A method for identifying and predicting yellow granulomatous cholecystitis based on preoperative enhanced CT images

Abstract

This invention relates to the field of tumor detection technology. It proposes a method for differential diagnosis and prediction of xanthogranulomatous cholecystitis (XGC) based on preoperative enhanced CT images. The method includes: acquiring enhanced CT images of the gallbladder wall in the arterial and portal venous phases; performing image segmentation and feature extraction; performing dimensionality reduction on the arterial phase features, portal venous phase features, and the combined dual-phase features to obtain corresponding feature datasets, which are then divided into training and testing sets; training three machine learning prediction models based on the three training sets to obtain the optimal machine learning prediction model; validating the performance of the optimal machine learning prediction model using the testing set to obtain a differential diagnosis prediction model; and inputting the feature data of the enhanced abdominal CT image to be predicted into the differential diagnosis prediction model for identification. This invention establishes a better differential diagnosis prediction model by analyzing the imaging characteristics of XGC and thick-walled GBC, providing a basis for treatment decisions.

Description

Technical Field

[0001] This invention relates to the field of tumor detection technology, and in particular to a method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images. Background Technology

[0002] Xanthogranulomatous cholecystitis (XGC) is a rare chronic inflammatory disease of the gallbladder, characterized by nodular, yellowish masses with inflammatory infiltration and abnormal thickening of the gallbladder wall. Its clinical presentation, laboratory findings, and CT / MRI scans are similar to those of thick-walled gallbladder cancer (GBC), leading to a high rate of misdiagnosis and potentially unnecessary extended resection. Due to the severe fibrosis and adhesions to adjacent organs present in XGC, its radiographic appearance and intraoperative findings, particularly those of thick-walled GBC, can easily be misdiagnosed as GBC. Misdiagnosis may result in unnecessary radical resection, extending the surgical resection area and increasing the incidence of postoperative complications. Therefore, a comprehensive, accurate, and scientific method for preoperative differentiation and prediction between XGC and thick-walled GBC is urgently needed.

[0003] In recent years, with the rapid development of machine learning algorithms (such as random forests, Bayesian networks, and classification and regression trees), they have been widely used in the diagnosis and prognostic assessment of gallbladder diseases. Zhou et al. established a random forest model based on CT image features for preoperative differentiation between XGC and GBC, demonstrating good diagnostic accuracy and providing assistance for clinical decision-making. Radiomics is increasingly being applied to disease diagnosis and prognostic assessment of patients with malignant tumors, effectively compensating for the shortcomings of visual identification and experience-based judgment in traditional imaging examinations, and improving predictive ability. Currently, CT-based radiomics features have been applied to the differential prediction of gallbladder adenoma and GBC, and their diagnostic efficacy is superior to that of radiologists.

[0004] However, there are currently no reports on the differentiation and prediction of XGC and thick-walled GBC based on CT radiomics. Summary of the Invention

[0005] To address the aforementioned problems, this invention aims to provide a method for differentiating and predicting xanthogranulomatous cholecystitis (XGC) based on preoperative enhanced CT images. By analyzing the imaging characteristics of XGC and thick-walled GBC, a better differential prediction model is established to effectively identify the imaging differences between XGC and thick-walled GBC, significantly reduce the misdiagnosis rate, avoid unnecessary radical resection surgery, and reduce patient trauma and waste of medical resources.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, this application discloses a method for the differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images, comprising the following steps: Step 1: Acquire contrast-enhanced CT images of the gallbladder wall in the arterial and portal venous phases of the patient's abdomen, respectively; Step 2: Perform image segmentation and feature extraction on the enhanced CT images of the arterial phase and portal venous phase respectively to obtain arterial phase features and portal venous phase features; Step 3: Perform three-step dimensionality reduction on the arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial and portal venous phase features, and divide the three feature datasets into training set and test set respectively; Step 4: Train the logistic regression, support vector machine, and lightweight gradient boosting machine models based on the three training sets respectively, select the best-performing training model, and verify the performance of the best-performing training model through the test set to obtain the discriminative prediction model. Step 5: Input the feature data of the enhanced CT images of the arterial phase and portal venous phase of the abdominal gallbladder wall of the patient to be predicted into the discrimination prediction model for discrimination, and obtain the discrimination prediction results.

