An auxiliary device for pathological diagnosis of liver nodule puncture biopsy based on artificial intelligence

Abstract

The application belongs to the technical field of pathological diagnosis, and discloses an auxiliary device for liver nodule puncture biopsy pathological diagnosis based on artificial intelligence, which comprises a liver nodule case collection module, a diagnosis model construction module, a main control module, a lesion information statistical module, an analysis module, a prediction module, a diagnosis report generation module and a display module. The analysis module compares the logical output value of the liver nodule lesion data to be analyzed with the sensibility curve, calculates and outputs accurate analysis results. Meanwhile, the prediction module uses K nearest neighbor algorithm and Bayesian algorithm to reduce misjudgment caused by insufficient professional experience of liver nodule lesions, and ensures the accuracy and efficiency of disease prediction of liver nodule lesion routine examination. In addition, the HnAIM model obtained by the diagnosis model construction module has high diagnosis accuracy.

Description

Technical Field

[0001] This invention belongs to the field of pathological diagnosis technology, and in particular relates to an auxiliary device for pathological diagnosis of liver nodules by puncture biopsy based on artificial intelligence. Background Technology

[0002] Hepatocellular nodular lesions (HNLs) comprise a heterogeneous group of diseases with diverse histological morphologies and significant clinical variability. Differential diagnosis of these lesions, especially high-grade dysplastic nodules (HGDNs) and well-differentiated hepatocellular carcinoma (WD-HCC), is extremely challenging, not to mention biopsy specimens. Our goal is to develop a deep learning model to address these challenges and improve the histopathological diagnosis of HNLs (WD-HCC, HGDN, LGDN, focal nodular hyperplasia, hepatocellular adenoma) and background liver tissue (nodular cirrhosis, normal liver tissue). However, current liver nodule biopsy pathological diagnosis relies heavily on manual judgment, which is highly subjective. Furthermore, given the limited amount of biopsy tissue and the limited observable content, the accuracy and repeatability of diagnosing liver nodular lesions are low, and the process is time-consuming and labor-intensive.

[0003] In summary, the problems with existing technologies are: the diagnostic criteria for liver nodules are complex and diverse, biopsy specimens are limited, and pathologists are subjective, resulting in low diagnostic repeatability, high rates of missed and misdiagnosed cases, and time and effort wasted. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides an auxiliary device for pathological diagnosis of liver nodules via puncture biopsy based on artificial intelligence.

[0005] This invention is implemented as follows: an artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy includes:

[0006] The system includes a liver nodule case collection module, a diagnostic model construction module, a main control module, a lesion information statistics module, an analysis module, a prediction module, a diagnostic report generation module, and a display module.

[0007] The liver nodule case collection module, connected to the diagnostic model construction module, is used to collect liver nodule case data;

[0008] The diagnostic model building module is connected to the liver nodule case acquisition module and the main control module. It is used to build an artificial intelligence diagnostic model of hepatocellular nodules based on the case through the model building program.

[0009] The main control module is connected to the diagnostic model construction module, lesion information statistics module, analysis module, prediction module, diagnostic report generation module, and display module, and is used to control the normal operation of each module.

[0010] The lesion information statistics module is connected to the main control module and is used to collect liver nodule lesion information through a statistical program.

[0011] The analysis module, connected to the main control module, is used to analyze liver nodule lesion data through an analysis program;

[0012] The prediction module, connected to the main control module, is used to predict liver nodule lesions through a prediction program;

[0013] The diagnostic report generation module, connected to the main control module, is used to generate diagnostic reports for liver nodule lesions.

[0014] The display module, connected to the main control module, is used to display liver nodule cases, lesion information statistics, analysis results, prediction results, and diagnostic reports.

[0015] Furthermore, the analysis method of the analysis module is as follows:

[0016] (1) Collect information on liver nodules and perform statistical classification on the information on liver nodules; use the preset training dataset of liver nodules to train the preset network architecture to obtain the trained model;

[0017] (2) The trained model is tested using a pre-set liver nodule lesion test dataset to obtain a sensitivity curve;

[0018] (3) Import the data of liver nodule lesions to be analyzed into the trained model and output the logical output value of the data of liver nodule lesions to be analyzed; compare the logical output value of the data of liver nodule lesions to be analyzed in the sensitivity curve, calculate and output the analysis result.

