A system for evaluating non-alcoholic fatty liver disease models based on artificial intelligence

Abstract

The present invention discloses a system for evaluating a non-alcoholic fatty liver disease model based on artificial intelligence, and relates to the technical field of non-alcoholic fatty liver disease evaluation; the system comprises: a data acquisition unit, a feature extraction unit, a fatty degeneration prediction unit and a liver fibrosis region determination unit; the liver fibrosis region determination unit is used to determine the liver fibrosis region of a non-alcoholic fatty liver disease patient according to a first fatty degeneration feature extracted by a fatty degeneration feature extraction module, a second fatty degeneration feature predicted by a fatty degeneration prediction unit, a lobular inflammation feature extracted by a lobular inflammation feature extraction module, a fibrosis feature extracted by a fibrosis feature extraction module, and a hepatocyte ballooning degeneration feature extracted by a hepatocyte ballooning degeneration feature extraction module; the present application can improve the efficiency of evaluating the non-alcoholic fatty liver disease condition.

Description

Technical Field

[0001] The present application relates to the technical field of non-alcoholic fatty liver disease assessment, and more specifically, to a system for assessing non-alcoholic fatty liver disease models based on artificial intelligence. Background Art

[0002] Nonalcoholic fatty liver disease refers to a clinical pathological syndrome characterized by excessive fat deposition in hepatocytes, excluding alcohol and other clear liver-damaging factors, and is an acquired metabolic stress liver injury closely related to insulin resistance and genetic susceptibility. It includes simple fatty liver (SFL), nonalcoholic steatohepatitis (NASH) and related cirrhosis.

[0003] In addition to directly causing decompensated cirrhosis, hepatocellular carcinoma, and recurrence of transplanted liver, non-alcoholic fatty liver disease can also affect the progression of other chronic liver diseases and participate in the onset of type 2 diabetes and atherosclerosis. Metabolic syndrome-related malignant tumors, arteriosclerotic cardiovascular and cerebrovascular diseases, and cirrhosis are important factors affecting the quality of life and life expectancy of patients with non-alcoholic fatty liver disease. For this reason, non-alcoholic fatty liver disease is a new challenge in the field of contemporary medicine, and the harm of non-alcoholic fatty liver disease to human health will continue to increase.

[0004] Currently, liver pathology images are usually analyzed manually to evaluate non-alcoholic fatty liver disease. However, problems with existing manual pathology evaluations include: a lack of professional doctors for non-alcoholic fatty liver disease, which results in patients not receiving timely treatment.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The embodiment of the present application provides a system for evaluating the non-alcoholic fatty liver disease model based on artificial intelligence to solve the above-mentioned technical problems.

[0007] The present application provides a system for evaluating a non-alcoholic fatty liver model based on artificial intelligence, comprising a data acquisition unit, a feature extraction unit, a fatty degeneration prediction unit, and a liver fibrosis area determination unit; wherein,

[0008] The data acquisition unit is used to acquire ultrasound imaging data, radiological imaging data and liver tissue pathology data of patients with non-alcoholic fatty liver disease;

[0009] The feature extraction unit includes a fatty degeneration feature extraction module, a lobular inflammation feature extraction module, a fibrosis feature extraction module and a hepatocyte ballooning feature extraction module;

[0010] The liver fibrosis area determination unit is used to determine the liver fibrosis area of ​​the non-alcoholic fatty liver disease patient based on the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning degeneration feature extracted by the hepatocyte ballooning degeneration feature extraction module.

[0011] Furthermore, the lobular inflammation feature extraction module is used to extract the lobular inflammation feature based on the data related to the inflammatory cell clusters in the ultrasound imaging data and the data related to the inflammatory cell clusters in the radiological imaging data; the fibrosis feature extraction module is used to extract the fibrosis feature based on the data related to fibrosis in the liver tissue pathology data; the hepatocyte ballooning degeneration feature extraction module is used to extract the hepatocyte ballooning degeneration feature based on the data related to ballooning degeneration in the liver tissue pathology data.

