Method and device for predicting thymus hyperplasia risk based on CT image features and medium

Abstract

The invention provides a method, equipment, medium and program product for predicting thymus hyperplasia risk based on CT image features, and relates to the field of intelligent medical treatment. The method comprises the following steps: acquiring a tested CT image; extracting target features in the CT image, and inputting the target features into the risk prediction model to obtain scores; and outputting an auxiliary prediction result of the thymus hyperplasia risk of the subject according to the score. When the score is greater than a first threshold value, outputting an auxiliary prediction result of a large thymus hyperplasia risk of the subject; and when the score is smaller than a first threshold value, outputting an auxiliary prediction result with a small thymus hyperplasia risk of the subject. According to the method, the thymus hyperplasia prediction model is established by using the CT image data through methods such as artificial intelligence modeling and the like. The method not only is beneficial to identifying MG patients, but also can provide important basis for clinical diagnosis and treatment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medicine, and more specifically, to a method, device, medium and program product for predicting the risk of thymic hyperplasia based on CT image features. Background Art

[0002] Myasthenia gravis (MG) is an autoimmune disease characterized by impaired conduction at the neuromuscular junction, primarily affecting acetylcholine receptors at the synapses of the neuromuscular junction. It is primarily an acetylcholine receptor-mediated autoimmune disease, dependent on cell-mediated immunity, and involving complement. The disease is closely related to the thymus gland, and its incidence has increased significantly over the past decade. According to incomplete clinical statistics, approximately 65% ​​to 80% of MG patients have thymic abnormalities, approximately 70% have concomitant thymic hyperplasia, and 10-15% have concomitant thymomas.

[0003] Thymic lesions are often a key factor in the development of myasthenia gravis. Therefore, early risk assessment of potential MG patients using thymic CT imaging provides a new approach and direction for early intervention and treatment. Summary of the Invention

[0004] The present invention aims to address at least one of the technical problems existing in the prior art. To this end, the present invention provides a method, device, medium, and program product for predicting the risk of thymic hyperplasia based on CT image features. The present method utilizes CT image data and, through artificial intelligence modeling and other methods, establishes a thymic hyperplasia prediction model. This not only aids in identifying MG patients but also provides important evidence for clinical diagnosis and treatment.

[0005] In a first aspect, the present application discloses a method for predicting the risk of thymic hyperplasia based on CT image features, the method comprising:

[0006] 101, obtaining CT images of the subject;

[0007] 102, extracting target features from the CT image, where the target features include any one or more of the following: original_glrlm_ShortRunEmphasis, original_gldm_DependenceEntropy, original_shape_Maximum2DDiameterColumn, original_glrlm_RunVariance, and original_glcm_lmc1;

[0008] 103, inputting the target feature into a risk prediction model to obtain a score;

[0009] 104. Output an auxiliary prediction result of the subject's risk of thymic hyperplasia based on the score; when the score is greater than a first threshold, output an auxiliary prediction result indicating that the subject has a high risk of thymic hyperplasia; when the score is less than the first threshold, output an auxiliary prediction result indicating that the subject has a low risk of thymic hyperplasia.

[0010] In some embodiments, the target features further include any one or more of the following: original_shape_SurfaceVolumeRatio, original_shape_Flatness, original_shape_Elongation, original_shape_Sphericity, original_glrlm_GrayLevelVariance, and original_ngtdm_Strength.

[0011] In some embodiments, the method for constructing the risk prediction model includes:

[0012] 201, obtaining CT images of training set samples and classification labels including healthy people and thymic hyperplasia people;

[0013] 202, extracting image features of the thymus region in the CT image;

[0014] 203, filtering out target features from the image features;

[0015] 204. Input the target feature into the machine learning model to obtain a predicted classification result, compare it with the classification label, optimize the model according to the comparison result, and obtain a constructed prediction model.

