A biomarker-based auxiliary diagnosis method for pleural effusion

Abstract

The present invention relates to the technical field of pleural effusion diagnosis, and discloses a biomarker-based auxiliary diagnosis method for pleural effusion. Biomarkers for auxiliary diagnosis of pleural effusion are screened; data processing is performed on the biomarkers to obtain standardized data of each biomarker; based on the standardized data of each biomarker, the weight of each biomarker in the disease to which the pleural effusion belongs is calculated; the standardized data of each biomarker and the weight of each biomarker in the disease to which the pleural effusion belongs are input into a trained auxiliary diagnosis model for pleural effusion to determine the disease to which the pleural effusion belongs; the number of hidden layers in the auxiliary diagnosis model for pleural effusion varies with the number of types of biomarkers input into the auxiliary diagnosis model for pleural effusion. The method can quickly and accurately diagnose pleural effusion, thereby improving the reliability of diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of pleural effusion diagnosis, and in particular to a biomarker-based auxiliary diagnosis method for pleural effusion. Background Art

[0002] Pleural effusions are abnormal collections of fluid within the pleural area, indicating an imbalance between the production and clearance of exudates. Excessive pleural effusions are a symptom of the underlying condition, not a sign of it. Cardiac, respiratory, and systemic diseases have all been associated with pleural effusions. Therefore, in addition to seeing a pulmonologist, patients with pericardial effusions may also see a rheumatologist, oncologist, thoracic surgeon, and hematologist. For optimal treatment, the cause of the pleural effusion must be identified.

[0003] Due to the complexity and diversity of pleural effusions, a single biomarker often fails to fully and accurately reflect their etiology. Etiology diagnosis is typically based on biomarkers that reflect the clinical features, morphology, and cellular biological properties of the pleural effusion. These biomarkers can be derived from the pleural effusion itself, such as proteins, enzymes, cytokines, and tumor markers, or they can indirectly reflect the status of the pleural effusion through testing in samples such as the patient's blood or urine.

[0004] As the neural network model becomes more sophisticated, it can be used to diagnose pleural effusion. However, due to the large number of biomarkers required for diagnosing pleural effusion, the following problems exist when using the neural network model to diagnose and predict pleural effusion:

[0005] 1. When a large number of biomarkers are used as input to a neural network model, it puts a lot of pressure on the accuracy of the model's feature learning, reducing the accuracy of predictions.

[0006] 2. Using a neural network model with a fixed number of layers for different types and numbers of biomarkers wastes model computing power and reduces prediction efficiency.

[0007] Therefore, there is an urgent need for a biomarker-based auxiliary diagnosis method for pleural effusion that can accurately diagnose pleural effusion and improve the reliability and prediction efficiency of the diagnostic results. Summary of the Invention

[0008] In order to solve the above technical problems, the present invention provides a biomarker-based auxiliary diagnosis method for pleural effusion, comprising the following steps:

[0009] S1: Screening of biomarkers for auxiliary diagnosis of pleural effusion;

[0010] S2: performing data processing on the biomarkers respectively to obtain standardized data of each biomarker;

[0011] S3: Calculating the weight of each biomarker in the disease to which pleural effusion belongs based on the standardized data of each biomarker;

[0012] S4: inputting the standardized data of each biomarker and the weight of each biomarker in the disease to which the pleural effusion belongs into the trained pleural effusion auxiliary diagnosis model to determine the disease to which the pleural effusion belongs;

[0013] The number of hidden layers in the pleural effusion auxiliary diagnosis model is determined by the following formula:

[0014] h=ka×log(1+b×n);

[0015] Where h represents the number of hidden layers in the pleural effusion auxiliary diagnosis model; n represents the number of biomarker types, k is the first constant, a is the second constant, and b is the third constant.

[0016] As an implementation method of the present application, the pleural effusion auxiliary diagnosis model is a convolutional neural network, including feature encoding, feature enhancement and feature decoding;

[0017] Among them, the attention module is used for feature enhancement, and the calculation formula of the attention module is:

[0018] ;

[0019] ;

[0020] in, represents the element multiplier, F represents the input feature map, (F) and (A C ) are channel attention module and spatial attention module respectively, The attention map generated for the channel attention module, Attention maps generated for the spatial attention module;

[0021] In addition, the residual is used to jump the link of the feature, specifically:

[0022] The residual block is formed by the filter;

[0023] Use the filter to connect with the residual, and then multiply it with the previous layer feature corresponding to the residual to obtain the deep feature corresponding to the residual;

[0024] The semantic category corresponding to the deep feature is determined by multiplying the deep feature and a preset convolution kernel.

