System for predicting weight and metabolism remission condition after weight loss operation

Abstract

The invention discloses a weight loss postoperative weight and metabolism relief condition prediction system in the technical field of radiomics, and the system comprises a data collection module which is used for collecting an abdominal CT image and clinical parameters of a patient from a hospital information system; the preprocessing module is used for carrying out image omics preprocessing on the CT image of the patient and extracting abdomen CT image features; the data analysis module is used for extracting key features from the image features by using a random forest algorithm; and the prediction module is used for synchronously predicting the body mass and triglyceride glucose index change after the weight reduction operation by using the trained convolutional neural network based on the key features and the clinical parameters. According to the method, key CT image features are extracted through a random forest algorithm, and the deep relevance between fat distribution and metabolism parameters is extracted by using a two-channel convolutional neural network, so that more accurate, comprehensive and real-time postoperative BMI and TyG index change results are obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of radiomics, and specifically relates to a prediction system for the weight and metabolic remission after bariatric surgery. Background Art

[0002] Bariatric surgery, also known as metabolic surgery, was initially mainly used to treat obesity. However, with the in-depth research, it has been found that such surgeries can not only effectively reduce weight, but also significantly improve obesity-related metabolic diseases, such as type 2 diabetes, hypertension, hyperlipidemia, etc. Currently, bariatric surgery has become one of the important means for treating obesity combined with metabolic diseases. Before the surgery, doctors need to predict the weight and metabolic conditions of patients after bariatric surgery. Similarly, after the patients undergo bariatric surgery, long-term assessment of weight and metabolic remission is also required.

[0003] Although existing bariatric prediction models can predict the improvement of single metabolic conditions (such as hypertension, type 2 diabetes), their pain points are becoming increasingly prominent: 1. Single indicators cannot fully reflect the overall metabolic conditions of patients. At the same time, the specific metabolic characteristics of adipose tissue may be the key influencing factors for the effect of bariatric surgery, and the existing prediction models have not fully integrated relevant information; 2. Existing models have difficulties in processing multi-dimensional index data. Conventional statistical methods cannot effectively process multi-dimensional data, and lack the ability of multi-output joint prediction.

[0004] Therefore, it is necessary to propose a prediction system for the weight and metabolic remission after bariatric surgery that can collect various clinical parameters and CT images of patients, comprehensively predict the overall metabolic conditions of patients and key adipose tissue-specific metabolic characteristics, and extract the deep correlation between adipose tissue distribution and metabolic parameters through multi-scale convolutional kernels to predict the changes in body mass index (BMI) and triglyceride glucose index (TyG) of patients after bariatric surgery. Summary of the Invention

[0005] In order to solve the above problems, the purpose of the present invention is to provide a prediction system for the weight and metabolic remission after bariatric surgery, which extracts key CT image features through the random forest algorithm, and uses a dual-channel convolutional neural network to extract the deep correlation between adipose tissue distribution and metabolic parameters, so as to obtain more accurate, comprehensive and more real-time results of postoperative BMI and TyG index changes.

[0006] In order to achieve the above purpose, the technical solution of the present invention is as follows: A prediction system for the weight and metabolic remission after bariatric surgery, including a data acquisition module, a preprocessing module, a data analysis module and a prediction module,

[0007] The data acquisition module is used to collect abdominal CT images and clinical parameters of patients from the hospital information system;

[0008] A preprocessing module for performing radiomics preprocessing on the patient's CT images and extracting abdominal CT image features;

[0009] A data analysis module for extracting key features from the image features using the random forest algorithm;

[0010] A prediction module for synchronously predicting the changes in body mass and triglyceride glucose index after weight loss surgery based on the key features and clinical parameters using a trained convolutional neural network.

