AI-based CT image automatic analysis system and method

Abstract

The invention relates to the technical field of image analysis, and discloses an AI-based CT image automatic analysis system, which comprises the following modules: a data acquisition and preprocessing module for acquiring original CT image data from CT equipment or an image storage system, an image segmentation module for segmenting different organs in a CT image, a feature extraction module for extracting features of different organs in the CT image, and an image analysis module for analyzing the features of different organs in the CT image. And the diagnosis and analysis module is used for carrying out real-time diagnosis on a patient, result visualization and report generation module and automatically generating a diagnosis report according to diagnosis and analysis results. According to the method, the connectivity of blood vessels can be kept when the AI system segments the CT image through the established topology structure, and the current segmentation process is optimized through the algorithm when the unreasonable part of the connectivity of the blood vessels exists in the current segmentation algorithm process through calculation and retrieval, so that the segmentation efficiency is improved. And the problem of blood vessel fracture possibly occurring in a traditional segmentation method is avoided.

Description

Technical Field

[0001] The present invention relates to the field of image analysis technology, and in particular to an AI-based CT image automatic analysis system and method. Background Art

[0002] Automatic CT image analysis is the process of automatically processing and analyzing computed tomography (CT) images using advanced artificial intelligence and machine learning technologies. This technology mainly relies on deep learning algorithms to quickly and accurately identify and classify abnormal structures and lesions in images, such as tumors, hemorrhages, pneumonia, etc. The automatic analysis system learns image features through training on a large number of labeled samples, and then realizes instant interpretation of new images. Compared with traditional manual analysis, automatic analysis not only improves the efficiency and accuracy of image reading, but also can provide doctors with auxiliary diagnosis suggestions and reduce the risk of misdiagnosis or missed diagnosis caused by human factors. With the development of technology, automatic CT image analysis has shown broad application prospects in the fields of medical imaging, tumor screening, and emergency image evaluation, and is driving medical services to develop in a more efficient and accurate direction.

[0003] In the prior art, image segmentation is a key step in the automatic analysis of CT images, and its accuracy directly affects the subsequent feature extraction and diagnostic analysis.

[0004] However, in actual applications, due to the large number and complex structure of blood vessels around different organs in the human body, traditional image segmentation methods often cause broken images at the connecting blood vessels during lesion and organ segmentation, resulting in inaccurate segmentation, which in turn affects the reliability of the final diagnosis results. In view of this, we propose an AI-based automatic CT image analysis system and method. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an AI-based CT image automatic analysis system and method to solve the problem.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI-based CT image automatic analysis system and method, comprising the following modules:

[0007] Data acquisition and preprocessing module: responsible for obtaining the original CT image data from the CT device or image storage system, and introducing the time series to sort the acquired images;

[0008] Image segmentation module: Segment different organs in CT images, and segment and locate lesions in CT images based on organ segmentation;

[0009] Feature extraction module: extracts the morphological features of lesions in CT images, including size, shape, boundary, and texture, and analyzes and identifies the texture features of lesions in images;

[0010] Diagnosis and analysis module: It uses integrated learning methods to integrate multiple different diagnostic models and conduct real-time diagnosis. It also combines the individual characteristics of the patient and other disease factors to conduct personalized disease analysis and prediction.

[0011] Result visualization and report generation module: The analysis results of CT images are presented to doctors in the form of three-dimensional reconstruction, image annotation, and heat map, and a diagnosis report is automatically generated based on the diagnosis and analysis results.

[0012] Preferably, the data acquisition and preprocessing module includes a data fusion processing module, an individual characteristic information entry module and a data preprocessing module. The data fusion processing module synchronously performs fusion acquisition with data from other medical devices including PET and MRI, and performs data combination analysis for the CT image scanning site. The individual characteristic information entry module automatically optimizes the acquisition parameters of the CT device through an intelligent algorithm based on the patient's individual age, body shape, disease type and scanning site characteristics. The data preprocessing module will sort and preprocess the image through a time series analysis algorithm when acquiring image information and individual characteristic information, and perform operations including denoising and contrast enhancement on the image.

[0013] Preferably, the image segmentation module uses a multi-scale segmentation algorithm to process CT images, and simultaneously segments and identifies multiple groups of lesions of different sizes in the CT scanning area. At the same time, for the segmentation and recognition algorithm, image segmentation is achieved through partial annotation data or indirect annotation information, and the model automatically learns other unlabeled image segmentation areas based on the identified areas. In addition, a topological structure is used to ensure that the segmented blood vessels are connected.