[0025] As an implementation method of this application, the calculation process of the channel attention module is:

[0026] ;

[0027] in, Represents the sigmoid activation function, MLP represents the dimensionality reduction operation of the multilayer perceptron; max-pool represents the maximum pooling operation, and avg-pool represents the average pooling operation.

[0028] As an implementation method of this application, the calculation process of the spatial attention module is:

[0029] ;

[0030] in Represents a convolution operation.

[0031] As one implementation of the present application, the biomarkers include: adenosine deaminase, lactate dehydrogenase, total protein, glucose, carcinoembryonic antigen, cytokeratin 19 fragment, body mass index, neuron-specific enolase, white blood cell count, monocyte count, multinuclear cell count, mesothelial cell count, specific gravity, pH value, color, gender, age, smoking status, body temperature, pulse rate, respiratory rate, blood pressure, and medical history.

[0032] As an implementation method of the present application, in S2, data processing is performed on the biomarkers respectively to obtain the standardized data of each biomarker:

[0033] If the biomarker value If is a positive indicator,

[0034] ;

[0035] If the biomarker value If it is a negative indicator,

[0036] ;

[0037] in, is the normalized data of biomarkers, max{ } represents the maximum value of the jth biomarker, min{ } represents the minimum value of the jth biomarker.

[0038] As an implementation method of the present application, the calculation of the weight of each biomarker in the disease to which pleural effusion belongs in S3 based on the standardized data of each biomarker includes the following steps:

[0039] S31: Calculate the information entropy of each biomarker using the entropy weight method based on the standardized data of each biomarker. The calculation formula is:

[0040] ;

[0041] Where n is the number of biomarker types, E j is the information entropy of the jth biomarker;

[0042] S32: Based on the information entropy of each biomarker, the weight of each biomarker in diagnosing the disease to which pleural effusion belongs is calculated using the following formula:

[0043] , and the sum of the weights of each biomarker in the same pleural effusion disease is 1;

[0044] Among them, W j is the weight of the jth biomarker in diagnosing the disease to which pleural effusion belongs.

[0045] As an implementation method of the present application, the number of hidden layers of the pleural effusion auxiliary diagnosis model is 30.

[0046] As an implementation method of the present application, when training the pleural effusion auxiliary diagnosis model, 80% of the sample data are randomly selected as the training set, and the remaining sample data are used as the test set.

[0047] As an implementation method of the present application, the diseases to which the pleural effusion belongs determined by the pleural effusion auxiliary diagnosis model are: malignant tumor combined with pleural effusion, parapulmonary effusion and congestive heart failure.

[0048] The embodiments of the present invention have the following technical effects:

[0049] 1. In this application, pleural effusion is diagnosed based on biomarkers using a pleural effusion auxiliary diagnosis model. The selected pleural effusion auxiliary diagnosis model is a convolutional neural network, including feature encoding, feature enhancement, and feature decoding. Among them, the attention module is used during feature enhancement to focus the training of the pleural effusion auxiliary diagnosis model on key areas and improve the accuracy of the model. The residual is used for feature jump links to improve the compatibility of feature learning, enhance the accuracy of the pleural effusion auxiliary diagnosis model, and further improve the sensitivity and specificity of pleural effusion auxiliary diagnosis.

[0050] 2. Based on the number of types of biomarkers, the number of hidden layers in the pleural effusion auxiliary diagnosis model is determined, so that the number of hidden layers varies with the number of types of biomarkers input into the pleural effusion auxiliary diagnosis model, thereby avoiding the problem of wasted computing power and low prediction efficiency caused by using a network model with a fixed number of hidden layers for different types and numbers of biomarkers.