[0011] The principle of the basic scheme is as follows: The abdominal CT images and clinical parameters of the patient are collected through the hospital information system. The abdominal CT images provide detailed fat distribution information, while the clinical parameters (such as preoperative BMI, blood glucose, blood lipids, etc.) reflect the overall metabolic status of the patient. Radiomics preprocessing is performed on the collected CT images, including steps such as denoising, enhancement, and segmentation, to extract the features of the abdominal CT images. These features may include fat area, fat density, fat distribution pattern, etc. The random forest algorithm is used to extract key features from the preprocessed image features. A trained convolutional neural network (CNN) is used to synchronously predict the changes in body mass (BMI) and triglyceride glucose index (TyG) after weight loss surgery based on the extracted key features and clinical parameters. The dual-channel convolutional neural network can capture the deep correlation between fat distribution and metabolic parameters, thereby improving the accuracy of prediction.

[0012] The beneficial effects of the basic scheme are as follows: 1. By combining the random forest algorithm and the convolutional neural network, the present invention can more accurately extract and analyze CT image features and clinical parameters, thereby improving the prediction accuracy of the changes in BMI and TyG index after weight loss surgery.

[0013] 2. The present invention not only considers the CT image features of the patient, but also combines clinical parameters, making the prediction results more comprehensive and capable of better reflecting the overall metabolic status of the patient.

[0014] 3. Due to the use of efficient machine learning and deep learning algorithms, the present invention can complete the prediction of the patient's long-term metabolic changes in a short time, providing sufficient decision-making support for clinicians.

[0015] 4. The present invention can perform personalized prediction according to the specific situation of the patient (such as preoperative BMI, etc.), providing a tailored treatment plan for each patient. By predicting the weight and metabolic changes after weight loss surgery, the present invention can assist clinicians in formulating more reasonable surgical plans and postoperative management plans, improving the success rate of surgery and patient satisfaction.

[0016] Furthermore, the clinical parameters collected by the data acquisition module include blood routine parameters, blood glucose parameters, liver and kidney function parameters, B vitamin level parameters, serum insulin parameters, serum C peptide parameters, glycated hemoglobin parameters, blood lipid parameters, serum dehydroepiandrosterone parameters, sex hormone-binding globulin parameters, sex hormone parameters, thyroid function test parameters, parathyroid hormone test parameters, and trace element level parameters.

[0017] The beneficial effects of the basic scheme are as follows: 1. The clinical parameters collected by the data acquisition module are extremely comprehensive. These parameters cover multiple aspects of the patient's metabolic state, providing rich input information for the prediction system and making the prediction results more comprehensive and accurate.

[0018] 2. The comprehensive clinical parameters combined with the CT image features provide a richer data basis for the random forest algorithm and the convolutional neural network. This enables the algorithm to more accurately capture and analyze the key factors affecting weight and metabolic changes after weight loss surgery, thereby improving the accuracy of prediction.

[0019] 3. By considering the individual differences of patients, such as gender, age, basal metabolic rate, and differences in various clinical parameters, the present invention can provide personalized prediction results for each patient. This helps clinicians formulate surgical plans and postoperative management plans that are more in line with the actual situation of the patients.

[0020] 4. By predicting the weight and metabolic changes after weight loss surgery, clinicians can detect potential problems early, such as metabolic disorders, malnutrition, etc., and take corresponding intervention measures. This helps prevent the occurrence of postoperative complications and promotes the recovery of patients.

[0021] 5. The prediction system proposed by the present invention not only has clinical application value, but also provides new ideas and methods for the research of weight loss surgery. By collecting and analyzing the clinical data and CT image features of a large number of patients, the mechanism of weight loss surgery can be further explored, providing data support for future research.

[0022] Furthermore, the preprocessing module includes an imaging genomics preprocessing unit and an image feature extraction unit.

[0023] The imaging genomics preprocessing unit is used to preprocess the abdominal CT images of patients.

[0024] The image feature extraction unit is used to extract image features related to clinical parameters from the preprocessed abdominal CT images.

