Deep learning-based children pneumonia lung image analysis and evaluation system and method

Abstract

The invention discloses a deep learning-based children pneumonia lung image analysis and evaluation system and method, relates to the technical field of children pneumonia analysis and evaluation, and aims to solve the technical problem that the accuracy of artificial diagnosis is difficult to guarantee in the prior art. Transmitting the data to a data calculation module; the data calculation module is connected to the lung shadow detection module and used for receiving the image of the lung shadow detection module and calculating pulmonary alveolar density, definition, quantity and echo intensity data; the image comparison module is used for comparing and analyzing data obtained by the data calculation module and judging the treatment condition of the child pneumonia patient; the evaluation module is used for evaluating whether the pneumonia of the child pneumonia patient is worsened or improved by applying a comprehensive evaluation algorithm according to a result of the image comparison module; and a treatment processing module. The method has the advantages that the lung image of the child is accurately recognized and judged, and the disease assessment accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of analysis and evaluation of childhood pneumonia, and more specifically, to a system and method for analyzing and evaluating lung images of childhood pneumonia based on deep learning. Background Art

[0002] In the field of diagnosis of childhood pneumonia, the disease judgment technology based on medical images is crucial. At present, the existing technology mainly relies on doctors to manually observe and analyze children's lung images (such as X-rays and CT images). Doctors rely on their own professional knowledge and experience to identify the pathological characteristics of the lungs in the images, and then judge whether the child has pneumonia and the severity of the disease. For example, doctors will observe whether the lung texture in the image is thickened, whether there are patchy shadows and other typical features to judge the condition. However, this manual analysis method has many shortcomings. In terms of disease judgment, manual interpretation is greatly affected by factors such as the doctor's personal experience and fatigue level. The diagnosis results between different doctors may be different, making it difficult to ensure the accuracy of the diagnosis. Especially when faced with a large amount of imaging data, doctors are prone to fatigue, and the risk of missed diagnosis and misdiagnosis increases significantly, which seriously affects the efficiency and quality of childhood pneumonia diagnosis and may delay the treatment of children.

[0003] In view of this, we propose a deep learning-based lung image analysis and evaluation system and method for children with pneumonia. Summary of the invention

[0004] The purpose of the present invention is to provide a deep learning-based lung image analysis and evaluation system and method for children with pneumonia, so as to solve the technical problem that the accuracy of manual diagnosis is difficult to ensure in the prior art.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a deep learning-based lung image analysis and evaluation system for children with pneumonia, comprising: The lung shadow detection module is used to take lung CT images of the child pneumonia patient being detected and transmit them to the data calculation module; A data calculation module is connected to the lung shadow detection module, and is used to receive images from the lung shadow detection module and calculate alveolar density, clarity, quantity and echo intensity data; The image comparison module compares and analyzes the data obtained by the data calculation module to determine the treatment status of children with pneumonia; The evaluation module uses a comprehensive evaluation algorithm to evaluate whether the pneumonia of children with pneumonia is getting worse or better based on the results of the image comparison module; The treatment module analyzes lung changes and lung shadow clarity based on the evaluation results of the evaluation module and generates treatment recommendations using a decision-making algorithm; A data recording module, storing the results of data calculation and image comparison, as well as the data generated by the evaluation module and the treatment processing module; The terminal device analyzes and summarizes the data from the data recording module, generates patient treatment data and change curves, and assists doctors in making treatment decisions and recording them.

[0006] Preferably, the method for calculating alveolar density, clarity, quantity and echo intensity data in the data calculation module is: After the data calculation module receives the image transmitted by the lung shadow detection module, it uses the image analysis algorithm based on the U-Net deep learning network to preprocess the image and obtain the preprocessed image. , and then use the trained U-Net network to Locate the alveolar region and mark each alveolar space, including: is the convolution kernel weight, is the bias term, is the activation function, is the pooling window; The data calculation module calculates the number of alveoli in the following way: After the alveolar region is located, the marked alveolar spaces are analyzed. The total number of marked alveolar spaces is By counting the alveoli in each alveolar space, the The number of alveoli in the alveolar space , then the total number of alveoli in the entire lung image ; The data calculation module calculates the alveolar density in the following way: Record the number of alveoli in each alveolar space and the area of ​​the alveolar space in the lung image The total lung image area is , after training with a large amount of sample data, the coefficients , alveolar density The calculation formula is: .

