Vancomycin AUC prediction method and device based on deep learning
Through the deep learning-based TML-ResNet model to process multi-dimensional feature information, the accuracy problem of vancomycin AUC prediction is solved, and more accurate treatment effect and risk assessment are achieved.
Patent Information
- Application Number
- CN202510313762.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-17
- Publication Date
- 2025-08-15
AI Technical Summary
There is a lack of methods for accurately predicting vancomycin AUC in the prior art, resulting in the inability to effectively monitor its accumulation in vivo, affecting the therapeutic effect and risk assessment of adverse reactions.
The vancomycin AUC prediction method based on deep learning is adopted to obtain multi-dimensional feature information and input it into the TML-ResNet model for prediction, including two-dimensional embedded layer, ResNet layer and fully connected layer, numerical and typed features are processed, missing values are filled and pixel size is adjusted according to the importance of the feature, and regression prediction is performed.
It improves the accuracy of vancomycin AUC prediction, can take into account individual differences in patients more comprehensively, and enhances the evaluation of treatment effect and judgment of adverse reaction risks.
Smart Images
Figure CN120496884A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a method and device for predicting vancomycin AUC based on deep learning. Background Art
[0002] Vancomycin is a broad-spectrum glycopeptide antibiotic commonly used to treat serious infections caused by Gram-positive bacteria such as methicillin-resistant Staphylococcus aureus, but its use requires strict blood concentration monitoring to avoid nephrotoxicity and ototoxicity.
[0003] The vancomycin AUC (Area Under the Curve) refers to the area under the plasma concentration-time curve of vancomycin. It measures the extent of drug accumulation in the blood during the process of vancomycin absorption, distribution, metabolism, and excretion. It considers the entire process of drug concentration changes over time, not just the concentration at a single point in time. The AUC is closely correlated with the therapeutic efficacy and adverse reactions of vancomycin. A higher AUC value indicates greater drug exposure in the body, which may increase efficacy but also the risk of adverse reactions. A lower AUC value may not achieve therapeutic effect.
[0004] Therefore, it is of great significance to provide a model that can accurately predict the vancomycin AUC in patients. Summary of the Invention
[0005] The present invention provides a vancomycin AUC prediction method and device based on deep learning to address the defects in the prior art.
[0006] The present invention provides a method for predicting vancomycin AUC based on deep learning, the method comprising: Obtaining multidimensional feature information of the user to be evaluated, the multidimensional feature information including physical sign information, current hospitalization and surgery information, current blood biochemistry test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; Inputting the multidimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
[0007] According to a deep learning-based vancomycin AUC prediction method provided by the present invention, the multidimensional feature information is input into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated, including: For each dimension of the multidimensional feature information, determining an embedding position of the feature information of each dimension in the two-dimensional image; At the embedding position corresponding to the feature information of each dimension, embedding processing is performed on the feature information of each dimension based on the Render function in the two-dimensional embedding layer to obtain a two-dimensional embedded image; The two-dimensional embedded image is input into the ResNet layer for feature extraction, and the vancomycin AUC prediction result of the user to be evaluated is output based on the fully connected layer.
[0008] According to a method for predicting vancomycin AUC based on deep learning provided by the present invention, the method further comprises: Determining missing values in the multidimensional feature information, and embedding the missing values into the two-dimensional image using a preset missing symbol; For feature information other than the missing values in the multidimensional feature information, numerical features are embedded into the two-dimensional image through numerical mapping, and categorical features are embedded into the two-dimensional image through symbol encoding.
[0009] According to a method for predicting vancomycin AUC based on deep learning provided by the present invention, the method further comprises: Determining the feature importance of feature information of each dimension in the multidimensional feature information; Based on the feature importance, determining the pixel size corresponding to the feature information of each dimension; At the embedding position corresponding to the feature information of each dimension, based on the Render function in the two-dimensional embedding layer, the feature information of each dimension is embedded according to the pixel size to obtain a two-dimensional embedded image.
[0010] According to a deep learning-based vancomycin AUC prediction method provided by the present invention, determining the feature importance of feature information of each dimension in the multidimensional feature information includes: Filling missing values in the multidimensional feature information based on the missing feature type to obtain target multidimensional feature information; The target multidimensional feature information is input into a gradient boosting decision tree model to obtain an importance ranking of feature information of each dimension in the target multidimensional feature information.
