A method for constructing an early diagnosis and typing prediction model for adrenocortical Cushing's syndrome based on radiomics and artificial intelligence and a treatment decision-making system

Abstract

The present application provides a method and system for constructing an early diagnosis and typing prediction model of adrenocortical Cushing's syndrome based on radiomics and artificial intelligence. After preprocessing adrenal CT images, the region of interest is delineated layer by layer and radiomics features are extracted. Based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features, the consistency evaluation of the radiomics features is carried out to determine the radiomics depth features passing the consistency evaluation. An early diagnosis model of adrenocortical Cushing's syndrome features is established according to the radiomics depth features, multi-dimensional indicators of adrenocortical Cushing's syndrome are obtained, and multi-modal fusion processing is carried out based on the adrenocortical Cushing's syndrome feature classification model and the multi-dimensional indicators, so as to obtain an early diagnosis and typing prediction model of adrenocortical Cushing's syndrome. According to the preoperative and postoperative biochemical metabolism indicators and image feature database of patients with adrenocortical Cushing's syndrome, the postoperative metabolic benefit prediction factors of patients are analyzed, and a treatment decision-making system for adrenocortical Cushing's syndrome is constructed.

Description

Technical Field

[0001] The present application relates to the fields of computer vision and imaging image analysis, and particularly relates to a method for constructing an early diagnosis and typing prediction model for adrenocortical Cushing's syndrome based on radiomics and artificial intelligence, and a treatment decision-making system. Background Art

[0002] Cushing's syndrome (CS) is a classic endocrine disease, which is a group of syndromes caused by long-term excessive secretion of cortisol by the adrenal cortex due to various etiologies, resulting in dysfunction of multiple systems and organs throughout the body. Adrenocortical Cushing's syndrome (ACS) is one of the main types of CS and an important cause of secondary hypertension, mainly including adrenocortical adenoma, adrenocortical carcinoma, primary bilateral macronodular adrenal hyperplasia (PBMAH), and primary pigmented nodular adrenocortical disease (PPNAD). The mortality rate of patients with Cushing's syndrome is 4 times higher than that of the normal population. Its most common complications are hypertension, diabetes, osteoporosis, and metabolic syndrome, and it significantly increases the risk of cardiovascular diseases. Currently, in clinical practice, the early screening targets for ACS come from the recognition of early symptoms and signs of patients. However, because most cases of ACS have insidious onset and complex and diverse clinical manifestations, and non-specialist doctors have insufficient understanding of this disease, most patients are not screened and diagnosed in a timely manner. On the other hand, the further screening and diagnosis steps for ACS are cumbersome, mainly relying on repeated hormone tests and various functional tests. Currently, most grass-roots institutions are unable to carry out relevant hormone tests, and they do not understand the specific practices and result judgments of specific screening tests, which further limits the early screening and diagnosis of the disease.

[0003] With the popularization of imaging examinations, the detection of adrenal incidentalomas has gradually become another important detection route for ACS. However, in routine CT examination reports, only basic imaging features of adrenal masses are provided, including size, CT value, etc. Such information has a certain role in determining the nature of adrenal masses, but it cannot make a determination of whether it is functional and the functional type. Existing studies all use a single software to simply analyze CT images and artificially propose a single parameter, without comprehensively considering all imaging features of ACS, and they are all single-center with a small sample size, so they cannot be popularized and applied. Moreover, the research results are only used for diagnosis and cannot propose treatment decisions.

[0004] The foregoing description is for the purpose of providing general background information and does not necessarily constitute prior art. Summary of the Invention

[0005] To solve the above problems, an embodiment of the present application provides a method for constructing an early diagnosis and typing prediction model for adrenocorticotropic Cushing's syndrome based on radiomics and artificial intelligence, and a treatment decision-making system, to achieve early and accurate identification of adrenocorticotropic Cushing's syndrome and provide a decision-making basis for the refined treatment of adrenocorticotropic Cushing's syndrome.