[0025] The beneficial effects of the basic solution are as follows: 1. The radiomics preprocessing unit in the preprocessing module is specifically responsible for professionally preprocessing the abdominal CT images of patients. This includes a series of operations such as denoising, enhancement, registration, and segmentation, aiming to improve the image quality and provide a clear and accurate image basis for subsequent feature extraction.

[0026] 2. The image feature extraction unit is responsible for extracting image features related to clinical parameters from the preprocessed abdominal CT images. These features may include the distribution, density, volume of adipose tissue, and the relative positional relationship with other tissue structures. By accurately extracting these features, rich and targeted information is provided for subsequent data analysis.

[0027] 3. Combining the results of radiomics preprocessing and feature extraction, as well as comprehensive clinical parameters, the random forest algorithm and convolutional neural network can more accurately capture and analyze the key factors affecting weight and metabolic changes after weight loss surgery. This further improves the accuracy of the prediction system and makes the prediction results more reliable.

[0028] Furthermore, the preprocessing of abdominal CT images includes image resampling, image filtering, image segmentation, image artifact removal, gray level quantization, and data augmentation.

[0029] The beneficial effects of the basic solution are as follows: 1. The high-quality image preprocessing steps provide an accurate and reliable image basis for subsequent image feature extraction. Combining comprehensive clinical parameters, through the random forest algorithm and convolutional neural network, the key factors affecting weight and metabolic changes after weight loss surgery can be more accurately captured and analyzed, improving the accuracy of prediction.

[0030] 2. Operations such as artifact removal and data augmentation in the image preprocessing steps enhance the system's resistance to interference factors such as noise and artifacts, improving the robustness of the system. This enables the system to maintain stable prediction performance when facing CT images from different sources and of different qualities.

[0031] Furthermore, the image features include mean value, standard deviation, skewness, kurtosis, gray level uniformity, clustering tendency, S-shaped curve fitting features, and the maximum two-dimensional plane diameter and three-dimensional volume of the adipose region.

[0032] The beneficial effects of the basic solution are as follows: 1. The extracted image features comprehensively cover multiple aspects of CT images, providing rich information for subsequent data analysis. Combining the extracted image features with clinical parameters provides more rich input information for the random forest algorithm and convolutional neural network. These features can more comprehensively reflect the patient's metabolic status and surgical effect, thus improving the accuracy of the prediction model.

[0033] 2. Since the extracted image features cover multiple aspects and are of higher quality after the preprocessing steps, the prediction model can exhibit stronger generalization ability when facing different patients and different surgical situations. The accurate and comprehensive prediction results provided by the prediction system can assist clinicians in more accurately evaluating the surgical risks of patients before surgery and more timely monitoring the recovery of patients after surgery. This helps improve the scientificity and accuracy of clinical decision-making.

[0034] Furthermore, the data analysis module includes a decision tree construction unit, a classification tree construction unit, and a key feature extraction unit.

[0035] The decision tree construction unit is used to generate a decision tree based on the extracted image features and clinical parameters, with each image feature and clinical parameter serving as a feature node.

[0036] The classification tree construction unit is used to classify the image features based on the feature nodes, and the number of constructed classification trees is equal to the number of feature nodes.

[0037] The key feature extraction unit is used to evaluate the best classification feature node based on the generated classification tree according to the Gini index, which is the key feature.

[0038] The beneficial effects of the basic solution are as follows: 1. The decision tree construction unit takes each image feature and clinical parameter as a feature node, and comprehensively integrates the multi-dimensional information of the patient by constructing a decision tree. This method not only considers the impact of a single feature on the prediction result, but also considers the interaction and correlation between features, improving the accuracy of prediction.

[0039] 2. The decision tree construction unit can flexibly adjust the selection and division rules of feature nodes according to different requirements and scenarios. This flexibility enables the system to adapt to different data sets and prediction tasks, improving the versatility and applicability of the system.