[0051] 3. In this application, the weights of each biomarker in diagnosing the disease associated with pleural effusion, obtained using the entropy weight method, are input into a trained pleural effusion auxiliary diagnosis model along with the standardized data for each biomarker to determine the disease associated with the pleural effusion. This takes into account the possibility that different types of pleural effusion-associated diseases may manifest differently on the same biomarker, further improving the reliability of the pleural effusion auxiliary diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific 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.

[0053] Figure 1 This is a flow chart of a biomarker-based auxiliary diagnosis method for pleural effusion provided by an embodiment of the present invention;

[0054] Figure 2 It is a structural schematic diagram of a pleural effusion auxiliary diagnosis model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are also within the scope of protection of the present invention.

[0056] The detection of biomarkers in pleural effusion can be used for clinical auxiliary diagnosis of related diseases. How to screen pleural effusion biomarkers with high specificity and sensitivity and improve the diagnostic efficiency of concurrent pleural effusion-related diseases has always been a hot topic and difficulty in research in this field. This application adopts a deep learning method to introduce artificial intelligence analysis into the clinical examination of pleural effusion diseases, aiming to study the biological characteristics of pleural effusion from a multidimensional perspective, select auxiliary clinical diagnostic indicators, and evaluate the clinical value of the screened biomarkers in auxiliary clinical diagnosis.

[0057] Specifically, the present invention provides a biomarker-based auxiliary diagnosis method for pleural effusion, such as Figure 1 As shown, the method includes the following steps:

[0058] S1: Screening of biomarkers for auxiliary diagnosis of pleural effusion;

[0059] In clinical practice, there are many biomarkers used to diagnose diseases such as pleural effusion. ROC curve analysis can be used to sort the biomarkers according to the AUC values, and the biomarkers with large AUC values ​​can be selected to accurately identify biomarkers sensitive to pleural effusion, avoiding errors caused by manual screening and reliance on clinical experience.

[0060] As one implementation of the present application, the biomarkers include: adenosine deaminase, lactate dehydrogenase, total protein, glucose, carcinoembryonic antigen, cytokeratin 19 fragment, body mass index, neuron-specific enolase, white blood cell count, monocyte count, multinuclear cell count, mesothelial cell count, specific gravity, pH value, color, gender, age, smoking status, body temperature, pulse rate, respiratory rate, blood pressure, and medical history.

[0061] The inventors collected data from 286 patients with pleural effusions at the Second Hospital of Tianjin Medical University between January 2021 and January 2023. Based on the treatment characteristics and clinical diagnosis and treatment guidelines for patients with concurrent pleural effusions, the three groups of diseases with concurrent pleural effusions were selected: malignant tumors with pleural effusions (MPE) (189 patients); parapulmonary effusions (PPE) (52 patients); and congestive heart failure (CHF) (45 patients). For the included subjects, medical history, symptoms and signs, electrocardiograms (ECGs), chest ultrasounds, and chest X-rays were collected. Laboratory tests also included white blood cell (WBC) and red blood cell (RBB) counts, as well as the specific density (RBB) and RBB count of the pleural effusion. (ECG), chest ultrasound and chest X-ray imaging examinations; laboratory tests are also performed: the number of white blood cells (WBC) and red blood cells (RBC) in the blood; the density of pleural effusion; the levels of four biomarkers such as Ada, TP, Glu and LDH are measured by electrochemiluminescence; the levels of three tumor markers such as CEA, Cyrfa21-1 and NSE are measured by immunochemiluminescence.

[0062] Based on the diagnostic and treatment guidelines for combined pleural effusions with malignant tumors, parapulmonary effusions, and congestive heart failure, general clinical data were collected from the subjects, including body temperature, pulse rate, respiratory rate, and systolic blood pressure. The results showed no statistically significant differences in these parameters among the three groups (p>0.05).