[0006] For this purpose, in one aspect of the present application, a method for constructing an early diagnosis model for adrenocorticotropic Cushing's syndrome based on radiomics and artificial intelligence is provided, including:

[0007] Obtain a plurality of adrenal CT images;

[0008] After preprocessing the adrenal CT images, layer by layer draw regions of interest and extract radiomics features, where the radiomics features include at least one of shape features, density features, texture features, gray-level co-occurrence matrix features, gray-level run-length matrix features, and gray-level size zone matrix features;

[0009] Based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features, conduct a consistency evaluation on the radiomics features to determine the radiomics depth features that pass the consistency evaluation;

[0010] Establish a feature classification model for adrenocorticotropic Cushing's syndrome according to the radiomics depth features;

[0011] Obtain multi-dimensional indicators of adrenocorticotropic Cushing's syndrome, and based on the feature classification model of adrenocorticotropic Cushing's syndrome and the multi-dimensional indicators, conduct multi-modal fusion processing to obtain an early diagnosis prediction model for adrenocorticotropic Cushing's syndrome.

[0012] Optionally, in combination with any of the above aspects, in another implementation manner of this aspect, the method further includes:

[0013] According to the preoperative and postoperative biochemical metabolism indicators and the image feature database of patients with adrenocorticotropic Cushing's syndrome and / or patients with subclinical Cushing's syndrome, use the rule-depth machine learning algorithm to analyze the postoperative metabolic benefit prediction factors of patients, and construct a treatment decision-making system for adrenocorticotropic Cushing's syndrome.

[0014] Optionally, in combination with any of the above aspects, in another implementation manner of this aspect, the preprocessing of the adrenal CT images includes:

[0015] Resample the adrenal CT images into voxels of a preset size by the linear interpolation method, and then normalize the gray levels of all CT images to the range of 0 to 255.

[0016] Optionally, in combination with any of the above aspects, in another implementation of this aspect, the consistency evaluation of the radiomics features based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features includes:

[0017] Determining the intra-class correlation coefficient according to the radiomics features extracted from the regions of interest outlined by the same operator twice successively, and determining the inter-class correlation coefficient according to the radiomics features extracted from the regions of interest outlined by different operators simultaneously;

[0018] In response to the intra-class correlation coefficient and / or the inter-class correlation coefficient being greater than or equal to a preset threshold, determining that the extracted radiomics features pass the consistency evaluation.

[0019] Optionally, in combination with any of the above aspects, in another implementation of this aspect, the determination of the radiomics depth features passing the consistency evaluation further includes:

[0020] Performing an independent samples t-test or Mann-Whitney U-test on the radiomics features of all training groups to exclude meaningless radiomics features;

[0021] Using the LASSO regression algorithm for dimensionality reduction and feature selection through variable selection and regularization, adjusting the regularization parameter according to the optimal principle, and selecting the radiomics features through 10-fold cross-validation;

[0022] Calculating the variance inflation factor for the selected radiomics features and excluding the features with a variance inflation factor greater than 10, thereby obtaining the radiomics depth features.

[0023] Optionally, in combination with any of the above aspects, in another implementation of this aspect, the establishment of an adrenocortical Cushing's syndrome feature classification model based on the radiomics depth features includes:

[0024] Taking the cross-entropy loss function as the loss function of the feature classification model and training the feature classification model using the Adam optimization algorithm.

[0025] Optionally, in combination with any of the above aspects, in another implementation of this aspect, after establishing the adrenocortical Cushing's syndrome feature classification model based on the radiomics depth features, it further includes:

[0026] Optimizing the feature classification model using a visual attention model, where the visual attention model is used to process the attention part of the pixels in the adrenal CT image and extract the region of interest.

[0027] Optionally, in combination with any of the above aspects, in another implementation of this aspect, the obtaining of the multi-dimensional indicators of adrenocortical Cushing's syndrome includes:

[0028] Obtaining the clinical indicators, metabolic indicators, and steroid hormone indicators of adrenocortical Cushing's syndrome based on the diagnostic data of patients with adrenocortical Cushing's syndrome.

