Lung pathology detection system, method and device applied to supervision place and medium

Abstract

The invention discloses a lung pathology detection system, method and device applied to a supervision place and a medium, and the system comprises a physiological information collection module which is used for collecting and processing the physiological information of a patient to obtain a first physiological feature parameter; the lung pathological feature classification module is used for processing the first physiological feature parameters through a lung pathological feature classification model to obtain a first lung pathological feature classification result; the lung feature image acquisition module is used for acquiring and processing a lung feature image of a patient to obtain a first lung feature parameter; and the lung pathology detection module is used for processing the first lung feature parameters and the lung pathology feature classification result through a lung pathology detection model to obtain a first lung pathology detection result. According to the method, the improved neural network model is used for replacing manual diagnosis, and automation, unmanned operation and high efficiency of the lung disease diagnosis process in a supervision place are achieved. The method can be widely applied to the technical field of lung disease intelligent detection.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent detection of lung diseases, and in particular to a lung pathological detection system, method, device and medium applied to supervision places. Background Art

[0002] For patients suspected of having lung diseases, the traditional diagnosis process is as follows:

[0003] 1. The doctor inquires about the patient, communicates with the patient, and asks about the patient's clinical symptoms and past medical history, such as whether there are symptoms such as fever, cough, expectoration, chest pain, shortness of breath, etc., and whether there are past medical histories such as chronic lung diseases, asthma, tuberculosis, etc. and family genetic histories such as lung cancer, asthma, etc.

[0004] 2. The doctor examines the patient visually, observes the chest shape and breathing pattern of the patient, and judges whether there are symptoms such as shortness of breath, chest deformity, clubbing of the fingertips, etc., and preliminarily judges the type of lung pathological characteristics of the patient.

[0005] 3. The doctor auscultates the patient to judge whether there are abnormal sounds such as moist rales, dry rales, wheezing sounds, etc. in the patient's lungs, and roughly judges the type of lung pathological characteristics of the patient according to the type of the patient's lung breath sounds, whether it belongs to infectious lung diseases, obstructive lung diseases or interstitial lung diseases.

[0006] 4. The doctor percusses the patient, and judges the density of the patient's lungs according to the sound emitted by tapping the chest, and further infers whether there are diseases such as consolidation, effusion or pneumothorax in the patient's lungs.

[0007] 5. The doctor comprehensively considers the results of inquiry, visual examination, auscultation and percussion, judges the type of lung pathological characteristics of the patient, determines the next examination items according to the type, and issues an examination form, such as performing sputum tests, blood routine tests, taking X-rays and chest CTs for further examinations.

[0008] 6. Analyze the examination results, determine the type of disease the patient has, and then determine the treatment plan.

[0009] In the above process, the doctor needs to have close contact with the patient, and some infectious lung diseases are highly contagious, such as novel coronavirus pneumonia, tuberculosis, etc. During the contact with the patient, the doctor may be infected, resulting in harm to personal safety; in addition, during the diagnosis process, limited by knowledge level and expression ability, the patient may not be able to accurately describe their symptoms, or may not be able to cooperate well with the doctor for diagnosis, thus affecting the diagnosis efficiency; and the processes of diagnosis and analysis of examination results both require manual full participation, consuming a large amount of human resources and taking a long time, resulting in low diagnosis efficiency and being unable to be applied to occasions that require large-scale and high-efficiency screening. Summary of the Invention

[0010] An object of the present invention is to solve at least to some extent one of the technical problems existing in the prior art.

[0011] To this end, an object of an embodiment of the present invention is to provide a pulmonary pathology detection system applied to a supervision place. By using a neural network model instead of manual diagnosis, the system improves the diagnosis efficiency, reduces the contact between medical staff and patients, ensures the safety of medical staff, and realizes high efficiency, automation and unmanned operation in the diagnosis process.

[0012] Another object of an embodiment of the present invention is to provide a pulmonary pathology detection method applied to a supervision place.

[0013] To achieve the above technical object, the technical solutions adopted in the embodiments of the present invention include:

[0014] In a first aspect, an embodiment of the present invention provides a pulmonary pathology detection system applied to a supervision place, which specifically includes:

[0015] A physiological information acquisition module, configured to acquire the physiological information of a patient, preprocess the physiological information, and obtain a first physiological characteristic parameter;

[0016] A pulmonary pathology feature classification module, configured to input the first physiological characteristic parameter into a trained pulmonary pathology feature classification model to obtain a first pulmonary pathology feature classification result;

[0017] A pulmonary feature image acquisition module, configured to acquire the pulmonary feature image of a patient, preprocess the pulmonary feature image, and obtain a first pulmonary feature parameter;

[0018] A pulmonary pathology detection module, configured to input the first pulmonary feature parameter and the pulmonary pathology feature classification result into a trained pulmonary pathology detection model to obtain a first pulmonary pathology detection result.