[0007] Preferably, the data calculation module calculates the alveolar clarity in the following manner: Using image texture extraction algorithm from preprocessed images To obtain lung texture information, assume The average gray value in the alveolar space is The number of alveoli in the alveolar space is , the standard deviation of its gray value is , define the fuzziness index; ; The alveolar blur level of the entire lung image is obtained by taking a weighted average of the blur level indices of all alveolar spaces. ; ; in, For the The weight of the alveolar space is calculated according to the degree of blur Judge the clarity of the alveoli. The lower the blur, the higher the clarity of the alveoli. The data calculation module calculates the alveolar echo intensity in the following way: Assume that the alveolar density of this test is , the alveolar density last detected was The alveolar clarity of this test is , the alveolar clarity detected last time was , defining the density change rate , resolution change rate , echo intensity The judgment logic is: when and hour, ; when and hour, .

[0008] Preferably, the method by which the image comparison module compares and analyzes the data obtained by the data calculation module is: Assume that the current data vector obtained by the data calculation module is ,in Represents alveolar density, clarity, quantity and echo intensity. The historical data vector set is , using cosine similarity Mahalanobis distance Make comprehensive judgments; ; ; in, Represents the covariance matrix of the data, by setting the similarity threshold and Mahalanobis distance threshold ,when and When , the current data and the historical data are considered to have similar treatment conditions; The method for the image comparison module to judge the treatment status of children with pneumonia is: For each indicator Setting weights ,and , calculate the current data vector With the historical optimal data vector The combined distance between , the formula is: ; in, yes The elements, according to the comprehensive distance Determine the treatment situation; when When the treatment condition is judged to be improved, is a pre-set improvement threshold; when When the treatment situation is judged to be worse, is a pre-set deterioration threshold; when The treatment situation was judged to be stable.

[0009] Preferably, the method in which the assessment module assesses whether the pneumonia of a child with pneumonia is worsening or improving is: Establishing an evaluation indicator system; Based on the judgment results of the image comparison module, the alveolar density is comprehensively considered , clarity ,quantity , echo intensity And the comprehensive distance of the image comparison module , assuming these indicators and The weights are , , , , ,and ; First, each indicator is normalized. , , , are the normalized functions of alveolar density, clarity, quantity, and echo intensity, respectively, and are mapped to Interval, for comprehensive distance Using normalization function ; ; in, and are the maximum and minimum values ​​of the comprehensive distance in the historical data respectively; Defining Assessment Scores ; Based on the evaluation score With the preset deterioration threshold and improvement threshold Compare and determine whether pneumonia is getting worse, getting better, or is stable; Based on the evaluation score Further divide the risk level into For low risk, For medium risk, For high risk, the corresponding score ranges are , , .

[0010] Preferably, the treatment module generates treatment suggestions using a decision algorithm in the following manner: Combined with the evaluation results and risk levels of the evaluation module, the clinical cases and treatment plans in the medical database are deeply mined. Suppose the case set stored in the medical database is , using the nearest neighbor algorithm to calculate the current evaluation score The distance from the evaluation scores of each case in the database , select the closest Cases, determine the recommended treatment plan by weighted voting ; ; ; in, For the most recent A collection of cases, is the indicator function, represents the treatment regimen variable, Indicates the first Treatment options for each case; Based on different risk levels and changes in the condition, emergency treatment measures are recommended for high-risk patients, suggestions for adjusting existing treatment plans are provided for medium-risk patients, and maintenance treatment or gradual reduction of treatment plans are given for low-risk patients.

[0011] Preferably, the data recording module stores alveolar density, clarity, quantity and echo intensity data and corresponding image images, records the patient's treatment status in detail in a structured table form, and stores the evaluation results of the evaluation module, risk level and treatment recommendations generated by the treatment processing module.

[0012] A method for analyzing and evaluating lung images of children with pneumonia based on deep learning, comprising the following steps: S1, image acquisition; Capture lung images through the lung shadow detection module; S2, image processing and data calculation; The acquired images are transmitted to the data calculation module, and the alveolar density, clarity, quantity and echo intensity data are calculated; S3, indicator analysis and disease diagnosis; Receive data on alveolar density, clarity, quantity and echo intensity, and use cosine similarity, Mahalanobis distance and comprehensive distance algorithms to determine the progression of the disease; S4. Disease assessment and risk stratification; Based on the results of the image contrast module, the assessment score is calculated to evaluate the worsening or improvement of pneumonia and determine the risk level; S5. Imaging evaluation and treatment decision-making; Based on the results of the evaluation module, a decision-making algorithm is used to generate treatment recommendations.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention constructs a pediatric pneumonia lung image analysis and evaluation system and method based on deep learning, and utilizes the powerful image feature extraction and analysis capabilities of deep learning to accurately identify and judge lung images. It can accurately extract lesion features, avoid the subjectivity and individual differences of manual interpretation, greatly improve the accuracy of disease judgment, and effectively solve the problem that the accuracy of manual diagnosis is difficult to ensure in the prior art.