[0012] Furthermore, the fatty degeneration feature extraction module is used to extract the first fatty degeneration feature according to the fatty degeneration related data in the ultrasonic imaging data and the fatty degeneration related data in the liver tissue pathology data.

[0013] Further, the data related to steatosis include the number, density, size, and spatial distribution data of macrovesicular steatosis and the number, density, size, and spatial distribution data of microvesicular steatosis; and extracting the first steatosis feature according to the data related to steatosis in the ultrasonic imaging data and the data related to steatosis in the liver tissue pathology data includes:

[0014] Constructing a global feature extraction network, a first local feature extraction network, and a second local feature extraction network; wherein the first local feature extraction network and the second local feature extraction network both adopt ResNet networks;

[0015] Inputting the quantity, density, size, and spatial distribution data of the macrovesicular steatosis into the first local feature extraction network to obtain macrovesicular steatosis features;

[0016] Inputting the quantity, density, size, and spatial distribution data of the microvesicular steatosis into the second local feature extraction network to obtain microvesicular steatosis features;

[0017] Performing a weighted operation on the macrovesicular steatosis feature and the microvesicular steatosis feature, and obtaining a steatosis fusion feature by a bilinear interpolation method;

[0018] The steatosis fusion feature is input into a global feature network to obtain the first steatosis feature.

[0019] Furthermore, the steatosis prediction unit is used for:

[0020] Inputting the first fatty degeneration feature into the embedding layer of the recurrent neural network to perform information dimension upgrading to obtain a dimension-upgraded feature;

[0021] Inputting the dimension-increased features into a first coding structure to obtain a first sample feature map;

[0022] Inputting the dimension-increased feature into a second coding structure to obtain a second sample feature map; wherein the first coding structure is different from the second coding structure;

[0023] Constructing a cross attention network; obtaining an updated first sample feature map and an updated second sample feature map through a multi-layer perceptron based on the first sample feature map and the second sample feature map;

[0024] Based on the updated first sample feature map and the updated second sample feature map, the second fatty degeneration feature is predicted.

[0025] Furthermore, the predicting of the second fatty degeneration feature based on the updated first sample feature map and the updated second sample feature map includes:

[0026] According to the updated first sample feature map and the updated second sample feature map, a first weight value of the updated first sample feature map is calculated based on the first neural network. and the second weight value of the updated second sample feature map ;

[0027] According to the first weight value of the updated first sample feature map and the second weight value of the updated second sample feature map , the second fatty degeneration feature is predicted.

[0028] Further, the determining of the liver fibrosis area of ​​the non-alcoholic fatty liver disease patient according to the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning degeneration feature extracted by the hepatocyte ballooning degeneration feature extraction module includes:

[0029] Preprocessing the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning feature extracted by the hepatocyte ballooning feature extraction module;

[0030] Add momentum factor a to the second neural network;

[0031] training the second neural network based on historical non-alcoholic fatty liver disease data;

[0032] The liver fibrosis area of ​​the non-alcoholic fatty liver disease patient is determined according to the second neural network obtained through training.

[0033] Furthermore, the global feature extraction network includes 3 convolution layers and 1 maximum pooling layer, the first convolution block includes 1 residual block, the second convolution block includes 4 residual blocks, and the third convolution block includes 6 residual blocks; each residual block contains 2 1×3 convolution kernels and 2 4×4 convolution kernels.

[0034] Furthermore, the data related to the inflammatory cell clusters include the number, density, size and spatial distribution data of the inflammatory cell clusters; the data related to fibrosis include the fibrosis area, the length, width, density and spatial distribution data of fibrous collagen; the data related to ballooning degeneration include the number, density, size and spatial distribution data of ballooning degeneration hepatocytes.

[0035] Furthermore, the data acquisition unit is also used to acquire radio frequency driving data of the non-alcoholic fatty liver patient, and the radio frequency driving data includes an attenuation coefficient and a backscattering coefficient.

