Lung lesion analysis system based on parameter reaction diagram and image generation system

Abstract

The lung analysis system based on the parameter response diagram and the image generation system is realized. The system is composed of a data collection and management module, an image registration module, a controllable image generation module, a parameter response diagram analysis module and a disease course development evaluation module. The data collection and management module is in butt joint with a hospital information system and CT image collection equipment. The image registration module receives a double-phase CT scanning image of a patient to obtain a registered double-phase CT image; the controllable image generation module receives a single inspiration phase CT scanning image of a patient as input to generate a registered expiration phase CT scanning image; the parameter response diagram analysis module obtains multiple parameter response diagram analysis and evaluation results of organic lesion evaluation of the lung of the patient; the disease course development evaluation module predicts the lung function condition of the patient. Therefore, the problems that when a parameter response diagram method is used, biphasic CT collection is difficult, and the ability to evaluate the disease course development of a patient is lacked are solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a lung lesion analysis system based on a parameter response diagram and an image generation system. Background Art

[0002] Chronic obstructive pulmonary disease (COPD) is a chronic respiratory disease characterized by airflow obstruction and lung dysfunction. Smoking is one of the main causes of the disease, but it is also related to air pollution, infection, chemical pollution, occupational dust and other factors.

[0003] COPD is a chronic disease in which patients may gradually lose lung function without noticing. Usually, when symptoms appear, the disease has entered a late stage, and the loss of lung function is difficult to reverse. Early detection of the disease and timely and effective intervention measures are critical, which can significantly slow the progression of the disease, improve the quality of life of patients, reduce the economic burden on patients and families, and reduce the risk of death caused by the disease. In addition, COPD is often accompanied by many serious complications (such as heart disease, osteoporosis, etc.), which impose a great burden on patients and society. Therefore, early diagnosis and treatment of COPD not only has important clinical value, but also has important significance for the health and quality of life of patients. In addition, regular review of COPD patients can help monitor disease progression, evaluate treatment effects, and improve patients' quality of life.

[0004] Chest CT plays an important role in the diagnosis and treatment of COPD patients. It can show the type, extent and distribution of lung lesions, thereby helping doctors evaluate the progression of the disease and the effectiveness of treatment. However, the current analysis of CT scan images is mainly carried out through naked eye observation and sensory evaluation by doctors. There is a lack of quantitative evaluation and quantitative comparison of specific lung lesions, making it difficult to accurately reflect the changes and progression of lung lesions. In addition, current CT examinations can only provide images of lung structures, and there is no association with the assessment of lung function and airflow obstruction. It is impossible to use the patient's organic lung lesion information in the CT examination results to estimate the patient's respiratory function performance and evaluate the progression of the disease.

[0005] In recent years, computer-aided diagnosis technology and quantitative analysis methods based on CT images have been widely used, such as parametric response mapping (PRM). Parametric response mapping is a lung disease assessment method based on high-resolution CT images. By registering and analyzing the end-inspiratory and end-expiratory images of the lungs, it can provide a more accurate quantitative assessment of the patient's emphysema, airway wall lesions, pulmonary vascular lesions, and treatment responsiveness. Therefore, parametric response mapping has the following value in the diagnosis and treatment of COPD:

[0006] 1. Help establish a direct correlation between organic and functional lung lesions in patients: PRM can quantitatively evaluate the lung tissue damage in different lung regions of patients, thereby determining their ventilation function, and further help determine the relationship between patients' lung function and organic lesions through subsequent algorithm research.

[0007] 2. Helps in early diagnosis of COPD: PRM can detect smaller organic lesions in the patient's lungs when the patient has no lung function, so it can help doctors diagnose COPD earlier and take timely measures to prevent further development of the disease.

[0008] 3. Monitor changes in patients’ conditions: PRM can quantitatively evaluate and compare the lung lesions shown in multiple CT examination results of COPD patients, thereby monitoring changes in patients’ conditions, evaluating the progression of the disease and the effectiveness of treatment.