[0016] In some embodiments, the method for screening the target features includes: screening features with significant differences between healthy people and thymic hyperplasia people; standardizing the features with significant differences to obtain standardized features; screening from the standardized features features that are not easily affected by batch effects, remove features with high missing rates and low variances, and remove brightness-related features to obtain features after the initial round of screening; and obtaining the target features from the features after the initial round of screening using the LASSO algorithm.

[0017] In some embodiments, the image features include any one or more of the following: shape features, first-order statistical features, grayscale co-occurrence matrix features, grayscale run length matrix features, grayscale block size matrix features, grayscale dependence matrix features, and neighborhood grayscale difference matrix features.

[0018] In some embodiments, the method used by the machine learning model includes any one or more of the following: RandomForest, Logistic Regression, SVM, XGBoost, CatBoost, LightGBM, Neural Network, Decision Tree, KNN, Naive Bayes; preferably Logistic Regression.

[0019] In some embodiments, between 201 and 202 , the step further includes performing preprocessing on the CT image including grayscale conversion, normalization, and contrast enhancement to obtain a preprocessed CT image.

[0020] A second aspect of the present application discloses a computer device, comprising: a memory and a processor; the memory is used to store a computer program; and the processor executes the computer program to implement the steps of the above method.

[0021] A third aspect of the present application discloses a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0022] A fourth aspect of the present application discloses a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0023] This application has the following beneficial effects:

[0024] 1. This application innovatively explores features in CT images associated with the risk of thymic hyperplasia and identifies 11 key features, which not only help identify MG patients but also provide important evidence for clinical diagnosis and treatment.

[0025] 2. This application innovatively discloses a method for predicting the risk of thymic hyperplasia based on CT image features. This method utilizes CT image data and uses artificial intelligence modeling and other methods to establish a thymic hyperplasia prediction model, thereby enabling risk assessment of potential MG patients based on CT images. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 This is a schematic diagram of the method flow provided by the first aspect of the embodiment of the present invention; Figure 2 Schematic diagram of a system for predicting the risk of thymic hyperplasia based on CT image features provided by the second aspect of an embodiment of the present invention;

[0028] Figure 3 is a schematic diagram of a computer device provided by an embodiment of the present invention;

[0029] Figure 4 is a schematic diagram of the architecture of an exemplary computing device provided by an embodiment of the present invention;

[0030] Figure 5 is a schematic diagram of a storage medium provided by an embodiment of the present invention;

[0031] Figure 6 This is a feature selection heat map screened by the LASSO algorithm provided in an embodiment of the present invention;

[0032] Figure 7 This is a bar chart of feature selection using the LASSO algorithm provided in an embodiment of the present invention;

[0033] Figure 8 This is a violin plot of the distribution of the target feature in the healthy group and the disease group provided by an embodiment of the present invention;

[0034] Figure 9 is a forest plot of the target feature odds ratio provided by an embodiment of the present invention;

[0035] Figure 10 This is a performance comparison chart of different machine learning models provided by the embodiments of the present invention in multiple experiments;

[0036] Figure 11 The ROC curve and AUC value of the Logistic Regression model provided by the embodiment of the present invention under the five-fold cross validation of the training set;

[0037] Figure 12 is the average SHAP value of the target feature provided by the embodiment of the present invention;

[0038] Figure 13 is the SHAP value distribution of the target feature pair model prediction result provided by the embodiment of the present invention;

[0039] Figure 14 This is an ICE analysis diagram of original_glrlm_ShortRunEmphasis provided in an embodiment of the present invention;

[0040] Figure 15 This is an ICE analysis diagram of original_gldm_DependenceEntropy provided by an embodiment of the present invention;

[0041] Figure 16This is an ICE analysis diagram of original_shape_Maximum2DDiameterColumn provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0043] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit "first" and "second" to be different types.