[0019] Furthermore, the method for testing the trained model using a preset liver nodule lesion test dataset is as follows:

[0020] Select one data sample from the liver nodule lesion test dataset as the current data sample;

[0021] The logical output value of the current data sample is used as the judgment threshold;

[0022] The judgment threshold is compared sequentially with the logical output value of each other data sample in the liver nodule lesion test dataset to obtain the judgment result of each other data sample in the liver nodule lesion test dataset.

[0023] Based on the judgment result of each other data sample in the liver nodule lesion test dataset and the diagnosis result corresponding to each other data sample, an confusion matrix corresponding to the logical output value of the current data sample is established;

[0024] Based on the confusion matrix corresponding to the logical output value of the current data sample, the sensitivity and specificity corresponding to the logical output value of the current data sample are calculated.

[0025] Return to the step of selecting a data sample from the liver nodule lesion test dataset as the current data sample, until the sensitivity and specificity corresponding to the logical output value of each data sample in the liver nodule lesion test dataset are calculated;

[0026] Based on the logical output value of each data sample in the liver nodule lesion test dataset, and the sensitivity and specificity corresponding to the logical output value of each data sample, a sensitivity curve is plotted.

[0027] Furthermore, the analysis method also includes:

[0028] Before training the preset network architecture using a preset liver nodule lesion training dataset and obtaining the trained model, the following steps are taken: selecting several data samples from the database as data samples in the liver nodule lesion training dataset to form the preset liver nodule lesion training dataset; and selecting the remaining data samples from the database as data samples in the liver nodule lesion test dataset to form the preset liver nodule lesion test dataset.

[0029] Furthermore, the analysis method also includes:

[0030] Each data sample includes unstructured data, structured data, and diagnostic results; the preset network architecture includes a first neural network architecture and a second neural network architecture.

[0031] The first neural network architecture is trained using unstructured data from the liver nodule lesion training dataset to obtain a first feature parameter, which is used to identify the feature information of the unstructured data; the second neural network architecture is trained using the first feature parameter and structured data from the liver nodule lesion training dataset to obtain the trained model.

[0032] The trained model is tested using unstructured and structured data from the liver nodule lesion test dataset to obtain the logical output value of the liver nodule lesion test dataset; a sensitivity curve is plotted based on the logical output value of the liver nodule lesion test dataset and the diagnostic results in the liver nodule lesion test dataset.

[0033] Furthermore, the unstructured data includes: omics data, image data, and signal data; the first neural network architecture includes deep belief networks, convolutional neural networks, and recurrent neural networks.

[0034] Furthermore, the prediction method of the prediction module is as follows:

[0035] 1) Access medical databases; obtain data on the sample population's gender, age, health status, liver function test items, and corresponding biochemical indicators; verify and calibrate the sample population data;

[0036] 2) Determine the baseline reference range and abnormal range of biochemical indicator data for different genders, ages and health conditions based on the medical knowledge base, and establish the first dataset accordingly;

[0037] 3) Obtain solutions for the symptoms corresponding to the first dataset and establish a second dataset that matches them; use the second dataset to train the classification model until its error is below the threshold and tends to stabilize, thus obtaining a well-trained classification model;

[0038] 4) Input the biochemical index data of the test subject into the classification model to obtain abnormal feature values; use the Bayesian algorithm and abnormal feature values ​​to predict the probability that the test subject has liver nodule lesions.

[0039] Furthermore, obtaining solutions for the symptoms corresponding to the first dataset and establishing a second dataset that matches them includes the following steps:

[0040] An abnormal data set was established based on the normal reference ranges of the liver function test items and corresponding biochemical indicator data.

[0041] Obtain the symptoms of liver nodules corresponding to the first dataset;

[0042] The corresponding solution is matched based on the symptoms of liver nodules and the medical knowledge base.