[0011] The present invention also provides a vancomycin AUC prediction device based on deep learning, comprising: The first vancomycin AUC prediction module is used to obtain multi-dimensional feature information of the user to be evaluated, wherein the multi-dimensional feature information includes human physical sign information, current hospitalization and surgery information, current blood biochemical test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; A second vancomycin AUC prediction module is configured to input the multidimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
[0012] According to the deep learning-based vancomycin AUC prediction device provided by the present invention, the second vancomycin AUC prediction module is further configured to: For each dimension of the multidimensional feature information, determining an embedding position of the feature information of each dimension in the two-dimensional image; At the embedding position corresponding to the feature information of each dimension, embedding processing is performed on the feature information of each dimension based on the Render function in the two-dimensional embedding layer to obtain a two-dimensional embedded image; The two-dimensional embedded image is input into the ResNet layer for feature extraction, and the vancomycin AUC prediction result of the user to be evaluated is output based on the fully connected layer.
[0013] According to the deep learning-based vancomycin AUC prediction device provided by the present invention, the second vancomycin AUC prediction module is further configured to: Determining missing values in the multidimensional feature information, and embedding the missing values into the two-dimensional image using a preset missing symbol; For feature information other than the missing values in the multidimensional feature information, numerical features are embedded into the two-dimensional image through numerical mapping, and categorical features are embedded into the two-dimensional image through symbol encoding.
[0014] According to the deep learning-based vancomycin AUC prediction device provided by the present invention, the second vancomycin AUC prediction module is further configured to: Determining the feature importance of feature information of each dimension in the multidimensional feature information; Based on the feature importance, determining the pixel size corresponding to the feature information of each dimension; At the embedding position corresponding to the feature information of each dimension, based on the Render function in the two-dimensional embedding layer, the feature information of each dimension is embedded according to the pixel size to obtain a two-dimensional embedded image.
[0015] According to the deep learning-based vancomycin AUC prediction device provided by the present invention, the second vancomycin AUC prediction module is further configured to: Filling missing values in the multidimensional feature information based on the missing feature type to obtain target multidimensional feature information; The target multidimensional feature information is input into a gradient boosting decision tree model to obtain an importance ranking of feature information of each dimension in the target multidimensional feature information.
[0016] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the vancomycin AUC prediction method based on deep learning as described above is implemented.
[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for predicting vancomycin AUC based on deep learning as described above is implemented.
[0018] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described methods for predicting vancomycin AUC based on deep learning.
[0019] The vancomycin AUC prediction method and device based on deep learning provided by the present invention obtains multidimensional feature information of the user to be evaluated, including human physical sign information, historical hospitalization and surgery information, historical blood biochemical test information, vancomycin daily dose, vancomycin blood oxygen concentration and concomitant medication information, and inputs this information into a vancomycin clearance prediction model based on TML-ResNet to obtain the vancomycin AUC prediction result for the user to be evaluated. Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying vancomycin AUC labels, and includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer for converting the input data into a two-dimensional embedded image, extracting image features and performing regression prediction. The present invention can more comprehensively consider the individual differences of patients and improve the accuracy of vancomycin AUC prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0021] Figure 1 Schematic diagram of the process of vancomycin AUC prediction method based on deep learning provided by the present invention; Figure 2 Schematic diagram of the structure of the vancomycin AUC prediction device based on deep learning provided by the present invention; Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0022] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.
[0023] It should be noted that, in the description of this application, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device comprising a list of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, article, or device comprising the element. Terms such as "upper" and "lower" indicate positions or location relationships based on those shown in the accompanying drawings and are intended solely for ease of description and simplification of the description. They are not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed, or operate in a specific orientation, and are therefore not to be construed as limiting this application. Unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be interpreted broadly, for example, to mean a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or internal communication between two elements. For those skilled in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0024] The terms "first," "second," and the like in this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that such terms are interchangeable where appropriate, so that embodiments of this application can be implemented in an order other than that illustrated or described herein. Furthermore, the terms "first," "second," and the like generally distinguish objects of a class and do not limit the number of objects; for example, the first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates an "or" relationship between the connected objects.