[0040] 3. The classification tree construction unit classifies the image features based on the feature nodes. By constructing multiple classification trees, a detailed division of the image features is achieved. This method helps to discover potential patterns and rules in the image features, providing strong support for subsequent key feature extraction.

[0041] 4. By constructing multiple classification trees and evaluating and selecting each classification tree, the system can find the best feature classification method, thereby improving the prediction accuracy. At the same time, the number of classification trees is equal to the number of feature nodes, ensuring the comprehensiveness and accuracy of classification.

[0042] 5. The key feature extraction unit evaluates the optimal classification feature nodes based on the Gini index. By quantifying the importance of feature nodes in the classification tree, it accurately identifies the key features that have the greatest impact on the prediction results. This approach not only helps simplify the prediction model and improve computational efficiency but also reveals the key factors affecting the weight and metabolic remission after weight loss surgery.

[0043] Furthermore, the convolutional neural network is a dual-channel convolutional neural network with two parallel convolutional layers that process data simultaneously.

[0044] The beneficial effects of the basic solution are as follows: 1. The two parallel convolutional layers of the dual-channel CNN can process image features and clinical parameter data simultaneously, greatly improving the speed and efficiency of data processing. This parallel processing ability enables the system to process a large number of CT images and clinical data faster and provide more timely prediction results for doctors.

[0045] 2. The two convolutional layers can extract different information from image features and clinical parameters respectively, and integrate this information through subsequent fusion layers to achieve complementarity and enhancement between features. This helps improve the accuracy of prediction because image features and clinical parameters often contain different information, and their fusion can provide more comprehensive patient information.

[0046] 3. By stacking multiple convolutional layers, the dual-channel CNN can extract multi-level features, including low-level edge and texture features and high-level semantic features. These features can play an important role in subsequent prediction processes and improve the accuracy of prediction.

[0047] 4. Techniques such as pooling layers and Dropout in the convolutional neural network help reduce overfitting and improve the generalization ability of the model. The dual-channel CNN further increases the complexity of the model by introducing two parallel convolutional layers, enabling the model to better adapt to different datasets and prediction tasks. Since the dual-channel CNN can process image features and clinical parameter data simultaneously, it can better adapt to different data distribution situations. Even in the case where there are noises or outliers at some feature nodes, the system can still maintain a high prediction performance.

[0048] 5. The dual-channel CNN realizes a comprehensive evaluation of the patient's condition by fusing the information in image features and clinical parameters. This comprehensive information utilization method helps improve the accuracy of prediction because image features and clinical parameters are often interrelated and jointly affect the patient's weight and metabolic remission. In subsequent prediction processes, the dual-channel CNN can automatically learn the combination methods between features and find the optimal feature combination strategy. This efficient feature combination method helps improve the accuracy of prediction and enables the system to better adapt to different patient situations.

[0049] Furthermore, the convolutional neural network is accelerated in convergence by the Adam optimizer. Adam is an optimization algorithm based on gradient descent, which updates parameters by calculating the first-moment estimate and the second-moment estimate.

[0050] The beneficial effects of the basic solution are as follows: 1. The Adam optimizer can adaptively adjust the learning rate of each parameter according to historical gradient information. In the initial stage of training, the learning rate is relatively large, which helps the model converge quickly. In the later stage of training, the learning rate gradually decreases, which helps the model adjust parameters more precisely, so as to find the global optimal solution. This adaptive learning rate mechanism significantly improves the training efficiency of the CNN and reduces the training time.

[0051] 2. Traditional gradient descent algorithms are prone to oscillation and jitter during training, resulting in an unstable training process. The Adam optimizer can more accurately estimate the update direction and step size of parameters by calculating the first-moment estimate and the second-moment estimate of the gradient, thus reducing oscillation and jitter and making the training process smoother.