[0019] Further, in an embodiment of the present invention, the physiological information acquisition module is specifically configured to:

[0020] Acquire the cough audio frequency of the patient through a voice acquisition device, sample, denoise and perform fast Fourier transform on the cough audio frequency to obtain the spectrogram of the cough audio frequency, and perform normalization processing and format conversion on the spectrogram;

[0021] Acquire the physiological indexes of the patient through a physiological monitoring device, and perform normalization processing and format conversion on the physiological indexes;

[0022] Use the spectrogram and the physiological indexes as the first physiological characteristic parameter.

[0023] Further, in an embodiment of the present invention, the lung pathological feature classification model includes:

[0024] A first input layer, which is used to output the first physiological feature parameters to subsequent layers;

[0025] A first convolutional layer, which is used to perform a convolutional operation on the first physiological feature parameters to obtain a first feature map;

[0026] A first pooling layer, which is used to perform downsampling on the first feature map;

[0027] A fully connected layer, which is used to flatten the pooled first feature map and output it to subsequent layers;

[0028] A first output layer, which is used to output the first lung pathological feature classification result.

[0029] Further, in an embodiment of the present invention, the lung pathological feature classification model is trained through the following steps:

[0030] Obtain physiological information sample data, and determine the lung pathological feature type labels corresponding to the physiological information sample data;

[0031] Preprocess the physiological information sample data to obtain second physiological feature parameters, and input the second physiological feature parameters into the lung pathological feature classification model to obtain a second lung pathological feature classification result;

[0032] Determine the first loss value of training according to the second lung pathological feature classification result and the lung pathological feature type labels;

[0033] Update the parameters of the lung pathological feature classification model according to the first loss value.

[0034] Further, in an embodiment of the present invention, the lung feature image acquisition module is specifically used for:

[0035] Obtain the lung feature image through a lung feature image acquisition device;

[0036] Perform adaptive scaling, filtering, normalization processing and format conversion on the lung feature image to obtain first lung feature parameters.

[0037] Further, in an embodiment of the present invention, the lung pathological detection model includes:

[0038] A second input layer, which is used to output the first lung feature parameters and the first lung pathological feature classification result to subsequent layers;

[0039] A second convolutional layer, which is used to perform a convolutional operation on the first lung feature parameter and the classification result of the lung pathological features to obtain a second feature map;

[0040] A dense convolutional layer, which is used to further extract features from the second feature map;

[0041] A second pooling layer, which is used to perform downsampling on the second feature map;

[0042] A conversion layer, which is used to perform further downsampling, activation, and convolutional operations on the output of the dense convolutional layer;

[0043] A second output layer, which is used to output the first lung pathological detection result.

[0044] Further, in an embodiment of the present invention, the lung pathological detection model is trained through the following steps:

[0045] Obtain lung feature image sample data, and determine the lung pathological detection result label and the corresponding lung pathological feature type corresponding to the lung feature image sample data;

[0046] Preprocess the lung feature image sample data to obtain a second lung feature parameter, and input the second lung feature parameter and the lung pathological feature type into the lung pathological detection model to obtain a second lung pathological detection result;

[0047] Determine the second loss value for training according to the second lung pathological detection result and the lung pathological detection result label;

[0048] Update the parameters of the lung pathological detection model according to the second loss value.

[0049] In a second aspect, an embodiment of the present invention provides a lung pathological detection method applied to a supervision place, which specifically includes:

[0050] Collect the physiological information of the patient, preprocess the physiological information to obtain a first physiological feature parameter;

[0051] Input the first physiological feature parameter into the trained lung pathological feature classification model to obtain a first lung pathological feature classification result;

[0052] Collect the lung feature image of the patient, preprocess the lung feature image to obtain a first lung feature parameter;

[0053] Input the first lung feature parameter and the classification result of the lung pathological features into the trained lung pathological detection model to obtain the first lung pathological detection result.

[0054] In a third aspect, an embodiment of the present invention provides a lung pathological detection device applied to a supervision place, including:

[0055] At least one processor;

[0056] At least one memory for storing at least one program;

[0057] When the at least one program is executed by the at least one processor, the at least one processor implements a lung pathological detection method applied to a supervision place as described above.

[0058] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, in which a program executable by a processor is stored, and the program executable by the processor is used to execute a lung pathological detection method applied to a supervision place as described above when executed by the processor.

[0059] The advantages and beneficial effects of the present invention will be partially given in the following description, partially will become obvious from the following description, or will be understood through the practice of the present invention:

[0060] In the embodiment of the present invention, by using a neural network model instead of manual diagnosis, the diagnosis efficiency is improved, and at the same time, the contact between medical staff and patients is reduced, ensuring the safety of medical staff, realizing the high efficiency, automation and unmanned of the diagnosis process, and can be used for rapid and large-scale screening in supervision places; at the same time, the present invention analyzes the physiological information of patients by adopting a double-layer neural network model, fully utilizes and integrates the physiological characteristic data and lung image characteristic data of patients, and improves the diagnosis accuracy; in addition, the present invention designs and improves the network structure of the lung pathological detection model, and effectively reduces the calculation parameters by improving the connection mode of the traditional dense convolutional network, improves the training speed and inference speed of the model while ensuring the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A schematic diagram of a lung pathological detection system applied to a supervision place provided by an embodiment of the present invention;