[0025] Figure 1 Schematic diagram of the process of vancomycin AUC prediction method based on deep learning provided by the present invention, such as Figure 1 As shown, the method includes steps 110 and 120. It should be noted that the execution entity of the deep learning-based vancomycin AUC prediction method provided in the embodiments of the present application can be a server, or other computer devices, such as mobile phones, tablet computers, laptop computers, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs). While the execution entity of the present application is a server for illustrative purposes, the deep learning-based vancomycin AUC prediction method of the present application is not limited to being executed by a server.
[0026] Step 110: Acquire multi-dimensional feature information of the user to be evaluated, wherein the multi-dimensional feature information includes physical sign information, current hospitalization and surgery information, current blood biochemical test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; Here, multidimensional feature information specifically refers to information about an individual in multiple dimensions that can represent individual characteristics and attributes. Specifically, the following data can be extracted from electronic medical records (EMRs), laboratory information systems (LISs), etc.: Physical information: age, weight, height, gender, gestational age, etc. Current history of hospitalization and surgery information: whether admitted to the ICU, whether receiving cardiothoracic surgery, etc.; Current blood biochemistry test information: uric acid, serum creatinine, blood urea nitrogen, glomerular filtration rate, hemoglobin, serum albumin, direct bilirubin, total bilirubin, total protein, procalcitonin, alanine aminotransferase, aspartate aminotransferase, white blood cell count, neutrophil count, lymphocyte count, red blood cell count, hematocrit, platelet count, C-reactive protein, etc.; Historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information. Here, historical concomitant medication information may include carbapenems, third-generation cephalosporins containing enzyme inhibitors or above, quinolones, aminoglycosides, tigecycline, metronidazole, fosfomycin, first- and second-generation cephalosporins, tetracyclines, polymyxins, loop diuretics, vasoactive drugs, etc.
[0027] Furthermore, since the data in the above data sources may include multiple useless dimensional features, i.e., feature information that has little or no impact on vancomycin AUC, in order to improve the accuracy of subsequent processing results, this embodiment can further filter out multidimensional feature information related to vancomycin AUC from the above data sources.
[0028] In some embodiments, a correlation analysis can be performed using vancomycin AUC as the target variable and relevant data in the human sign information, current hospitalization and surgery information, current blood biochemical test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information corresponding to the user to be evaluated as covariates. For example, statistical analysis methods can be used to determine whether there is a correlation between the independent variables by testing the independent variables, including Mann-Whitney U test method, variance test method, chi-square test method, Fish test method, and XGBoost algorithm.
[0029] Specifically, the Mann-Whitney U test method refers to a non-parametric statistical method, which compares the ranks of two groups of data and ultimately calculates a P value to determine whether there is a significant difference between the two groups of data. The variance test method refers to a statistical method for testing whether two or more population variances are equal. The chi-square test method can be specifically used to test whether the degree of deviation between the observed value and the theoretical value is significant, thereby determining whether there is a significant correlation between the two feature variables. The Fish test method refers to a statistical method for analyzing the relationship between two categorical variables, which tests the independence between the two feature variables by comparing the degree of deviation between the observed data and the expected data. The XGBoost algorithm refers to a machine learning algorithm based on a gradient boosting tree, which can specifically perform regression, sorting, and other aspects. In this embodiment, at least one of the above methods can be used for feature screening, without limitation.
[0030] Step 120: Input the multi-dimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; The vancomycin clearance prediction model based on TML-ResNet is trained based on multi-dimensional feature sample information carrying vancomycin AUC labels.
[0031] It should be noted that TML-ResNet (Tabular data Machine Learning - ResidualNetwork) is a deep learning model that combines tabular data (structured data) with residual networks (ResNet). Its core structure mainly includes a two-dimensional embedding layer, a ResNet layer, and a fully connected layer.
[0032] The 2D embedding layer converts the input data into a 2D embedded image. Specifically, for each input multidimensional feature, its features are projected into a 2D embedding space, forming an image. Assuming the tabular data has four feature columns, the feature values of each sample are embedded in a different position in the image. For example, numerical features are written directly as numbers, and categorical features are written as corresponding strings.
[0033] The ResNet layer is used to extract image features from the 2D embedded image. Specifically, a pre-trained ResNet model (such as ResNet-154) is used to extract features from the generated 2D embedded image. The ResNet model uses residual connections to solve the vanishing gradient problem in deep networks, making the network more efficient in learning and training.
[0034] The fully connected layer is used for regression prediction based on the extracted image features. Specifically, the fully connected layer maps the features extracted by the ResNet module to a specific regression target, that is, predicting the AUC value of vancomycin.