[0043] Furthermore, the trained classification model is trained through the following steps:

[0044] The second dataset is standardized and normalized to obtain the first feature set of the second dataset;

[0045] The first feature set is deduplicated and merged to obtain the second feature set;

[0046] The second feature set is classified using the K-nearest neighbor algorithm until the error is below the threshold and tends to stabilize;

[0047] The step of classifying the second feature set using the K-nearest neighbor algorithm until the error is below a threshold and tends to stabilize includes the following steps:

[0048] The second feature set is divided into a training set, a test set, and a validation set;

[0049] Calculate the distance between each data point in the test set and the data in the training set;

[0050] Randomly select any data point in the test set as the sample label, and select the K points from the training set that are closest to the sample label;

[0051] Determine the frequency of occurrence of the category of the smallest K points, and take the category with the highest frequency as the predicted category of the test data;

[0052] Iterate through all the test set data and repeat the above steps until the classification error is below the threshold and tends to stabilize;

[0053] The distance between each data point in the test set and the data in the training set is calculated using Euclidean distance or Manhattan distance.

[0054] Furthermore, the step of inputting the biochemical index data of the subject into the classification model to obtain abnormal feature values; and using the Bayesian algorithm and abnormal feature values ​​to predict the probability that the subject has liver nodule lesions includes the following steps:

[0055] Abnormal feature values ​​are divided into two categories: excessively high and excessively low.

[0056] Match the types of liver nodules corresponding to abnormal feature values ​​to obtain a set of candidate liver nodules.

[0057] The probability of each candidate liver nodule lesion and the probability of the remaining candidate liver nodules lesions are calculated based on the Bayesian algorithm.

[0058] The candidate liver nodule with the highest probability is used as the predictor of whether the subject has liver nodule lesions.

[0059] The advantages and positive effects of this invention are as follows: This invention trains a pre-defined network architecture using a pre-defined training dataset of liver nodules to obtain a trained model; it then tests the trained model using a pre-defined test dataset of liver nodules to obtain a sensitivity curve; the data to be analyzed from liver nodules is imported into the trained model, and the logical output value of the data is output; the logical output value of the data to be analyzed is compared with the sensitivity curve to calculate and output accurate analysis results; simultaneously, the prediction module utilizes the K-nearest neighbor algorithm and Bayesian algorithm to fully explore the correlation between liver nodule lesion examination items, symptoms, and possible solutions in the medical knowledge base; this reduces misjudgments of liver nodules due to insufficient professional experience, saves time in routine examinations, and improves the efficiency of medical institutions. Triage efficiency is improved, serving as an auxiliary means for screening liver nodules and ensuring the accuracy and efficiency of routine examinations for predicting liver nodules. Furthermore, the HnAIM model obtained from the diagnostic model construction module provided in this application has a high diagnostic accuracy (external validation AUC value reaches 93.5%), effectively assisting in pathological diagnosis. For liver nodule biopsy tissue, the model demonstrates good diagnostic efficacy, reaching the diagnostic level of senior pathologists. For clinically challenging cases, this model, in collaboration with physicians, can improve the consistency rate of pathological diagnosis of liver nodules. Through WSI panoramic images and pie charts, the distribution and proportion of NC, LGDN, HGDN, and WDHCC on the WSI can be intuitively observed, highlighting the multi-stage evolution of hepatocellular carcinoma. Based on the proportion of lesions at different stages, targeted risk assessments can be conducted for patients, providing effective disease prevention and management. Attached Figure Description

[0060] Figure 1 This is a structural block diagram of an auxiliary device for pathological diagnosis of liver nodules based on artificial intelligence, provided in an embodiment of the present invention.

[0061] Figure 2 This is a flowchart of the analysis module analysis method provided in the embodiment of the present invention.

[0062] Figure 3 This is a flowchart of the prediction method of the prediction module provided in the embodiment of the present invention.

[0063] Figure 1 The module consists of: 1. Liver nodule case collection module; 2. Diagnostic model construction module; 3. Main control module; 4. Lesion information statistics module; 5. Analysis module; 6. Prediction module; 7. Diagnostic report generation module; and 8. Display module. Detailed Implementation

[0064] To further understand the invention's content, features, and effects, the following embodiments are provided, and detailed descriptions are given in conjunction with the accompanying drawings.