[0009] As a new lung imaging analysis method, parametric response map (PRM) has shown broad application prospects and clinical significance in the diagnosis and treatment of COPD. However, PRM has not been widely used in clinical practice. One of the reasons is that it is difficult to obtain the biphasic lung CT scan images required at the end of deep inspiration and normal expiration. For the elderly or seriously ill patients, there may be insufficient exhalation or inability to hold the breath for a long time. In addition, the existing PRM analysis method cannot include the historical CT scan results of multiple single inspiratory phases of COPD patients in the analysis to comprehensively evaluate the progression of the patient's lung lesions. Summary of the invention

[0010] In order to solve the problem that the existing parameter response graph analysis method can only be used for the inspiratory-expiratory biphasic chest CT image pair of a single examination, and the resulting difficulty in collecting biphasic CT, lack of ability to evaluate the course of the patient's disease, and inability to use it, the present invention first proposes a lung lesion analysis method based on parameter response graphs and compatible with single-phase and biphasic chest CT scan images. The method judges the patient's lung lesion status and development trend by analyzing the inspiratory-expiratory biphasic chest CT image pair and the single inspiratory phase chest CT image. For the inspiratory-expiratory biphasic chest CT image pair, the method first aligns it to obtain a registered biphasic CT image pair; for the single inspiratory phase CT image, the method uses an image generation method to generate a missing expiratory phase CT image that has been aligned with the inspiratory phase CT image based on the single inspiratory phase CT image, and also obtains a registered biphasic CT image pair; finally, the method analyzes the patient's lung lesion status on the registered biphasic CT image pair through an analysis method based on the parameter response graph, and judges the degree, distribution and development trend of various distributed lesions. Compared with the existing lung lesion analysis methods based on parameter response graphs, this method is not only limited to the analysis of paired expiratory-inspiratory biphasic CT scans, but can also realize the parameter response graph analysis of single expiratory phase CT commonly used in clinical practice by supplementing the missing inspiratory phase CT for single expiratory phase CT.

[0011] In order to complete the missing inspiratory phase CT image for single expiratory phase CT, the present invention proposes a clinical text-guided expiratory phase CT image generation method. The method takes the patient's electronic medical record, respiratory symptom assessment, pulmonary function test results and other clinical texts as well as the patient's inspiratory phase chest CT scan image as input, and can learn the relationship between the patient's respiratory function clinical phenotype reflected in the clinical text and the patient's lung lesion imaging phenotype reflected in the CT image, thereby converting a given inspiratory phase CT image into an expiratory phase CT image.

[0012] Based on the lung lesion analysis method and the expiratory phase CT image generation method, the present invention proposes a lung lesion analysis system based on a parameter response map and an image generation system, which is composed of a data collection and management module, an image registration module, a controllable image generation module and a parameter response map analysis module:

[0013] The data collection and management module is connected to the hospital information system and the CT image collection device, collects patient information and historical CT examination results and inputs them into the disease progression assessment module. According to different types of CT scans, the biphasic CT images are sent to the image registration module, and the single inspiratory phase CT images are sent to the controllable image generation module;

[0014] The image registration module receives the patient's bi-phase CT scan image, registers the expiratory phase CT scan image at the end of deep exhalation with the inspiratory phase CT scan image at the end of inspiration, so that the same lung tissue is in the same spatial position in the two different CT images, and the registered bi-phase CT image is input into the controllable image generation module for its training, and is also input into the parameter response map analysis module;

[0015] The controllable image generation module is based on the clinical text-guided expiratory phase CT image generation method. After being trained using the registered biphasic CT images and the corresponding patient clinical texts as training data, it can learn the relationship between the patient's respiratory function clinical phenotype and the lung parenchymal lesions and imaging phenotypes. In actual use, it can generate an expiratory phase CT image registered with the inspiratory phase chest CT image based on the patient's single inspiratory phase chest CT image and clinical texts such as electronic medical records, respiratory symptom assessments, and pulmonary function test results.