[0035] Here, TML-ResNet is trained using multidimensional feature samples labeled with vancomycin AUC. This sample information includes multidimensional feature information from historical patients (such as physical signs, current hospitalization and surgery information, current blood biochemistry test results, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information), and the corresponding vancomycin AUC. The trained TML-ResNet-based vancomycin clearance prediction model can be used to predict vancomycin AUC based on the multidimensional feature information of the user to be evaluated.
[0036] It should be understood that the data processing process in the process of training TML-ResNet using multidimensional feature sample information is the same as the data processing process in the process of using the multidimensional feature information of the user to be evaluated to perform predictions in the vancomycin clearance prediction model based on TML-ResNet below, and will not be described in detail here.
[0037] The deep learning-based vancomycin AUC prediction method proposed in this embodiment obtains multidimensional feature information of the user to be evaluated, including physical signs, historical hospitalization and surgery information, historical blood biochemical test information, daily vancomycin dose, vancomycin blood oxygen concentration, and concomitant medication information. This information is input into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated. The vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying vancomycin AUC labels. It includes a two-dimensional embedding layer, a ResNet layer, and a fully connected layer for converting data into a two-dimensional embedded image, extracting image features, and performing regression prediction. The present invention can more comprehensively consider individual differences among patients and improve the accuracy of vancomycin AUC prediction.
[0038] It should be noted that each implementation method of the present application can be freely combined, the order can be changed, or it can be executed separately, and does not need to rely on or depend on a fixed execution order.
[0039] In some embodiments, the multi-dimensional feature information is input into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated, including: The multi-dimensional feature information is input into the vancomycin clearance prediction model based on TML-ResNet to obtain the vancomycin AUC prediction result of the user to be evaluated, including: For each dimension of the multidimensional feature information, determining an embedding position of the feature information of each dimension in the two-dimensional image; At the embedding position corresponding to the feature information of each dimension, embedding processing is performed on the feature information of each dimension based on the Render function in the two-dimensional embedding layer to obtain a two-dimensional embedded image; The two-dimensional embedded image is input into the ResNet layer for feature extraction, and the vancomycin AUC prediction result of the user to be evaluated is output based on the fully connected layer.
[0040] Here, each feature in the multidimensional feature information (such as age, creatinine clearance, and weight) is mapped to a fixed position in the two-dimensional image, forming a structured layout. It should be understood that each feature occupies a separate block in the image (e.g., divided into a grid), ensuring that the model can clearly distinguish between different features.
[0041] In one example, regions can be assigned based on feature type (numerical, categorical) or feature importance. For example, key physiological indicators (such as renal function parameters) occupy the center of the image, while secondary parameters (such as gender) are assigned to the edge.
[0042] Furthermore, the Render function in the two-dimensional embedding layer can convert the features of each dimension into pixel values in the image, forming a two-dimensional embedded image that can be processed by CNN.
[0043] In one example, for numerical features, the original values can be normalized to the range [0, 255] (e.g., mapping creatinine clearance from 20-120 to grayscale values). These values can then be embedded in a 2D image as numbers or bar charts at designated locations (e.g., "60" can be directly written to the corresponding grid). For categorical features, the categorical values (e.g., "male," "female") can be converted to corresponding color codes and then embedded in the 2D image as the corresponding colors at designated locations.
[0044] Next, the two-dimensional embedded image is fed into the ResNet layer for feature extraction. The ResNet layer uses residual connections and convolution operations to extract features from the image. These features include edges, textures, and shapes, effectively representing multidimensional feature information. The image feature extraction process performed by the ResNet layer is consistent with existing techniques and will not be detailed here.
[0045] Finally, based on the features extracted by the ResNet layer, regression prediction is performed through the fully connected layer to output the vancomycin AUC prediction result for the user to be evaluated.
[0046] In another example, further comprising: Determining missing values in the multidimensional feature information, and embedding the missing values into the two-dimensional image using a preset missing symbol; For feature information other than the missing values in the multidimensional feature information, numerical features are embedded into the two-dimensional image through numerical mapping, and categorical features are embedded into the two-dimensional image through symbol encoding.