[0065] The structure of the present invention will now be described in detail with reference to the accompanying drawings.

[0066] like Figure 1 As shown in the embodiment of the present invention, the auxiliary device for pathological diagnosis of liver nodules based on artificial intelligence includes: a liver nodule case acquisition module 1, a diagnostic model construction module 2, a main control module 3, a lesion information statistics module 4, an analysis module 5, a prediction module 6, a diagnostic report generation module 7, and a display module 8.

[0067] The liver nodule case acquisition module 1 is connected to the diagnostic model construction module 2 and is used to collect data from 5 liver nodule cases.

[0068] The diagnostic model construction module 2 is connected to the liver nodule case acquisition module 1 and the main control module 3. It is used to construct an artificial intelligence diagnostic model for hepatocellular nodules based on the case through the model construction program.

[0069] The main control module 3 is connected to the diagnostic model construction module 2, lesion information statistics module 4, analysis module 5, prediction module 6, diagnostic report generation module 7, and display module 8, and is used to control the normal operation of each module.

[0070] The lesion information statistics module 4 is connected to the main control module 3 and is used to collect liver nodule lesion information through a statistical program.

[0071] Analysis module 5, connected to main control module 3, is used to analyze liver nodule lesion data through an analysis program;

[0072] Prediction module 6, connected to main control module 3, is used to predict liver nodule lesions through a prediction program;

[0073] The diagnostic report generation module 7, connected to the main control module 3, is used to generate diagnostic reports for liver nodule lesions.

[0074] Display module 8, connected to main control module 3, is used to display liver nodule cases, lesion information statistics, analysis results, prediction results, and diagnostic reports.

[0075] like Figure 2 As shown, the analysis method of analysis module 5 provided by this invention is as follows:

[0076] S101: Collect information on liver nodules and perform statistical classification on the information on liver nodules; train the preset network architecture using the preset liver nodule training dataset to obtain the trained model.

[0077] S102, The trained model is tested using a preset liver nodule lesion test dataset to obtain a sensitivity curve;

[0078] S103, import the data of liver nodule lesions to be analyzed into the trained model, and output the logical output value of the data of liver nodule lesions to be analyzed; compare the logical output value of the data of liver nodule lesions to be analyzed in the sensitivity curve, calculate and output the analysis result.

[0079] The present invention provides a method for testing the trained model using a preset liver nodule lesion test dataset:

[0080] Select one data sample from the liver nodule lesion test dataset as the current data sample;

[0081] The logical output value of the current data sample is used as the judgment threshold;

[0082] The judgment threshold is compared sequentially with the logical output value of each other data sample in the liver nodule lesion test dataset to obtain the judgment result of each other data sample in the liver nodule lesion test dataset.

[0083] Based on the judgment result of each other data sample in the liver nodule lesion test dataset and the diagnosis result corresponding to each other data sample, an confusion matrix corresponding to the logical output value of the current data sample is established;

[0084] Based on the confusion matrix corresponding to the logical output value of the current data sample, the sensitivity and specificity corresponding to the logical output value of the current data sample are calculated.

[0085] Return to the step of selecting a data sample from the liver nodule lesion test dataset as the current data sample, until the sensitivity and specificity corresponding to the logical output value of each data sample in the liver nodule lesion test dataset are calculated;

[0086] Based on the logical output value of each data sample in the liver nodule lesion test dataset, and the sensitivity and specificity corresponding to the logical output value of each data sample, a sensitivity curve is plotted.

[0087] The analytical method provided by this invention also includes:

[0088] Before training the preset network architecture using a preset liver nodule lesion training dataset and obtaining the trained model, the following steps are taken: selecting several data samples from the database as data samples in the liver nodule lesion training dataset to form the preset liver nodule lesion training dataset; and selecting the remaining data samples from the database as data samples in the liver nodule lesion test dataset to form the preset liver nodule lesion test dataset.