[0014] Preferably, the feature extraction module, in addition to extracting features from conventional morphology and texture in CT images, will simultaneously mine the causal relationship features between lesions and surrounding tissues in CT images, and develop corresponding optimization algorithms to screen out the most representative and discriminative features from a large number of extracted features, and remove redundant and irrelevant features.

[0015] Preferably, the diagnosis and analysis module adopts a cross-disease diagnosis model and performs comprehensive AI diagnostic analysis on multiple characteristic information points marked in CT images. By designing a general feature extraction and diagnosis framework, targeted training is carried out in combination with the characteristics of different diseases. At the same time, a dynamic risk assessment model is established to update the patient's disease risk assessment in real time based on the results of each CT imaging examination of the patient, the interval time, and treatment status factors.

[0016] Preferably, the result visualization and report generation module uses AR technology to superimpose the analysis results of the CT image on the patient's real body parts. The doctor can intuitively see the actual position, size and shape of the lesion in the patient's body, as well as its relationship with the surrounding tissues through the AR device. In addition, when generating a written diagnosis report, a voice report generation function is also provided. The system generates a detailed report in the form of voice and sends it to the physician based on the diagnosis results and analysis.

[0017] The AI-based automatic CT image analysis method includes the following steps:

[0018] S1: Data collection and preprocessing

[0019] Establish connections with the hospital's CT, PET, MRI equipment and image storage systems to obtain original image data. At the same time, based on the basic information of individual patients, including age, body shape, suspected disease type, and characteristics of the scanned area, automatically adjust the acquisition parameters of the CT equipment, including optimizing the scan layer thickness, scan range, and exposure time;

[0020] S2: Image Segmentation

[0021] The preprocessed CT image data is input into the pre-trained organ segmentation model to segment different organs in the CT image, and after the segmentation is completed, the image data containing the organs is further input into the lesion segmentation model;

[0022] S3: Feature extraction

[0023] For the segmented lesion images, the algorithm is used to extract the morphological features of the lesions, and the texture features of the lesions are extracted by analyzing the gray-level co-occurrence matrix and local binary pattern texture feature analysis method of the lesion images;

[0024] S4: Diagnosis and Analysis

[0025] The extracted lesion morphological features, texture features and other fusible features are used as input and sent to the pre-trained disease diagnosis model to diagnose the disease on the CT image, determine whether the disease exists and the type of disease, and use the historical CT image data and current image data features to predict the development trend of the disease;

[0026] S5: Results visualization and report generation

[0027] The analysis results of CT images are displayed through three-dimensional reconstruction, image annotation, and heat map visualization technology. Based on the diagnosis and analysis results, a detailed diagnosis report is automatically generated using natural language processing technology. At the same time, the generated text version of the diagnosis report is converted into a voice version and sent to the physician.

[0028] Preferably, in the S2 image segmentation, the segmentation task is regarded as a decision-making process, and a reinforcement learning algorithm is used to allow the segmentation algorithm to try different segmentation actions in different areas of the CT image. Rewards or penalties are given by comparing the actual segmentation results after each action or based on evaluation indicators, and the image segmentation algorithm is self-optimized and improved through reward and penalty weighting.

[0029] Preferably, when extracting texture features at the lesion in the S3 feature extraction, features related to specific biomarkers will be simultaneously extracted from the CT image, including for cancer diagnosis, extracting image features related to specific antigen expression on the surface of tumor cells, and using correlation analysis with known biomarkers.

[0030] Preferably, a hybrid intelligent diagnosis system will be constructed in the S4 diagnosis and analysis, combining a rule-based expert system and a deep learning model. In the process, the rule-based expert system will be used to preliminarily screen and classify the CT images to exclude some cases that obviously do not meet the characteristics of a certain disease, and then the preliminarily processed image data will be input into the deep learning model for further AI diagnostic analysis.

[0031] The present invention provides an AI-based CT image automatic analysis system and method, which has the following beneficial effects:

[0032] 1. The present invention establishes a topological structure that enables the AI ​​system to maintain the connectivity of blood vessels when segmenting CT images. When unreasonable blood vessel connectivity is found in the current segmentation algorithm during calculation and retrieval, the current segmentation process is optimized through the algorithm, thereby avoiding the problem of blood vessel breakage that may occur in traditional segmentation methods.