[0052] 3. The Adam optimizer combines the momentum mechanism and uses the first-moment estimate of the gradient to accelerate convergence and reduce oscillation. This momentum mechanism enables the model to better cope with the influence of noise and outliers during training and improves the training stability. When calculating the first-moment estimate and the second-moment estimate of the gradient, the Adam optimizer adopts a bias correction technique to eliminate the bias in the initial iteration. This bias correction technique enables the model to obtain a more accurate parameter update direction in the initial stage of training, further improving the training stability.

[0053] 4. The Adam optimizer helps the model avoid falling into local minima by dynamically adjusting the learning rate and combining the momentum mechanism. During training, the model can explore the parameter space more widely, so as to find a better global solution and improve the performance of the model. Since the Adam optimizer can accelerate convergence and improve training stability, the model can better learn the internal laws of the data during training, thus enhancing the generalization ability of the model. This enables the model to still maintain a high prediction accuracy when facing new and unseen data.

[0054] Furthermore, image filtering includes using Gaussian filters and median filters to reduce noise and artifacts in the image.

[0055] The beneficial effects of the basic solution are as follows: 1. The Gaussian filter can effectively reduce Gaussian noise in the image through its smoothing characteristics, making the image clearer. The median filter has a good suppression effect on impulse noise such as salt-and-pepper noise and can remove isolated noise points in the image.

[0056] 2. The quality of the filtered image data is improved, providing more accurate and reliable input data for the subsequent Convolutional Neural Network (CNN). High-quality input data helps the CNN better learn image features, thereby improving the prediction accuracy of the model. At the same time, reducing noise and artifacts can also reduce the risk of model overfitting.

[0057] Furthermore, data augmentation includes rotating, translating, and flipping the images to increase the number of samples for image feature extraction.

[0058] The beneficial effects of the basic solution are as follows: 1. By performing operations such as rotating, translating, and flipping the images, a large number of image samples with different angles, positions, and orientations can be generated. These newly added samples enrich the diversity of the training data, enabling the model to learn more different feature combinations and variation patterns. Abundant training data helps the model learn more generalized feature representations and improve the model's adaptability to unseen data. In the prediction of post-bariatric surgery weight and metabolic remission, this means that the model can more accurately predict the postoperative effects of different patients.

[0059] 2. Enhancing model robustness: Data augmentation techniques can simulate various changes in the actual application of images, such as shooting angles and position offsets. By exposing the model to these changes during training, the model's robustness to these changes can be enhanced, reducing the risk of overfitting. Reducing overfitting means that the performance of the model on the training set and the test set is closer, thereby improving the practicality and reliability of the model. In the prediction of post-bariatric surgery weight and metabolic remission, this helps improve the accuracy and stability of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the prediction system for post-bariatric surgery weight and metabolic remission in an embodiment of the present invention.

[0061] Figure 2 It is a schematic diagram of the preprocessing module in an embodiment of the present invention.

[0062] Figure 3 It is a schematic diagram of the data analysis module in an embodiment of the present invention.

[0063] Figure 4 It is a flowchart of the prediction system for post-bariatric surgery weight and metabolic remission in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0064] The following is a further detailed description through specific embodiments:

[0065] Example 1

[0066] Basically as shown in the appendix Figure 1 , Figure 2 ,Figure 3 and Figure 4 As shown in Figure 4 , a prediction system for post - weight - loss body weight and metabolic remission, characterized in that: it includes a data acquisition module for collecting abdominal CT images and clinical parameters of patients from a hospital information system. The clinical parameters collected by the data acquisition module include blood routine parameters, blood glucose parameters, liver and kidney function parameters, B - vitamin level parameters, serum insulin parameters, serum C - peptide parameters, glycated hemoglobin parameters, blood lipid parameters, serum dehydroepiandrosterone parameters, sex hormone - binding globulin parameters, sex hormone parameters, thyroid function test parameters, parathyroid hormone test parameters, and trace element level parameters.