[0007] Furthermore, in step 3, the three-step dimensionality reduction process includes: The first step is to remove statistical values. P Features with a value >0.05 are used to obtain preliminary screening features; The second step is to remove features with a correlation coefficient greater than 0.9 from the initial feature selection to obtain the secondary feature selection. The third step involves using feature dimensionality reduction and embedding selection methods to perform final filtering on the secondary filtered features, resulting in tertiary filtered features, or three feature datasets.

[0008] Furthermore, the screening method for the three screenings is any one of principal component analysis, random forest, and LASSO regression.

[0009] Furthermore, the three feature datasets are: arterial phase important radiomics features, portal venous phase important radiomics features, and dual-phase important radiomics features.

[0010] Furthermore, in step 4, the training model with the best performance is the one with the highest average AUC and accuracy after 5-fold cross-validation.

[0011] Secondly, this application discloses a method for the differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images, including: The image acquisition module is used to acquire enhanced CT images of the gallbladder wall in the arterial and portal venous phases of the patient's abdomen, respectively. The feature extraction module is used to perform image segmentation and feature extraction on enhanced CT images of the arterial phase and portal venous phase, respectively, to obtain arterial phase features and portal venous phase features; The dimensionality reduction and data partitioning module is used to perform three-step dimensionality reduction on arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial phase features and portal venous phase features, and to partition the three feature datasets into training set and test set respectively. The training and selection module is used to train the logistic regression, support vector machine and lightweight gradient boosting machine models respectively, select the best-performing training model, and verify the performance of the best-performing training model through the test set to obtain the discriminative prediction model. The discrimination prediction module is used to input the feature data of enhanced CT images of the arterial phase and portal venous phase of the abdominal gallbladder wall of the patient to be predicted into the discrimination prediction model for discrimination and to obtain the discrimination prediction result.

[0012] Thirdly, this application discloses an electronic device, including a processor and a memory, wherein computer instructions are stored in the memory, and the processor is used to execute the computer instructions stored in the memory to implement the above-mentioned method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images.

[0013] Fourthly, this application discloses a computer-readable storage medium storing computer instructions thereon, wherein the computer instructions are used to cause a computer to execute the above-mentioned method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention significantly improves the accuracy of differentiating between XGC (adenocarcinoma) and thick-walled GBC (gallbladder cancer) by constructing a cholecystitis differential prediction model based on preoperative enhanced CT images.

[0015] Specifically: Improved diagnostic accuracy: The logistic regression model based on LASSO regression screening effectively identifies the radiological differences between XGC and thick-walled GBC by optimizing feature selection, significantly reducing the misdiagnosis rate, avoiding unnecessary radical resection surgery, and reducing patient trauma and waste of medical resources.

[0016] Multimodal information integration: The model integrates radiomic features of enhanced CT (including arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial and portal venous phase features) with preoperative clinical data, breaking through the limitations of traditional empirical qualitative analysis and providing more comprehensive prognostic information for clinical decision-making.

[0017] Clinical Application Value: This method achieves a technological breakthrough in the differential diagnosis and prediction of cholecystitis through algorithm optimization and multi-feature fusion, providing a more accurate basis for clinical treatment decisions. It can serve as an auxiliary diagnostic tool for radiologists, improving the efficiency of identifying inflammatory lesions of the cholecystitis. It is particularly suitable for primary healthcare institutions, demonstrating significant social and economic benefits. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the research process of the present invention.

[0019] Figure 2 The ROC curves of XGC and thick-walled GBC were used to differentiate between two radiologists in this invention.