[0029] Optionally, in combination with any of the above aspects, in another implementation of this aspect, the multi-modal fusion processing based on the adrenocortical Cushing's syndrome feature classification model and the multi-dimensional indicators includes:

[0030] Establishing a first prediction result based on the feature classification model and a second prediction result based on the multi-dimensional indicators;

[0031] Performing multi-modal fusion processing on the first prediction result and the second prediction result according to at least one of the maximum value fusion or average value fusion algorithms.

[0032] In another aspect of the present application, there is also provided a precise typing prediction model for adrenocortical Cushing's syndrome based on radiomics and artificial intelligence, and the typing prediction model is obtained according to the construction method described in any of the foregoing aspects.

[0033] As described above, the method for constructing an early diagnosis and subtype prediction model of adrenocortical Cushing's syndrome based on radiomics and artificial intelligence in the present application, after preprocessing the adrenal CT images, delineates the regions of interest layer by layer and extracts radiomics features, evaluates the consistency of the radiomics features based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features, determines the radiomics depth features passing the consistency evaluation, establishes a feature classification model for adrenocortical Cushing's syndrome according to the radiomics depth features, obtains multi-dimensional indicators of adrenocortical Cushing's syndrome, and performs multi-modal fusion processing based on the feature classification model for adrenocortical Cushing's syndrome and the multi-dimensional indicators, so as to obtain an early diagnosis and subtype prediction model for adrenocortical Cushing's syndrome. In the above manner, by combining artificial intelligence and radiomics technologies, an intelligent imaging reporting system for early screening of adrenocortical Cushing's syndrome (ACS) is developed, reliable ACS imaging features can be extracted, and a simple, accurate and easy-to-popularize intelligent imaging reporting recognition system for early screening of ACS is developed. At the same time, the image features extracted by radiomics are combined with the multi-dimensional indicators of ACS, relying on a multi-modal fusion artificial intelligence algorithm platform, to conduct early diagnosis research on ACS and establish a precise subtype prediction system, improving the objectivity and accuracy of ACS lesion diagnosis. In addition, according to the preoperative and postoperative biochemical metabolism indicators and imaging feature database of ACS patients, a rule-depth machine learning algorithm is used to analyze the postoperative metabolic benefit prediction factors of patients, and an ACS treatment decision-making system is constructed to provide a decision-making basis for its refined treatment.

[0034] The above summary is provided to introduce in a simplified form some concepts that will be further described in detail in the following detailed description. The above summary is neither intended to identify the key features or essential features of the claimed subject matter, nor is it intended to be used to help determine the scope of the claimed subject matter. The claimed subject matter of this application is not limited to embodiments that solve any or all of the disadvantages noted in the background art. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts. These drawings and the textual description are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments.

[0036] Figure 1Schematic flowchart of the method for constructing an early diagnosis and typing prediction model of adrenocortical Cushing's syndrome based on radiomics and artificial intelligence provided by the embodiments of the present application;

[0037] Figure 2 Schematic diagram for constructing a feature classification model of adrenocortical Cushing's syndrome provided by the embodiments of the present application;

[0038] Figure 3 Schematic diagram for constructing a typing prediction model of adrenocortical Cushing's syndrome provided by the embodiments of the present application;

[0039] Figure 4 Schematic diagram for constructing a treatment decision-making system for adrenocortical Cushing's syndrome provided by the embodiments of the present application. Detailed implementation manners

[0040] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0041] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including that element. It should be further understood that, as used in this article, the singular forms "a", "an" and "the" are also intended to include the plural forms, unless the context indicates otherwise. Furthermore, the terms "or", "and / or", "including at least one of the following" and the like used in this article can be interpreted inclusively, or mean any one or any combination. An exception to this definition will occur only when the combination of elements, functions, steps or operations is inherently mutually exclusive in some way.