[0033] 2. In terms of feature extraction, the present invention not only extracts conventional morphological and texture features, but also mines the causal relationship characteristics between the lesion and the surrounding tissue through AI algorithms, which helps to improve the accuracy of diagnosis and early detection of the disease.

[0034] 3. The establishment of a dynamic risk assessment model can update the disease risk assessment in real time according to the patient's specific situation, providing doctors with more accurate treatment recommendations and prognosis assessments.

[0035] 4. The introduction of the voice report generation function enables doctors to receive diagnosis results through hearing, especially in emergency situations, to quickly obtain diagnostic information and improve medical efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is the module diagram of the AI-based CT image automatic analysis system;

[0037] Figure 2It is a schematic diagram of the data acquisition and preprocessing module of the present invention;

[0038] Figure 3 This is a flow chart of the AI-based automatic CT image analysis method. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Example:

[0041] Please refer to the attached Figure 1 -Attached Figure 2 The embodiment of the present invention provides an AI-based CT image automatic analysis system, comprising the following modules:

[0042] Data acquisition and preprocessing module: responsible for obtaining the original CT image data from the CT device or image storage system, and introducing the time series to sort the acquired images;

[0043] Image segmentation module: Segment different organs in CT images, and segment and locate lesions in CT images based on organ segmentation;

[0044] Feature extraction module: extracts the morphological features of lesions in CT images, including size, shape, boundary, and texture, and analyzes and identifies the texture features of lesions in images;

[0045] Diagnosis and analysis module: It uses integrated learning methods to integrate multiple different diagnostic models and conduct real-time diagnosis. It also combines the individual characteristics of the patient and other disease factors to conduct personalized disease analysis and prediction.

[0046] Result visualization and report generation module: The analysis results of CT images are presented to doctors in the form of three-dimensional reconstruction, image annotation, and heat map, and a diagnosis report is automatically generated based on the diagnosis and analysis results.

[0047] The data acquisition and preprocessing module includes a data fusion processing module, an individual characteristic information entry module and a data preprocessing module. The data fusion processing module synchronously performs fusion acquisition with data from other medical devices including PET and MRI, and performs data combination analysis for the CT image scanning site. The individual characteristic information entry module automatically optimizes the acquisition parameters of the CT device through an intelligent algorithm based on the patient's individual age, body shape, disease type and scanning site characteristics. The data preprocessing module will sort and preprocess the image through a time series analysis algorithm when acquiring image information and individual characteristic information, and perform operations including denoising and contrast enhancement on the image.

[0048] The image segmentation module uses a multi-scale segmentation algorithm to process CT images, and simultaneously segment and identify multiple groups of lesions of different sizes in the CT scanning area. At the same time, for the segmentation and recognition algorithm, image segmentation is achieved through partial annotation data or indirect annotation information, and the model automatically learns other unlabeled image segmentation areas based on the identified areas. In addition, a topological structure is used to ensure that the segmented blood vessels have connectivity. The following algorithms are used in the image segmentation module:

[0049] The blood vessel region in the CT image is regarded as a graph G = (V, E), where V is a set of nodes (which can be key points in the blood vessel, such as branch points, endpoints, etc.), and E is an edge set (line segments connecting nodes, indicating the connected parts of the blood vessel). To detect whether the blood vessels are connected, the depth-first search (DFS) or breadth-first search (BFS) algorithm can be used. Taking the depth-first search as an example, its recursive formula is as follows:

[0050]

[0051] Where v is the current node, visited(v) indicates whether node v has been visited, and adj(v) indicates the set of nodes adjacent to node v. By performing a depth-first search starting from a starting node, if all nodes in the graph can be traversed, it means that the blood vessels are connected;

[0052] In fact, an array is created to record whether each node has been visited. It is initialized to a state where all nodes are unvisited. The nodes are v1, v2, ..., v 10 , assuming that this array is named visited, its length is the same as the size of the node set V, and its initial value is false. For example, for the above 10 nodes, the initial state of the visited array is [false,false,false,false,false,false,false,false,false,false];