[0089] The analytical method provided by this invention also includes:

[0090] Each data sample includes unstructured data, structured data, and diagnostic results; the preset network architecture includes a first neural network architecture and a second neural network architecture.

[0091] The first neural network architecture is trained using unstructured data from the liver nodule lesion training dataset to obtain a first feature parameter, which is used to identify the feature information of the unstructured data; the second neural network architecture is trained using the first feature parameter and structured data from the liver nodule lesion training dataset to obtain the trained model.

[0092] The trained model is tested using unstructured and structured data from the liver nodule lesion test dataset to obtain the logical output value of the liver nodule lesion test dataset; a sensitivity curve is plotted based on the logical output value of the liver nodule lesion test dataset and the diagnostic results in the liver nodule lesion test dataset.

[0093] The unstructured data provided by this invention includes: omics data, image data, and signal data; the first neural network architecture includes a deep belief network, a convolutional neural network, and a recurrent neural network.

[0094] like Figure 3 As shown, the prediction method of the prediction module 6 provided by the present invention is as follows:

[0095] S201, retrieve medical database; obtain data on the sample population's gender, age, health status, liver function test items, and corresponding biochemical indicators; verify and calibrate the sample population data;

[0096] S202, Based on the medical knowledge base, determine the benchmark reference range and abnormal range of biochemical indicator data under different genders, ages and health conditions, and establish the first dataset based on them;

[0097] S203, obtain the solutions for the symptoms corresponding to the first dataset, and establish a second dataset that matches them; use the second dataset to train the classification model until its error is below the threshold and tends to stabilize, and obtain the trained classification model.

[0098] S204, input the biochemical index data of the subject into the classification model to obtain abnormal feature values; use the Bayesian algorithm and abnormal feature values ​​to predict the probability that the subject has liver nodule lesions.

[0099] The solution provided by this invention for obtaining the symptoms corresponding to the first dataset and establishing a second dataset that matches them includes the following steps:

[0100] An abnormal data set was established based on the normal reference ranges of the liver function test items and corresponding biochemical indicator data.

[0101] Obtain the symptoms of liver nodules corresponding to the first dataset;

[0102] The corresponding solution is matched based on the symptoms of liver nodules and the medical knowledge base.

[0103] The pre-trained classification model provided by this invention is trained through the following steps:

[0104] The second dataset is standardized and normalized to obtain the first feature set of the second dataset;

[0105] The first feature set is deduplicated and merged to obtain the second feature set;

[0106] The second feature set is classified using the K-nearest neighbor algorithm until the error is below the threshold and tends to stabilize;

[0107] The step of classifying the second feature set using the K-nearest neighbor algorithm until the error is below a threshold and tends to stabilize includes the following steps:

[0108] The second feature set is divided into a training set, a test set, and a validation set;

[0109] Calculate the distance between each data point in the test set and the data in the training set;

[0110] Randomly select any data point in the test set as the sample label, and select the K points from the training set that are closest to the sample label;

[0111] Determine the frequency of occurrence of the category of the smallest K points, and take the category with the highest frequency as the predicted category of the test data;

[0112] Iterate through all the test set data and repeat the above steps until the classification error is below the threshold and tends to stabilize;

[0113] The distance between each data point in the test set and the data in the training set is calculated using Euclidean distance or Manhattan distance.

[0114] The present invention provides a method for inputting biochemical index data of a subject into a classification model to obtain abnormal feature values; and for predicting the probability of a subject having liver nodule lesions using a Bayesian algorithm and abnormal feature values, including the following steps:

[0115] Abnormal feature values ​​are divided into two categories: excessively high and excessively low.

[0116] Match the types of liver nodules corresponding to abnormal feature values ​​to obtain a set of candidate liver nodules.

[0117] The probability of each candidate liver nodule lesion and the probability of the remaining candidate liver nodules lesions are calculated based on the Bayesian algorithm.

[0118] The candidate liver nodule with the highest probability is used as the predictor of whether the subject has liver nodule lesions.