[0036] Based on the data acquisition unit, feature extraction unit, fatty degeneration prediction unit and liver fibrosis area determination unit included in the system provided by the present application, automatic assessment of non-alcoholic fatty liver disease can be achieved, and the efficiency of assessing non-alcoholic fatty liver disease can be improved. Since the participation of professional doctors of non-alcoholic fatty liver disease is not required, patients can receive timely treatment. Specifically, it has the following beneficial effects: Comprehensive data integration: By acquiring ultrasound imaging data, radiological imaging data and liver tissue pathology data, patients with non-alcoholic fatty liver disease can be evaluated from different angles and levels, which increases the comprehensiveness and accuracy of diagnosis; Multi-feature extraction: Through different feature extraction modules, the features of fatty degeneration, lobular inflammation, fibrosis and hepatocyte ballooning are extracted respectively, which helps to understand each aspect of the lesion more carefully and provide richer information for diagnosis; Accurate determination of liver fibrosis area: Using the above-mentioned extracted features, the liver fibrosis area determination unit can integrate this information to determine the specific area of ​​liver fibrosis, which is crucial for assessing the severity of the disease and formulating a treatment plan; Combination of predicted and actual data: By combining the predicted second fatty degeneration feature with the actually extracted first fatty degeneration feature, more accurate comparison and verification can be performed to improve the reliability of diagnosis; Potential research value: This method can also be used as a research tool to help scientists better understand the pathological mechanism and development process of patients with non-alcoholic fatty liver disease; Improve the quality of patient care: By identifying and treating patients with non-alcoholic fatty liver disease earlier and more accurately, the quality of life of patients can be improved and complications associated with the disease may be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are used to provide a further understanding of the embodiments of the present invention and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0038] Figure 1 is a structural diagram of an optional system for evaluating a non-alcoholic fatty liver disease model based on artificial intelligence according to an embodiment of the present application;

[0039] Figure 2 This is an optional flow chart for extracting the first fatty degeneration feature according to an embodiment of the present application.

[0040] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0042] Alternatively, if Figure 1 As shown, the present application provides a system for evaluating a non-alcoholic fatty liver model based on artificial intelligence, comprising a data acquisition unit 101, a feature extraction unit 102, a fatty degeneration prediction unit 103, and a liver fibrosis area determination unit 104; wherein,

[0043] The data acquisition unit 101 is used to acquire ultrasound imaging data, radiological imaging data and liver tissue pathology data of patients with non-alcoholic fatty liver disease;

[0044] The feature extraction unit 102 includes a fatty degeneration feature extraction module, a lobular inflammation feature extraction module, a fibrosis feature extraction module, and a hepatocyte ballooning feature extraction module;

[0045] The liver fibrosis area determination unit 104 is used to determine the liver fibrosis area of ​​the patient with non-alcoholic fatty liver disease based on the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit 103, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning degeneration feature extracted by the hepatocyte ballooning degeneration feature extraction module.

[0046] Based on the data acquisition unit, feature extraction unit, fatty degeneration prediction unit and liver fibrosis area determination unit included in the system provided by the present application, automatic assessment of non-alcoholic fatty liver disease can be achieved, and the efficiency of assessing non-alcoholic fatty liver disease can be improved. Since the participation of professional doctors of non-alcoholic fatty liver disease is not required, patients can receive timely treatment. Specifically, it has the following beneficial effects: Comprehensive data integration: By acquiring ultrasound imaging data, radiological imaging data and liver tissue pathology data, patients with non-alcoholic fatty liver disease can be evaluated from different angles and levels, which increases the comprehensiveness and accuracy of diagnosis; Multi-feature extraction: Through different feature extraction modules, the features of fatty degeneration, lobular inflammation, fibrosis and hepatocyte ballooning are extracted respectively, which helps to understand each aspect of the lesion more carefully and provide richer information for diagnosis; Accurate determination of liver fibrosis area: Using the above-mentioned extracted features, the liver fibrosis area determination unit can integrate this information to determine the specific area of ​​liver fibrosis, which is crucial for assessing the severity of the disease and formulating a treatment plan; Combination of predicted and actual data: By combining the predicted second fatty degeneration feature with the actually extracted first fatty degeneration feature, more accurate comparison and verification can be performed to improve the reliability of diagnosis; Potential research value: This method can also be used as a research tool to help scientists better understand the pathological mechanism and development process of patients with non-alcoholic fatty liver disease; Improve the quality of patient care: By identifying and treating patients with non-alcoholic fatty liver disease earlier and more accurately, the quality of life of patients can be improved and complications associated with the disease may be reduced.