[0042] It should be understood that although the terms first, second, third, etc. may be used herein to describe various parameters or modules, these parameters or modules should not be limited to these terms. These terms are only used to distinguish parameters or modules of the same type from each other. For example, without departing from the scope of this document, the first parameter may also be referred to as the second parameter, and similarly, the second parameter may also be referred to as the first parameter. Depending on the context, the words "if", "when" as used herein may be interpreted as "when...", "when...", "in response to determining", or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detected (stated condition or event)" may be interpreted as "when determined", "in response to determining", "when detecting (stated condition or event)", or "in response to detecting (stated condition or event)". In addition, components, features, and elements with the same name in different embodiments of this application may have the same meaning or different meanings, and their specific meanings need to be determined according to their explanations in the specific embodiment or further in combination with the context in the specific embodiment.

[0043] It should be understood that although the steps in the flowcharts in the embodiments of this application are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this document, the execution of these steps has no strict order limit, and they can be executed in other orders. Moreover, at least a part of the steps in the figure may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0044] It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit the scope of the rights of this application.

[0045] Artificial intelligence technologies represented by radiomics and deep learning are increasingly widely used in the medical field and have shown excellent performance in disease recognition, lesion classification, and other fields. Radiomics can obtain more abundant quantitative features that are difficult to extract by the human eye by making full use of the digital information of imaging images, providing important value for clinical precision treatment and comprehensive quantitative evaluation. Artificial intelligence algorithms are good at finding features, building models, and finally making predictions in large data sets, and have more advantages when facing problems with many features and large amounts of data. However, current research on adrenocorticotropic hormone-dependent Cushing's syndrome (ACS) is still limited to the recognition of single imaging features such as adrenal adenoma lesions and changes in the diameter and volume of the contralateral adrenal gland, and there is no report on the research of deeply extracting the imaging information of ACS and combining it with other relevant clinical data.

[0046] Please refer to Figure 1 , which shows the process of a method for constructing an early diagnosis and typing prediction model of adrenocortical Cushing's syndrome based on radiomics and artificial intelligence. Specifically, it includes the following steps:

[0047] S1: Obtain multiple adrenal CT images;

[0048] As Figure 2 shown, all adrenal CT images in the CT examination database of the hospital can be screened as the source of training data.

[0049] S2: After preprocessing the adrenal CT images, layer by layer draw the region of interest and extract radiomics features, where the radiomics features include at least one of shape features, density features, texture features, gray-level co-occurrence matrix features, gray-level run-length matrix features, and gray-level size zone matrix features;

[0050] Among them, preferably, the preprocessing of the adrenal CT images can be achieved through the following steps: Resample the adrenal CT images into voxels of a preset size by the linear interpolation method, and then normalize the gray levels of all CT images to the range of 0 to 255.

[0051] Specifically, the AK software (Artificial Intelligence Kit) can be used to preprocess all CT images. The images are resampled into voxels of 1×1×1mm by linear interpolation, and then the gray levels of all CT images are normalized to 0 to 255. Resampling and gray-level normalization are performed to eliminate the heterogeneity of CT scan parameters. Then, region of interest (ROI) segmentation and radiological feature extraction are carried out.

[0052] ROI drawing uses the ITK-SNAP software to manually draw each layer of the adenoma, the adrenal tissue on the diseased side and the contralateral side layer by layer to create a three-dimensional region. After matching the ROI with the corresponding CT images in the AK software, the AK software is used to extract radiomics features. In patients with ACS, due to the autonomous secretion of excessive corticosteroids by adrenal tumors, the level of adrenocorticotropic hormone (ACTH) is chronically suppressed, resulting in atrophy of the normal adrenal glands on the diseased side and the contralateral side except for the tumor. It shows unique imaging features compared with non-functional adrenal adenomas or other functional adrenal tumors. The features extracted by radiomics include shape features, density features, texture features, gray-level co-occurrence matrix (GLCM) features, gray-level run-length matrix (RLM) features, and gray-level size zone matrix (GLSZM) features. For feature selection, according to the evaluation criteria, a suitable subset is directly selected from the feature set, or the original features are linearly / non-linearly combined to generate a new feature set, and then a suitable subset is selected from the feature set.