[0062] Figure 2 A structural diagram of a lung pathological feature classification model provided by an embodiment of the present invention;

[0063] Figure 3 A structural diagram of a lung pathological detection model applied to a supervision place provided by an embodiment of the present invention;

[0064] Figure 4 A structural diagram of a dense convolutional layer provided by an embodiment of the present invention;

[0065] Figure 5 A structural diagram of a conversion layer provided by an embodiment of the present invention;

[0066] Figure 6 A schematic diagram of a lung pathological detection method applied to a supervision place provided by an embodiment of the present invention;

[0067] Figure 7 A structural diagram of a computer device provided by an embodiment of the present invention. Detailed implementation manners

[0068] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention. For the step numbers in the following embodiments, they are only set for the convenience of description and explanation, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0069] In the description of the present invention, the meaning of "a plurality" is two or more. If the first and second are described, it is only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features or implicitly indicating the sequence of the indicated technical features. In addition, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field of the present invention.

[0070] Figure 1 A schematic diagram of a lung pathological detection system provided by an embodiment of the present invention. Refer to Figure 1 , an embodiment of the present invention provides a lung pathological detection system applied to a supervision place, including:

[0071] A physiological information acquisition module, configured to acquire the physiological information of a patient, preprocess the physiological information, and obtain a first physiological characteristic parameter;

[0072] A lung pathological feature classification module, configured to input the first physiological characteristic parameter into a trained lung pathological feature classification model to obtain a first lung pathological feature classification result;

[0073] A lung feature image acquisition module, configured to acquire the lung feature image of a patient, preprocess the lung feature image, and obtain a first lung feature parameter;

[0074] The lung pathology detection module is used to input the first lung characteristic parameters and the classification results of lung pathology characteristics into the trained lung pathology detection model to obtain the first lung pathology detection result.

[0075] Specifically, in the physiological information collection module of this embodiment, the physiological information of the patient mainly includes physiological indicators such as body temperature, blood oxygen saturation, vital capacity, etc. and audio information such as the patient's cough sound. This embodiment collects the above physiological information to replace the inquiry, auscultation, inspection and percussion of the doctor in the traditional diagnosis process to a certain extent. During the process of collecting the above physiological information, the patient can operate the instrument independently without the presence of a doctor to guide. For example, the patient's body temperature data is collected through a thermometer, the blood oxygen saturation of the patient is quickly measured through a pulse oximeter, the vital capacity of the patient is measured through a spirometer, and the cough sound of the patient is collected through a recording device. These instruments are simple to operate and can be completed independently by the patient without the need for contact between the doctor and the patient.

[0076] Among the above physiological information, body temperature can indicate whether the patient has a fever, thus suggesting whether the patient has a lung inflammation and inferring whether the pathological type of the patient's lungs is an infectious lung disease; blood oxygen saturation can suggest whether the patient has hypoxemia, and further infer whether the pathological type of the patient's lungs is an obstructive lung disease or an interstitial lung disease; vital capacity reflects the lung function status of the patient. By detecting the vital capacity of the patient, it can help to judge whether the patient has a problem of decreased lung function and further infer whether the pathological type of the patient's lungs is an obstructive lung disease; cough sounds can be roughly divided into types such as moist rales, dry rales, wheezing sounds, etc. Among them, moist rales suggest that the pathological type of the patient's lungs may be an infectious lung disease, dry rales suggest that the pathological type of the patient's lungs may be an interstitial lung disease, and wheezing sounds suggest that the pathological type of the patient's lungs may be an obstructive lung disease. Inputting this physiological information into the trained lung pathology characteristic classification model can obtain the inference result of the pathological type of the patient's lungs without manual analysis by a doctor.

[0077] Therefore, by realizing the unmanned operation of the process of collecting and analyzing physiological information, we can reduce the contact between the patient and the doctor, reduce the risk of the doctor being infected, and at the same time avoid the misleading of the doctor caused by the limitations of the patient's knowledge level and expression ability, reduce the risk of misdiagnosis, and the collection process can be completed without the cooperation of the patient, reducing the difficulty of the diagnosis process; while using a neural network model to analyze the physiological information of the patient can greatly improve the diagnosis speed, reduce the workload of the doctor, and avoid misdiagnosis caused by human factors to a certain extent.

[0078] In the lung pathological feature classification module of this embodiment, by collecting the physiological information of patients in historical cases and the corresponding lung pathological types, a training dataset can be obtained for training a neural network model that classifies the lung pathological features of patients based on their physiological information. During the classification process, the convolutional layer in the neural network can extract high-dimensional features in the cough sound frequency, identify different types of cough sound patterns (such as moist rales, dry rales, wheezing sounds, etc.), and perform feature fusion in combination with other physiological feature parameters (such as body temperature, blood oxygen saturation, vital capacity). Finally, the type of the patient's lung pathological features is output. In this embodiment, the types of lung pathological features altogether include four categories: infectious lungs, obstructive lung diseases, interstitial lung diseases, and other lung diseases (mainly including tumors and pulmonary vascular diseases). Based on the above physiological information as the judgment basis, the lung pathological feature classification model of this embodiment can better complete the classification task, and there are significant substantial differences between each type, with a relatively high discrimination degree.