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0045] Figure 1 1 is a flow chart of a method for predicting the risk of thymic hyperplasia based on CT image features provided by an embodiment of the present invention. Specifically, the method includes the following steps: 101: obtaining a CT image of a subject;

[0046] In some embodiments, the terms "subject," "person to be tested," or "test sample" as used herein refer to any animal (e.g., mammal), including but not limited to humans, non-human primates, rodents, etc., that is to be the recipient of a particular treatment. Generally, the terms "subject" and "patient" are used interchangeably herein when referring to a human subject. Preferably, the subject is a human.

[0047] 102: Extract target features from the CT image, where the target features include any one or more of the following: original_glrlm_ShortRunEmphasis, original_gldm_DependenceEntropy, original_shape_Maximum2DDiameterColumn, original_glrlm_RunVariance, and original_glcm_lmc1; in some embodiments, the target features also include any one or more of the following: original_shape_SurfaceVolumeRatio, original_shape_Flatness, original_shape_Elongation, original_shape_Sphericity, original_glrlm_GrayLevelVariance, and original_ngtdm_Strength.

[0048] 103: Inputting the target feature into a risk prediction model to obtain a score;

[0049] In some embodiments, the method for constructing the risk prediction model includes: 201, obtaining CT images of training set samples and classification labels including healthy people and people with thymic hyperplasia; in some embodiments, between 201 and 202, further comprising preprocessing the CT images including grayscale conversion, normalization, and contrast enhancement to obtain preprocessed CT images. 202, extracting image features of the thymus region in the CT images; in some embodiments, the image features include any one or more of the following: shape features, first-order statistical features, grayscale co-occurrence matrix features, grayscale run length matrix features, grayscale block size matrix features, grayscale dependency matrix features, and neighborhood grayscale difference matrix features. 203. Filter out target features from the image features. In some embodiments, the method for screening the target features includes: screening features with significant differences between healthy people and people with thymic hyperplasia; standardizing the features with significant differences to obtain standardized features; screening the standardized features that are less susceptible to batch effects, remove features with high missing rates and low variances, and remove brightness-related features to obtain features after the initial screening; and filtering the target features from the features after the initial screening using the LASSO algorithm. 204. Input the target features into a machine learning model to obtain a predicted classification result, compare it with the classification label, optimize the model based on the comparison result, and obtain a constructed prediction model. In some embodiments, the method used by the machine learning model includes any one or more of the following: Random Forest, Logistic Regression, SVM, XGBoost, CatBoost, LightGBM, Neural Network, Decision Tree, KNN, Naive Bayes; preferably LogisticRegression.

[0050] 104: Outputting an auxiliary prediction result of the subject's risk of thymic hyperplasia based on the score; when the score is greater than a first threshold, outputting an auxiliary prediction result indicating that the subject has a high risk of thymic hyperplasia; when the score is less than the first threshold, outputting an auxiliary prediction result indicating that the subject has a low risk of thymic hyperplasia.

[0051] In some embodiments, the auxiliary prediction results output based on the scoring include but are not limited to paper or electronic report forms. The results are only obtained by the intelligent machine based on the analysis of the relevant data of the subject, and are only used as a reference for medical staff, and are not the final diagnosis result of the subject.

[0052] In some embodiments, the first threshold is obtained through training of training set samples, and can be a specific threshold or an interval range. The specific form is not specifically limited in this embodiment.

[0053] Figure 3is a schematic diagram of a computer device provided by an embodiment of the present invention, such as Figure 3 As shown, the device 2000 may include: one or more processors 2010, and one or more memories 2020; wherein the memories store computer-readable codes, and when the computer-readable codes are run by the one or more processors, they may execute the method described above.

[0054] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. It can implement or execute the various methods, operations, and logic block diagrams disclosed in the embodiments of the present disclosure. The general-purpose processor can be a microprocessor or any conventional processor, etc., and can be an X86 architecture or an ARM architecture.

[0055] In general, various example embodiments of the present disclosure may be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. When various aspects of the disclosed embodiments are illustrated or described as block diagrams, flow charts, or using some other graphical representation, it will be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0056] For example, the method or apparatus according to the embodiment of the present disclosure may also be implemented by Figure 4 The architecture of the computing device 3000 shown in FIG. Figure 4 As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage device in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the method provided in the present disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 4 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 4 One or more components of a computing device are shown.