[0047] In this embodiment, during the data preprocessing stage, all features are traversed to identify features with missing values. For example, if the value after the "creatinine clearance" field is empty, a unified symbol (such as "NA" or "Missing") is assigned to the missing value and written to a specific location in the image. For example, if "creatinine clearance" is missing, "NA" is marked in the corresponding area of the image. Each feature has a fixed area in the image (such as a grid layout), and the missing symbol can occupy the edge of the area. Furthermore, the font and color of the missing symbol can be unified (such as a gray "NA") to avoid interfering with the model's recognition of other features.
[0048] After encoding missing values, the remaining valid features are encoded differently depending on their type. Specifically, for numeric features, their values are directly converted to the numeric form in the image. For example, the numeric feature "55" can be rendered as the string "55". For categorical features, their values are converted to specific symbols.
[0049] In some embodiments, further comprising: Determining the feature importance of feature information of each dimension in the multidimensional feature information; Based on the feature importance, determining the pixel size corresponding to the feature information of each dimension; At the embedding position corresponding to the feature information of each dimension, based on the Render function in the two-dimensional embedding layer, the feature information of each dimension is embedded according to the pixel size to obtain a two-dimensional embedded image.
[0050] It should be understood that feature importance refers to the relative importance of each feature in the prediction task. Feature importance analysis can determine which features have the greatest impact on the prediction results. In this embodiment, machine learning methods (such as XGBoost and Random Forest) are used to calculate the importance score of each feature. These methods can assess feature importance based on the contribution of the feature to the model's prediction results.
[0051] Next, the pixel size of each feature in the two-dimensional image is determined based on its importance. Features with higher importance are assigned larger pixel sizes, while features with lower importance are assigned smaller pixel sizes. For example, features can be classified into different importance levels based on their importance scores. For example, the feature importance scores can be normalized to a range of 0 to 1, and the pixel size can be determined based on the normalized scores. Higher importance scores are associated with larger pixel sizes.
[0052] Finally, at the embedding position of the feature information of each dimension in the two-dimensional image, the feature information of each dimension is rendered according to the pixel size using the Render function in the two-dimensional embedding layer.
[0053] For example, in one example, the features rendered by the Render function are as follows: ; in, is the feature information of the i-th dimension The coordinate position of the j-th feature in the two-dimensional image, is the jth eigenvalue of the feature information of the i-th dimension, and the Render function normalizes the eigenvalue Norm( )∈[0,1] is converted to RGB pixels, pixel size Positively correlated with feature importance.
[0054] In some embodiments, determining the feature importance of feature information of each dimension in the multi-dimensional feature information includes: Filling missing values in the multidimensional feature information based on the missing feature type to obtain target multidimensional feature information; The target multidimensional feature information is input into a gradient boosting decision tree model to obtain an importance ranking of feature information of each dimension in the target multidimensional feature information.
[0055] Here, missing data is filled based on its type. For example, for numerical features, the mean, median, or mode can be used; for categorical features, the mode or a specific category can be used. This filled data is called the target multidimensional feature information. This target multidimensional feature information can then be input into a gradient boosting decision tree model (such as XGBoost, LightGBM, CatBoost, and other algorithms) to calculate the importance of each feature and obtain a feature importance ranking.
[0056] The deep learning-based vancomycin AUC prediction method proposed in this example uses a two-dimensional embedding layer in the TML-ResNet-based vancomycin clearance prediction model to more accurately predict vancomycin AUC values by converting multidimensional feature information into a two-dimensional embedded image and leveraging the powerful feature extraction capabilities of the ResNet layer. This not only automatically handles categorical data and missing values, but also adapts to datasets of varying sizes, improving prediction accuracy and robustness.
[0057] Based on any of the above embodiments, the present invention further provides a vancomycin AUC prediction device based on deep learning. The vancomycin AUC prediction device based on deep learning described below and the vancomycin AUC prediction method based on deep learning described above can refer to each other. Figure 2 Schematic diagram of the structure of the vancomycin AUC prediction device based on deep learning provided by the present invention, such as Figure 2As shown, it includes: a first vancomycin AUC prediction module 210 and a second vancomycin AUC prediction module 220.