[0079] Therefore, in this embodiment, by designing the form of the training dataset, the physiological information of the patients to be collected and all the types of lung pathological features are determined. Then, a lung pathological feature classification model is trained and used to quickly and automatically classify the types of the patients' lung pathological features, eliminating the process of manual analysis by physicians, saving human resources, and improving the diagnostic efficiency at the same time. In addition, in this embodiment, the numerical physiological information and audio physiological information of the patients are collected simultaneously, realizing the multi-modalization of the input data and improving the robustness and classification accuracy of the model.

[0080] In the lung feature image acquisition module of this embodiment, the devices for acquiring the lung feature images of patients mainly include X-ray machines and CT scanners. During the acquisition process, the patient does not need to operate the instrument and only needs to go to the designated position according to the guidance to complete the acquisition operation. Therefore, the unmanned operation of the diagnostic process can be maintained. In the traditional diagnostic process, physicians will determine the next examination items for the patient based on the comprehensive results of inquiry, auscultation, inspection, and percussion, generally including blood routine tests, sputum tests, chest X-rays, and chest CTs. Among them, the imaging speed of chest X-rays and chest CTs is relatively fast, but the analysis process of their images takes a long time. The traditional analysis method requires physicians to perform qualitative and quantitative analysis on image features such as the shape, edge, texture, and blurring degree of the images, seriously reducing the diagnostic efficiency of lung diseases and having relatively high requirements for the professional level of physicians. The process of manual analysis is prone to omissions, resulting in misdiagnosis.

[0081] In this embodiment, by performing processing such as noise reduction, scaling, normalization, and format conversion on the lung feature images, the image data can be converted into a form that is easy for the neural network model to analyze, which is beneficial to improving the detection speed and detection accuracy of the lung pathological detection model in the subsequent steps.

[0082] In the lung pathology detection module of this embodiment, a large number of chest X-ray images and chest CT images of patients with lung diseases, as well as their corresponding lung pathology feature types and specific lung disease types, are collected as the training data for the lung pathology detection model. During the training process, the input data is the preprocessed chest X-ray images or chest CT images, as well as the manually annotated lung pathology feature types. The convolutional neural network is used to automatically extract image features and output the specific lung disease type. In this embodiment, in addition to the lung feature images, the input data also includes their corresponding lung pathology feature types, where the latter is the output of the lung pathology feature classification model and is used to assist in judging the specific lung disease type of the patient. For some lung diseases with similar lung image features but different lung pathology feature types, pre-determining the lung pathology feature type to which the patient belongs can effectively improve the accuracy of the lung pathology detection model. For example, silicosis belonging to interstitial lung disease and pulmonary edema belonging to other types of lung diseases both show diffuse infiltration shadows in their lung feature images. If the input data only contains lung feature images, the lung pathology detection model is very likely unable to distinguish between the two, resulting in misjudgment. If the input data includes the prior lung pathology feature type, the lung pathology detection model will be able to accurately output the recognition result.

[0083] In addition, in this embodiment, the network structure of the lung pathology detection model is specifically optimized for multi-modal input data and application scenarios that pursue high efficiency. In the traditional Dense Convolutional Network (DenseNet), the input of each layer will participate in the splicing of subsequent layers, not just the input of the next layer. This connection method can enhance the reuse rate of feature information, enabling the model to utilize both low-level and high-level features simultaneously when performing deep-level convolutions, improving the accuracy of the model. However, due to the increase in the number of parameters, the training speed and inference speed of the neural network model using this structure will both decrease significantly, requiring more computing resources and having higher performance requirements for training devices and deployment devices. This embodiment optimizes the connection method of the traditional dense convolutional network, making the output of the first layer in the middle of the dense convolutional network not participate in the splicing of subsequent input layers, and at the same time controlling the number of layers of the convolutional neural network at a lower order of magnitude, effectively reducing the calculation parameters, improving the training speed and inference speed of the model while ensuring the accuracy of the model;

[0084] In addition, in the traditional DenseNet, the transition layer is mainly responsible for downsampling the feature map to reduce its size, usually only including pooling operations. In this embodiment, in addition to using the max pooling layer for downsampling operations, an activation layer and a convolutional layer are added. The former is used to enhance the model's ability to handle non-linear problems, and the latter is used to reduce feature loss during the downsampling process of the model, further improving the utilization rate of feature information while reducing the data dimension and enhancing the model performance.

[0085] Further as an optional implementation manner, the physiological information acquisition module is specifically configured to:

[0086] S101. Collect the cough audio of the patient through a voice acquisition device, sample, denoise, and perform fast Fourier transform on the cough audio to obtain the spectrogram of the cough audio, and perform normalization processing and format conversion on the spectrogram;

[0087] S102. Collect the physiological indicators of the patient through a physiological monitoring device, and perform normalization processing and format conversion on the physiological indicators;

[0088] S103. Use the spectrogram and the physiological indicators as the first physiological feature parameters.