[0057] The embodiment of the present invention further provides a computer-readable storage medium, such as Figure 5 As shown, it is a schematic diagram of a storage medium 4000 provided in an embodiment of the present invention, and computer-readable instructions 4010 are stored on the computer storage medium 4020. When the computer-readable instructions 4010 are executed by the processor, the method according to the embodiment of the present disclosure described with reference to the above figures can be executed. The computer-readable storage medium in the embodiment of the present disclosure can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. The non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory. It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0058] The embodiments of the present disclosure further provide a computer program product or system, including a computer program, which implements the steps of the above method when executed by a processor.

[0059] In some embodiments, this embodiment further discloses a system for predicting the risk of thymic hyperplasia based on CT image features, such as Figure 2 As shown, the system includes:

[0060] An image acquisition module 301 is used or configured to acquire a CT image of a subject;

[0061] A feature extraction module 302 is used or configured to extract target features from the CT image, where the target features include any one or more of the following: original_glrlm_ShortRunEmphasis, original_gldm_DependenceEntropy, original_shape_Maximum2DDiameterColumn, original_glrlm_RunVariance, and original_glcm_lmc1;

[0062] A score calculation module 303 is used or configured to input the target feature into a risk prediction model to obtain a score;

[0063] The result prediction module 304 is used or configured to output an auxiliary prediction result of the subject's risk of thymic hyperplasia based on the score; when the score is greater than a first threshold, output an auxiliary prediction result that the subject has a high risk of thymic hyperplasia; when the score is less than the first threshold, output an auxiliary prediction result that the subject has a low risk of thymic hyperplasia. Specific embodiment:

[0065] 1. Existing data and samples: 624 CT image samples, including 315 healthy subjects and 309 confirmed MG patients.

[0066] 2. Analysis Objective: After identifying the location of the thymus from the patient's CT image data, extract image features, and build models for healthy and disease groups to obtain a risk assessment model that can be used to distinguish between people with thymic hyperplasia and healthy people.

[0067] 3. Analysis results:

[0068] 3.1 Analysis content and main conclusions:

[0069] The following features are used to effectively predict myasthenia gravis, achieving an AUC ROC of 0.91 in a reserved test dataset. The model's performance is: original_glcm_Imc1, original_gldm_DependenceEntropy, original_glrlm_GrayLevelVariance, original_glrlm_RunVariance, original_glrlm_ShortRunEmphasis, original_ngtdm_Strength, original_shape_Elongation, original_shape_Flatness, original_shape_Maximum2DDiameterColumn, original_shape_Sphericity, and original_shape_SurfaceVolumeRatio. These features are extracted from multiple dimensions of the image, including the gray-level co-occurrence matrix, gray-level length dependency matrix, gray-level run length matrix, non-uniform texture descriptor, and morphological features. These features comprehensively reflect the texture, morphology, and complexity of the tumor. Through these multi-dimensional features, the model is able to capture the imaging manifestations of myasthenia gravis, providing effective imaging evidence for early clinical diagnosis and prognosis.

[0070] After careful selection and optimization during model training, the final model achieved an AUC ROC of 0.91 in predicting myasthenia gravis, demonstrating the model's efficiency and reliability in processing imaging features and predicting disease. This result suggests that imaging-based analysis methods have significant potential for early diagnosis of myasthenia gravis, particularly when combining multiple imaging features, demonstrating strong predictive capabilities.

[0071] 3.2 Image preprocessing:

[0072] In order to enable subsequent image analysis to proceed smoothly, it is necessary to perform a series of preprocessing operations on the images.

[0073] Grayscale Conversion: Many image processing algorithms are designed for grayscale images because grayscale images reduce computational complexity while retaining key information in the image. Compared to color images, grayscale images only contain data from one channel, making subsequent processing more efficient.