[0063] An analysis of 23 biomarkers in the blood and pleural effusion of the above-mentioned research subjects showed that the differences in six biomarkers, namely TP, ADA, CEA, CYFRA211, NSE, and MNC%, among the three groups of pleural effusion diseases were statistically significant (P<0.05), while the differences in three physical characteristics of pleural effusion, namely color, transparency, and specific gravity, and the other six biomarkers (including WBC, PNC%, MTC%, pH level, GLU, and LDH) were not statistically significant (p>0.05). Furthermore, the ROC curve was used to analyze the clinical value of the indexes in assisting the diagnosis of the above three types of diseases. The results showed that the AUC values ​​of each index were: CEA (0.778), CYFRA21-1 (0.768), TP (0.737), NSE (0.718), SG (0.694), WBC (0.666), MTC % (0.614), ADA (0.566), pH (0.469), MNC % (0.487), GLU (0.543), PNC % (0.455), and MTC % (0.443). Further analysis of pleural effusions in patients with congestive heart failure revealed the following area under the receiver operating characteristic (ROC) curve analysis: pH (0.768), ADA (0.702), TP (0.772), LDH (0.693), CEA (0.784), CYRFA21-1 (0.731), NSE (0.765), WBC (0.645), MNC % (0.701), PNC % (0.564), and MTC % (0.664). The Shapiro-Wilk normality test was used to verify normality. The distribution of biomarkers in pleural effusions is presented as median and interquartile range. For categorical variables, the chi-square test and Fisher's exact test were used for comparisons between groups; for continuous variables, the nonparametric Kruskal-Wallis test was used (because variables were not normally distributed). If the Kruskal-Wallis test result is significant, a post hoc analysis can be performed using the Dunn test. To determine the clinical utility of the biomarker in the study group and to assess its performance in distinguishing pleural effusions, we reviewed the marker values ​​using receiver operating characteristic (ROC) analysis. A P value < 0.05 indicates statistical significance. An ROC plot summarizes the performance of a curve classifier at all possible thresholds and is a commonly used graph. When varying the threshold for assigning observations to a particular class, sensitivity values ​​are plotted against the false-positive rate or certainty-1 value or true-positive rate.

[0064] S2: performing data processing on the biomarkers respectively to obtain standardized data of each biomarker;

[0065] Since each indicator has a different measurement unit and a different range of values, the data must be normalized. Therefore, the following method will be used to normalize the data. After this operation, all data will be in the range of [0, 1]. Specifically:

[0066] If the biomarker value If is a positive indicator,

[0067] ;

[0068] If the biomarker value If it is a negative indicator,

[0069] ;

[0070] in, is the normalized data of biomarkers, max{ } represents the maximum value of the jth biomarker, min{ } represents the minimum value of the jth biomarker.

[0071] By processing the values ​​of the biomarkers separately, standardized data of each biomarker is obtained, which avoids the problem of data inconsistency caused by different measurement units and different value ranges for each biomarker, facilitates subsequent calculations, and improves computing power.

[0072] S3: Calculating the weight of each biomarker in the disease to which pleural effusion belongs based on the standardized data of each biomarker;

[0073] Information entropy is an important factor in determining the weight of biomarker evaluations; a high information entropy indicates that the biomarker's composite score contains a large amount of information and therefore has a higher weight. Entropy can be used to calculate the degree of disorder and its application in information systems.

[0074] As an implementation method of the present application, the calculation of the weight of each biomarker in the disease to which pleural effusion belongs in S3 based on the standardized data of each biomarker includes the following steps:

[0075] S31: Calculate the information entropy of each biomarker using the entropy weight method based on the standardized data of each biomarker. The calculation formula is:

[0076] ;

[0077] Where n is the number of biomarker types, E j is the information entropy of the jth biomarker;

[0078] S32: Based on the information entropy of each biomarker, the weight of each biomarker in diagnosing the disease to which pleural effusion belongs is calculated using the following formula:

[0079] , and the sum of the weights of each biomarker in the same pleural effusion disease is 1;

[0080] Among them, W j is the weight of the jth biomarker in diagnosing the disease to which pleural effusion belongs.

[0081] A high information entropy value indicates that the data contains a large number of valuable features, so the model should be given a higher weight.

[0082] Different types of pleural effusions may manifest differently on the same biomarker, and therefore different biomarkers play different roles in diagnosing pleural effusions. In this application, the weights of each biomarker in diagnosing pleural effusions are obtained using the entropy weight method, and are input into a trained pleural effusion auxiliary diagnosis model together with the standardized data of each biomarker to determine the disease to which the pleural effusion belongs; this makes the output results of the pleural effusion auxiliary diagnosis model more accurate.

