptoms, and various test results. Disease monitoring, after a patient's diagnosis, continuously tracks changes in lung function indicators using a time series model to predict the risk of disease progression and promptly adjust treatment plans. Consultation recommendations generate multidisciplinary consultation recommendations, promoting collaboration and communication between different departments and improving diagnostic and treatment efficiency.

[0043] When applying:

[0044] (1) System deployment

[0045] The AI-assisted LAM diagnostic system of the present invention is deployed on a cloud server or a local server of a medical institution to ensure that the system can run stably and has sufficient computing resources and storage capabilities.

[0046] (2) Data collection and transmission

[0047] Medical institutions collect multimodal data from patients using various medical devices (such as CT scanners and pathology slide scanners) and upload the data to the system. Encryption technology is used during data transmission to ensure data security and privacy.

[0048] (3) Data preprocessing

[0049] After receiving the data, the system preprocesses the imaging data, pathological data, and biomarker data according to the methods in the above technical solution, extracts key features, and provides high-quality data input for subsequent diagnostic models.

[0050] (4) Diagnostic model operation

[0051] The preprocessed data is input into the AI diagnostic model, which performs calculations and analyses based on the trained parameters and outputs the diagnostic results, subtype predictions, and complication risk assessments of LAM.

[0052] (V) Risk assessment and decision support

[0053] The dynamic risk assessment and decision support module evaluates the patient's risk of disease onset and risk of disease progression based on the diagnosis results and the patient's historical data, and generates corresponding treatment recommendations and consultation suggestions.

[0054] (6) Intelligent referral

[0055] The intelligent referral system recommends the optimal referral route based on the patient's condition and location, and sends the patient's relevant information to the target hospital. Upon receiving this information, the target hospital can prepare for diagnosis and treatment in advance.

[0056] Example 1

[0057] For example, a patient suspected of LAM came to a primary care hospital. The hospital collected chest imaging data using a CT scanner, performed a lung biopsy and VEGF-D testing, and uploaded the patient's symptom history and pulmonary function test results to the AI-assisted LAM diagnostic system.

[0058] The system preprocesses the imaging data, extracts the features of lung cystic lesions, and analyzes them using the 3DResNet model; processes the pathological sections and uses the Transformer model to identify the features of smooth muscle-like cells; and combines VEGF-D test values and clinical data to perform a comprehensive diagnosis through a multi-task learning model.

[0059] The diagnosis indicated a high likelihood of LAM, a sporadic form, and a moderate risk of pneumothorax complications. The Dynamic Risk Assessment and Decision Support Module generated treatment recommendations based on the diagnosis, recommending conservative treatment and close monitoring of lung function indicators.

[0060] Due to the limited medical resources of primary care hospitals, the intelligent referral system recommends referrals to nearby higher-level hospitals for further diagnosis and treatment based on the patient's location and the severity of their condition. Upon receiving the patient's information, the higher-level hospital prepares for the consultation in advance, ensuring that the patient receives a timely professional diagnosis and treatment upon arrival. This system reduces the patient's diagnostic cycle from the traditional several months to just a few days, improving both efficiency and accuracy, thus freeing up valuable time for treatment.

[0061] A reading medium of an AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system includes a storage module and a processor execution module. The processor execution module is disposed on the storage module. The processor execution module executes the following steps: (1) receiving a first set of medical images. (2) determining whether the first set of medical images satisfies a first condition. (3) in response to determining that the first set of medical images does not satisfy the first condition, after acquiring a second set of medical images, using the first set of medical images and the second set of medical images as input to obtain a first diagnostic result for lymphangioleiomyomatosis (LAM). The first set of medical images includes positron emission tomography (PET) / computed tomography (CT) images.

[0062] A reading medium of an AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system is provided, wherein a processor executor module executes the following steps: (1) receiving a third set of medical images; (2) determining whether the third set of medical images satisfies a second condition; and (3) in response to determining that the third set of medical images satisfies the second condition, using the third set of medical images as input to obtain a second diagnostic result for LAM. The third set of medical images is magnetic resonance imaging (MRI) images.