[0014] 2. Compared with manual interpretation, the present invention greatly improves the diagnostic efficiency. By analyzing and evaluating a large number of images in a short period of time, it can provide doctors with timely diagnostic suggestions, reduce the time that children have to wait for diagnostic results, and further solve the problem of low efficiency of manual diagnosis when facing a large amount of image data.

[0015] 3. The present invention can quantitatively analyze the pathological features in lung images, which can not only determine whether pneumonia is present, but also more accurately assess the severity of the disease, providing doctors with a more detailed and accurate basis for formulating personalized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the system framework of the present invention; Figure 2 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0017] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0018] Embodiment 1, as Figure 1 As shown, a deep learning-based lung image analysis and evaluation system for children with pneumonia includes: The lung shadow detection module is used to take lung CT images of the child pneumonia patient being detected and transmit them to the data calculation module; A data calculation module is connected to the lung shadow detection module, and is used to receive images from the lung shadow detection module and calculate alveolar density, clarity, quantity and echo intensity data; The image comparison module compares and analyzes the data obtained by the data calculation module to determine the treatment status of children with pneumonia; The evaluation module uses a comprehensive evaluation algorithm to evaluate whether the pneumonia of children with pneumonia is getting worse or better based on the results of the image comparison module; The treatment module analyzes lung changes and lung shadow clarity based on the evaluation results of the evaluation module and generates treatment recommendations using a decision-making algorithm; A data recording module, storing the results of data calculation and image comparison, as well as the data generated by the evaluation module and the treatment processing module; The terminal device analyzes and summarizes the data from the data recording module, generates patient treatment data and change curves, and assists doctors in making treatment decisions and recording them.

[0019] In an embodiment of the present invention, the method for calculating alveolar density, clarity, quantity and echo intensity data in the data calculation module is: After receiving the images transmitted by the lung shadow detection module, the data calculation module pre-processes the images using the image analysis algorithm based on the U-Net deep learning network and performs convolution operations on the images. , max pooling operation And interpolation upsampling, get the preprocessed image ; During the training process, the model continuously adjusts these parameters based on the feedback of the loss function to minimize the difference between the predicted results and the true labels; The mean square error loss function For example, for a The training set of samples is used, and the loss function is ,in, is the true label, is the model prediction value, and the loss function is calculated with respect to and The gradient of , using optimization algorithms such as gradient descent to update them, the update formula is , ; Using the Sigmoid function, ; In the maximum pooling operation, the pooling window Used to determine the size of the pooling area on the input feature map, which is 2×2; Then use the trained U-Net network to Locate the alveolar region and mark each alveolar space, including: is the convolution kernel weight, is the updated convolution kernel weight, is the bias term, is the updated bias term, is the learning rate, Represents the loss function Convolution kernel weights The gradient of Represents the loss function for the bias term The gradient of is the activation function, Represents the output value of the Sigmoid activation function, is the input of the activation function, is a natural constant, is the pooling window; The data calculation module calculates the number of alveoli in the following way: After the alveolar region is located, the marked alveolar spaces are analyzed. The total number of marked alveolar spaces is By counting the alveoli in each alveolar space, the The number of alveoli in the alveolar space , then the total number of alveoli in the entire lung image ; The data calculation module calculates the alveolar density in the following way: Record the number of alveoli in each alveolar space and the area of ​​the alveolar space in the lung image The total lung image area is , after training with a large amount of sample data, the coefficients , alveolar density The calculation formula is: .

[0020] In an embodiment of the present invention, the data calculation module calculates the alveolar clarity in the following manner: Using image texture extraction algorithm from preprocessed images To obtain lung texture information, assume The average gray value in the alveolar space is The number of alveoli in the alveolar space is , the standard deviation of its gray value is , define the fuzziness index ; ; The alveolar blur level of the entire lung image is obtained by taking a weighted average of the blur level indices of all alveolar spaces. ; ; in, For the The weight of the alveolar space is calculated according to the degree of blur Judge the clarity of the alveoli. The lower the blur, the higher the clarity of the alveoli. The data calculation module calculates the alveolar echo intensity in the following way: Assume that the alveolar density of this test is , the alveolar density last detected was The alveolar clarity of this test is , the alveolar clarity detected last time was , defining the density change rate , resolution change rate , echo intensity The judgment logic is: when and hour, ; when and hour, .