[0016] The parameter response map analysis module receives the inspiration-expiration biphasic CT image pairs generated by the image registration module and the controllable image generation module, divides the whole lung into normal lung tissue, emphysema, functional small airway damage and COPD areas using a voxel classification method, and calculates the volume proportion of different types of lung tissue in the whole lung and different lung segments, thereby obtaining a parameter response map analysis evaluation result for evaluating the patient's organic lung lesions; given several historical CT examination results of the same patient, the parameter response map analysis module can provide the type, degree, distribution and development trend of the patient's lung lesions by comparing and analyzing the parameter response map analysis evaluation results, thereby providing a basis for doctors to make auxiliary diagnosis and treatment decisions.

[0017] The specific implementations of each module in the pulmonary lesion analysis system are as follows:

[0018] The specific implementation method of the image registration module is as follows: receiving the expiratory phase CT scan image of the patient at the end of deep exhalation and the CT scan image at the end of inspiration, taking the CT image at the end of inspiration as the target image, and taking the expiratory phase CT image as the original image to input into the diffusion model, the diffusion model learns the correspondence between the features in the patient's biphasic CT image and the biphasic CT image by learning the conditional scoring function; then inputting the output result of the diffusion network into the deformation network to be converted into a registration field that can be directly used for non-rigid transformation of the image, the obtained registration field is directly applied to the expiratory phase CT image by the spatial transformation layer, and the registered image is obtained after non-rigid transformation.

[0019] The controllable image generation module is based on the expiratory phase CT image generation method guided by the clinical text. The specific implementation of the method is as follows: it is composed of a clinical text encoding model and an expiratory phase CT image generation model. The clinical text encoding model is used to encode the information related to the clinical phenotype of the patient's respiratory function in the patient's electronic medical record, respiratory symptom assessment and pulmonary function test results into a feature vector; the image generation model includes two processes, diffusion and inverse diffusion. In the diffusion process, the inspiratory phase CT image will first be converted into a feature vector in the latent space by the encoder ε in the variational autoencoder, and the noise will be gradually added through the diffusion process. After that, the inspiratory phase CT feature vector with the noise added will be gradually generated by the clinical text feature inverse diffusion process while removing the noise and generating a feature vector representing the expiratory phase CT image, and the decoder in the variational autoencoder will be used to generate a feature vector representing the expiratory phase CT image. Converted into an actual expiratory phase CT image. The clinical text-guided expiratory phase CT image generation method introduces clinical text describing the patient's respiratory function into the image generation model. Unlike the existing image generation method, the text in this method does not directly indicate the performance of the generated image, but provides auxiliary information for the model to convert from expiratory phase CT to inspiratory phase CT. The model establishes the connection between the clinical phenotype and imaging phenotype of the respiratory function of patients with lung diseases by learning the relationship between the biphasic CT image and the clinical text, thereby realizing the generation of expiratory phase chest CT images based on clinical texts such as the patient's electronic medical record, respiratory symptom assessment, and pulmonary function test results, and inspiratory phase chest CT images.

[0020] The training of the expiratory CT generation model (ECTGen) and the reasoning process in actual use are composed of the above-mentioned inspiratory CT image diffusion process and the inverse diffusion process guided by clinical text. Its training goal can be expressed as:

[0021]

[0022] y:={y1,y2,y3,…},

[0023] Where x represents the patient's inspiratory CT image, y1, y2, y3, ... represent clinical texts such as the patient's electronic medical record, pulmonary function test results, respiratory function assessment form, etc., τ θ is the clinical text encoding model, which is obtained by pre-training on a respiratory disease-related text dataset. ∈ represents the noise added in the diffusion process and obeys the Gaussian distribution, ∈ θ (z t ,t,τ θ(y)) represents the Denoising U-Net model for removing noise, and the model is based on the diffusion step t, the inspiratory phase CT image feature vector z at this step t and the clinical text code τ θ (y) as input. The meaning of the above training objectives is to provide auxiliary information for the model through clinical texts such as patient electronic medical records, pulmonary function test results, and respiratory function assessment forms. The model gradually removes the noise added in the diffusion process during the reverse diffusion process, and generates expiratory phase CT images based on clinical texts and inspiratory phase CT images.