[0063] The processor-executor module is associated with a database. The processor-executor module queries the database for multiple drugs corresponding to the first diagnostic result, and sorts the multiple drugs according to the ranking score of each drug. The sorted multiple drugs are output. The processor-executor module queries the database for a series of treatment plans corresponding to the first diagnostic result, and sorts the series of treatment plans according to the ranking score of each treatment plan. The sorted series of treatment plans are output. The processor-executor module queries the database for multiple drug names corresponding to the first diagnostic result, and sorts the multiple drug names according to the ranking score of each drug name. The sorted multiple drug names are output. The processor-executor module queries the database for multiple manufacturer information corresponding to the first diagnostic result, and sorts the multiple manufacturer information according to the ranking score of each manufacturer information. The sorted multiple manufacturer information is output. The processor-executor module queries the database for multiple generic drug names corresponding to the first diagnostic result, and sorts the multiple generic drug names according to the ranking score of each generic drug name. The sorted multiple generic drug names are output. The processor-executor module queries the database for multiple product information corresponding to the first diagnostic result, and sorts the multiple product information according to the ranking score of each product information. The sorted multiple product information is output. The processor-executor module queries a database for multiple chemical drugs corresponding to the first diagnostic result. The multiple chemical drugs are sorted according to the ranking score of each chemical drug. The sorted multiple chemical drugs are output. The processor-executor module queries a database for multiple molecular structural formulas corresponding to the first diagnostic result. The multiple molecular structural formulas are sorted according to the ranking score of each molecular structural formula. The sorted multiple molecular structural formulas are output.

[0064] Application principles, such as Figure 1 As shown,

[0065] A computer-readable storage medium stores instructions thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0066] S101: Receive a first set of medical images;

[0067] S102: Determine whether the first group of medical images meets a first condition;

[0068] S103: In response to determining that the first group of medical images does not meet the first condition, after acquiring the second group of medical images, the first group of medical images and the second group of medical images are used as input to obtain a first diagnostic result for lymphangioleiomyomatosis (LAM); wherein the first group of medical images includes positron emission tomography (PET) / X-ray computed tomography (CT) images.

[0069] First, a PET / CT image of the patient is obtained through a PET / CT scanner, and then a determination is made as to whether the image meets the criteria. If not, another set of images, i.e., the second set of medical images, is collected again. Finally, the two sets of images are input together into the model for training, thereby obtaining the first diagnostic result for LAM.

[0070] The first set of medical images also includes low-dose chest CT images.

[0071] Since LAM is a rare diffuse lung disease, it is difficult to find enough cases in existing databases to build a model. Therefore, this embodiment provides a method to use low-dose chest CT images to supplement the data of normal people, thereby improving the accuracy of the model.

[0072] The first set of medical images also includes conventional dose chest CT images.

[0073] Since LAM is a rare diffuse lung disease, it is difficult to find enough cases in existing databases to build a model. Therefore, this embodiment provides a method to use conventional dose chest CT images to supplement the data of normal people, thereby improving the accuracy of the model.

[0074] The computer-readable storage medium stores instructions thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0075] receiving a third set of medical images;

[0076] determining whether the third set of medical images satisfies a second condition;

[0077] In response to determining that the third set of medical images meets the second condition, the third set of medical images is used as input to obtain a second diagnosis result for LAM; wherein the third set of medical images are magnetic resonance imaging (MRI) images.

[0078] First, the patient's MRI images are obtained through an MRI scanner, and then it is determined whether the images meet the standards. If they do, this set of images is directly input into the model for training to obtain a second diagnostic result for LAM.

[0079] The third set of medical images also includes enhanced MRI images.

[0080] Since LAM is a rare diffuse lung disease, it is difficult to find enough cases in existing databases to build a model. Therefore, this embodiment provides a method of using enhanced MRI images to supplement data from normal subjects, thereby improving the accuracy of the model.

[0081] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0082] S301: Querying a database for a plurality of drugs corresponding to the first diagnosis result;

[0083] S302: Sort the multiple drugs according to the ranking score of each drug;

[0084] S303: Output the sorted drugs.

[0085] After the model outputs the first diagnostic result for LAM, relevant drugs can be searched from the database based on the result, and then sorted according to the ranking score of each drug, and finally output multiple sorted drugs.

[0086] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0087] S401: Querying a database for a series of treatment plans corresponding to the first diagnosis result;

[0088] S402: Sort the series of treatment plans at hand according to the ranking score of each treatment plan;

[0089] S403: Output a series of sorted treatment plans.

[0090] After the model outputs the first diagnostic result for LAM, a series of related treatment plans can be searched from the database based on the result, and then sorted according to the ranking score of each treatment plan, and finally output the sorted series of treatment plans.

[0091] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0092] S501: Querying a database for multiple drug names corresponding to the first diagnosis result;

[0093] S502: Sort the multiple drug names according to the ranking score of each drug name;

[0094] S503: Output the sorted names of multiple medicines.

[0095] After the model outputs the first diagnostic result for LAM, the relevant drug names can be searched from the database based on the result, and then sorted according to the ranking score of each drug name, and finally output multiple sorted drug names.

[0096] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0097] S601: Querying a database for multiple manufacturer information corresponding to the first diagnosis result;

[0098] S602: Sort the plurality of manufacturer information according to the ranking score of each manufacturer information;

[0099] S603: Output the sorted information of multiple manufacturers.

[0100] After the model outputs the first diagnostic result for LAM, relevant manufacturer information can be searched from the database based on the result, and then sorted according to the ranking score of each manufacturer information, and finally output the sorted multiple manufacturer information.