[0067] A pre - processing module for performing radiomics pre - processing on the CT images of patients and extracting abdominal CT image features. The pre - processing module includes a radiomics pre - processing unit for pre - processing the abdominal CT images of patients. The pre - processing of abdominal CT images includes image resampling, image filtering, image segmentation, image artifact removal, gray - level quantization, and data augmentation. Image filtering includes using a Gaussian filter and a median filter to reduce image noise and artifacts. Data augmentation includes rotating, translating, and flipping the images to increase the number of samples for image feature extraction; an image feature extraction unit for extracting image features related to clinical parameters from the pre - processed abdominal CT images. The image features include mean value, standard deviation, skewness, kurtosis, gray - level uniformity, clustering tendency, S - curve fitting features, and the maximum two - dimensional plane diameter and three - dimensional volume of the fat area.

[0068] A data analysis module for extracting key features from the image features using a random forest algorithm. The data analysis module includes a decision tree construction unit for generating a decision tree based on the extracted image features and clinical parameters, with each image feature and clinical parameter as a feature node; a classification tree construction unit for classifying the image features based on the feature nodes, and the number of constructed classification trees is equal to the number of feature nodes; a key feature extraction unit for evaluating the best classification feature node based on the generated classification trees according to the Gini index, which is the key feature.

[0069] A prediction module for synchronously predicting changes in body mass and triglyceride - glucose index after weight - loss surgery using a trained convolutional neural network based on the key features and clinical parameters.

[0070] The specific implementation process is as follows: The data acquisition module collects the abdominal CT images and clinical parameters of patients from the hospital information system. The collected clinical parameters include, but are not limited to, blood routine parameters, blood glucose parameters, liver and kidney function parameters, B vitamin level parameters, serum insulin parameters, serum C peptide parameters, glycated hemoglobin parameters, blood lipid parameters, serum dehydroepiandrosterone parameters, sex hormone-binding globulin parameters, sex hormone parameters, thyroid function test parameters, parathyroid hormone test parameters, and trace element level parameters. These comprehensive data on the metabolic conditions of patients will provide better data support for subsequent analysis and processing.

[0071] As Figure 4 shown, the radiomics preprocessing unit of the preprocessing module preprocesses the abdominal CT images of patients, including image resampling, image filtering, image segmentation, image artifact removal, gray quantization, and data augmentation. Among them, Gaussian filter and median filter are used in image filtering to reduce the noise and artifacts of the images. The Gaussian filter reduces Gaussian noise through smoothing, while the median filter can effectively remove impulse noise such as salt-and-pepper noise; Data augmentation rotates, translates, and flips the images to increase the number of samples for image feature extraction and improve the generalization ability of the model; Image resampling unifies CT images with different resolutions or under different acquisition conditions to the same resolution and format for subsequent processing and analysis; Image segmentation: Segments different tissues (such as fat, muscle, bone, etc.) in the abdominal CT images for separate extraction and analysis; Image artifact removal removes artifacts caused by equipment or shooting conditions to improve the authenticity of the images; Gray quantization quantizes the gray values of the images for subsequent feature extraction and analysis. The image feature extraction unit extracts image features related to clinical parameters from the preprocessed abdominal CT images. These features include, but are not limited to, mean value, standard deviation, skewness, kurtosis, gray level co-occurrence matrix features, clustering tendency, and S-shaped curve fitting features, etc. In addition, features such as the maximum two-dimensional plane diameter and three-dimensional volume of the fat area are extracted to reflect the fat distribution of the patient. These features can more comprehensively reflect the metabolic status and surgical effect of the patient, thereby improving the accuracy of the prediction model.

[0072] The decision tree construction unit of the data analysis module generates a decision tree based on the extracted image features and clinical parameters. Each image feature and clinical parameter is used as a feature node. The decision tree recursively selects the best feature node for splitting until the stopping condition is met (such as reaching the maximum depth, too few node samples, etc.). The classification tree construction unit classifies the image features based on the feature nodes, constructs multiple classification trees. The number of constructed classification trees is equal to the number of feature nodes. Each classification tree independently classifies the image features to form different evaluation paths. The key feature extraction unit evaluates the best classification feature nodes based on the generated classification trees using the Gini index. The smaller the Gini index, the greater the contribution of the feature node in classification. Select multiple feature nodes with the smallest Gini index as key features for the subsequent prediction module. The random forest algorithm used by the data analysis module is an ensemble learning method that improves the accuracy and stability of prediction by constructing multiple decision trees. It can handle high-dimensional data and is not prone to overfitting. By extracting key features, it can simplify the model complexity and improve the prediction efficiency.