[0053] Select any node from the node set V as the starting node to start the depth-first search. In this example, select node v1 as the starting node;

[0054] Call the depth-first search function DFS(v_1) for the starting node v1. Inside the DFS function, first mark the current node as visited, that is, visited[1] = true (assuming that the index of node v1 in the array is 1)

[0055] Then, traverse all adjacent nodes of the current node v1. By searching the edge set E, the nodes adjacent to v1 can be determined. Suppose there are three adjacent nodes v2, v3, and v4. For each adjacent node, check whether it has been visited. If it has not been visited, recursively call the DFS function for the adjacent node. For example, for node v2, because visited[2] = false (assuming that the index of node v2 in the array is 2), DFS(v_2) is called. Inside the DFS(v_2) function, v2 is also marked as visited, and then the adjacent nodes of v2 are traversed, and the DFS function is continued to be recursively called according to the above rules until a node is encountered whose adjacent nodes have been visited;

[0056] When the depth-first search process is completed (that is, there are no new unvisited nodes for recursive calls to the DFS function), check all elements in the visited array. If all elements in the visited array are true, it means that starting from the starting node, all nodes in the graph can be traversed through the depth-first search, then it can be determined that the segmented blood vessels are connected. In this example, if the final visited array becomes [true, true, true, true, true, true, true, true, true, true], it means that the entire blood vessel structure is topologically connected, that is, the segmentation result is reasonable in terms of connectivity;

[0057] Assume that an error occurs during the segmentation process, causing a section of the blood vessel to be disconnected incorrectly. For example, there should be an edge connecting nodes v5 and v6, but due to the segmentation error, this edge is not correctly identified;

[0058] Eventually, the visited array may display a state similar to [true, true, true, true, true, false, false, false, false, false], which indicates that the segmented blood vessels are not connected and the segmentation algorithm or results need to be further checked and corrected.

[0059] In addition to extracting features from conventional morphology and texture in CT images, the feature extraction module will simultaneously mine the causal relationship features between lesions and surrounding tissues in CT images, and develop corresponding optimization algorithms to select the most representative and discriminative features from a large number of extracted features and remove redundant and irrelevant features. The following algorithms are involved in the process:

[0060] Suppose two time series are X t (represents the time series of lesion-related indicators) and Y t (represents the time-varying sequence of indicators related to surrounding tissues) First, t and Y t Perform a stationary test to ensure that the data is stationary or becomes a stationary series after transformation such as difference. Then, establish an autoregressive model (AR) to predict X. t and Y t , for X t p-order autoregressive model AR(p);

[0061]

[0062] Where c is a constant term, α i is the autoregressive coefficient, ε t is the error term;

[0063] Similarly, for Y t Establish the q-order autoregressive model AR(q);

[0064] Next, in the t The hysteresis value of X is established t The prediction model is:

[0065]

[0066] where β j Yes t The coefficient of the lagged term of ;

[0067] The statistic F for the causality test is:

[0068]

[0069] RSS0 contains only X t The residual sum of squares of the model with its own lag term (i.e., the first AR(p) model above), RSS1 is the residual sum of squares including Y t The residual sum of squares of the model with the lag term (i.e. the third model above), T is the number of samples, r is the introduced Y t The number of lag terms;

[0070] If the calculated F statistic is greater than the given critical value, the null hypothesis is rejected and Y t X t There is a causal relationship, that is, changes in the relevant indicators of the surrounding tissues will cause changes in the relevant indicators of the lesion to a certain extent;

[0071] Assume that there is a series of lung CT image data, which is taken at different time points for a patient with lung disease, and the volume of the lung lesion is taken as X. t (The lesion volume is measured at regular intervals) and the blood perfusion volume of the lung tissue around the lesion is taken as Y t (Use specific imaging analysis techniques to obtain blood perfusion data at each time point)

[0072] After the stability test, we established X t The AR(2) model and Y t AR(3) model, and then include Y t The hysteresis value is established in the case of X t prediction models;

[0073] Assuming that the calculated F statistic is greater than the critical value, this indicates that changes in the blood perfusion volume of the tissue surrounding the lung lesion have a causal relationship with changes in the lesion volume. We can further extract features that specifically reflect this causal relationship based on the correlation coefficient, such as the time interval between the change in blood perfusion volume and the change in lesion volume, the correlation between the amplitude of the change, and other features, and use them as part of the causal relationship between the lesion and the surrounding tissue.