[0101] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0102] S701: Querying a database for multiple generic drug names corresponding to the first diagnosis result;

[0103] S702: Sort the plurality of generic drug names according to the ranking score of each generic drug name;

[0104] S703: Output the sorted common names of multiple drugs.

[0105] After the model outputs the first diagnostic result for LAM, the relevant generic names of drugs can be searched from the database based on the result, and then sorted according to the ranking score of each generic name of the drug, and finally output multiple sorted generic names of the drug.

[0106] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0107] S801: Querying a database for multiple product information corresponding to the first diagnosis result;

[0108] S802: Sort the plurality of product information according to the ranking score of each product information;

[0109] S803: Output the sorted product information.

[0110] After the model outputs the first diagnostic result for LAM, relevant product information can be searched from the database based on the result, and then sorted according to the ranking score of each product information, and finally output the sorted multiple product information.

[0111] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0112] S901: Querying a database for multiple chemical drugs corresponding to the first diagnosis result;

[0113] S902: Sort the multiple chemical drugs according to the ranking score of each chemical drug;

[0114] S903: Output the sorted multiple chemical drugs.

[0115] After the model outputs the first diagnostic result for LAM, relevant chemical drugs can be searched from the database based on the result, and then sorted according to the ranking score of each chemical drug, and finally output multiple sorted chemical drugs.

[0116] A computer-readable storage medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform at least the following steps:

[0117] S1001: Querying a database for multiple molecular structural formulas corresponding to the first diagnosis result;

[0118] S1002: Sort the multiple molecular structural formulas according to the ranking score of each molecular structural formula;

[0119] S1003: Outputting the sorted molecular structural formulas.

[0120] After the model outputs the first diagnostic result for LAM, it can search for relevant molecular structural formulas in the database based on the result, then sort each molecular structural formula according to its ranking score, and finally output multiple sorted molecular structural formulas.

[0121] In summary, an embodiment of the present disclosure provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, causes the processor to perform at least the following steps:

[0122] receiving a first set of medical images;

[0123] determining whether the first group of medical images satisfies a first condition;

[0124] In response to determining that the first group of medical images does not meet the first condition, after acquiring a second group of medical images, the first group of medical images and the second group of medical images are used as input to obtain a first diagnostic result for lymphangioleiomyomatosis (LAM); wherein the first group of medical images includes positron emission tomography (PET) / X-ray computed tomography (CT) images.

[0125] Therefore, the present disclosure achieves the purpose of using artificial intelligence to assist in the diagnosis of LAM, and improves the work efficiency and accuracy of doctors.

[0126] In some embodiments, the present disclosure further provides a method comprising:

[0127] receiving a first set of medical images;

[0128] determining whether the first group of medical images satisfies a first condition;

[0129] In response to determining that the first group of medical images does not meet the first condition, after acquiring a second group of medical images, the first group of medical images and the second group of medical images are used as input to obtain a first diagnostic result for lymphangioleiomyomatosis (LAM); wherein the first group of medical images includes positron emission tomography (PET) / X-ray computed tomography (CT) images.

[0130] In some embodiments, the present disclosure further provides an apparatus comprising:

[0131] A receiving unit, configured to receive a first set of medical images;

[0132] a determining unit, configured to determine whether the first group of medical images satisfies a first condition;

[0133] A processing unit is configured to, in response to determining that the first set of medical images does not satisfy the first condition, use the first set of medical images and the second set of medical images as input after acquiring a second set of medical images to obtain a first diagnostic result for lymphangioleiomyomatosis (LAM); wherein the first set of medical images includes positron emission tomography (PET) / X-ray computed tomography (CT) images.

[0134] The present disclosure also provides a device, comprising:

[0135] A memory, configured to store program codes and transmit the program codes to a processor;

[0136] The processor is configured to execute instructions in the program code so that the device executes the method provided in any one of the above embodiments.

[0137] The above are only some embodiments of the present invention. For those skilled in the art, several modifications and improvements can be made without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the invention.