[0083] S4: inputting the standardized data of each biomarker and the weight of each biomarker in the disease to which the pleural effusion belongs into the trained pleural effusion auxiliary diagnosis model to determine the disease to which the pleural effusion belongs;

[0084] The number of hidden layers in the pleural effusion auxiliary diagnosis model is determined by the following formula:

[0085] h=ka×log(1+b×n);

[0086] Where h represents the number of hidden layers in the pleural effusion auxiliary diagnosis model; n represents the number of biomarker types; k is the first constant, which represents the maximum number of hidden layers in the absence of biomarkers; a is the second constant, which represents the maximum reduction in the number of hidden layers; and b is the third constant, which represents the maximum reduction in the number of hidden layers.

[0087] The trend of the above formula is that the greater the number of biomarker types, the fewer the number of hidden layers in the pleural effusion auxiliary diagnosis model, and the fewer the number of biomarker types, the more the number of hidden layers in the pleural effusion auxiliary diagnosis model. The number of hidden layers in the pleural effusion auxiliary diagnosis model is determined based on the number of biomarker types, so that the number of hidden layers varies with the number of biomarker types input into the pleural effusion auxiliary diagnosis model. This avoids the problem of wasted computing power and low prediction efficiency caused by using a network model with a fixed number of hidden layers for different numbers of biomarkers.

[0088] This study constructed a pleural effusion auxiliary diagnosis model and analyzed the auxiliary detection value of the aforementioned pleural effusion biomarkers from multiple dimensions. The pleural effusion auxiliary diagnosis model, comprising 30 hidden layers, was designed based on the characteristics of excessive pleural effusion, the clinical features of patients with pleural effusion, biochemical biomarkers, tumor markers, and different cell types present in the pleural effusion. The clinical applicability of the multidimensional analysis of clinically detected biomarkers was demonstrated. This research contributes to improving the diagnostic efficiency of complex pleural effusion-related diseases and has potential for clinical application.

[0089] As an implementation method of the present application, the pleural effusion auxiliary diagnosis model is a convolutional neural network, including feature encoding, feature enhancement and feature decoding;

[0090] The pleural effusion auxiliary diagnosis model consists of an input layer, a hidden layer, and an output layer. The input layer consists of the standardized data of 23 biomarkers and the weight of each biomarker in the disease to which pleural effusion belongs. The standardized data of each biomarker and its weight in the disease to which pleural effusion belongs are as follows: Figure 2 In this example, x1, x2, ..., x23 are represented by the input data. Input data is converted to output data by applying information weights, adding biases, and passing them through activation functions at each layer. All biomarkers are converted into mathematical information and fed into the algorithm. Algorithm parameters (biases and weights) are mathematically adjusted to reduce the error between actual results and output values. After training, the deep learning algorithm is complete and evaluated on test data. Each convolutional layer consists of many units, each of which performs a convolutional transformation on the previous layer by multiplying it with a filter. The neural network transforms the input information through convolution with the filter and returns a new signal as output. This signal reduces features that the filter does not care about, retaining only the key features. Through convolution, key features are extracted and passed to the convolutional layer. Each analysis and training layer outputs a relevant biomarker, ultimately summarizing the calculation results of important biomarkers for auxiliary diagnosis of pleural effusion diseases.

[0091] Figure 2 Schematic diagram of the structure of a pleural effusion auxiliary diagnosis model provided by an embodiment of the present invention. The pleural effusion auxiliary diagnosis model is a convolutional neural network. The hidden layer of the pleural effusion auxiliary diagnosis model implements feature encoding, feature enhancement, and feature decoding, automatically extracts pleural effusion biomarker features, and performs end-to-end training.

[0092] During feature encoding, the encoder has four layers, each of which repeats two 3x3 convolutions. Each convolution is followed by a data normalization layer and an activation function (ReLU). Due to the excessive amount of data, network performance is unstable. Finally, downsampling is performed through a 2×2 max pooling layer with a stride of 2, doubling the number of feature channels.

[0093] After encoding the feature data, the underlying feature dataset is obtained, and an attention module (CBAM) is introduced between the two-step convolution operations, as shown in Figure 2 As shown in the attention module.