[0074] Normalization: Normalization adjusts the pixel values ​​in an image to a uniform range (usually 0 to 255) to reduce the impact of image brightness differences. This process helps improve the consistency and robustness of the algorithm when processing different images.

[0075] Contrast enhancement: By stretching or compressing the pixel value distribution of an image, the boundaries and details of the target area (such as the thymus) are made more distinct. This helps improve the effectiveness of subsequent algorithms (such as edge detection or segmentation algorithms).

[0076] 3.3 Image feature extraction and intergroup distribution analysis: For the thymus region, 108 common image features were calculated. These features can be classified into the following categories:

[0077] (1) Shape Features: Shape features describe the geometric shape of the target area and are usually independent of the grayscale value of the image. These features mainly reflect information such as the size, shape, and surface area of ​​the lesion or structure. (2) First Order Statistics: First Order Statistics describe the distribution characteristics of the grayscale values ​​of pixels or voxels in the image, reflecting the overall intensity and density information of the image. (3) Gray Level Co-occurrence Matrix Features (GLCM): GLCM features describe the grayscale connection information between pixel pairs in the image, capturing the local texture characteristics of the image. (4) Gray Level Run Length Matrix Features (GLLM): GLRLM features describe the length information of the continuous arrangement of pixels with the same grayscale value in the image, mainly used to analyze the texture structure of the image. (5) Gray Level Size Zone Matrix Features (GLSZM): GLSZM features describe the size information of the continuous region with the same grayscale value in the image, capturing the regional homogeneity in the image. (6) Gray Level Dependence Matrix Features (GLDM): GLDM features describe the dependency of a pixel on its surrounding pixels with the same or similar gray levels. (7) Neighborhood Gray Tone Difference Matrix Features (NGTDM): NGTDM features describe the degree of grayscale difference between a voxel in an image and its surrounding voxels. They are mainly used to analyze the local texture characteristics of an image.

[0078] 4.MG Risk AI Modeling:

[0079] 4.1 Data Preprocessing

[0080] To enable subsequent modeling calculations and avoid interference with the numerical calculations of some machine learning algorithms, numerical data is normalized based on the mean and variance. The purpose of feature standardization is to scale all numerical variables to the same scale, thereby preventing certain variables from having an unfair impact on the model due to excessively large or small numerical ranges. Standard normalization, used in this study, is a method that converts data into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0081] 4.2 Feature Screening

[0082] Because some imaging features are susceptible to batch effects during CT scans (such as first-order statistical features representing overall intensity and density), which can mislead subsequent AI modeling results (the model may learn differences between batches), further feature screening and model training were conducted only on 74 imaging features related to morphology, size, plaque, and cord region recognition that are less susceptible to batch effects. Furthermore, some features with high missing rates and low variance were removed. To minimize the impact of environmental monitoring factors on the model's performance in both diseased and healthy samples, some brightness-related features were removed. Ultimately, 74 features were selected for modeling.

[0083] 74 image features are still a large number and often suffer from multicollinearity, which can reduce the stability and predictive accuracy of the trained model. Excessive features also hinder model interpretation. Therefore, we first used 100 rounds of simulation experiments combined with the Lasso algorithm to screen the features. Ultimately, 11 of the 74 image features were selected for inclusion in the model. Figure 6 This chart shows a feature selection heatmap. Each column represents a LASSO selection, and each row represents a feature. Blue indicates that the feature was selected in this experiment. The more blue a feature is, the more frequently it was selected, indicating that it is more important. Figure 7 A feature selection bar chart is displayed. Based on the results of 100 modeling experiments, the number of times each feature was selected in the experiment is counted, and the features selected in 90% or more of the experiments are selected. That is, the features selected by the model 90 or more times in 100 experiments are selected.