[0047] Further, the lobular inflammation feature extraction module is used to extract the lobular inflammation features based on the data related to the inflammatory cell clusters in the ultrasound imaging data and the data related to the inflammatory cell clusters in the radiological imaging data;

[0048] A fibrosis feature extraction module is used to extract fibrosis features based on fibrosis-related data in liver tissue pathology data;

[0049] The hepatocyte ballooning degeneration feature extraction module is used to extract the hepatocyte ballooning degeneration features based on the ballooning degeneration related data in the liver tissue pathology data.

[0050] Furthermore, the fatty degeneration feature extraction module is used to extract a first fatty degeneration feature based on the fatty degeneration related data in the ultrasound imaging data and the fatty degeneration related data in the liver tissue pathology data.

[0051] Based on the embodiments provided in the present application, the following beneficial effects are achieved: improving diagnostic accuracy: the fatty degeneration feature extraction module combines ultrasound and pathology data to more accurately identify and quantify fat accumulation in the liver; the lobular inflammation feature extraction module combines ultrasound and radiographic data to help identify the location and extent of inflammation; in-depth analysis of pathological features: the fibrosis feature extraction module and the hepatocyte ballooning feature extraction module focus on extracting key features from pathology data, which helps to assess disease progression and potential liver structural changes.

[0052] Furthermore, the data related to steatosis include the number, density, size, and spatial distribution data of macrovesicular steatosis and the number, density, size, and spatial distribution data of microvesicular steatosis; Figure 2 As shown, according to the data related to fatty degeneration in the ultrasound imaging data and the data related to fatty degeneration in the liver tissue pathology data, the first fatty degeneration feature is extracted, including:

[0053] S201, constructing a global feature extraction network, a first local feature extraction network, and a second local feature extraction network; wherein the first local feature extraction network and the second local feature extraction network both adopt ResNet networks;

[0054] S202, inputting the data of the number, density, size, and spatial distribution of macrovesicular steatosis into a first local feature extraction network to obtain features of macrovesicular steatosis;

[0055] S203, inputting the quantity, density, size, and spatial distribution data of microvesicular steatosis into a second local feature extraction network to obtain microvesicular steatosis features;

[0056] S204, performing a weighted operation on the macrovesicular steatosis feature and the microvesicular steatosis feature, and obtaining a steatosis fusion feature by a bilinear interpolation method;

[0057] S205, inputting the steatosis fusion feature into the global feature network to obtain a first steatosis feature.

[0058] Furthermore, the steatosis prediction unit is used to:

[0059] The first fatty degeneration feature is input into the embedding layer of the recurrent neural network to perform information dimension upgrading, thereby obtaining a dimension-upgraded feature;

[0060] Inputting the dimension-increased features into a first encoding structure to obtain a first sample feature map;

[0061] Inputting the dimension-increased features into a second coding structure to obtain a second sample feature map; wherein the first coding structure is different from the second coding structure;

[0062] Construct a cross attention network; obtain an updated first sample feature map and an updated second sample feature map through a multi-layer perceptron based on the first sample feature map and the second sample feature map;

[0063] Based on the updated first sample feature map and the updated second sample feature map, a second fatty degeneration feature is predicted.