[0074] The diagnosis and analysis module adopts a cross-disease diagnosis model and performs comprehensive AI diagnosis and analysis on multiple feature information points marked in CT images. By designing a general feature extraction and diagnosis framework, targeted training is carried out in combination with the characteristics of different diseases. At the same time, a dynamic risk assessment model is established to update the patient's disease risk assessment in real time based on the results of each CT imaging examination of the patient, the interval time, and treatment status factors.

[0075] The result visualization and report generation module uses AR technology to superimpose the analysis results of CT images on the patient's real body parts. Doctors can use AR devices to intuitively see the actual position, size and shape of the lesions in the patient's body, as well as their relationship with surrounding tissues. In addition, when generating a written diagnosis report, a voice report generation function is also provided. The system generates a detailed report in the form of voice and sends it to the physician based on the diagnosis results and analysis.

[0076] Please refer to the attached Figure 3 , the AI-based automatic CT image analysis method includes the following steps:

[0077] S1: Data collection and preprocessing

[0078] Establish connections with the hospital's CT, PET, MRI equipment and image storage systems to obtain original image data. At the same time, based on the basic information of individual patients, including age, body shape, suspected disease type, and characteristics of the scanned area, automatically adjust the acquisition parameters of the CT equipment, including optimizing the scan layer thickness, scan range, and exposure time;

[0079] S2: Image Segmentation

[0080] The preprocessed CT image data is input into the pre-trained organ segmentation model to segment different organs in the CT image, and after the segmentation is completed, the image data containing the organs is further input into the lesion segmentation model;

[0081] S3: Feature extraction

[0082] For the segmented lesion images, the algorithm is used to extract the morphological features of the lesions, and the texture features of the lesions are extracted by analyzing the gray-level co-occurrence matrix and local binary pattern texture feature analysis method of the lesion images;

[0083] S4: Diagnosis and Analysis

[0084] The extracted lesion morphological features, texture features and other fusible features are used as input and sent to the pre-trained disease diagnosis model to diagnose the disease on the CT image, determine whether the disease exists and the type of disease, and use the historical CT image data and current image data features to predict the development trend of the disease;

[0085] S5: Results visualization and report generation

[0086] The analysis results of CT images are displayed through three-dimensional reconstruction, image annotation, and heat map visualization technology. Based on the diagnosis and analysis results, a detailed diagnosis report is automatically generated using natural language processing technology. At the same time, the generated text version of the diagnosis report is converted into a voice version and sent to the physician.

[0087] In the S2 image segmentation, the segmentation task is regarded as a decision-making process. The reinforcement learning algorithm is used to allow the segmentation algorithm to try different segmentation actions in different areas of the CT image. Rewards or penalties are given based on the actual segmentation results after each action or based on evaluation indicators, and the image segmentation algorithm is self-optimized and improved through reward and penalty weighting.

[0088] When extracting texture features at the lesion in the S3 feature extraction, features related to specific biomarkers will be simultaneously extracted from the CT image, including for cancer diagnosis, extracting image features related to specific antigen expression on the surface of tumor cells, and using correlation analysis with known biomarkers.

[0089] In the S4 diagnosis and analysis, a hybrid intelligent diagnosis system will be constructed, combining a rule-based expert system and a deep learning model. In the process, the rule-based expert system is first used to preliminarily screen and classify CT images to exclude some cases that obviously do not meet the characteristics of a certain disease. Then, the preliminarily processed image data is input into the deep learning model for further AI diagnosis and analysis. The deep learning model algorithm constructed in the process is as follows:

[0090] Convolutional Layer:

[0091] Let the input feature map be X (with dimension H in ×W in ×C in , where H in Indicates height, W in Indicates width, C in represents the number of channels), the convolution kernel is W (the dimension is h×w×C in ×C out , where h represents the height of the convolution kernel, w represents the width of the convolution kernel, and C out represents the number of output channels), the bias is b (the dimension is C out ), the output Y of the convolutional layer (dimension H out ×W out ×C out ) is calculated as:

[0092]

[0093] Where i, j, and k are the indexes of the output feature map Y in height, width, and channel dimensions, respectively; padding represents the padding size (used to control the size of the output feature map); and stride represents the step size (controls the stride of the convolution kernel moving on the input feature map);