[0073] When performing the Gini index evaluation, the variable importance score is denoted as VIM, and the Gini index is denoted as GI. When extracting n image features, constructing I decision trees, and dividing into C categories, the Gini index calculation formula for the feature node q of the i-th decision tree is,

[0074]

[0075] In the formula, p qc represents the proportion of category c in node q,

[0076] feature X n The importance of the i-th tree node q, that is, the change in the Gini index before and after the branching of node q is,

[0077]

[0078] In the formula, and represent the Gini indices of the two new nodes after branching respectively.

[0079] When there are I trees, the nodes where feature X n appears in the decision tree i are the set Q. Then the importance of X n in the i-th tree is,

[0080]

[0081] Then normalize all the obtained importance scores.

[0082]

[0083] The prediction module uses a convolutional neural network (CNN) for model training. The training data includes key features and clinical parameters, and the training objective is to simultaneously predict the changes in body mass index (BMI) and triglyceride glucose (TyG) index after bariatric surgery. After training, new patient data (including abdominal CT images and clinical parameters) is input into the trained CNN model. The model extracts key features and clinical parameters from the input data, makes predictions, and outputs the prediction results, including the changes in body mass and triglyceride glucose index after bariatric surgery.

[0084] Example 2

[0085] The difference from the above example is as shown in the appended Figure 1 、 Figure 3 and Figure 4 : The convolutional neural network is a two-channel convolutional neural network with two parallel convolutional layers that process data simultaneously. The convolutional neural network is accelerated for convergence by the Adam optimizer. Adam is an optimization algorithm based on gradient descent that updates parameters by calculating first-moment estimates and second-moment estimates.

[0086] The specific implementation process is as follows: During the construction of the two-channel convolutional neural network, two convolutional layers are defined and the outputs of the two channels are fused, and concatenation or other fusion techniques can be used.

[0087] During the optimization of the convolutional neural network, the Adam optimizer smooths the gradient by accumulating the exponentially weighted moving average of the gradient (i.e., momentum) to reduce oscillations. The formula is as follows:

[0088] m t =β1·m t-1 +(1-β1)·g t

[0089] In the formula, m t is the first-moment estimate, β1 is the momentum parameter, and g t is the gradient;

[0090] The Adam optimizer also accumulates the exponentially weighted moving average of the squared gradient for adjusting the learning rate of each parameter. The formula is as follows:

[0091]

[0092] In the formula, v t is the second-moment estimate, and β2 is the decay rate;

[0093] In addition, at the beginning of training, there may be biases in the momentum and second-moment estimates. The Adam optimizer introduces a bias correction mechanism:

[0094]

[0095] The parameter update formula is

[0096]

[0097] where α is the learning rate and ∈ is a small value (such as 10-8) used to avoid division-by-zero errors.

[0098] The Adam optimizer calculates an adaptive learning rate for each parameter, so it can adjust the learning rate more flexibly at different training stages. Moreover, the combination of the momentum and second-moment estimates of the Adam optimizer enables it to converge faster than the standard gradient descent method in most cases, especially performing well in a noisy gradient environment. The dual-channel convolutional neural network after convergence by the Adam optimizer can obtain more accurate prediction results based on the patient's key image features and comprehensive clinical parameters.