[0064] Based on the embodiments provided in the present application, the following beneficial effects are achieved: information dimension upgrading: the original first fatty degeneration feature is upgraded through the embedding layer of the recurrent neural network, which helps to extract richer feature representations and increase the model's ability to understand data; multi-coding structure analysis: the upgraded feature is processed using two different coding structures, which can extract features from different angles and increase the diversity and comprehensiveness of the features; feature map generation: the generation of the first sample feature map and the second sample feature map provides multi-dimensional input data for the subsequent cross-attention network, which helps to more comprehensively understand the characteristics of fatty degeneration; cross-attention mechanism: the construction of a cross-attention network can enhance the model's understanding of the relationship between different feature maps, and focus on the features that are more important for prediction through the attention mechanism; feature update: the first sample feature map and the second sample feature map are updated through a multi-layer perceptron, which can further optimize the feature representation and improve the relevance and prediction of the features. accuracy; Feature fusion: Prediction based on the updated feature map can achieve feature fusion and complementarity, and improve the model's predictive ability for complex medical phenomena; Improve prediction accuracy: The prediction of the second fatty degeneration feature is based on features that have been processed and optimized multiple times, which helps to improve the accuracy and reliability of the prediction; Model generalization ability: The combination of different encoding structures and cross-attention enhances the model's generalization ability for new data, enabling it to better adapt to different data distributions; Computational efficiency: The combination of recurrent neural networks and multi-layer perceptrons can optimize computational efficiency and reduce the consumption of computing resources while maintaining prediction accuracy; Data-driven insights: This method provides a data-driven way to gain insight into the characteristics of fatty degeneration, which helps medical researchers better understand the pathological mechanisms of patients with non-alcoholic fatty liver disease; Auxiliary diagnosis and treatment planning: Accurate prediction of the second fatty degeneration feature can assist doctors in making more accurate diagnoses and formulating personalized treatment plans.

[0065] Further, based on the updated first sample feature map and the updated second sample feature map, a second fatty degeneration feature is predicted, including:

[0066] According to the updated first sample feature map and the updated second sample feature map, a first weight value of the updated first sample feature map is calculated based on the first neural network. and the second weight value of the updated second sample feature map ;

[0067]

[0068]

[0069] in, is the first feature vector corresponding to the updated first sample feature map; is the second feature vector corresponding to the updated second sample feature map; , , is the trainable parameter of the first neural network; [ ] represents the connection operation;

[0070] According to the first weight value of the updated first sample feature map and the second weight value of the updated second sample feature map , a second fatty degeneration feature was predicted.

[0071] Based on the embodiments provided in the present application, the following beneficial effects are achieved: feature map update: by updating the first sample feature map and the second sample feature map, more accurate information on liver pathological changes can be captured, providing a richer data basis for subsequent analysis; weight value calculation: using the first neural network to calculate the weight values ​​of the first and second sample feature maps, these weight values ​​can reflect the importance and contribution of different features in the feature maps; parameter training: the parameters in the first neural network can be automatically adjusted through the learning process to better adapt to data characteristics and improve prediction accuracy; feature vector integration: integrating the first and second feature vectors through the connection operation, which helps the model to simultaneously consider the feature information from the two samples and enhance the model's ability to capture the relationship between features; predicting the second fatty degeneration feature: based on the calculated weight value, the model can predict the second fatty degeneration feature of the liver. This prediction includes information such as the degree and distribution of fatty degeneration, which is of great significance for diagnosis and treatment; Model adaptability: Since the model parameters are trainable, the model can continuously optimize itself according to new data, improving its adaptability and prediction accuracy to complex pathological changes; Diversity of mathematical functions: The use of cosine and tangent functions may help extract periodic and nonlinear information in feature maps, which is beneficial for understanding and predicting complex biomedical phenomena; Quantification of feature importance: By calculating weight values, the importance of different features in the prediction process can be quantified, which helps further feature selection and model optimization.

[0072] Further, according to the first steatosis feature extracted by the steatosis feature extraction module, the second steatosis feature predicted by the steatosis prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning feature extracted by the hepatocyte ballooning feature extraction module, the liver fibrosis area of ​​the non-alcoholic fatty liver patient is determined, including:

[0073] Preprocessing the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning feature extracted by the hepatocyte ballooning feature extraction module;

[0074] Among them, preprocessing may include but is not limited to normalization, denoising, feature selection and other steps.

[0075] The Momentum Factor Add a second neural network; where

[0076]

[0077] in, is the momentum factor; is the algorithm learning rate; represents the weights of the second neural network;

[0078] Adding momentum to the second neural network , which can accelerate the learning process, reduce shocks during training, and improve the convergence speed and stability of the model. The momentum update formula shows that the combination of momentum factor and learning rate helps to smooth the weight update process while maintaining learning efficiency.