[0024] The parameter response map analysis module receives the registered paired patient biphasic CT images generated by the image registration module or the controllable image generation module, and performs a lung lesion analysis based on the parameter response map, thereby realizing a lung lesion analysis that is not limited by single-phase or biphasic CT. The CT parameter response map is a quantitative CT analysis method based on the comparison between biphasic CT images and voxel-by-voxel threshold classification. The analysis method based on the CT parameter response map described in this patent divides the whole lung into normal lung tissue, emphysema, functional small airway damage and pulmonary fibrosis areas by comparing the CT values ​​of the same part of the lung tissue under two different states of inspiration and exhalation, and the four different categories of lung tissue are marked in red, yellow, green and purple, respectively. The specific classification criteria are as follows:

[0025] (1) <-950 HU in the inspiratory phase and <-856 HU in the expiratory phase represent the emphysema area, which is indicated by red on the image;

[0026] (2) Inspiratory phase ≥ -950HU and expiratory phase < -856HU represent small airway lesions, which are indicated by yellow on the image;

[0027] (3) ≥-950HU in the inspiratory phase and ≥-856HU in the expiratory phase represent the normal area, which is indicated by green on the image;

[0028] (4) -600 to -250 HU during the inspiratory or expiratory phase represents the area of ​​pulmonary fibrosis, which is represented by purple on the image.

[0029] The existing parameter response graph analysis method only includes three classifications: normal, emphysema, and functional small airway damage. This method introduces the classification criteria of pulmonary fibrosis into the parameter response graph classification criteria to achieve a comprehensive assessment of the lung lesions of various types of patients.

[0030] In addition, the module includes a whole lung and lung segment segmentation model on a CT image and a result comparison and analysis model. The whole lung and different lung segment segmentation results will be combined with the parameter response map analysis results to estimate the volume proportion of different types of lung tissue in the whole lung and different lung segments. This helps doctors quantitatively evaluate the damage to lung tissue in different parts and provides a basis for subsequent evaluation of the patient's respiratory function based on the degree of organic damage to the lungs. The result comparison and analysis model compares the CT examination result images of the same patient and the corresponding parameter response map analysis results, and gives the corresponding change trend, evaluation visualization report and text report, thereby showing detailed information on the patient's lung lesions, including the degree, distribution and development trend of emphysema, small airway lesions and pulmonary fibrosis, etc., to provide doctors with a basis for auxiliary diagnosis and treatment decisions.

[0031] The technical effects to be achieved by the present invention are:

[0032] 1. The present invention proposes a clinical text-guided expiratory phase CT image generation method, which takes clinical texts such as patient electronic medical records, respiratory symptom assessments, pulmonary function test results, and patient inspiratory phase chest CT scan images as inputs, and can learn the relationship between the clinical phenotype of the patient's respiratory function reflected in the clinical text and the patient's lung lesion imaging phenotype reflected in the CT image, and between functional lesions and organic lesions. According to the patient's respiratory function and the lung condition in the inspiratory state, the lung condition in the expiratory state is estimated, thereby converting a given inspiratory phase CT image into an expiratory phase CT image.

[0033] 2. A method for analyzing lung lesions based on parameter response graphs and compatible with single-phase and dual-phase chest CT scan images. For existing expiratory-inspiratory dual-phase chest CT images, the method generates a registered dual-phase CT image pair through image registration; for chest CT images with only a single inspiratory phase, the method uses the expiratory phase CT image generation method guided by the clinical text to complete the missing expiratory phase CT image registered with the inspiratory phase to form a dual-phase CT image pair; finally, the analysis method analyzes the registered dual-phase CT image pair generated by dual-phase or single-phase chest CT based on the parameter response graph to determine the patient's lung lesion status and development trend. Compared with existing analysis methods, the method proposed in the present invention is not limited to patients who must undergo dual-phase CT examinations.