[0021] In an embodiment of the present invention, the method by which the image comparison module compares and analyzes the data obtained by the data calculation module is: Assume that the current data vector obtained by the data calculation module is ,in Represents alveolar density, clarity, quantity and echo intensity. The historical data vector set is , using cosine similarity Mahalanobis distance Make comprehensive judgments; ; ; in, Represents the covariance matrix of the data, by setting the similarity threshold and Mahalanobis distance threshold ,when and When , the current data and the historical data are considered to have similar treatment conditions; The method for the image comparison module to judge the treatment status of children with pneumonia is: For each indicator Setting weights ,and , calculate the current data vector With the historical optimal data vector The comprehensive distance between the historical data vectors with the best treatment effect , the formula is: ; in, yes The elements, according to the comprehensive distance Determine the treatment situation; when When the treatment condition is judged to be improved, is a pre-set improvement threshold; when When the treatment situation is judged to be worse, is a pre-set deterioration threshold; when The treatment situation was judged to be stable.

[0022] In an embodiment of the present invention, the method in which the evaluation module evaluates whether the pneumonia of a child with pneumonia is worsening or improving is: Establishing an evaluation indicator system; Based on the judgment results of the image comparison module, the alveolar density is comprehensively considered , clarity ,quantity , echo intensity And the comprehensive distance of the image comparison module , assuming these indicators and The weights are , , , , ,and ; First, each indicator is normalized. , , , are the normalized functions of alveolar density, clarity, quantity, and echo intensity, respectively, and are mapped to Interval, for comprehensive distance Using normalization function ; ; in, and are the maximum and minimum values ​​of the comprehensive distance in the historical data respectively; Defining Assessment Scores ; Based on the evaluation score With the preset deterioration threshold and improvement threshold Compare and determine whether pneumonia is getting worse, getting better, or is stable; Based on the evaluation score Further divide the risk level into For low risk, For medium risk, For high risk, the corresponding score ranges are , , .

[0023] In an embodiment of the present invention, the method in which the treatment module generates treatment suggestions using the decision algorithm is: Combined with the evaluation results and risk levels of the evaluation module, the clinical cases and treatment plans in the medical database are deeply mined. Suppose the case set stored in the medical database is , using the nearest neighbor algorithm to calculate the current evaluation score The distance from the evaluation scores of each case in the database , select the closest Cases, determine the recommended treatment plan by weighted voting ; ; ; in, For the most recent A collection of cases, is the indicator function, represents the treatment regimen variable, Indicates the first Treatment options for each case; Based on different risk levels and changes in the condition, emergency treatment measures are recommended for high-risk patients, suggestions for adjusting existing treatment plans are provided for medium-risk patients, and maintenance treatment or gradual reduction of treatment plans are given for low-risk patients.

[0024] In an embodiment of the present invention, the data recording module stores alveolar density, clarity, quantity and echo intensity data and corresponding image images, records the patient's treatment status in detail in a structured table form, and stores the evaluation results of the evaluation module, risk level and treatment recommendations generated by the treatment processing module.

[0025] Embodiment 2, as Figure 2 As shown, a method for analyzing and evaluating lung images of children with pneumonia based on deep learning includes the following steps: S1, image acquisition; Capture lung images through the lung shadow detection module; S2, image processing and data calculation; The acquired images are transmitted to the data calculation module, and the alveolar density, clarity, quantity and echo intensity data are calculated; S3, indicator analysis and disease diagnosis; Receive data on alveolar density, clarity, quantity and echo intensity, and use cosine similarity, Mahalanobis distance and comprehensive distance algorithms to determine the progression of the disease; S4. Disease assessment and risk stratification; Based on the results of the image contrast module, the assessment score is calculated to evaluate the worsening or improvement of pneumonia and determine the risk level; S5. Imaging evaluation and treatment decision-making; Based on the results of the evaluation module, a decision-making algorithm is used to generate treatment recommendations.

[0026] The embodiments of the present invention disclose preferred embodiments, but are not limited thereto. A person skilled in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not deviate from the spirit of the present invention, they are all within the protection scope of the present invention.

Claims

1. A deep learning-based lung image analysis and evaluation system for children with pneumonia, characterized in that: include: The lung shadow detection module is used to take lung CT images of the child pneumonia patient being detected and transmit them to the data calculation module; A data calculation module is connected to the lung shadow detection module, and is used to receive images from the lung shadow detection module and calculate alveolar density, clarity, quantity and echo intensity data; The image comparison module compares and analyzes the data obtained by the data calculation module to determine the treatment status of children with pneumonia; The evaluation module uses a comprehensive evaluation algorithm to evaluate whether the pneumonia of children with pneumonia is getting worse or better based on the results of the image comparison module; The treatment module analyzes lung changes and lung shadow clarity based on the evaluation results of the evaluation module and generates treatment recommendations using a decision-making algorithm; The data recording module stores the results of data calculation and image comparison, as well as the data generated by the evaluation module and the treatment processing module.