[0119] In operation, this invention first collects liver nodule case data through the liver nodule case collection module 1; then, through the diagnostic model construction module 2, it constructs an artificial intelligence diagnostic model for hepatocellular nodules based on the cases using a model construction program; next, the main control module 3 uses the lesion information statistics module 4 to collect liver nodule lesion information using a statistical program; the analysis module 5 uses the analysis program to analyze the liver nodule lesion data; and the prediction module 6 uses the prediction program to predict liver nodule lesions. Then, the diagnostic report generation module 7 generates a liver nodule lesion diagnostic report; finally, the display module 8 displays the liver nodule cases, lesion information statistics results, analysis results, prediction results, and diagnostic report.

[0120] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall fall within the scope of the technical solution of the present invention.

Claims

1. An auxiliary device for pathological diagnosis of liver nodules via puncture biopsy based on artificial intelligence, characterized in that, The artificial intelligence-based auxiliary equipment for pathological diagnosis of liver nodules via puncture biopsy includes: The system includes a liver nodule case collection module, a diagnostic model construction module, a main control module, a lesion information statistics module, an analysis module, a prediction module, a diagnostic report generation module, and a display module. The liver nodule case collection module, connected to the diagnostic model construction module, was used to collect 462 cases of surgically resected liver tissue specimens from five types of liver nodule lesions: well-differentiated hepatocellular carcinoma (WDHCC), high-grade dysplastic nodule (HGDN), low-grade dysplastic nodule (LGDN), focal nodular hyperplasia (FNH), and hepatocellular adenoma (HCA), as well as two background liver tissue specimens: cirrhosis (NC) and relatively normal liver tissue (NNL). The diagnostic model building module is connected to the liver nodule case acquisition module and the main control module. It is used to build an artificial intelligence diagnostic model of hepatocellular nodules based on the case through the model building program. The main control module is connected to the diagnostic model construction module, lesion information statistics module, analysis module, prediction module, diagnostic report generation module, and display module, and is used to control the normal operation of each module. The lesion information statistics module is connected to the main control module and is used to collect liver nodule lesion information through a statistical program. The analysis module, connected to the main control module, is used to analyze liver nodule lesion data through an analysis program; The prediction module, connected to the main control module, is used to predict liver nodule lesions through a prediction program; The diagnostic report generation module, connected to the main control module, is used to generate diagnostic reports for liver nodule lesions. The display module, connected to the main control module, is used to display liver nodule cases, lesion information statistics, analysis results, prediction results, and diagnostic reports; The analysis module uses the following analysis method: (1) Collect information on liver nodules and perform statistical classification on the information on liver nodules; use the preset training dataset of liver nodules to train the preset network architecture and obtain the trained model. (2) The trained model is tested using a pre-set liver nodule lesion test dataset to obtain a sensitivity curve; (3) Import the data of liver nodule lesions to be analyzed into the trained model and output the logical output value of the data of liver nodule lesions to be analyzed; compare the logical output value of the data of liver nodule lesions to be analyzed in the sensitivity curve, calculate and output the analysis result.

2. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 1, characterized in that, The method for testing the trained model using a pre-set liver nodule lesion test dataset is as follows: Select one data sample from the liver nodule lesion test dataset as the current data sample; The logical output value of the current data sample is used as the judgment threshold; The judgment threshold is compared sequentially with the logical output value of each other data sample in the liver nodule lesion test dataset to obtain the judgment result of each other data sample in the liver nodule lesion test dataset. Based on the judgment result of each other data sample in the liver nodule lesion test dataset and the diagnosis result corresponding to each other data sample, an confusion matrix corresponding to the logical output value of the current data sample is established; Based on the confusion matrix corresponding to the logical output value of the current data sample, the sensitivity and specificity corresponding to the logical output value of the current data sample are calculated. Return to the step of selecting a data sample from the liver nodule lesion test dataset as the current data sample, until the sensitivity and specificity corresponding to the logical output value of each data sample in the liver nodule lesion test dataset are calculated; Based on the logical output value of each data sample in the liver nodule lesion test dataset, and the sensitivity and specificity corresponding to the logical output value of each data sample, a sensitivity curve is plotted.

3. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 1, characterized in that, The analytical method further includes: Before training the preset network architecture using a preset liver nodule lesion training dataset and obtaining the trained model, the following steps are taken: selecting several data samples from the database as data samples in the liver nodule lesion training dataset to form the preset liver nodule lesion training dataset; and selecting the remaining data samples from the database as data samples in the liver nodule lesion test dataset to form the preset liver nodule lesion test dataset.

4. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 1, characterized in that, The analytical method further includes: Each data sample includes unstructured data, structured data, and diagnostic results; the preset network architecture includes a first neural network architecture and a second neural network architecture. The first neural network architecture is trained using unstructured data from the liver nodule lesion training dataset to obtain a first feature parameter, which is used to identify the feature information of the unstructured data; the second neural network architecture is trained using the first feature parameter and structured data from the liver nodule lesion training dataset to obtain the trained model. The trained model is tested using unstructured and structured data from the liver nodule lesion test dataset to obtain the logical output value of the liver nodule lesion test dataset; a sensitivity curve is plotted based on the logical output value of the liver nodule lesion test dataset and the diagnostic results in the liver nodule lesion test dataset.

5. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 4, characterized in that, The unstructured data includes: omics data, image data, and signal data; the first neural network architecture includes deep belief networks, convolutional neural networks, and recurrent neural networks.

6. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 1, characterized in that, The prediction module uses the following prediction method: 1) Access medical databases; obtain data on the sample population's gender, age, health status, liver function test items, and corresponding biochemical indicators; verify and calibrate the sample population data; 2) Determine the baseline reference range and abnormal range of biochemical indicator data for different genders, ages and health conditions based on the medical knowledge base, and establish the first dataset accordingly; 3) Obtain solutions for the symptoms corresponding to the first dataset and establish a second dataset that matches them; use the second dataset to train the classification model until its error is below the threshold and tends to stabilize, thus obtaining a well-trained classification model; 4) Input the biochemical index data of the test subject into the classification model to obtain abnormal feature values; use the Bayesian algorithm and abnormal feature values ​​to predict the probability that the test subject has liver nodule lesions.

7. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 6, characterized in that, The steps of obtaining solutions for the symptoms corresponding to the first dataset and establishing a second dataset that matches them include the following: An abnormal data set was established based on the normal reference ranges of the liver function test items and corresponding biochemical indicator data. Obtain the symptoms of liver nodules corresponding to the first dataset; The corresponding solution is matched based on the symptoms of liver nodules and the medical knowledge base.

8. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 6, characterized in that, The trained classification model is trained through the following steps: The second dataset is standardized and normalized to obtain the first feature set of the second dataset; The first feature set is deduplicated and merged to obtain the second feature set; The second feature set is classified using the K-nearest neighbor algorithm until the error is below the threshold and tends to stabilize; The step of classifying the second feature set using the K-nearest neighbor algorithm until the error is below a threshold and tends to stabilize includes the following steps: The second feature set is divided into a training set, a test set, and a validation set; Calculate the distance between each data point in the test set and the data in the training set; Randomly select any data point in the test set as the sample label, and select the K points from the training set that are closest to the sample label; Determine the frequency of occurrence of the category of the smallest K points, and take the category with the highest frequency as the predicted category of the test data; Iterate through all the test set data and repeat the above steps until the classification error is below the threshold and tends to stabilize; The distance between each data point in the test set and the data in the training set is calculated using Euclidean distance or Manhattan distance.

9. The artificial intelligence-based auxiliary device for pathological diagnosis of liver nodules via puncture biopsy as described in claim 6, characterized in that, The process of inputting the biochemical index data of the subject into the classification model to obtain abnormal feature values, and using the Bayesian algorithm and abnormal feature values ​​to predict the probability that the subject has liver nodule lesions, includes the following steps: Abnormal feature values ​​are divided into two categories: excessively high and excessively low. Match the types of liver nodules corresponding to abnormal feature values ​​to obtain a set of candidate liver nodules. The probability of each candidate liver nodule lesion and the probability of the remaining candidate liver nodules lesions are calculated based on the Bayesian algorithm. The candidate liver nodule with the highest probability is used as the predictor of whether the subject has liver nodule lesions.

Images