[0020] Figure 3 The images are typical of XGC and thick-walled GBC of the present invention, wherein (A) shows diffuse thickening of the gallbladder wall in patients with XGC; (BD) shows continuous mucosal lines, intramural nodules and "sandwich sign" in patients with XGC; (E) shows patients with XGC complicated by gallstones and liver involvement; and (F) shows enlarged peripheral lymph nodes in patients with XGC.

[0021] Figure 4 The figures show the ROC curves of machine learning diagnostic prediction models under different phases using different radiomics feature screening methods of this invention. Among them, (AC) is the ROC curve of the diagnostic model based on arterial phase, portal venous phase and dual phase features under the principal component analysis feature screening method; (DF) is the ROC curve of the diagnostic model based on arterial phase, portal venous phase and dual phase features under the random forest feature screening method; and (GI) is the ROC curve of the diagnostic model based on arterial phase, portal venous phase and dual phase features under the LASSO regression feature screening method.

[0022] Figure 5 This is a ranking diagram of the importance of radiomics features in the arterial phase, portal venous phase, and bi-phase random forest of this invention, where (A) is the arterial phase; (B) is the portal venous phase; and (C) is the bi-phase. Detailed Implementation

[0023] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0024] Please see Figures 1-5 This application discloses a method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images, comprising the following steps: Step 1: Acquire contrast-enhanced CT images of the gallbladder wall in the arterial and portal venous phases of the patient's abdomen, respectively; The arterial phase lasted 25-35 seconds after intravenous injection of the non-ionic contrast agent, while the venous phase lasted 65-70 seconds.

[0025] Step 2: Perform image segmentation and feature extraction on the enhanced CT images of the arterial phase and portal venous phase respectively to obtain arterial phase features and portal venous phase features; Image segmentation was performed manually by two radiologists with six years of experience in abdominal imaging, using 3D Slicer version 4.11 to delineate the region of interest (ROI). Step 3: Perform three-step dimensionality reduction on the arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial and portal venous phase features, and divide the three feature datasets into training set and test set respectively; Step 3, the three-step dimensionality reduction process includes: The first step is to remove values ​​that have no or minimal impact on the classification results (statistical values). P Features with a value >0.05 were used to obtain preliminary screening features; The second step is to remove features with high redundancy (correlation coefficient greater than 0.9) from the initial feature selection to obtain the secondary feature selection. The third step involves using various feature dimensionality reduction and embedding selection methods in the secondary screening to obtain the tertiary screening features, namely important radiomics features.

[0026] Furthermore, feature reduction and embedding selection methods are used to perform final screening on the secondary screening features, resulting in tertiary screening features, thus yielding three feature datasets. The screening method for the tertiary screening is any one of principal component analysis, random forest, and LASSO regression. The three feature datasets are: arterial phase important radiomics features, portal venous phase important radiomics features, and dual-phase important radiomics features.

[0027] Step 4: Train the logistic regression, support vector machine, and lightweight gradient boosting machine models based on the three training sets respectively, select the best-performing training model, and verify the performance of the best-performing training model through the test set to obtain the discriminative prediction model. Specifically, the best-performing training model is the one with the highest average AUC and accuracy after 5-fold cross-validation.

[0028] Step 5: Input the feature data of the enhanced CT images of the arterial phase and portal venous phase of the abdominal gallbladder wall of the patient to be predicted into the discrimination prediction model for discrimination, and obtain the discrimination prediction results.

[0029] This application also discloses a method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images, implemented using the above method, including: The image acquisition module is used to acquire enhanced CT images of the gallbladder wall in the arterial and portal venous phases of the patient's abdomen, respectively. The feature extraction module is used to perform image segmentation and feature extraction on enhanced CT images of the arterial phase and portal venous phase, respectively, to obtain arterial phase features and portal venous phase features; The dimensionality reduction and data partitioning module is used to perform three-step dimensionality reduction on arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial phase features and portal venous phase features, and to partition the three feature datasets into training set and test set respectively. The training and selection module is used to train the logistic regression, support vector machine and lightweight gradient boosting machine models respectively, select the best-performing training model, and verify the performance of the best-performing training model through the test set to obtain the discriminative prediction model. The discrimination prediction module is used to input the feature data of enhanced CT images of the arterial phase and portal venous phase of the abdominal gallbladder wall of the patient to be predicted into the discrimination prediction model for discrimination and to obtain the discrimination prediction result.