[0094] Assume that the feature map X of the input CT image has a size of 32×32×3 and the convolution kernel W has a size of 3×3×3×16 (i.e., h=3, w=3, C in =3,C out =16), padding=1, stride=1;

[0095] First calculate the size of the output feature map:

[0096]

[0097] Then for any point Y in the output feature map Y ijk , calculated according to the above convolutional layer formula, for example, for the point i=1,j=2,k=3:

[0098]

[0099] Pooling layer (taking maximum pooling as an example):

[0100] Let the input feature map be X (with dimension H in ×W in ×C in ), the pooling window size is h pool × pool , stride pool , the feature map is Y (the dimension is H out ×W out ×C out );

[0101]

[0102] Assume that the input feature map X size is 32×32×16, the pooling window size is 2×2, the step size is 2, and the padding pool =0, calculate the size of the feature map:

[0103]

[0104] For any point Y in the output feature map Y ijk , calculated according to the maximum pooling formula, for example, for the point i = 3, j = 4, k = 5:

[0105]

[0106] Fully connected layer:

[0107] Assume the input vector is x (dimension is n), the weight matrix is ​​W (dimension is n×m), the bias vector is b (dimension is m), and the output vector y of the fully connected layer is calculated as:

[0108] y=Wx+b

[0109] For example, suppose that after the previous convolutional layer, pooling layer and other operations, an input x is flattened into a one-dimensional vector with a dimension of 100, the weight matrix W has a dimension of 100×5, and the bias vector b has a dimension of 5;

[0110] Then the output vector y of the fully connected layer is calculated as follows:

[0111] y=Wx+b

[0112]

[0113] Finally, the fully connected layer outputs a vector that can represent the probability distribution of different diseases for the final diagnosis.

[0114] Example:

[0115] Continuing with the example of the rule-based expert system above, after a preliminary judgment that the patient may have lung cancer, the pre-processed CT image data of the patient (after normalization, cropping, etc.) is input into a pre-trained deep learning model. Assume that after a series of operations such as the convolutional layer, pooling layer, and fully connected layer mentioned above, the input image data finally outputs a vector y=0.1,0.6,0.3], where it is assumed that the probability of having pneumonia is 0.1, the probability of having lung cancer is 0.6, and the probability of having other diseases is 0.3;

[0116] Based on this output result, combined with the preliminary judgment of the previous rule-based expert system, it is further determined that the patient is more likely to have lung cancer, thereby completing a more accurate AI diagnostic analysis.

[0117] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based CT image automatic analysis system, characterized in that: Includes the following modules: Data acquisition and preprocessing module: responsible for obtaining the original CT image data from the CT device or image storage system, and introducing the time series to sort the acquired images; Image segmentation module: Segment different organs in CT images, and segment and locate lesions in CT images based on organ segmentation; Feature extraction module: extracts the morphological features of lesions in CT images, including size, shape, boundary, and texture, and analyzes and identifies the texture features of lesions in images; Diagnosis and analysis module: It uses integrated learning methods to integrate multiple different diagnostic models and conduct real-time diagnosis. It also combines the individual characteristics of the patient and other disease factors to conduct personalized disease analysis and prediction. Result visualization and report generation module: The analysis results of CT images are presented to doctors in the form of three-dimensional reconstruction, image annotation, and heat map, and a diagnosis report is automatically generated based on the diagnosis and analysis results.

2. The AI-based CT image automatic analysis system according to claim 1, characterized in that: The data acquisition and preprocessing module includes a data fusion processing module, an individual characteristic information entry module and a data preprocessing module. The data fusion processing module synchronously performs fusion acquisition with data from other medical devices including PET and MRI, and performs data combination analysis for the CT image scanning site. The individual characteristic information entry module automatically optimizes the acquisition parameters of the CT device through an intelligent algorithm based on the patient's individual age, body shape, disease type and scanning site characteristics. The data preprocessing module will sort and preprocess the image through a time series analysis algorithm when acquiring image information and individual characteristic information, and perform operations including denoising and contrast enhancement on the image.