[0094] Among them, the attention module is used for feature enhancement, and the calculation formula of the attention module is:

[0095] ;

[0096] ;

[0097] in, represents the element multiplier, F represents the input feature map, (F) and (A C ) are channel attention module and spatial attention module respectively, The attention map generated for the channel attention module, Attention maps generated for the spatial attention module;

[0098] The channel attention module collects feature map information through both average and max pooling operations, then performs feature dimensionality reduction using a multi-layer perceptron (MLP). Finally, the two features are added together, weighted using a sigmoid activation function, and multiplied by the input feature map to produce a scaled new feature. The computational process of the channel attention module is as follows:

[0099] ;

[0100] in, Represents the sigmoid activation function, MLP represents the dimensionality reduction operation of the multilayer perceptron; max-pool represents the maximum pooling operation, and avg-pool represents the average pooling operation.

[0101] The spatial attention module first performs max pooling and average pooling along the channel axis and merges the results. It then applies spatial attention mapping to the concatenated feature maps using a 7×7 convolutional layer and a sigmoid activation function. The computational process of the spatial attention module is as follows:

[0102] ;

[0103] in Represents a convolution operation. For example, a convolution operation with a convolution size of 7×7 may be selected.

[0104] Each layer of the feature decoding part includes an upsampled feature map, followed by a 2×2 convolution layer, which halves the number of feature channels. The jump link in U-Net connects the features of the encoder and decoder. The encoder features are data calculated in the early layers of the network and are shallow features, while the decoder features are calculated in the deep layers of the network. The resulting data must go through multiple layers of calculation before it can be summarized as deep features. In this way, there is a semantic gap in the features merged at both ends of the jump link. Therefore, incompatible feature sets will interfere with the fusion process. This part introduces the Res path, which uses residuals to replace ordinary jump links. Specifically:

[0105] The residual block is formed by the filter;

[0106] Use the filter to connect with the residual, and then multiply it with the previous layer feature corresponding to the residual to obtain the deep feature corresponding to the residual;

[0107] The semantic category corresponding to the deep feature is determined by multiplying the deep feature and a preset convolution kernel.

[0108] For example, each residual block consists of a 3×3 filter and a 1×1 filter. These filters are used in convolutional layers, where they are connected to the residuals and then fused by element-wise multiplication. The four residual paths from shallow to deep layers use 4, 3, 2, and 1 residual blocks, respectively. This allows the deep network to not only replenish the spatial information lost in pooling operations but also maximize the fusion of these features. The final layer uses a 1×1 convolution to map each component's feature vector to its corresponding semantic category.

[0109] Based on biomarkers, pleural effusion is diagnosed using a pleural effusion auxiliary diagnosis model. The selected pleural effusion auxiliary diagnosis model is a convolutional neural network, including feature encoding, feature enhancement, and feature decoding. Among them, the attention module is used during feature enhancement to focus the training of the pleural effusion auxiliary diagnosis model on key areas, thereby improving the accuracy of the model. In addition, residuals are used for feature jump links to improve the compatibility of feature learning, enhance the accuracy of the pleural effusion auxiliary diagnosis model, and further improve the sensitivity and specificity of pleural effusion auxiliary diagnosis.

[0110] As an implementation method of the present application, when training the pleural effusion auxiliary diagnosis model, 80% of the sample data are randomly selected as the training set, and the remaining sample data are used as the test set.

[0111] When training the pleural effusion auxiliary diagnosis model, the algorithm is used to optimize the cross entropy loss function and iterate. The loss is monitored on the test set to reduce overfitting of the deep model. The model with the highest accuracy is stored and used to generate classification performance on the test set. The designed algorithm randomly selects 80% of the data as the training set and the remaining 20% ​​as the test set to avoid excessive dependence of the network model on the biomarker type and improve the robustness of the model. The embodiments of the present application are implemented under the PyTorch framework and performed on a computer equipped with an Intel Core i9-7900X processor and a TITANV GPU. The operating system is Ubuntu 18.04.4 LTS.

[0112] When the pleural effusion auxiliary diagnosis model trained using this method is used for auxiliary diagnosis of male effusion, the accuracy rate can reach 78%.

[0113] As an implementation method of the present application, the diseases to which the pleural effusion belongs determined by the pleural effusion auxiliary diagnosis model are: malignant tumor combined with pleural effusion, parapulmonary effusion and congestive heart failure. Figure 2 As shown in y1, y2 and y3, these are also the most important diseases of pleural effusion at present.