[0079] A second neural network was trained based on historical NAFLD data;

[0080] Based on the trained second neural network, the liver fibrosis area of ​​patients with non-alcoholic fatty liver disease is determined.

[0081] Based on the embodiments provided by the present application, the following beneficial effects are achieved: Multi-feature extraction: Through different feature extraction modules, the pathological changes of the liver can be comprehensively analyzed. This includes: the first fatty degeneration feature: reflects the degree of fat accumulation in the liver; the second fatty degeneration feature: obtained through the prediction unit, which may contain further development information of fatty degeneration; lobular inflammation feature: reflects the presence and degree of liver inflammation; fibrosis feature: directly related to the development of liver fibrosis; hepatocyte ballooning feature: reflects a morphological change of liver cell damage; feature preprocessing: preprocessing the extracted features can improve the quality of the data and the prediction accuracy of the model; training of the second neural network: using historical non-alcoholic fatty liver data to train the second neural network can enable the model to learn the inherent laws and feature representations of the disease, thereby improving the generalization ability of new data; determination of liver fibrosis area: the second neural network obtained through training can analyze new patient data and predict the presence and range of liver fibrosis areas, which is of great significance for clinical diagnosis and formulation of treatment plans.

[0082] Overall, this approach can provide more precise and personalized guidance for the diagnosis and treatment of non-alcoholic fatty liver disease by comprehensively utilizing multiple liver pathological features and combining them with advanced machine learning technology.

[0083] Furthermore, the global feature extraction network includes 3 convolutional layers and 1 maximum pooling layer, the first convolutional block includes 1 residual block, the second convolutional block includes 4 residual blocks, and the third convolutional block includes 6 residual blocks; each residual block contains 2 1×3 convolution kernels and 2 4×4 convolution kernels. Furthermore, the data related to inflammatory cell clusters include the number, density, size and spatial distribution data of inflammatory cell clusters.

[0084] Furthermore, the data related to fibrosis include the fibrosis area, the length, width, density and spatial distribution data of fibrous collagen; the data related to ballooning degeneration include the number, density, size and spatial distribution data of ballooning degeneration hepatocytes; the data acquisition unit is also used to obtain radio frequency drive data of patients with non-alcoholic fatty liver disease, and the radio frequency drive data includes attenuation coefficient and backscattering coefficient.