[0058] The first vancomycin AUC prediction module 210 is configured to obtain multi-dimensional feature information of the user to be evaluated, wherein the multi-dimensional feature information includes physical sign information, current hospitalization and surgery information, current blood biochemical test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; The second vancomycin AUC prediction module 220 is configured to input the multi-dimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
[0059] The deep learning-based vancomycin AUC prediction device provided in an embodiment of the present invention obtains multidimensional feature information of the user to be evaluated, including physical signs, historical hospitalization and surgery information, historical blood biochemical test information, daily vancomycin dose, vancomycin blood oxygen concentration, and concomitant medication information. This information is input into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated. The TML-ResNet-based vancomycin clearance prediction model is trained based on multidimensional feature sample information carrying vancomycin AUC labels and includes a two-dimensional embedding layer, a ResNet layer, and a fully connected layer for converting data into a two-dimensional embedded image, extracting image features, and performing regression prediction. The present invention can more comprehensively consider individual differences among patients and improve the accuracy of vancomycin AUC prediction.
[0060] In some embodiments, the second vancomycin AUC prediction module 220 is further configured to: For each dimension of the multidimensional feature information, determining an embedding position of the feature information of each dimension in the two-dimensional image; At the embedding position corresponding to the feature information of each dimension, embedding processing is performed on the feature information of each dimension based on the Render function in the two-dimensional embedding layer to obtain a two-dimensional embedded image; The two-dimensional embedded image is input into the ResNet layer for feature extraction, and the vancomycin AUC prediction result of the user to be evaluated is output based on the fully connected layer.
[0061] In some embodiments, the second vancomycin AUC prediction module 220 is further configured to: Determining missing values in the multidimensional feature information, and embedding the missing values into the two-dimensional image using a preset missing symbol; For feature information other than the missing values in the multidimensional feature information, numerical features are embedded into the two-dimensional image through numerical mapping, and categorical features are embedded into the two-dimensional image through symbol encoding.
[0062] In some embodiments, the second vancomycin AUC prediction module 220 is further configured to: Determining the feature importance of feature information of each dimension in the multidimensional feature information; Based on the feature importance, determining the pixel size corresponding to the feature information of each dimension; At the embedding position corresponding to the feature information of each dimension, based on the Render function in the two-dimensional embedding layer, the feature information of each dimension is embedded according to the pixel size to obtain a two-dimensional embedded image.
[0063] In some embodiments, the second vancomycin AUC prediction module 220 is further configured to: Filling missing values in the multidimensional feature information based on the missing feature type to obtain target multidimensional feature information; The target multidimensional feature information is input into a gradient boosting decision tree model to obtain an importance ranking of feature information of each dimension in the target multidimensional feature information.
[0064] Figure 3 An example of a physical structure diagram of an electronic device is shown below. Figure 3 As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, the communications interface 320, and the memory 330 communicate with each other via the communication bus 340. The processor 310 may call logic instructions in the memory 330 to execute a deep learning-based vancomycin AUC prediction method, which includes: Obtaining multidimensional feature information of the user to be evaluated, the multidimensional feature information including physical sign information, current hospitalization and surgery information, current blood biochemistry test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; Inputting the multidimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
[0065] Furthermore, the logic instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0066] In another aspect, the present invention further provides a computer program product, comprising a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the vancomycin AUC prediction method based on deep learning provided by the above methods, the method comprising: Obtaining multidimensional feature information of the user to be evaluated, the multidimensional feature information including physical sign information, current hospitalization and surgery information, current blood biochemistry test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; Inputting the multidimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
[0067] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the vancomycin AUC prediction method based on deep learning provided by the above methods, the method comprising: Obtaining multidimensional feature information of the user to be evaluated, the multidimensional feature information including physical sign information, current hospitalization and surgery information, current blood biochemistry test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; Inputting the multidimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
[0068] The device embodiments described above are merely illustrative. 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, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.
[0069] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.
[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical coding feature diagrams therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A vancomycin AUC prediction method based on deep learning, characterized in that: The method comprises: Obtaining multidimensional feature information of the user to be evaluated, the multidimensional feature information including physical sign information, current hospitalization and surgery information, current blood biochemistry test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; Inputting the multidimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
2. The method for predicting vancomycin clearance based on deep learning according to claim 1, characterized in that: The multi-dimensional feature information is input into the vancomycin clearance prediction model based on TML-ResNet to obtain the vancomycin AUC prediction result of the user to be evaluated, including: For each dimension of the multidimensional feature information, determining an embedding position of the feature information of each dimension in the two-dimensional image; At the embedding position corresponding to the feature information of each dimension, embedding processing is performed on the feature information of each dimension based on the Render function in the two-dimensional embedding layer to obtain a two-dimensional embedded image; The two-dimensional embedded image is input into the ResNet layer for feature extraction, and the vancomycin AUC prediction result of the user to be evaluated is output based on the fully connected layer.