[0034] 3. The present invention proposes a pulmonary lesion analysis system based on a parameter response diagram and an image generation system. The system, based on the expiratory phase CT image generation method and the pulmonary lesion analysis method, can not only analyze a single examination result of a patient, but also analyze multiple examination results of the same patient, including single-phase or double-phase.

[0035] The CT scan results are compared and analyzed, and text and visual analysis reports are generated to evaluate the extent, distribution and development trend of different types of lung lesions, quantitatively monitor the occurrence and development of the disease, and serve as an effective tool for evaluating the severity and course of the disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Pulmonary lesion analysis system architecture based on parametric response graph and image generation system;

[0037] Figure 2 Diagram of the method for generating expiratory CT images guided by clinical text;

[0038] Figure 3 Image registration module structure; DETAILED DESCRIPTION

[0039] The following is a preferred embodiment of the present invention and combined with the accompanying drawings to further describe the technical solution of the present invention, but the present invention is not limited to this embodiment.

[0040] The present invention proposes a lung lesion analysis system based on a parameter response diagram and an image generation system.

[0041] The system is mainly composed of data collection and management module, image registration module, controllable image generation module, parameter response map analysis module and disease progression assessment module. Its overall architecture is as follows: Figure 1 Shown

[0042] Module 1 Data Collection and Management Module: This module connects with the hospital information system and CT image collection equipment to collect patient information and historical CT examination results. According to the different types of CT scans (single-phase or bi-phase CT), the CT images are sent to different subsequent system modules for image processing, parameter response diagram analysis and disease progression evaluation.

[0043] Module 2 Image Registration Module: This module receives the patient's biphasic CT scan images, and registers the expiratory phase CT scan image at the end of deep exhalation with the inspiratory phase CT scan image at the end of inspiration, so that the same lung tissue is in the same spatial position in two different CT images. In this way, by analyzing the difference in CT values ​​between the CT images of the same part of the lung tissue under two different lung states of inspiration and exhalation, an assessment of the lung inflation and pathological conditions can be obtained. At the same time, the registered biphasic CT images generated by this module will be used for the training of the module three.

[0044] Module three: controllable image generation module: This module is based on the clinical text-guided expiratory phase CT image generation method and the Stale Diffusion model architecture. It is trained with the registered bi-phase CT image pairs generated by the image registration module and the clinical texts such as the patient's electronic medical records, respiratory symptom assessment, and pulmonary function test results corresponding to the images. After training, it can directly generate corresponding, registered expiratory phase CT images based on clinical texts and single inspiratory phase CT images. The generated bi-phase CT image pairs can be used in the parameter response map analysis module to analyze the lung lesions.

[0045] Module 4 Parameter Response Map Analysis Module: This module accepts the registered biphasic CT images generated by modules 2 and 3, and uses voxel classification to divide the whole lung into normal lung tissue, emphysema, functional small airway damage and pulmonary fibrosis areas. The four different types of lung tissue are represented in red, yellow, green and purple in the CT image, which is convenient for doctors to evaluate the damage to lung tissue in different parts. At the same time, this module calculates the volume proportion of different types of lung tissue in the whole lung and different lung segments to quantitatively evaluate the severity of overall lung damage.

[0046] Data Collection and Management Module:

[0047] The data collection and management module is connected to the hospital information system, CT examination equipment and pulmonary function examination equipment to collect and manage the patient's electronic medical records, respiratory symptom assessments, historical CT examination results and corresponding pulmonary function examination results: the single inspiratory phase CT image and the corresponding electronic medical records, pulmonary function examination results, and respiratory symptom assessment results will be sent to the controllable image generation module to complete the missing expiratory phase CT image and generate a two-phase image pair; the two-phase CT image pair (including the patient's original one and the one generated by this method) will be sent to the image registration module, and the registered two-phase CT image pair will be sent to the parameter response diagram analysis module.