[0089] Specifically, after collecting the cough sound frequency of the patient through voice collection devices such as microphones and radios in step S101, first, the original sound signal is sampled according to a preset sampling rate (e.g., 16 kHz) to ensure the quality and integrity of the sound data. After sampling, since environmental noise may interfere with the sound signal, it is also necessary to perform noise reduction processing on the collected cough sound frequency. The noise reduction processing can be carried out through various methods, such as a low-pass filter (to remove high-frequency noise) or adaptive filtering (to process specific environmental noise), so as to retain the key features of the cough sound and reduce the interference of environmental sounds, improving the data quality. Then, the noise-reduced sound signal will undergo a Fast Fourier Transform (FFT) to convert the sound signal from the time domain to the frequency domain, generating a corresponding spectrogram. Through the FFT, each frequency component in the cough sound can be clearly displayed, facilitating the distinction of different types of cough sounds (such as moist rales, dry rales, or wheezing). After obtaining the spectrogram, it is also necessary to perform normalization processing on the spectrogram to eliminate the differences in data scales between different patients or different devices, enabling the data to be better processed in the neural network. Normalization usually converts the data to a certain range (such as a mean of 0 and a variance of 1), thereby reducing the instability caused by eigenvalue differences. Finally, after format conversion, the processed spectrogram is input into the neural network in a suitable format for subsequent analysis. The process of format conversion includes scaling or cropping the size of the spectrogram to a unified size, and then converting the spectrogram into a matrix or tensor format suitable for neural network input.

[0090] In step S102, physiological indicators of the patient are collected through physiological monitoring devices such as thermometers, pulse oximeters, and spirometers, including the patient's body temperature, blood oxygen saturation, vital capacity, etc. In this embodiment, the basis for selecting the physiological indicators to be measured is whether these indicators can well indicate the type of pulmonary pathological characteristics to which the patient belongs. The physiological indicators also need to undergo normalization processing to ensure data consistency during the fusion process of multiple modal data. Similarly, they can be converted to a numerical range with a mean of 0 and a variance of 1, which is consistent with the numerical range of the spectrogram. After normalization processing, it is also necessary to perform format conversion on them, converting these numerical data into feature vectors or matrices that can be directly input into the neural network for processing. Multiple indicators can be directly concatenated to form a one-dimensional vector, or they can be padded to expand them into a matrix or tensor with the same dimension as the spectrogram, ensuring that different types of data can undergo unified feature learning and processing.

[0091] After processing the spectrogram and physiological indicators through the above steps, in this embodiment, in step S103, the obtained spectrogram and physiological indicator data will be combined into the patient's first physiological characteristic parameter by means of splicing fusion or weighted fusion, providing an input for the subsequent lung pathological detection model. This acquisition, processing, and fusion of multimodal data can effectively utilize the audio feature data of the patient's cough sound and the physiological indicator feature data of the patient, improving the robustness and accuracy of the model.

[0092] As Figure 2 shown, further as an optional implementation manner, the lung pathological feature classification model includes:

[0093] A first input layer for outputting the first physiological characteristic parameter to subsequent layers;

[0094] A first convolutional layer for performing a convolution operation on the first physiological characteristic parameter to obtain a first feature map;

[0095] A first pooling layer for performing a downsampling operation on the first feature map;

[0096] A fully connected layer for flattening the pooled first feature map and outputting it to subsequent layers;

[0097] A first output layer for outputting the first lung pathological feature classification result.

[0098] Specifically, the input layer passes the input data to the subsequent convolutional layer for feature extraction. The convolutional layer contains several convolutional kernels, which are used to extract the frequency changes, temporal characteristics in the audio spectrogram, and relevant features in the physiological index data. After multiple convolutional layers are stacked and processed, a feature map is generated. The pooling layer is used to perform downsampling on the feature map output by the convolutional layer, reducing the spatial dimension of the feature map, reducing the computational amount, and at the same time retaining key features. Through pooling, the size of the feature map can be reduced, making it more compact and reducing the risk of overfitting. After multiple convolutions and poolings, the output feature map is flattened into a one-dimensional vector (this operation can also be completed by adding a flatten layer in the fully connected layer). The one-dimensional vector is connected to the output layer through the fully connected layer. The fully connected layer contains several neuron nodes, and each node is connected to all the outputs of the previous layer, thus realizing a linear transformation, compressing multi-dimensional features into low-dimensional features, and at the same time integrating the information of each feature channel. The features processed by the fully connected layer are input to the output layer. In this embodiment, the output layer uses the Softmax activation function to output the probability distribution of each pathological feature, and takes the type of lung pathological feature with the highest probability as the classification result. Through this neural network model, the preliminary classification of the type of lung pathological feature to which the patient belongs can be completed, making full use of the patient's physiological information, improving the accuracy of diagnosis, and using the prior type of lung pathological feature as the input of the subsequent lung pathological detection model can improve the detection accuracy of the lung pathological detection model.