[0053] S3: Evaluate the consistency of the radiomics features based on the intra-class correlation coefficient (ICC) and inter-class correlation coefficient of the radiomics features, and determine the radiomics depth features that pass the consistency evaluation;

[0054] Preferably, the evaluating the consistency of the radiomics features based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features includes:

[0055] Determine the intra-class correlation coefficient according to the radiomics features extracted from the regions of interest (ROIs) outlined by the same operator twice successively, and determine the inter-class correlation coefficient according to the radiomics features extracted from the ROIs outlined by different operators simultaneously;

[0056] In response to the intra-class correlation coefficient and / or the inter-class correlation coefficient being greater than or equal to a preset threshold, determine that the extracted radiomics features pass the consistency evaluation.

[0057] Specifically, randomly select 20 CT images, and ask radiologist A and radiologist B to outline the ROIs, where the ROIs are adrenal adenomas in the CT images. Four weeks later, radiologist A outlines the ROIs of these 20 CT images again. Perform the intra-observer ICC test on the radiomics feature data extracted from the ROIs outlined by radiologist A twice successively, and perform the inter-observer ICC test on the radiomics feature data extracted from the ROIs outlined by radiologist A and radiologist B. An ICC value greater than 0.75 indicates no difference in the measurements between the two, and the consistency evaluation is passed. Exclude the radiomics feature variables with an intra-class ICC value and / or an inter-class ICC value less than 0.75. The remaining radiomics feature variables are considered to pass the consistency evaluation and can be included in the construction of the model. Finally, radiologist A completes the outlining of all the CT image ROIs.

[0058] S4: Establish a feature classification model for adrenocortical Cushing's syndrome based on the radiomics depth features.

[0059] Preferably, the determining the radiomics depth features that pass the consistency evaluation further includes:

[0060] Perform an independent samples t-test or Mann-Whitney U test on the radiomics features of all training groups to exclude meaningless radiomics features;

[0061] Use the LASSO regression algorithm for dimensionality reduction and feature selection through variable selection and regularization, adjust the regularization parameter according to the optimal principle, and select the radiomics features with 10-fold cross-validation;

[0062] Calculate the variance inflation factor for the selected radiomics features, and exclude the features with a variance inflation factor greater than 10, so as to obtain the radiomics depth features.

[0063] Specifically, the extracted features are normalized by z-score for stable analysis. The specific feature screening process is as follows: First, exclude the radiomics features with an intra-group ICC value and / or an inter-group ICC value less than 0.75; Second, perform an independent samples t-test or Mann-Whitney U test on all the radiomics features of the adrenal mass in the training group, and exclude those with no significance; Third, use LASSO regression for dimensionality reduction and feature selection through variable selection and regularization to improve the prediction accuracy and generate an interpretable statistical model, adjust the regularization parameter (λ) according to the optimal principle, and select features by 10-fold cross-validation; Finally, calculate the variance inflation factor (VIF) for the LASSO regression and the selected features, and exclude the features with VIF>10 to avoid potential severe collinearity. Finally, use Logistic regression to linearly combine the obtained imaging features and calculate the radiomics score (Rad-Score).

[0064] The establishment of an adrenal Cushing's syndrome feature classification model based on the radiomics depth features includes:

[0065] Use the cross-entropy loss function as the loss function of the feature classification model, and train the feature classification model using the Adam optimization algorithm.

[0066] Specifically, after obtaining the image features, we send them into a classifier composed of a pooling layer and a fully connected layer to obtain the final classification result. We use the commonly used cross-entropy loss in the classification model as the loss function of the model. To fully train the model, we select the Adam optimization algorithm as the optimizer for model training. The Adam optimizer combines the advantages of the AdaGrad and RMSProp optimization algorithms, comprehensively considers the first-order moment estimation and second-order moment estimation of the gradient, and calculates the update step size.

[0067] S5: Obtain multi-dimensional indicators of adrenal Cushing's syndrome, and perform multi-modal fusion processing based on the adrenal Cushing's syndrome feature classification model and the multi-dimensional indicators to obtain an early diagnosis and typing prediction model for adrenal Cushing's syndrome.