[0085] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A system for evaluating non-alcoholic fatty liver disease model based on artificial intelligence, characterized in that: It includes a data acquisition unit, a feature extraction unit, a fatty degeneration prediction unit and a liver fibrosis area determination unit; wherein, The data acquisition unit is used to acquire ultrasound imaging data, radiological imaging data and liver tissue pathology data of patients with non-alcoholic fatty liver disease; The feature extraction unit includes a fatty degeneration feature extraction module, a lobular inflammation feature extraction module, a fibrosis feature extraction module and a hepatocyte ballooning feature extraction module; The liver fibrosis region determination unit is used to determine the liver fibrosis region of the non-alcoholic fatty liver disease patient according to the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning degeneration feature extracted by the hepatocyte ballooning degeneration feature extraction module; The fatty degeneration feature extraction module is used to extract the first fatty degeneration feature according to the fatty degeneration related data in the ultrasonic imaging data and the fatty degeneration related data in the liver tissue pathology data; The data related to steatosis include the number, density, size, and spatial distribution data of macrovesicular steatosis and the number, density, size, and spatial distribution data of microvesicular steatosis; extracting the first steatosis feature according to the data related to steatosis in the ultrasonic imaging data and the data related to steatosis in the liver tissue pathology data includes: Constructing a global feature extraction network, a first local feature extraction network, and a second local feature extraction network; wherein the first local feature extraction network and the second local feature extraction network both adopt ResNet networks; Inputting the quantity, density, size, and spatial distribution data of the macrovesicular steatosis into the first local feature extraction network to obtain macrovesicular steatosis features; Inputting the quantity, density, size, and spatial distribution data of the microvesicular steatosis into the second local feature extraction network to obtain microvesicular steatosis features; Performing a weighted operation on the macrovesicular steatosis feature and the microvesicular steatosis feature, and obtaining a steatosis fusion feature by a bilinear interpolation method; Inputting the steatosis fusion feature into a global feature extraction network to obtain the first steatosis feature; The fatty degeneration prediction unit is used for: Inputting the first fatty degeneration feature into the embedding layer of the recurrent neural network to perform information dimension upgrading to obtain a dimension-upgraded feature; Inputting the dimension-increased features into a first coding structure to obtain a first sample feature map; Inputting the dimension-increased feature into a second coding structure to obtain a second sample feature map; wherein the first coding structure is different from the second coding structure; Constructing a cross attention network; obtaining an updated first sample feature map and an updated second sample feature map through a multi-layer perceptron based on the first sample feature map and the second sample feature map; Based on the updated first sample feature map and the updated second sample feature map, predicting the second fatty degeneration feature; The predicting the second fatty degeneration feature based on the updated first sample feature map and the updated second sample feature map includes: According to the updated first sample feature map and the updated second sample feature map, a first weight value of the updated first sample feature map is calculated based on the first neural network. and the second weight value of the updated second sample feature map ; ； in, is the first feature vector corresponding to the updated first sample feature map; is the second feature vector corresponding to the updated second sample feature map; , , is the trainable parameter of the first neural network; [ ] represents the connection operation; According to the first weight value of the updated first sample feature map and the second weight value of the updated second sample feature map , predicting the second fatty degeneration feature; The method of determining the liver fibrosis area of ​​the non-alcoholic fatty liver disease patient according to the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning degeneration feature extracted by the hepatocyte ballooning degeneration feature extraction module comprises: Preprocessing the first fatty degeneration feature extracted by the fatty degeneration feature extraction module, the second fatty degeneration feature predicted by the fatty degeneration prediction unit, the lobular inflammation feature extracted by the lobular inflammation feature extraction module, the fibrosis feature extracted by the fibrosis feature extraction module, and the hepatocyte ballooning feature extracted by the hepatocyte ballooning feature extraction module; The momentum factor a is added to the second neural network; wherein, in, is the momentum factor; is the algorithm learning rate; represents the weights of the second neural network; training the second neural network based on historical non-alcoholic fatty liver disease data; The liver fibrosis area of ​​the non-alcoholic fatty liver disease patient is determined according to the second neural network obtained through training.

2. The system for evaluating non-alcoholic fatty liver disease model based on artificial intelligence according to claim 1, characterized in that: The lobular inflammation feature extraction module is used to extract the lobular inflammation feature according to the data related to the inflammatory cell clusters in the ultrasonic imaging data and the data related to the inflammatory cell clusters in the radiological imaging data; The fibrosis feature extraction module is used to extract the fibrosis feature according to the fibrosis-related data in the liver tissue pathology data; The hepatocyte ballooning degeneration feature extraction module is used to extract the hepatocyte ballooning degeneration feature according to the ballooning degeneration related data in the liver tissue pathology data.

3. The system for evaluating non-alcoholic fatty liver disease model based on artificial intelligence according to claim 1, characterized in that: The global feature extraction network includes 3 convolution layers and 1 maximum pooling layer, the first convolution block includes 1 residual block, the second convolution block includes 4 residual blocks, and the third convolution block includes 6 residual blocks; each residual block contains 2 1×3 convolution kernels and 2 4×4 convolution kernels.

4. The system for evaluating non-alcoholic fatty liver disease model based on artificial intelligence according to claim 2, characterized in that: The data related to the inflammatory cell clusters include the number, density, size and spatial distribution data of the inflammatory cell clusters; The fibrosis-related data include fibrosis area, length, width, density and spatial distribution data of fibrous collagen; The ballooning degeneration-related data include the number, density, size and spatial distribution data of ballooning degeneration hepatocytes.

5. The system for evaluating non-alcoholic fatty liver disease model based on artificial intelligence according to claim 1, characterized in that: The data acquisition unit is further used to acquire radio frequency driving data of the non-alcoholic fatty liver patient, and the radio frequency driving data includes an attenuation coefficient and a backscattering coefficient.

Images