[0099] Further as an optional implementation manner, the lung pathological feature classification model is obtained through the following steps:

[0100] A1. Obtain physiological information sample data and determine the label of the type of lung pathological feature corresponding to the physiological information sample data;

[0101] A2. Preprocess the physiological information sample data to obtain second physiological feature parameters, input the second physiological feature parameters into the lung pathological feature classification model, and obtain a second lung pathological feature classification result;

[0102] A3. Determine the first loss value of the training according to the second lung pathological feature classification result and the label of the type of lung pathological feature;

[0103] A4. Update the parameters of the lung pathological feature classification model according to the first loss value.

[0104] Specifically, in this embodiment, the physiological information sample data mainly comes from patient data in actual diagnosis. First, data such as body temperature, blood oxygen saturation, vital capacity, and cough sound frequency in the patient's case are extracted. According to the confirmed disease type of the patient, corresponding pulmonary pathological category labels (i.e., the four types of interstitial lung disease, infectious lung disease, obstructive lung disease, and other lungs) are labeled for these data. The one-hot encoding is used to encode these label types, and the labeled data is used as the training data set for training the pulmonary pathological feature classification model. In step A2, first, the preprocessing method of the sample data is kept consistent with the processing method in step S101 to ensure the quality and consistency of the input data. In step A3, a loss function (in this embodiment, since it is a multi-classification problem, the cross-entropy loss function can be used) is used to calculate the gap between the classification result output by the model and the true label data. According to the error between the second pulmonary pathological feature classification result and the corresponding pulmonary pathological feature type label, a loss value is obtained. In step A4, according to the calculated loss value, the gradient descent algorithm is used to update the trainable parameters in the model, including the weights of the convolutional kernels, the weights and bias terms of the fully connected layers, etc. In each training iteration, the parameters of the model are adjusted in the direction of reducing the loss value, and finally, the classification performance of the model on the sample data is gradually improved.

[0105] Further as an optional implementation manner, the pulmonary feature image acquisition module is specifically used for:

[0106] S201. Obtain pulmonary feature images through a pulmonary feature image acquisition device;

[0107] S202. Perform adaptive scaling, filtering, normalization processing, and format conversion on the pulmonary feature images to obtain first pulmonary feature parameters.

[0108] Specifically, in this embodiment, the pulmonary feature image acquisition device mainly includes an X-ray machine and a CT scanner. The acquired pulmonary feature images are mainly chest X-ray films and chest CT images. The former can effectively display pathological changes such as shadows, inflammations, effusions, and tumors in the patient's lungs and is widely used in the preliminary screening of lung diseases. The latter can provide pulmonary anatomical structure images with higher resolution, can effectively identify smaller lesions, and has a higher fineness. In this embodiment, first, the input pulmonary feature images need to be uniformly scaled to images of 224*224, and then the images are denoised (Gaussian filtering or median filtering can be used). Subsequently, the pixel values of the images are normalized, and each pixel value is normalized from an integer range (such as 0 to 255) to a standard range of floating-point numbers (such as 0 to 1 or -1 to 1) to prevent the feature values from being too large and affecting the model training effect. Then, the image format is converted into a tensor or matrix form suitable for input to the deep learning model to obtain the first pulmonary feature parameters.

[0109] As Figure 3 shown, as a further optional embodiment, the lung pathology detection model includes:

[0110] A second input layer, which is used to output the first lung feature parameters and the classification result of the first lung pathology features to the subsequent layers;

[0111] A second convolutional layer, which is used to perform a convolutional operation on the first lung feature parameters and the classification result of the lung pathology features to obtain a second feature map;

[0112] A dense convolutional layer, which is used to perform further feature extraction on the second feature map;

[0113] A second pooling layer, which is used to perform downsampling on the second feature map;

[0114] A conversion layer, which is used to perform further downsampling, activation, and convolutional operations on the output of the dense convolutional layer;

[0115] A second output layer, which is used to output the first lung pathology detection result.

[0116] Specifically, the input layer passes the input data to the subsequent convolutional layer for feature extraction. In this embodiment, the role of the convolutional layer is to initially extract the feature information in the lung feature image. The result of each convolution will undergo a non-linear transformation through the activation layer and a downsampling operation through the max pooling layer. The finally obtained feature map is used to output to the dense convolutional layer for further feature extraction operations. In this embodiment, the lung pathology detection model contains two dense convolutional layers, and the outputs inside are concatenated by 1*1 and 3*3 convolutional feature maps respectively, as Figure 4 shown. Each dense convolutional layer stacks four dense convolutional blocks. Each convolutional block includes a normalization layer, an activation layer, a convolutional layer, and a connection layer. Among them, the convolutional output feature map of the first layer in the middle of each dense convolutional layer does not participate in the concatenation of the input of the subsequent convolutional layer, effectively reducing the number of parameters of the model. In the traditional dense convolutional network, the conversion layer only performs a pooling operation on the output of the dense convolutional layer, while in this embodiment, as Figure 5As shown, the conversion layer includes a normalization layer, an activation layer, a convolutional layer, and an average pooling layer. In addition to the conventional pooling operation, activation and convolution operations are also performed. The former is used to enhance the model's ability to handle non-linear problems, and the latter is used to reduce feature loss during downsampling. While reducing the data dimension, it further improves the utilization rate of feature information. The combination of the two enables the lung pathology detection model of this embodiment to still have a high recognition accuracy while reducing the number of dense convolutional layers and the number of internal parameters of the dense convolutional layer. In this embodiment, the output layer also uses the softmax activation function to generate the probability distribution of each lung disease type and complete the classification task.