[0068] Among them, preferably, the obtaining of the multi-dimensional indicators of adrenal Cushing's syndrome includes:

[0069] Obtain the clinical indicators, metabolic indicators, and steroid hormone indicators of adrenal Cushing's syndrome according to the diagnostic data of patients with adrenal Cushing's syndrome.

[0070] Such as Figure 3As shown in the figure, considering that there are multiple subtypes of adrenocorticotropic Cushing's syndrome (ACS) and different treatment methods for different subtypes, we deeply explored the subtype characteristics of ACS patients. We combined the image features extracted from the developed adrenocorticotropic Cushing's syndrome feature classification model with multi-dimensional indicators of ACS (such as one or more of clinical indicators, metabolic indicators, and steroid hormone indicators), and relied on a multi-modal fusion artificial intelligence algorithm platform to establish an early diagnosis and subtype prediction model for ACS.

[0071] Furthermore, the multi-modal fusion processing based on the adrenocorticotropic Cushing's syndrome feature diagnosis model and the multi-dimensional indicators includes:

[0072] Establishing a first prediction result based on the feature diagnosis and subtype model and a second prediction result based on the multi-dimensional indicators;

[0073] Performing multi-modal fusion processing on the first prediction result and the second prediction result according to at least one of the maximum value fusion or average value fusion algorithms.

[0074] Specifically, the multi-modal model consists of two branches, namely Figure 3 The intelligent imaging system for the diagnosis of ACS based on adrenal images obtained in [reference] and multi-dimensional clinical indicators, steroid hormones, and metabolic indicators respectively obtain two prediction results, and then the final prediction result is obtained through methods such as maximum value fusion (max-fusion) and / or average value fusion (averaged-fusion). The previously trained model is used as the initial parameter of the multi-modal model, and then the entire model is trained.

[0075] In this embodiment, a method for constructing an early diagnosis and classification prediction model of adrenocortical Cushing's syndrome based on radiomics and artificial intelligence. After preprocessing the adrenal CT images, the region of interest is delineated layer by layer and radiomics features are extracted. Based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features, the consistency of the radiomics features is evaluated, and the radiomics depth features passing the consistency evaluation are determined. According to the radiomics depth features, a feature classification model of adrenocortical Cushing's syndrome is established, and multi-dimensional indicators of adrenocortical Cushing's syndrome are obtained. Based on the feature classification model of adrenocortical Cushing's syndrome and the multi-dimensional indicators, multi-modal fusion processing is performed to obtain a diagnosis and classification prediction model of adrenocortical Cushing's syndrome. In the above manner, by combining artificial intelligence and radiomics technology, an intelligent imaging reporting system for early screening of adrenocortical Cushing's syndrome (ACS) is developed, which can extract reliable ACS imaging features, and a simple, accurate and easy-to-popularize intelligent imaging reporting recognition system for early screening of ACS is developed. At the same time, the image features extracted by radiomics are combined with the multi-dimensional indicators of ACS, and relying on a multi-modal fusion artificial intelligence algorithm platform, early diagnosis research of ACS is carried out and a precise classification prediction system is established to improve the objectivity and accuracy of ACS lesion diagnosis.

[0076] On the basis of the above embodiment, the diagnosis and classification prediction model of adrenocortical Cushing's syndrome can also be optimized, and the following optional methods can be used to implement it.

[0077] As an optional implementation method, increase the data volume for model training. Although the imaging database established in the early stage of the hospital is relatively complete, there are still problems of insufficient data volume and uneven sample distribution in the image data set. Therefore, the data volume for modeling can be further increased to fundamentally improve the generalization ability of the model and improve the usability of the model in primary hospitals.