[0084] After the above analysis and selection, we finally selected the following 11 features, as shown in Table 1;

[0085] Table 1

[0086]

[0087] In addition, for the above 11 target features, the numerical distribution in the healthy group (Group 0) and the disease group (Group 1) was analyzed, as shown in the following figure: Figure 8 As shown. After feature screening, drawing a violin plot can effectively show the distribution of each feature in different groups, thereby helping to further evaluate its potential in distinguishing different categories. The violin plot can not only show the distribution pattern of the features, but also reveal the central tendency, outliers and variability of the data. Compared with the traditional box plot, the violin plot provides more intuitive distribution information, especially when the data has a multimodal distribution. The density curve it displays can help us better understand the distribution characteristics of the data. The violin plot not only helps to identify potential key features, but also reveals outliers and abnormal distributions in the data, providing support for subsequent data cleaning and modeling.

[0088] We also made an important feature advantage ratio forest plot for the above 11 target features, such as Figure 9 As shown. Forest Plot is a graphical tool that is often used to display the effect size and confidence interval of multiple research results or multiple analyses. In feature analysis, Forest Plot can be used to display the coefficient and confidence interval of each feature in the Logistic Regression model. Forest Plot visualizes the coefficient of each feature and its 95% confidence interval, allowing researchers to quickly identify the features that have the greatest impact on model predictions. By observing the positive and negative signs of the coefficients, you can clearly understand the positive and negative effects of each feature on the target variable. The width of the confidence interval shows the uncertainty of the estimate, and a narrower confidence interval indicates a more precise estimate of the feature's impact.

[0089] Odds Ratio (horizontal axis): An odds ratio of 1 (red dashed line) indicates that the feature has no significant effect on the target variable (neutral effect). An odds ratio > 1 indicates that the feature is positively correlated with the target variable (increasing the likelihood of predicting the value). An odds ratio < 1 indicates that the feature is negatively correlated with the target variable (decreasing the likelihood of predicting the value).

[0090] Point Estimates and Confidence Intervals: Each point represents a point estimate of the odds ratio (the effect of the feature calculated by the model). Error bars (black horizontal lines) represent 95% confidence intervals, which measure the statistical robustness of the feature's odds ratio. If the confidence interval crosses the red dashed line (i.e., 1), the feature is likely not statistically significant.

[0091] Significance judgment: If the error bar is completely to the right or left of the red line, it means that the effect of the feature on the target variable is statistically significant. If the error bar crosses the red line, it means that the effect of the feature may not be significant.

[0092] Coefficient sign: A positive coefficient indicates that the feature increases the probability of the target event, while a negative coefficient indicates that it decreases. Coefficient magnitude: The larger the absolute value of the coefficient, the more significant the impact of the feature on the model output.

[0093] 4.3 AI Model Algorithm Selection

[0094] After using LASSO to screen features, using other more powerful models can often further improve the model's prediction effect. In this study, we selected ten models to analyze the data set. We first divided the entire data set into a training set and a test set in an 8:2 ratio, with 499 cases divided into a training set and 125 cases divided into a test set. On the training set, we conducted multiple experimental training on different machine learning models. The results are as follows: Figure 10 Figure 2 shows a comparison of the performance of different machine learning models across multiple experiments. The boxplots show the distribution of ROC-AUC values ​​for each model (Random Forest, Logistic Regression, SVM, XGBoost, CatBoost, LightGBM, Neural Network, DecisionTree, KNN, and Naive Bayes). Each box represents the distribution range of the model across the experiments, and the dots represent the specific ROC-AUC value for each experiment. The center of the box represents the median, and the upper and lower boundaries represent the quartiles (Q1 and Q3). The lines indicate the overall data range, providing information on the stability and dispersion of the model across experiments.