[0099] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0100] The above are only embodiments of the present invention. Specific structures and common knowledge such as characteristics well known in the art are not described in detail herein. Those of ordinary skill in the art know all the common general technical knowledge in the technical field to which the invention belongs before the application date or priority date, can know all the prior art in this field, and have the ability to apply the conventional experimental means before this date. Those of ordinary skill in the art can, under the inspiration given in this application, complete and implement this solution in combination with their own abilities. Some typical well-known structures or well-known methods should not become an obstacle for those of ordinary skill in the art to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can still be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be subject to the content of its claims, and the specific implementation manners described in the specification can be used to interpret the content of the claims.

Claims

1. A prediction system for the weight and metabolic remission after bariatric surgery, characterized in that: It includes a data acquisition module, a preprocessing module, a data analysis module, and a prediction module. The data acquisition module is used to acquire the abdominal CT images and clinical parameters of patients from the hospital information system. The preprocessing module is used to perform radiomics preprocessing on the CT images of patients and extract abdominal CT image features. The data analysis module is used to extract key features from the image features using the random forest algorithm. The prediction module is used to synchronously predict the changes in body mass and triglyceride glucose index after weight loss surgery based on the key features and clinical parameters using a trained convolutional neural network.

2. The prediction system for weight and metabolic remission after bariatric surgery according to claim 1, wherein: The clinical parameters collected by the data acquisition module include blood routine parameters, blood glucose parameters, liver and kidney function parameters, B vitamin level parameters, serum insulin parameters, serum C peptide parameters, glycated hemoglobin parameters, blood lipid parameters, serum dehydroepiandrosterone parameters, sex hormone-binding globulin parameters, sex hormone parameters, thyroid function test parameters, parathyroid hormone test parameters, and trace element level parameters.

3. The prediction system for post-bariatric weight and metabolic remission according to claim 2, characterized in that: The preprocessing module includes a radiomics preprocessing unit and an image feature extraction unit. The radiomics preprocessing unit is used to preprocess the abdominal CT images of patients. The image feature extraction unit is used to extract image features related to clinical parameters from the preprocessed abdominal CT images.

4. The prediction system for weight and metabolic remission after bariatric surgery according to claim 3, characterized in that: The preprocessing of the abdominal CT images includes image resampling, image filtering, image segmentation, image artifact removal, gray level quantization, and data augmentation.

5. The prediction system for post-bariatric surgery weight and metabolic remission according to claim 4, wherein: The image features include mean value, standard deviation, skewness, kurtosis, gray level uniformity, clustering tendency, S-shaped curve fitting features, and the maximum two-dimensional plane diameter and three-dimensional volume of the fat area.

6. The prediction system for post - bariatric surgery weight and metabolic remission according to claim 5, characterized in that: The data analysis module includes a decision tree construction unit, a classification tree construction unit, and a key feature extraction unit. The decision tree construction unit is used to generate a decision tree based on the extracted image features and clinical parameters, with each image feature and clinical parameter as a feature node. The classification tree construction unit is used to classify the image features based on the feature nodes, and the number of constructed classification trees is equal to the number of feature nodes. The key feature extraction unit is used to evaluate the best classification feature node based on the generated classification tree using the Gini index, which is the key feature.

7. The prediction system for the weight and metabolic remission after bariatric surgery according to claim 6, wherein: The convolutional neural network is a two-channel convolutional neural network with two parallel convolutional layers that process data simultaneously.

8. The prediction system for post-bariatric surgery weight and metabolic remission according to claim 7, wherein: The convolutional neural network is accelerated for convergence by the Adam optimizer. Adam is an optimization algorithm based on gradient descent that updates parameters by calculating the first-order moment estimate and the second-order moment estimate.

9. The prediction system for post-bariatric surgery weight and metabolic remission according to claim 8, wherein: Image filtering includes using a Gaussian filter and a median filter to reduce the noise and artifacts in the image.

10. The prediction system for weight and metabolic remission after bariatric surgery according to claim 9, wherein: Data augmentation includes rotating, translating, and flipping the image to increase the number of samples for image feature extraction.

Images