[0078] As another optional implementation method, enhance the data in deep learning. Deep neural networks perform well in many tasks, but these networks usually require a large amount of data to avoid overfitting. Unfortunately, a large amount of data cannot be obtained in many scenarios, such as medical image analysis. A large amount of medical data can be obtained through one or more of data augmentation methods such as geometric transformation, color transformation, noise injection, random erasing, scaling transformation, moving, and flipping transformation, to increase the size and quality of the training data set and build a stable deep learning model.

[0079] As yet another optional implementation method, a visual attention model is used to optimize the feature classification model, and the visual attention model is used to process the attention part pixels in the adrenal CT image and extract the region of interest.

[0080] Specifically, a visual attention model refers to a model that uses a computer to simulate the human visual attention system and extracts the attention-grabbing foci that can be observed by the human eye in an image. We added an attention mechanism to the basic feature extraction module to learn the parts of the current image that require human attention for observation and processing. Each time, the network processes the pixels in the attention part based on the position to be focused on learned from the current state, rather than all the pixels of the image. Through the attention model mechanism, the image parts of the regions of interest, such as lesions and atrophied adrenal glands, can be extracted from the collected adrenal gland images.

[0081] Furthermore, after building the model, the following method can be used for model verification: After preprocessing and annotating the collected external imaging data, extract the deep features of the adrenal gland images, and verify according to the established model above to evaluate parameters such as discrimination, calibration, and the DCA curve, so as to verify the effectiveness of the existing model and give a targeted improvement training direction.

[0082] Furthermore, as mentioned above, after judging the ACS classification according to the model, subsequent treatment strategies can be formulated. In this application, a treatment assistance decision-making system for ACS will be further explored to provide a decision-making basis for the refined treatment of adrenal Cushing's syndrome. Specifically, as Figure 4 shown, the method further includes the following steps:

[0083] S6: According to the preoperative and postoperative biochemical metabolism indexes and the imaging feature database of patients with adrenal Cushing's syndrome and / or subclinical Cushing's syndrome, use a rule-depth machine learning algorithm to analyze the postoperative metabolic benefit prediction factors of the patients and construct a treatment decision-making system for adrenal Cushing's syndrome.

[0084] Specifically, based on the previously established complete follow-up database, according to the preoperative biochemical metabolism indexes, steroid hormones, imaging features of ACS patients (mainly SCS) and the changes in postoperative steroid hormones and metabolic indexes, use a rule-depth machine learning algorithm to analyze the postoperative metabolic benefit prediction factors of ACS (mainly SCS), and construct an ACS treatment decision-making system in combination with domestic and foreign guidelines and multidisciplinary management. This system can comprehensively consider the patient's comorbidities such as glucose and lipid metabolism, hypertension, osteoporosis, and heart, as well as the postoperative benefit probability, and determine the treatment method to achieve refined treatment. This integrated ACS diagnosis classification and decision-making system significantly improves the detection rate of ACS, simplifies the ACS diagnosis process, significantly saves medical costs, and realizes precise treatment.

[0085] On the other hand, the present application also provides a classification diagnosis model for adrenocortical Cushing's syndrome based on radiomics and artificial intelligence, and the classification diagnosis model is obtained according to the construction method described in any of the foregoing aspects. It has the same technical features as the construction method of the classification diagnosis model for adrenocortical Cushing's syndrome based on radiomics and artificial intelligence provided in the above embodiments, so it can also solve the same technical problems and achieve the same technical effects, which will not be elaborated herein.

[0086] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0087] In the present application, for the description of the same or similar term concepts, technical solutions and / or application scenarios, generally only the first occurrence is described in detail. When it appears repeatedly later, for the sake of brevity, it is generally not described again. When understanding the technical solutions and other contents of the present application, for the same or similar term concepts, technical solutions and / or application scenarios that are not described in detail later, reference may be made to the relevant detailed descriptions before.

[0088] In the present application, the descriptions of the various embodiments have their own emphases. For the parts not described or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0089] The technical features of the technical solutions of the present application can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present application.

[0090] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the essence of the technical solution of the present application or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above and includes several instructions for causing a terminal device to execute the methods of each embodiment of the present application.