[0117] Further as an optional implementation manner, the lung pathology detection model is trained through the following steps, which specifically include:

[0118] B1. Obtain lung feature image sample data, and determine the lung pathology detection result label and the corresponding lung pathology feature type corresponding to the lung feature image sample data;

[0119] B2. Preprocess the lung feature image sample data to obtain the second lung feature parameter, input the second lung feature parameter and the lung pathology feature type into the lung pathology detection model, and obtain the second lung pathology detection result;

[0120] B3. Determine the second loss value of the training according to the second lung pathology detection result and the lung pathology detection result label;

[0121] B4. Update the parameters of the lung pathology detection model according to the second loss value.

[0122] Specifically, in this embodiment, the lung feature image sample data mainly comes from patient data in actual diagnosis. First, collect the lung X-ray images or CT images of different patients, and according to the diagnosed lung diseases of the patients, label the corresponding disease type tags (such as tuberculosis, silicosis, pulmonary edema, etc.) for these data. Use one-hot encoding to encode each lung disease, and use the labeled data as the training dataset for training the lung pathology detection model. In step B2, first preprocess the sample data, and the preprocessing method is the same as that in step S103 to ensure the quality and consistency of the input data. In step B3, use a loss function (in this embodiment, since it is a multi-classification problem to be solved, the cross-entropy loss function can be used) to calculate the gap between the classification result output by the model and the true label data, and obtain the loss value according to the error between the second lung pathology detection result and the corresponding lung disease type tag. In step B4, according to the calculated loss value, use the gradient descent algorithm to update the trainable parameters in the model, including the weights of the convolutional kernels, the weights of the dense convolutional layers, the weights and bias terms of the fully connected layers, etc. In each training iteration, the parameters of the model will be adjusted in the direction of reducing the loss value, and finally the classification performance of the model on the sample data will be gradually improved.

[0123] Referring to Figure 6 , an embodiment of the present invention provides a lung pathology detection method applied to a supervision place, which specifically includes:

[0124] S301. Collect the physiological information of the patient, preprocess the physiological information, and obtain the first physiological feature parameter;

[0125] S302. Input the first physiological feature parameter into the trained lung pathology feature classification model to obtain the first lung pathology feature classification result;

[0126] S303. Collect the lung feature image of the patient, preprocess the lung feature image, and obtain the first lung feature parameter;

[0127] S304. Input the first lung feature parameter and the lung pathology feature classification result into the trained lung pathology detection model to obtain the first lung pathology detection result.

[0128] The content in the above method embodiments is applicable to the system embodiments of the present invention. The functions specifically implemented by the system embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0129] Referring to Figure 7 , an embodiment of the present invention provides a computer device, including:

[0130] At least one processor;

[0131] At least one memory for storing at least one program;

[0132] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned method for detecting pulmonary pathology applied to a supervised place.

[0133] The content in the above method embodiments is applicable to the device embodiments of the present invention. The functions specifically implemented by the device embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0134] The embodiments of the present invention also provide a computer-readable storage medium, in which a processor-executable program is stored. The processor-executable program is used to execute the above-mentioned method for detecting pulmonary pathology applied to a supervised place when executed by a processor.

[0135] A computer-readable storage medium according to an embodiment of the present invention can execute a method for detecting pulmonary pathology applied to a supervised place provided by an embodiment of the method of the present invention, can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0136] The embodiments of the present invention also disclose a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes Figure 6 the method shown.

[0137] In some alternative embodiments, the functions / operations mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the functions / operations involved, two consecutive blocks shown may actually be executed substantially simultaneously, or the above blocks can sometimes be executed in the reverse order. In addition, the embodiments presented and described in the flowcharts of the present invention are provided by way of example for the purpose of providing a more comprehensive understanding of the technology. The disclosed method is not limited to the operations and logical processes presented herein. Alternative embodiments are foreseeable, in which the order of various operations is changed and the sub-operations described as part of a larger operation are executed independently.

[0138] In addition, although the present invention has been described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the above functions and / or features may be integrated in a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It should also be understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present invention. Rather, considering the attributes, functions, and internal relationships of the various functional modules in the devices disclosed herein, the actual implementation of the modules will be understood within the ordinary skills of an engineer. Therefore, those skilled in the art can implement the present invention as set forth in the claims without undue experimentation. It should also be understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0139] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the above methods in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0140] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0141] More specific examples (nonexhaustive list) of computer-readable media include the following: an electrical connection (electronic device) having one or more wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable media can even be paper or other suitable media on which the above programs can be printed, because the above programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, deciphering, or otherwise processing as appropriate, and then storing them in a computer memory.

[0142] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0143] In the above description of this specification, the descriptions referring to the terms "one embodiment / example", "another embodiment / example", or "certain embodiments / examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0144] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the claims and their equivalents.