3. The method for predicting vancomycin clearance based on deep learning according to claim 2, characterized in that: The method further comprises: Determining missing values in the multidimensional feature information, and embedding the missing values into the two-dimensional image using a preset missing symbol; For feature information other than the missing values in the multidimensional feature information, numerical features are embedded into the two-dimensional image through numerical mapping, and categorical features are embedded into the two-dimensional image through symbol encoding.
4. The method for predicting vancomycin clearance based on deep learning according to claim 2, characterized in that: The method further comprises: Determining the feature importance of feature information of each dimension in the multidimensional feature information; Based on the feature importance, determining the pixel size corresponding to the feature information of each dimension; At the embedding position corresponding to the feature information of each dimension, based on the Render function in the two-dimensional embedding layer, the feature information of each dimension is embedded according to the pixel size to obtain a two-dimensional embedded image.
5. The method for predicting vancomycin clearance based on deep learning according to claim 4, characterized in that: Determining the feature importance of feature information of each dimension in the multidimensional feature information includes: Filling missing values in the multidimensional feature information based on the missing feature type to obtain target multidimensional feature information; The target multidimensional feature information is input into a gradient boosting decision tree model to obtain an importance ranking of feature information of each dimension in the target multidimensional feature information.
6. A vancomycin AUC prediction device based on deep learning, characterized in that: The device comprises: The first vancomycin AUC prediction module is used to obtain multi-dimensional feature information of the user to be evaluated, wherein the multi-dimensional feature information includes human physical sign information, current hospitalization and surgery information, current blood biochemical test information, historical vancomycin daily dose, historical vancomycin blood oxygen concentration, and historical concomitant medication information; A second vancomycin AUC prediction module is configured to input the multidimensional feature information into a vancomycin clearance prediction model based on TML-ResNet to obtain a vancomycin AUC prediction result for the user to be evaluated; Among them, the vancomycin clearance prediction model based on TML-ResNet is trained based on multidimensional feature sample information carrying the vancomycin AUC label; the vancomycin clearance prediction model based on TML-ResNet includes a two-dimensional embedding layer, a ResNet layer and a fully connected layer, the two-dimensional embedding layer is used to convert the input data into a two-dimensional embedded image, the ResNet layer is used to extract image features from the two-dimensional embedded image, and the fully connected layer is used to perform regression prediction based on the image features.
7. The vancomycin AUC prediction device based on deep learning according to claim 6, characterized in that: The second vancomycin AUC prediction module is further used to: For each dimension of the multidimensional feature information, determining an embedding position of the feature information of each dimension in the two-dimensional image; At the embedding position corresponding to the feature information of each dimension, embedding processing is performed on the feature information of each dimension based on the Render function in the two-dimensional embedding layer to obtain a two-dimensional embedded image; The two-dimensional embedded image is input into the ResNet layer for feature extraction, and the vancomycin AUC prediction result of the user to be evaluated is output based on the fully connected layer.
8. The vancomycin AUC prediction device based on deep learning according to claim 7, characterized in that: The second vancomycin AUC prediction module is further used to: Determining missing values in the multidimensional feature information, and embedding the missing values into the two-dimensional image using a preset missing symbol; For feature information other than the missing values in the multidimensional feature information, numerical features are embedded into the two-dimensional image through numerical mapping, and categorical features are embedded into the two-dimensional image through symbol encoding.
9. The vancomycin AUC prediction device based on deep learning according to claim 7, characterized in that: The second vancomycin AUC prediction module is further used to: Determining the feature importance of feature information of each dimension in the multidimensional feature information; Based on the feature importance, determining the pixel size corresponding to the feature information of each dimension; At the embedding position corresponding to the feature information of each dimension, based on the Render function in the two-dimensional embedding layer, the feature information of each dimension is embedded according to the pixel size to obtain a two-dimensional embedded image.
10. The vancomycin AUC prediction device based on deep learning according to claim 9, characterized in that: The second vancomycin AUC prediction module is further used to: Filling missing values in the multidimensional feature information based on the missing feature type to obtain target multidimensional feature information; The target multidimensional feature information is input into a gradient boosting decision tree model to obtain an importance ranking of feature information of each dimension in the target multidimensional feature information.