[0048] Image Registration Module

[0049] The image registration module receives the CT scan image of the patient's expiratory phase at the end of deep exhalation and the CT scan image at the end of inspiration, takes the CT image at the end of inspiration as a fixed image, and takes the CT image of the expiratory phase as a moving image, and uses the time-varying diffeomorphism with mean square metric (TVMSQC) method in the medical registration software ANTs (Advanced Normalization Tools) to achieve the registration of the lung CT scan images of the inspiratory phase and the expiratory phase. The time-varying diffeomorphism registration method is a registration method based on a physical model, assuming that the transformation between images is driven by a continuous velocity field determined by a control function. This method uses an iterative optimization algorithm with mean square error as a similarity index to find the optimal control function, and calculates the corresponding velocity field and displacement field, and then transforms the moving image to the space of the fixed image to obtain the registered image.

[0050] Controllable image generation module

[0051] The controllable image is based on the expiratory phase CT image generation method guided by the clinical text. The specific implementation method of the method is: it is composed of a clinical text encoding model and an expiratory phase CT image generation model. The clinical text encoding model is used to encode the information related to the patient's respiratory function clinical phenotype in the patient's electronic medical record, respiratory symptom assessment and pulmonary function test results into a feature vector; the image generation model includes two processes, diffusion and inverse diffusion. In the diffusion process, the inspiratory phase CT image will first be converted into a feature vector in the latent space by the encoder ε in the variational autoencoder, and the noise will be gradually added through the diffusion process. After that, the inspiratory phase CT feature vector with added noise is gradually generated by the clinical text feature inverse diffusion process while removing the noise and generating a feature vector representing the expiratory phase CT image, and the decoder in the variational autoencoder Converted into an actual expiratory phase CT image. The clinical text-guided expiratory phase CT image generation method introduces clinical text describing the patient's respiratory function into the image generation model. Unlike the existing image generation method, the text in this method does not directly indicate the performance of the generated image, but provides auxiliary information for the model to convert from expiratory phase CT to inspiratory phase CT. The model establishes the connection between the clinical phenotype and imaging phenotype of the respiratory function of patients with lung diseases by learning the relationship between the biphasic CT image and the clinical text, thereby realizing the generation of expiratory phase chest CT images based on clinical texts such as the patient's electronic medical record, respiratory symptom assessment, and pulmonary function test results, and inspiratory phase chest CT images.

[0052] The training of the expiratory CT generation model (ECTGen) and the reasoning process in actual use are composed of the above-mentioned inspiratory CT image diffusion process and the inverse diffusion process guided by clinical text. Its training goal can be expressed as:

[0053]

[0054] y:={y1,y2,y3,…},

[0055] Where x represents the patient's inspiratory CT image, y1, y2, y3, ... represent clinical texts such as the patient's electronic medical record, pulmonary function test results, respiratory function assessment form, etc., τ θ is the clinical text encoding model, which is obtained by pre-training on a respiratory disease-related text dataset. ∈ represents the noise added in the diffusion process and obeys the Gaussian distribution, ∈ θ (z t ,t,τ θ (y)) represents the Denoising U-Net model for removing noise, and the model is based on the diffusion step t, the inspiratory phase CT image feature vector z at this step t and the clinical text code τ θ (y) as input. The meaning of the above training objectives is to provide auxiliary information for the model through clinical texts such as patient electronic medical records, pulmonary function test results, and respiratory function assessment forms. The model gradually removes the noise added in the diffusion process during the reverse diffusion process, and generates expiratory phase CT images based on clinical texts and inspiratory phase CT images.