[0030] This application also discloses an electronic device, including a processor and a memory, wherein computer instructions are stored in the memory, and the processor is used to execute the computer instructions stored in the memory to implement the above-mentioned method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images.

[0031] This application also discloses a computer-readable storage medium storing computer instructions thereon, wherein the computer instructions are used to cause a computer to execute the above-described method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images.

[0032] Example The reliability of the proposed method for differentiating and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images is verified below with specific embodiments. The method is used to differentiate and predict xanthogranulomatous cholecystitis from thick-walled gallbladder cancer. I. Research Subjects: This study included patients who received treatment at a large medical center between January 2011 and December 2020 and were pathologically confirmed to have XGC or GBC. Inclusion criteria: (1) preoperative enhanced CT scan of the upper abdomen in the hospital; (2) CT images showing thick-walled GBC; (3) complete clinicopathological features and serum biomarker data. Exclusion criteria: (1) CT images showing distant metastasis; (2) GBC patients who received neoadjuvant therapy before surgery; (3) patients with poor image quality.

[0033] II. Image Acquisition (corresponding to step 1: acquire contrast-enhanced CT images of the patient's abdominal gallbladder wall in the arterial and portal venous phases respectively) Each patient underwent an enhanced abdominal CT scan within one month prior to surgery using a 256-slice CT scanner. Scanning parameters included: tube voltage 120 kVp, tube current 250 mA, slice thickness 5.0 mm, and reconstruction interval 5.0 mm. Scans were performed at 25–35 s (arterial phase) and 65–70 s (portal venous phase) after intravenous injection of a non-ionic contrast agent. A schematic diagram of the study workflow is shown below. Figure 1 .

[0034] Table 1 compares the CT imaging features of XGC and thick-walled GBC. Significant differences were found between the two in gallbladder wall thickening patterns, mucosal lines, intramural nodules, gallstones, abnormal enhancement of adjacent liver parenchyma, and enlarged surrounding lymph nodes (P<0.05). The AUCs predicted by two radiologists based on CT imaging features for differentiating XGC and thick-walled GBC were 0.6735 (95% confidence interval: 0.5768–0.7701, P<0.001) and 0.6867 (95% confidence interval: 0.5902–0.7833, P<0.001), respectively. Delong's test showed no statistically significant difference in AUC between the two radiologists (Z=0.251, P=0.8017). Figure 2 Typical images of XGC and thick-walled GBC are shown below. Figure 3 .

[0035] Table 1. Baseline and imaging features of patients with XGC and thick-walled GBC III. Image Segmentation and Feature Extraction (corresponding to step 2: Perform image segmentation and feature extraction on the enhanced CT images of the arterial phase and portal venous phase respectively to obtain arterial phase features and portal venous phase features) 1502 radiomics features were extracted from manually segmented ROIs in the arterial and portal venous phases, respectively. After two-step screening, 135 features were obtained in the arterial phase and 138 features were obtained in the portal venous phase. In the third step, principal component analysis, random forest, and LASSO regression were used for final dimensionality reduction, resulting in 31, 30, and 35 features in the arterial phase, and 30, 30, and 9 features in the portal venous phase, respectively.

[0036] To evaluate the value of dual-phase features in differentiating XGC from thick-walled GBC, 3504 features from the arterial and portal venous phases were merged and subjected to a three-step dimensionality reduction (corresponding to step 3: performing three-step dimensionality reduction on arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial and portal venous phase features, and dividing the resulting three feature datasets into training and testing sets respectively). After two steps of selection, 257 features were obtained. In the third step, principal component analysis, random forest, and LASSO regression were used for final dimensionality reduction, resulting in 41, 30, and 14 features, respectively. The importance ranking of the radiomics features selected by random forest is shown in Figure S1, and the features selected by LASSO regression are shown in Table 2.