[0114] It should be noted that the terms used in the present invention are only for describing specific embodiments and are not intended to limit the scope of this application. As shown in the present specification, unless the context clearly indicates an exception, the words "one", "a", "a kind of" and / or "the" do not specifically refer to the singular and may also include the plural. The terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method or device comprising a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, the elements defined by the sentence "comprise a..." do not exclude the presence of other identical elements in the process, method or device comprising the elements.

[0115] It should also be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention. Unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the internal parts of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0116] 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A biomarker-based auxiliary diagnosis method for pleural effusion, characterized in that: The steps include: S1: Screening of biomarkers for auxiliary diagnosis of pleural effusion; S2: performing data processing on the biomarkers respectively to obtain standardized data of each biomarker; S3: Calculating the weight of each biomarker in the disease to which pleural effusion belongs based on the standardized data of each biomarker; S4: inputting the standardized data of each biomarker and the weight of each biomarker in the disease to which the pleural effusion belongs into the trained pleural effusion auxiliary diagnosis model to determine the disease to which the pleural effusion belongs; The number of hidden layers in the pleural effusion auxiliary diagnosis model is determined by the following formula: h=ka×log(1+b×n); Wherein, h represents the number of hidden layers in the pleural effusion auxiliary diagnosis model; n represents the number of biomarker types. The more biomarker types there are, the fewer hidden layers there are in the pleural effusion auxiliary diagnosis model; the fewer biomarker types there are, the more hidden layers there are in the pleural effusion auxiliary diagnosis model. k is the first constant, representing the maximum number of hidden layers in the absence of biomarkers; a is the second constant; and b is the third constant. The pleural effusion auxiliary diagnosis model is a convolutional neural network, including feature encoding, feature enhancement and feature decoding; Among them, the attention module is used for feature enhancement; and the residual is used for feature jump links; During feature encoding, the encoder has four layers, and each layer repeats two 3x3 convolution operations.

2. A biomarker-based pleural effusion auxiliary diagnosis method according to claim 1, characterized in that: The calculation formula of the attention module is: ； ； in, represents the element multiplier, F represents the input feature map, (F) and (A C ) are channel attention module and spatial attention module respectively, The attention map generated for the channel attention module, Attention maps generated for the spatial attention module.

3. The biomarker-based auxiliary diagnosis method for pleural effusion according to claim 2, characterized in that: The calculation process of the channel attention module is: ； in, Represents the sigmoid activation function, MLP represents the dimensionality reduction operation of the multilayer perceptron; max-pool represents the maximum pooling operation, and avg-pool represents the average pooling operation.

4. A biomarker-based auxiliary diagnosis method for pleural effusion according to claim 3, characterized in that: The calculation process of the spatial attention module is: ； in Represents a convolution operation.

5. The biomarker-based auxiliary diagnosis method for pleural effusion according to claim 1, characterized in that: The biomarkers include: adenosine deaminase, lactate dehydrogenase, total protein, glucose, carcinoembryonic antigen, cytokeratin 19 fragment, body mass index, neuron-specific enolase, white blood cell count, monocyte count, multinuclear cell count, mesothelial cell count, specific gravity, pH value, color, gender, age, smoking status, body temperature, pulse rate, respiratory rate, blood pressure, and medical history.

6. The biomarker-based auxiliary diagnosis method for pleural effusion according to claim 1, characterized in that: In S2, data processing is performed on the biomarkers to obtain standardized data of each biomarker: If the biomarker value If is a positive indicator, ； If the biomarker value If it is a negative indicator, ； in, is the normalized data of biomarkers, max{ } represents the maximum value of the jth biomarker, min{ } represents the minimum value of the jth biomarker.

7. The biomarker-based auxiliary diagnosis method for pleural effusion according to claim 1, characterized in that: When training the pleural effusion auxiliary diagnosis model, 80% of the sample data are randomly selected as the training set, and the remaining sample data are used as the test set.

8. The biomarker-based auxiliary diagnosis method for pleural effusion according to claim 1, characterized in that: The disease to which the pleural effusion belongs determined by the pleural effusion auxiliary diagnosis model is: malignant tumor combined with pleural effusion, parapulmonary effusion or congestive heart failure.

Images