[0056] Parameter Response Diagram Analysis Module

[0057] The parameter response map analysis module receives the registered paired patient biphasic CT images generated by the image registration module or the controllable image generation module, and performs a lung lesion analysis based on the parameter response map, thereby realizing a lung lesion analysis that is not limited by single-phase or biphasic CT. The CT parameter response map is a quantitative CT analysis method based on the comparison between biphasic CT images and voxel-by-voxel threshold classification. The analysis method based on the CT parameter response map described in this patent divides the whole lung into normal lung tissue, emphysema, functional small airway damage and pulmonary fibrosis areas by comparing the CT values ​​of the same part of the lung tissue under two different states of inspiration and exhalation, and the four different categories of lung tissue are marked in red, yellow, green and purple, respectively. The specific classification criteria are as follows:

[0058] (1) <-950 HU in the inspiratory phase and <-856 HU in the expiratory phase represent the emphysema area, which is indicated by red on the image;

[0059] (2) Inspiratory phase ≥ -950HU and expiratory phase < -856HU represent small airway lesions, which are indicated by yellow on the image;

[0060] (3) ≥-950HU in the inspiratory phase and ≥-856HU in the expiratory phase represent the normal area, which is indicated by green on the image;

[0061] (4) -600 to -250 HU during the inspiratory or expiratory phase represents the area of ​​pulmonary fibrosis, which is represented by purple on the image.

[0062] In addition, the module includes a whole lung and lung segment segmentation model on a CT image and a result comparison and analysis model. The whole lung and different lung segment segmentation results will be combined with the parameter response map analysis results to estimate the volume proportion of different types of lung tissue in the whole lung and different lung segments. This helps doctors quantitatively evaluate the damage to lung tissue in different parts and provides a basis for subsequent evaluation of the patient's respiratory function based on the degree of organic damage to the lungs. The result comparison and analysis model compares the CT examination result images of the same patient and the corresponding parameter response map analysis results, and gives the corresponding change trend, evaluation visualization report and text report, thereby showing detailed information on the patient's lung lesions, including the degree, distribution and development trend of emphysema, small airway lesions and pulmonary fibrosis, etc., to provide doctors with a basis for auxiliary diagnosis and treatment decisions.

Claims

1. A pulmonary lesion analysis system based on a parameter response graph and an image generation system, characterized in that: The pulmonary lesion analysis method based on parameter response diagram and compatible with single-phase and dual-phase chest CT scan images was used to analyze the expiratory-inspiratory dual-phase chest CT image pairs and single inspiratory phase chest CT images. The analysis system is based on the lung lesion analysis method and is composed of a data collection and management module, an image registration module, a controllable image generation module, and a parameter response map analysis module. For an exhalation-inhalation biphasic chest CT image, the image registration module will first register the two images, and then the parameter response map analysis module will analyze the registered image pair; for a single inhalation phase chest CT image, the controllable image generation module will supplement the missing exhalation phase CT image, and the parameter response map analysis module will analyze the generated exhalation-inhalation biphasic chest CT image pair; The data collection and management module is connected with the hospital information system, CT examination equipment and pulmonary function examination equipment to collect and manage the patient's electronic medical records, respiratory symptom assessments, historical CT examination results and corresponding pulmonary function test results: the single inspiratory phase CT image and the corresponding electronic medical records, pulmonary function test results and respiratory symptom assessment results will be sent to the controllable image generation module to complete the missing expiratory phase CT image and generate a two-phase image pair; The bi-phase CT image pair will be sent to the image registration module; the registered bi-phase CT image pair will be sent to the parameter response map analysis module; The image registration module receives the patient's bi-phase CT scan image, registers the expiratory phase CT scan image at the end of deep exhalation with the inspiratory phase CT scan image at the end of inspiration, so that the same lung tissue is in the same spatial position in two different CT images, and obtains the registered bi-phase CT image, which is input into the controllable image generation module for its training, and is also input into the parameter response map analysis module; The controllable image generation module is based on a clinical text-guided expiratory phase CT image generation method and a StableDiffusion model architecture. The text encoder encodes the clinical text containing the patient's electronic medical record, respiratory symptom assessment results, and pulmonary function test results into indicative text features, and the image encoder encodes the patient's single inspiratory phase CT into an image potential representation. Under the conditions of the indicative text and the image potential representation, the expiratory phase CT is generated. The parameter response map analysis module receives the expiratory-inspiratory phase CT image pairs generated by the image registration module and the controllable image generation module, and uses a threshold-based voxel classification method to divide the whole lung into normal lung tissue, emphysema, and functional small airway damage areas, and calculates the volume proportions of the three different types of lung tissue in the whole lung and different lung segments, and represents the three different types of lung tissue in red, yellow, and green in the CT image. Finally, for the same patient, the results of previous examinations are compared and analyzed to determine the changes in each type of lung tissue reflected in the patient's previous examinations.