[0037] Table 2. Screening results of radiomics features IV. Machine Learning Model Establishment and Evaluation (corresponding to steps 4 and 5: Train logistic regression, support vector machine, and lightweight gradient booster models based on three training sets respectively, select the best-performing training model, and verify the performance of the best-performing training model through the test set to obtain the discrimination prediction model; input the feature data of the enhanced CT images of the arterial phase and portal venous phase of the abdominal gallbladder wall of the patient to be predicted into the discrimination prediction model for discrimination, and obtain the discrimination prediction result.) Based on the radiomics features selected through the three-step dimensionality reduction process, nine machine learning prediction models were constructed using arterial phase, portal venous phase, and dual-phase features, combined with logistic regression, support vector machine, and lightweight gradient boosting machine algorithms. Table 3 shows the average AUC and accuracy of each model in the training set for identifying and predicting XGC and thick-walled GBC using 5-fold cross-validation. The model constructed based on dual-phase radiomics features selected using LASSO regression showed better average AUC and accuracy than models constructed using arterial and portal venous phases alone, and also better than models constructed based on dual-phase features selected using principal component analysis and random forest. Overall, the logistic regression model based on LASSO regression had an average AUC of 0.9052 in the training set and 0.9590 in the test set, outperforming other models and the AUC values ​​predicted by radiologists.

[0038] Table 3. Mean AUC and accuracy of machine learning models in distinguishing XGC and thick-walled GBC on the training set using 5-fold cross-validation. In the test set, the AUCs of the logistic regression, support vector machine, and lightweight gradient booster models constructed based on dual-phase radiomics features selected by LASSO regression were 0.959, 0.951, and 0.940, respectively. These AUCs were all higher than those of models constructed individually for the arterial and portal venous phases, and also superior to models constructed based on dual-phase features selected by principal component analysis and random forest. The ROC curves are shown below. Figure 4 The AUC of machine learning diagnostic prediction models at different time phases under different radiomics feature selection methods showed that the logistic regression model based on dual-phase radiomics features selected by LASSO regression exhibited better predictive performance, significantly outperforming models built using arterial or portal venous phase features, and also outperforming diagnostic prediction models constructed using radiomics features selected by other machine learning methods or modeling methods. Therefore, the radiomics feature selection method is an important factor affecting the performance of prediction models, and dual-phase radiomics features can compensate for the deficiencies of single-phase features in diagnostic model construction.

[0039] Statistical analysis Statistical analysis was performed using SPSS 25.0. Normally distributed continuous data were expressed as mean ± standard deviation, and the two-sample t-test was used for comparisons between groups. Skewedly distributed continuous data were expressed as median (range), and the Mann-Whitney U test was used for comparisons between groups. The χ² test was used for comparisons between groups of categorical data. The Delong test was used to compare the AUC value of the machine learning model based on LASSO regression screening with the AUC value predicted by two radiologists. A p-value < 0.05 was considered statistically significant. The results showed that the AUC value of the machine learning model based on LASSO regression screening was superior to the diagnostic AUC value of the radiologists, indicating that the model's diagnostic ability was better than that of the radiologists.

[0040] Ethical Statement This study was approved by the Ethics Committee of the First Affiliated Hospital of Xi'an Jiaotong University (No.: XJTU1AF2022LSK-089). Data collection and use complied with medical ethics principles and the requirements of the Declaration of Helsinki, and did not pose any risk or harm to the health, safety, or privacy of the participants. All enrolled patients and their families signed written informed consent forms before enrollment in the study.