[0095] 4.4 Final model training

[0096] Cross-validation is a commonly used evaluation method during model training. By dividing the training set into multiple subsets (folds) and performing multiple rounds of training and validation, it effectively assesses the model's generalization ability. We performed 5-fold cross-validation on the training set, plotted the ROC-AUC curve, and took the mean and corresponding 95% confidence interval. We also obtained the ROC-AUC curve on the test set to test the model's generalization ability. Figure 11The ROC curves and AUC values ​​for the Logistic Regression model under 5-fold cross-validation of the training set are shown. Each curve represents the ROC curve for a fold, and the adjacent AUC value shows the model's discriminative ability for each fold. The differences between the ROC curves for all folds are represented by the standard deviation in gray shading, indicating the stability of the model's performance. The thick blue line is the mean ROC curve, and the gray area is the 95% confidence interval. The AUC is 0.95±0.03, indicating that the model performs well overall in the 5-fold validation, with high ability to discriminate between positive and negative samples and good consistency between folds.

[0097] The confidence intervals (grey shaded areas) in the figure show that the logistic regression model exhibited high stability in five-fold cross-validation, with minimal variation in ROC AUC between folds, indicating consistent model performance across different data partitions. The model demonstrated both high accuracy and high reliability. The graph shows that the logistic regression model achieved the best ROC-AUC value on this dataset, and its performance was relatively stable across all folds of cross-validation. A ROC-AUC (area under the receiver operating characteristic curve) analysis on the test set yielded an AUC of 0.91.

[0098] The weights and intercepts of the 11 features are shown in Table 2:

[0099] Table 2

[0100]

[0101] The intercept is a constant term used in regression analysis and other predictive models. It represents the expected value of the dependent variable (output variable) when all independent variables (input variables) are zero.

[0102] 4.6 Model Interpretability Analysis

[0103] SHAP analysis:

[0104] The SHAP value is a tool that helps us understand how machine learning models make decisions. It is based on a mathematical concept called the Shapley value, which measures the contribution of each feature to the final prediction result. This is particularly useful for dealing with complex models (such as deep learning models or integrated models) because these models are often like "black boxes" and difficult to interpret directly. By calculating the average contribution of each feature to the prediction, the SHAP value can tell us which features are most important in the model and can also explain the specific prediction process of a single sample. In this way, researchers can not only understand how the model works as a whole, but also deeply analyze the specific situation of each sample. Using the SHAP value, the features that contribute most to the model prediction can be identified, thereby helping to discover key biomarkers related to the disease. The importance ranking of the 11 target features is as follows Figure 12 and Figure 13 shown. Figure 12 This chart shows the average SHAP value for each feature in the model, indicating its importance to the prediction results. The horizontal axis represents the average SHAP value, with larger values ​​indicating a greater influence of the feature on the model's predictions. Features are sorted from top to bottom by importance, with higher-ranked features being more critical to the model. Lower SHAP values ​​(such as Strength) indicate that these features have less influence on the model. This chart helps us identify the most important features in the model, thereby improving model interpretability. Figure 13 This chart shows the SHAP value distribution of each feature on the model's predictions. The horizontal axis represents the feature's impact on the prediction (SHAP value), while the vertical axis represents the feature name, sorted by importance. Colors range from blue to red, indicating higher or lower feature values. SHAP values ​​to the right of the baseline indicate that the feature increases the predicted value, while SHAP values ​​to the left of the baseline indicate that the feature decreases the predicted value. The color and SHAP value distribution reveal the specific impact of different feature values ​​on the prediction. For example, high values ​​of original_glrlm_ShortRunEmphasis significantly increase the predicted value, while high values ​​of original_glrlm_GrayLevelVariance significantly decrease the predicted value.

[0105] ICE Analysis:

[0106] Individual Conditional Expectation (ICE) analysis is a tool used to explain machine learning model predictions. It demonstrates the impact of changes in specific features on individual sample predictions, helping to understand how the model makes different predictions based on those features. ICE analysis can reveal how different samples respond to feature changes, providing a more nuanced, individual-level explanation. Choosing to perform ICE analysis can help uncover nonlinear relationships between individual features and predictions, reveal interaction effects, and effectively improve model transparency. It is particularly suitable for scenarios requiring personalized explanations. Its primary advantage is its ability to capture changes in predictions for each sample and reveal differences between samples, providing deeper insights for model tuning and decision-making. Figure 14 、 Figure 15 、 Figure 16 ICE analysis graphs for original_glrlm_ShortRunEmphasis, original_gldm_DependenceEntropy, and original_shape_Maximum2DDiameterColumn are shown. ICE analysis graphs for the remaining target features are also available, but are not shown in this application due to space limitations.