[0145] The above has specifically described the preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art can also make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A pulmonary pathological detection system applied to supervision places, characterized in that Including: A physiological information acquisition module, configured to acquire physiological information of a patient, preprocess the physiological information, and obtain first physiological characteristic parameters; A pulmonary pathological feature classification module, configured to input the first physiological characteristic parameters into a trained pulmonary pathological feature classification model to obtain a first pulmonary pathological feature classification result; A pulmonary feature image acquisition module, configured to acquire a pulmonary feature image of the patient, preprocess the pulmonary feature image, and obtain first pulmonary feature parameters; A pulmonary pathological detection module, configured to input the first pulmonary feature parameters and the first pulmonary pathological feature classification result into a trained pulmonary pathological detection model to obtain a first pulmonary pathological detection result.

2. The pulmonary pathological detection system applied to a supervision place according to claim 1, wherein The physiological information acquisition module is specifically configured to: Acquire the cough audio frequency of the patient through a voice acquisition device, sample, denoise, and perform fast Fourier transform on the cough audio frequency to obtain a spectrogram of the cough audio frequency, and perform normalization processing and format conversion on the spectrogram; Acquire the physiological indicators of the patient through a physiological monitoring device, and perform normalization processing and format conversion on the physiological indicators; Use the spectrogram and the physiological indicators as the first physiological characteristic parameters.

3. The pulmonary pathological detection system applied to a supervision place according to claim 1, characterized in that, The pulmonary pathological feature classification model includes: A first input layer, which is configured to output the first physiological characteristic parameters to subsequent layers; A first convolutional layer, which is configured to perform a convolutional operation on the first physiological characteristic parameters to obtain a first feature map; A first pooling layer, which is configured to perform a downsampling operation on the first feature map; A fully connected layer, which is configured to flatten the first feature map after pooling and output it to subsequent layers; A first output layer, which is configured to output the first pulmonary pathological feature classification result.

4. The pulmonary pathological detection system applied to a supervision place according to claim 1, wherein, The pulmonary pathological feature classification model is trained through the following steps: Obtain physiological information sample data and determine the pulmonary pathological feature type labels corresponding to the physiological information sample data; Preprocess the physiological information sample data to obtain second physiological characteristic parameters, input the second physiological characteristic parameters into the pulmonary pathological feature classification model, and obtain a second pulmonary pathological feature classification result; Determine a first loss value for training according to the second pulmonary pathological feature classification result and the pulmonary pathological feature type labels; Update the parameters of the pulmonary pathological feature classification model according to the first loss value.

5. The pulmonary pathological detection system applied to a supervision place according to claim 1, characterized in that The pulmonary feature image acquisition module is specifically configured to: Obtain the pulmonary feature image through a pulmonary feature image acquisition device; Perform adaptive scaling, filtering, normalization processing, and format conversion on the pulmonary feature image to obtain the first pulmonary feature parameters.

6. The pulmonary pathological detection system applied to a supervision place according to claim 1, wherein The pulmonary pathological detection model includes: A second input layer, which is configured to output the first pulmonary feature parameters and the first pulmonary pathological feature classification result to subsequent layers; A second convolutional layer, which is configured to perform a convolutional operation on the first pulmonary feature parameters and the pulmonary pathological feature classification result to obtain a second feature map; Dense convolutional layer, which is used to perform further feature extraction on the second feature map; Second pooling layer, which is used to perform downsampling operation on the second feature map; Conversion layer, which is used to perform further downsampling, activation and convolution operations on the output of the dense convolutional layer; Second output layer, which is used to output the first pulmonary pathological detection result.

7. The pulmonary pathological detection system applied to a supervision place according to claim 1, characterized in that, The pulmonary pathological detection model is obtained through the following steps: Obtain pulmonary feature image sample data, and determine the pulmonary pathological detection result label and corresponding pulmonary pathological feature type corresponding to the pulmonary feature image sample data; Preprocess the pulmonary feature image sample data to obtain second pulmonary feature parameters, input the second pulmonary feature parameters and the pulmonary pathological feature type into the pulmonary pathological detection model to obtain a second pulmonary pathological detection result; Determine the second loss value of training according to the second pulmonary pathological detection result and the pulmonary pathological detection result label; Update the parameters of the pulmonary pathological detection model according to the second loss value.

8. A lung pathological detection method applied to a supervision place, characterized in that, Including the following steps: Collect the physiological information of the patient, preprocess the physiological information to obtain the first physiological feature parameters; Input the first physiological feature parameters into the trained pulmonary pathological feature classification model to obtain the first pulmonary pathological feature classification result; Collect the pulmonary feature image of the patient, preprocess the pulmonary feature image to obtain the first pulmonary feature parameters; Input the first pulmonary feature parameters and the pulmonary pathological feature classification result into the trained pulmonary pathological detection model to obtain the first pulmonary pathological detection result.

9. A pulmonary pathological detection device applied to a supervision place, characterized in that, Including: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a method for pulmonary pathological detection applied to a supervision place as claimed in claim 8.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that, The program executable by the processor is used to execute a method for pulmonary pathological detection applied to a supervision place as claimed in claim 8 when executed by the processor.

Images