[0041] Results (corresponding to step 6: Input the feature data of the abdominal enhanced CT image to be predicted into the discrimination prediction model for discrimination, and obtain the discrimination prediction result.) CT imaging features Table 1 compares the CT imaging features of XGC and thick-walled GBC. Significant differences were found between the two in gallbladder wall thickening patterns, mucosal lines, intramural nodules, gallstones, abnormal enhancement of adjacent liver parenchyma, and enlarged surrounding lymph nodes (P<0.05). The AUCs predicted by two radiologists based on CT imaging features for differentiating XGC and thick-walled GBC were 0.6735 (95% confidence interval: 0.5768–0.7701, P<0.001) and 0.6867 (95% confidence interval: 0.5902–0.7833, P<0.001), respectively. Delong's test showed no statistically significant difference in AUC between the two radiologists (Z=0.251, P=0.8017). Figure 2 Typical images of XGC and thick-walled GBC are shown below. Figure 3 .

[0042] The foregoing description illustrates the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images, characterized in that, Includes the following steps: Step 1: Acquire contrast-enhanced CT images of the gallbladder wall in the arterial and portal venous phases of the patient's abdomen, respectively; Step 2: Perform image segmentation and feature extraction on the enhanced CT images of the arterial phase and portal venous phase respectively to obtain arterial phase features and portal venous phase features; Step 3: Perform three-step dimensionality reduction on the arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial and portal venous phase features, and divide the three feature datasets into training set and test set respectively; Step 4: Train the logistic regression, support vector machine, and lightweight gradient boosting machine models based on the three training sets respectively, select the best-performing training model, and verify the performance of the best-performing training model through the test set to obtain the discriminative prediction model. Step 5: Input the feature data of the enhanced CT images of the arterial phase and portal venous phase of the abdominal gallbladder wall of the patient to be predicted into the discrimination prediction model for discrimination, and obtain the discrimination prediction results.

2. The method for differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images according to claim 1, characterized in that, Step 3, the three-step dimensionality reduction process includes: The first step is to remove statistical values. P Features with a value >0.05 are used to obtain preliminary screening features; The second step is to remove features with a correlation coefficient greater than 0.9 from the initial feature selection to obtain the secondary feature selection. The third step involves using feature dimensionality reduction and embedding selection methods to perform final filtering on the secondary filtered features, resulting in tertiary filtered features, or three feature datasets.

3. The method for differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images according to claim 2, characterized in that, The screening method for the three screenings is any one of principal component analysis, random forest, and LASSO regression.

4. The method for differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images according to claim 2, characterized in that, The three feature datasets are: arterial phase important radiomics features, portal venous phase important radiomics features, and dual-phase important radiomics features.

5. The method for differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images according to claim 1, characterized in that, In step 4, the best-performing training model is the one with the highest average AUC and accuracy after 5-fold cross-validation.

6. A method for differential diagnosis and prediction of xanthogranulomatous cholecystitis based on preoperative enhanced CT images, characterized in that, Implemented using the method described in any one of claims 1-5, comprising: The image acquisition module is used to acquire enhanced CT images of the gallbladder wall in the arterial and portal venous phases of the patient's abdomen, respectively. The feature extraction module is used to perform image segmentation and feature extraction on enhanced CT images of the arterial phase and portal venous phase, respectively, to obtain arterial phase features and portal venous phase features; The dimensionality reduction and data partitioning module is used to perform three-step dimensionality reduction on arterial phase features, portal venous phase features, and dual-phase features obtained by mixing arterial phase features and portal venous phase features, and to partition the three feature datasets into training set and test set respectively. The training and selection module is used to train the logistic regression, support vector machine and lightweight gradient boosting machine models respectively, select the best-performing training model, and verify the performance of the best-performing training model through the test set to obtain the discriminative prediction model. The discrimination prediction module is used to input the feature data of enhanced CT images of the arterial phase and portal venous phase of the abdominal gallbladder wall of the patient to be predicted into the discrimination prediction model for discrimination and to obtain the discrimination prediction result.

7. An electronic device comprising a processor and a memory, characterized in that, The memory stores computer instructions, and the processor is used to run the computer instructions stored in the memory to implement the method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images as described in any one of claims 1-5.

8. A computer-readable storage medium having computer instructions stored thereon, wherein, The computer instructions are used to cause the computer to execute a method for identifying and predicting xanthogranulomatous cholecystitis based on preoperative enhanced CT images as described in any one of claims 1-5.

Images