3. The AI-based CT image automatic analysis system according to claim 1, characterized in that: The image segmentation module uses a multi-scale segmentation algorithm to process CT images, and simultaneously segments and identifies multiple groups of lesions of different sizes in the CT scanning area. At the same time, for the segmentation and recognition algorithm, image segmentation is achieved through partial annotation data or indirect annotation information, and the model automatically learns other unlabeled image segmentation areas based on the identified areas. In addition, a topological structure is used to ensure that the segmented blood vessels are connected.

4. The AI-based CT image automatic analysis system according to claim 1, characterized in that: In addition to extracting features from conventional morphology and texture in CT images, the feature extraction module will simultaneously mine the causal relationship features between lesions and surrounding tissues in CT images, and develop corresponding optimization algorithms to screen out the most representative and discriminative features from a large number of extracted features and remove redundant and irrelevant features.

5. The AI-based automatic CT image analysis system according to claim 1, characterized in that: The diagnosis and analysis module adopts a cross-disease diagnosis model and performs comprehensive AI diagnosis and analysis on multiple feature information points marked in CT images. By designing a general feature extraction and diagnosis framework, targeted training is carried out in combination with the characteristics of different diseases. At the same time, a dynamic risk assessment model is established to update the patient's disease risk assessment in real time based on the results of each CT imaging examination of the patient, the interval time, and treatment status factors.

6. The AI-based automatic CT image analysis system according to claim 1, characterized in that: The result visualization and report generation module uses AR technology to superimpose the analysis results of CT images on the patient's real body parts. Doctors can use AR devices to intuitively see the actual position, size and shape of the lesions in the patient's body, as well as their relationship with surrounding tissues. In addition, when generating a written diagnosis report, a voice report generation function is also provided. The system generates a detailed report in the form of voice and sends it to the physician based on the diagnosis results and analysis.

7. The AI-based automatic CT image analysis method according to claim 1, characterized in that: The following steps are involved: S1: Data collection and preprocessing Establish connections with the hospital's CT, PET, MRI equipment and image storage systems to obtain original image data. At the same time, based on the basic information of individual patients, including age, body shape, suspected disease type, and characteristics of the scanned area, automatically adjust the acquisition parameters of the CT equipment, including optimizing the scan layer thickness, scan range, and exposure time; S2: Image Segmentation The preprocessed CT image data is input into the pre-trained organ segmentation model to segment different organs in the CT image, and after the segmentation is completed, the image data containing the organs is further input into the lesion segmentation model; S3: Feature extraction For the segmented lesion images, the algorithm is used to extract the morphological features of the lesions, and the texture features of the lesions are extracted by analyzing the gray-level co-occurrence matrix and local binary pattern texture feature analysis method of the lesion images; S4: Diagnosis and Analysis The extracted lesion morphological features, texture features and other fusible features are used as input and sent to the pre-trained disease diagnosis model to diagnose the disease on the CT image, determine whether the disease exists and the type of disease, and use the historical CT image data and current image data features to predict the development trend of the disease; S5: Results visualization and report generation The analysis results of CT images are displayed through three-dimensional reconstruction, image annotation, and heat map visualization technology. Based on the diagnosis and analysis results, a detailed diagnosis report is automatically generated using natural language processing technology. At the same time, the generated text version of the diagnosis report is converted into a voice version and sent to the physician.

8. The AI-based automatic CT image analysis method according to claim 7, characterized in that: In the S2 image segmentation, the segmentation task is regarded as a decision-making process. The reinforcement learning algorithm is used to allow the segmentation algorithm to try different segmentation actions in different areas of the CT image. Rewards or penalties are given based on the actual segmentation results after each action or based on evaluation indicators, and the image segmentation algorithm is self-optimized and improved through reward and penalty weighting.

9. The AI-based automatic CT image analysis method according to claim 7, characterized in that: When extracting texture features at the lesion in the S3 feature extraction, features related to specific biomarkers will be simultaneously extracted from the CT image, including for cancer diagnosis, extracting image features related to specific antigen expression on the surface of tumor cells, and using correlation analysis with known biomarkers.

10. The AI-based automatic CT image analysis method according to claim 7, characterized in that: The S4 diagnosis and analysis will construct a hybrid intelligent diagnosis system that combines a rule-based expert system and a deep learning model. In the process, the rule-based expert system will be used to preliminarily screen and classify the CT images to exclude some situations that obviously do not meet the characteristics of a certain disease. The preliminarily processed image data will then be input into the deep learning model for further AI diagnostic analysis.

Images