[0107] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0108] In general, various example embodiments of the present disclosure may be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. When various aspects of the disclosed embodiments are illustrated or described as block diagrams, flow charts, or using some other graphical representation, it will be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0109] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0110] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0111] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0112] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0113] The exemplary embodiments of the present disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will appreciate that various modifications and combinations may be made to these embodiments or their features without departing from the principles and spirit of the present disclosure, and such modifications should fall within the scope of the present disclosure.

Claims

1. A method for predicting the risk of thymic hyperplasia based on CT image features, characterized in that: The method comprises: 101, obtaining CT images of the subject; 102, extracting target features from the CT image, where the target features include any one or more of the following: original_glrlm_ShortRunEmphasis, original_gldm_DependenceEntropy, original_shape_Maximum2DDiameterColumn, original_glrlm_RunVariance, and original_glcm_lmc1; 103, inputting the target feature into a risk prediction model to obtain a score; 104. Output an auxiliary prediction result of the subject's risk of thymic hyperplasia based on the score; when the score is greater than a first threshold, output an auxiliary prediction result indicating that the subject has a high risk of thymic hyperplasia; when the score is less than the first threshold, output an auxiliary prediction result indicating that the subject has a low risk of thymic hyperplasia.

2. The method for predicting the risk of thymic hyperplasia based on CT image features according to claim 1, characterized in that: The target features also include any one or more of the following: original_shape_SurfaceVolumeRatio, original_shape_Flatness, original_shape_Elongation, original_shape_Sphericity, original_glrlm_GrayLevelVariance, and original_ngtdm_Strength.

3. The method for predicting the risk of thymic hyperplasia based on CT image features according to claim 1, characterized in that: The method for constructing the risk prediction model includes: 201, obtaining CT images of training set samples and classification labels including healthy people and thymic hyperplasia people; 202, extracting image features of the thymus region in the CT image; 203, filtering out target features from the image features; 204. Input the target feature into the machine learning model to obtain a predicted classification result, compare it with the classification label, optimize the model according to the comparison result, and obtain a constructed prediction model.

4. The method for predicting the risk of thymic hyperplasia based on CT image features according to claim 3, characterized in that: The method for screening the target features includes: screening features with significant differences between healthy people and thymic hyperplasia people; standardizing the features with significant differences to obtain standardized features; screening from the standardized features features that are not easily affected by batch effects, remove features with high missing rates and low variances, and remove brightness-related features to obtain features after the initial round of screening; and obtaining the target features from the features after the initial round of screening using the LASSO algorithm.

5. The method for predicting the risk of thymic hyperplasia based on CT image features according to claim 3, characterized in that: The image features include any one or more of the following: shape features, first-order statistical features, grayscale co-occurrence matrix features, grayscale run length matrix features, grayscale block size matrix features, grayscale dependency matrix features, and neighborhood grayscale difference matrix features.

6. The method for predicting the risk of thymic hyperplasia based on CT image features according to claim 3, characterized in that: The method used by the machine learning model includes any one or more of the following: Random Forest, Logistic Regression, SVM, XGBoost, CatBoost, LightGBM, Neural Network, Decision Tree, KNN, Naive Bayes; preferably Logistic Regression.

7. The method for predicting the risk of thymic hyperplasia based on CT image features according to claim 3, characterized in that: The step between 201 and 202 further includes performing preprocessing on the CT image including grayscale conversion, normalization, and contrast enhancement to obtain a preprocessed CT image.

8. A computer device, characterized in that: The device comprises: a memory and a processor; the memory is used to store a computer program; and the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images