2. A pulmonary lesion analysis system based on a parameter response graph and an image generation system as claimed in claim 1, characterized in that: The generation method is based on the Stable Diffusion model framework, and uses clinical texts such as patient electronic medical records, respiratory symptom assessments, and pulmonary function test results as control information, and retrains the Stable Diffusion model together with the patient's biphasic chest CT scan images, so that the model learns the relationship between the patient's respiratory function clinical phenotype and the imaging phenotype in the process of diffusing random noise from the inspiratory phase CT and inversely diffusing the random noise to generate the expiratory phase CT. The controllable image generation module is based on the clinical text-guided expiratory phase CT image generation method, and is trained with the registered bi-phase CT image pairs generated by the image registration module and the clinical texts such as the patient's electronic medical records, respiratory symptom assessment, and pulmonary function test results corresponding to the images. After the training is completed, the corresponding, registered expiratory phase CT images can be directly generated based on the clinical text and the single inspiratory phase CT image, and the generated bi-phase CT image pairs can be used in the parameter response diagram analysis module to analyze the lung lesions.

3. A pulmonary lesion analysis system based on a parameter response graph and an image generation system as claimed in claim 2, characterized in that: The specific implementation method of the image registration module is to receive the expiratory phase CT scan image of the patient at the end of deep exhalation and the CT scan image at the end of inspiration, take the CT image at the end of inspiration as the target image, take the expiratory phase CT image as the original image and input the diffusion model DiffuseMorph to complete the image registration task, the diffusion model learns the correspondence between the features in the patient's bi-phase CT image and the bi-phase CT image through the conditional scoring function; then, the output result of the diffusion network is input into the deformation network to be converted into a registration field that can be directly used for non-rigid transformation of the image, the obtained registration field is directly applied to the expiratory phase CT image by the spatial transformation layer, and the registered image is obtained after non-rigid transformation.

4. A pulmonary lesion analysis system based on a parameter response graph and an image generation system as claimed in claim 3, characterized in that: The parameter response map analysis module receives the registered paired patient bi-phase CT images generated by the image registration module or the controllable image generation module, and performs analysis based on the CT parameter response map on the patient's chest CT examination results and image sequences not limited to bi-phase or single-phase; The analysis method based on CT parameter response diagram divides the whole lung into normal lung tissue, emphysema, functional small airway damage and pulmonary fibrosis areas by comparing the CT values ​​of the same part of lung tissue in two different states of inspiration and expiration, and the four different types of lung tissue are marked in red, yellow, green and purple respectively. The specific classification criteria are as follows: (1) <-950 HU in the inspiratory phase and <-856 HU in the expiratory phase represent the emphysema area, which is indicated by red on the image; (2) Inspiratory phase ≥ -950HU and expiratory phase < -856HU represent small airway lesions, which are indicated by yellow on the image; (3) ≥-950HU in the inspiratory phase and ≥-856HU in the expiratory phase represent the normal area, which is indicated by green on the image; (4) -600 to -250 HU in the inspiratory or expiratory phase represents the area of ​​pulmonary fibrosis, which is indicated by purple on the image; The parameter response graph analysis module includes a result comparison analysis model, which compares the CT examination result images of the same patient and the corresponding parameter response graph analysis results, and gives corresponding change trend analysis and evaluation results; the parameter response graph analysis module generates a visual result report, showing detailed information on the patient's lung lesions, including the degree, distribution and development trend of emphysema, small airway lesions and pulmonary fibrosis.

Images