[0091] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the contents of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A method for constructing an early diagnosis, typing prediction model and treatment decision-making system for adrenocortical Cushing's syndrome based on radiomics and artificial intelligence, characterized in that, it includes: Obtain multiple adrenal CT images; After preprocessing the adrenal CT images, layer by layer delineate the region of interest and extract radiomics features, and the radiomics features include shape features, density features, texture features, gray-level co-occurrence matrix features, gray-level run-length matrix features and gray-level size zone matrix features; Based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features, conduct consistency evaluation on the radiomics features to determine the radiomics depth features that pass the consistency evaluation; Establish a feature classification model for adrenocortical Cushing's syndrome according to the radiomics depth features, and optimize the feature classification model by using a visual attention model, and the visual attention model is used to process the attention part pixels in the adrenal CT image and extract the region of interest; Obtain multi-dimensional indicators of adrenocortical Cushing's syndrome, and perform multi-modal fusion processing based on the adrenocortical Cushing's syndrome feature classification model and the multi-dimensional indicators to obtain an early diagnosis and typing prediction model for adrenocortical Cushing's syndrome; According to the preoperative and postoperative biochemical metabolism indicators and image feature database of patients with adrenocortical Cushing's syndrome and / or subclinical Cushing's syndrome, use the rule-depth machine learning algorithm to analyze the postoperative metabolic benefit prediction factors of patients, and construct an adrenocortical Cushing's syndrome treatment decision-making system; Among them, the multi-modal fusion processing based on the adrenocortical Cushing's syndrome feature classification model and the multi-dimensional indicators includes: Establish a first prediction result based on the feature classification model and a second prediction result based on the multi-dimensional indicators; Perform multi-modal fusion processing on the first prediction result and the second prediction result according to at least one of the maximum value fusion or average value fusion algorithms.

2. The method according to claim 1, characterized in that, The preprocessing of the adrenal CT image includes: Resample the adrenal CT image into voxels of a preset size by the linear interpolation method, and then normalize the gray levels of all CT images to the range of 0 to 255.

3. The method according to claim 1, characterized in that, The consistency evaluation of the radiomics features based on the intra-class correlation coefficient and inter-class correlation coefficient of the radiomics features includes: Determine the intra-class correlation coefficient according to the radiomics features extracted from the regions of interest delineated by the same processor twice successively, and determine the inter-class correlation coefficient according to the radiomics features extracted from the regions of interest delineated by different processors simultaneously; In response to the intra-class correlation coefficient and / or the inter-class correlation coefficient being greater than or equal to a preset threshold, determine that the extracted radiomics features pass the consistency evaluation.

4. The method according to claim 3, characterized in that, The determination of the radiomics depth features that pass the consistency evaluation further includes: Independent sample t-tests or Mann-Whitney U tests were performed on the radiomics features of all training groups to exclude meaningless radiomics features; The LASSO regression algorithm was used for dimensionality reduction and feature selection through variable selection and regularization. The regularization parameter was adjusted according to the optimal principle, and radiomics features were selected by 10-fold cross-validation; The variance inflation factor was calculated for the selected radiomics features, and features with a variance inflation factor greater than 10 were excluded to obtain the radiomics depth features.

5. The method according to claim 1, wherein, The establishment of an adrenocorticotropic Cushing's syndrome feature classification model based on the radiomics depth features includes: Taking the cross-entropy loss function as the loss function of the feature classification model, and training the feature classification model using the Adam optimization algorithm.

6. The method according to claim 1, wherein, The acquisition of multi-dimensional indicators of adrenocorticotropic Cushing's syndrome includes: Clinical indicators, metabolic indicators, and steroid hormone indicators of adrenocorticotropic Cushing's syndrome were obtained based on the diagnostic data of patients with adrenocorticotropic Cushing's syndrome.

7. A construction system for an early diagnosis, typing prediction model, and treatment decision-making system of adrenocorticotropic Cushing's syndrome based on radiomics and artificial intelligence, wherein, The construction system is obtained according to the construction method described in any one of claims 1-6.

Images