Claims

1. An AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system, installed on a server, characterized in that ,The LAM diagnostic system includes a multimodal data acquisition and preprocessing module responsible for collecting ,various types of data, an AI diagnostic model construction module ,connected with the multimodal data acquisition and preprocessing module for data ,transmission and identifying LAM related features, a dynamic risk assessment ,decision support module connected with the AI diagnostic model construction module for data ,transmission, and an intelligent referral system module connected with the AI diagnostic model construction module for data ,transmission and recommending the optimal referral path; The data collected by the multimodal data acquisition and preprocessing module include imaging data, pathological data, biomarker data and clinical data; The image data is preprocessed by using a U-Net model to remove artifacts in the image and extract key features of lung cystic lesions, such as the size, number, and distribution density of the lesions; The preprocessing of the pathological data is to use the Mask R-CNN model to identify the HMB-45 positive cell area in the slice; The preprocessing of the biomarker data is to combine the time series analysis method to observe the dynamic change trend of VEGF-D detection value over time; The AI diagnostic model building modules include image analysis, pathology analysis, and multi-task learning; The image analysis uses a 3D ResNet model to automatically label and classify cystic lesions in the lungs, accurately identifying LAM-related features and providing high-quality data input for subsequent diagnostic models. The pathological analysis is to use the Transformer model to learn the morphological characteristics of smooth muscle-like cells to assist in pathological diagnosis; The multi-task learning method is used, and the model simultaneously predicts the subtype of LAM and the risk of complications in patients; The model training is carried out using a large amount of clinical data to ensure the accuracy and generalization ability of the model; The dynamic risk assessment decision support module calculates and analyzes the data input according to the trained parameters and outputs the diagnosis results, subtype predictions and complication risk assessments of LAM; The dynamic risk assessment decision support module includes early warning, disease monitoring and consultation recommendations; The early warning is based on the patient's genetic data, symptoms and various examination results. By building a risk assessment model, the patient's risk of disease is given an early warning. The disease monitoring mentioned above is to continuously track the changes in the patient's lung function indicators through a time series model after the patient is diagnosed, predict the risk of disease worsening, and adjust the treatment plan in a timely manner; The consultation suggestion is to generate a multidisciplinary consultation suggestion, promote collaboration and communication between different departments, and improve the efficiency of diagnosis and treatment; The intelligent referral system module refers to a system that recommends the optimal referral path based on the patient's geographical location, severity of the disease, and the diagnosis and treatment capabilities of surrounding hospitals based on risk assessment results and the use of an optimization algorithm.

2. The AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system according to claim 1 is characterized in that , The imaging data is chest CT or HRCT; The pathological data are lung biopsy sections; The biomarker data is the VEGF-D detection value; The clinical data include patient symptom records and lung function test results.

3. The AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system according to claim 1 is characterized in that ,The server is a cloud server or a local server of a medical institution.

4. The reading medium of the AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system according to claim 1, characterized in that ,the reading medium includes a storage module and a processor execution module; The processor execution module is arranged on the storage module; The processor executor module performs the following steps: (1) receiving a first set of medical images; (2) determining whether the first group of medical images satisfies a first condition; (3) in response to determining that the first set of medical images does not satisfy the first condition, after acquiring a second set of medical images, using the first set of medical images and the second set of medical images as input to obtain a first diagnosis result for lymphangioleiomyomatosis (LAM); The first set of medical images includes positron emission tomography (PET) / computed tomography (CT) images.

5. The reading medium of the AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system according to claim 4, characterized in that ,The processor executor module executes the following steps: (1) receiving a third set of medical images; (2) determining whether the third set of medical images satisfies a second condition; (3) in response to determining that the third set of medical images satisfies the second condition, using the third set of medical images as input to obtain a second diagnosis result for LAM; The third set of medical images are magnetic resonance imaging (MRI) images.

6. The reading medium of the AI-assisted lymphangioleiomyomatosis (LAM) diagnostic system according to claim 4 or 5, characterized in that ,the processor executor module is associated with a database; The processor executor module queries a database for a plurality of drugs corresponding to the first diagnosis result, and sorts the plurality of drugs according to a ranking score of each drug; Output multiple sorted drugs; The processor executor module queries the database for a series of treatment plans corresponding to the first diagnosis result; sorts the series of treatment plans at hand according to the ranking score of each treatment plan; and outputs the sorted series of treatment plans; The processor executor module queries the database for multiple drug names corresponding to the first diagnosis result; and sorts the multiple drug names according to the ranking score of each drug name; Output multiple sorted drug names; The processor executor module queries the database for multiple manufacturer information corresponding to the first diagnostic result; sorts the multiple manufacturer information according to the ranking score of each manufacturer information; and outputs the sorted multiple manufacturer information; The processor executor module queries a database for multiple generic drug names corresponding to the first diagnosis result; and sorts the multiple generic drug names according to the ranking score of each generic drug name; Output multiple sorted generic names of drugs; The processor executor module queries the database for multiple product information corresponding to the first diagnosis result; and sorts the multiple product information according to the ranking score of each product information; Output sorted product information; The processor executor module queries the database for multiple chemical drugs corresponding to the first diagnosis result; sorting the plurality of chemical drugs according to the ranking score of each chemical drug; Output multiple sorted chemical drugs; The processor executor module queries the database for multiple molecular structural formulas corresponding to the first diagnostic result; and sorts the multiple molecular structural formulas according to the ranking score of each molecular structural formula; Output multiple sorted molecular structures.

Images