2. According to claim 1, a deep learning-based lung image analysis and evaluation system for children with pneumonia, characterized in that: The method for calculating alveolar density, clarity, quantity and echo intensity data in the data calculation module is: After the data calculation module receives the image transmitted by the lung shadow detection module, it uses the image analysis algorithm based on the U-Net deep learning network to preprocess the image and obtain the preprocessed image. , and then use the trained U-Net network to Locate the alveolar region and mark each alveolar space; The method for calculating the number of alveoli by the data calculation module is: After the alveolar region is located, the marked alveolar spaces are analyzed. The total number of marked alveolar spaces is By counting the alveoli in each alveolar space, the The number of alveoli in the alveolar space , then the total number of alveoli in the entire lung image ; The data calculation module calculates the alveolar density in the following way: Record the number of alveoli in each alveolar space and the area of ​​the alveolar space in the lung image The total lung image area is , after training with a large amount of sample data, the coefficients , alveolar density The calculation formula is: 。 3. According to claim 2, a deep learning-based lung image analysis and evaluation system for children with pneumonia, characterized in that: The data calculation module calculates the alveolar clarity in the following way: Using image texture extraction algorithm from preprocessed images To obtain lung texture information, assume The average gray value in the alveolar space is The number of alveoli in the alveolar space is , the standard deviation of its gray value is , define the fuzziness index ; ； The alveolar blur level of the entire lung image is obtained by taking a weighted average of the blur level indices of all alveolar spaces. ; ； in, For the The weight of the alveolar space is calculated according to the degree of blur Judge the clarity of the alveoli. The lower the blur, the higher the clarity of the alveoli. The data calculation module calculates the alveolar echo intensity in the following way: Assume that the alveolar density of this test is , the alveolar density last detected was The alveolar clarity of this test is , the alveolar clarity detected last time was , defining the density change rate , resolution change rate , echo intensity The judgment logic is: when and hour, ; when and hour, .

4. According to claim 3, a deep learning-based lung image analysis and evaluation system for children with pneumonia is characterized in that: The method of comparing and analyzing the data obtained by the data calculation module by the image comparison module is as follows: Assume that the current data vector obtained by the data calculation module is ,in Represents alveolar density, clarity, quantity and echo intensity. The historical data vector set is , using cosine similarity Mahalanobis distance Make comprehensive judgments; ； ； in, Represents the covariance matrix of the data, by setting the similarity threshold and Mahalanobis distance threshold ,when and When , the current data and the historical data are considered to have similar treatment conditions; The method for the image comparison module to judge the treatment status of children with pneumonia is: For each indicator Setting weights ,and , calculate the current data vector With the historical optimal data vector The combined distance between , the formula is: ； in, yes The elements, according to the comprehensive distance Determine the treatment situation; when When the treatment condition is judged to be improved, is a pre-set improvement threshold; when When the treatment situation is judged to be worse, is a pre-set deterioration threshold; when The treatment situation was judged to be stable.

5. According to claim 4, a deep learning-based lung image analysis and evaluation system for children with pneumonia, characterized in that: The method by which the assessment module assesses whether the pneumonia of a child with pneumonia is worsening or improving is: Establishing an evaluation indicator system; Based on the judgment results of the image comparison module, the alveolar density is comprehensively considered , clarity ,quantity , echo intensity And the comprehensive distance of the image comparison module , assuming these indicators and The weights are , , , , ,and ; First, each indicator is normalized. , , , are the normalized functions of alveolar density, clarity, quantity, and echo intensity, respectively, and are mapped to Interval, for comprehensive distance Using normalization function ; ； in, and are the maximum and minimum values ​​of the comprehensive distance in the historical data respectively; Defining Assessment Scores ; Based on the evaluation score With the preset deterioration threshold and improvement threshold Compare and determine whether pneumonia is getting worse, getting better, or is stable; Based on the evaluation score Further divide the risk level into For low risk, For medium risk, For high risk, the corresponding score ranges are , , .

6. A deep learning-based pediatric pneumonia lung image analysis and evaluation system according to claim 5, characterized in that: The method in which the treatment processing module uses the decision algorithm to generate treatment suggestions is: Combined with the evaluation results and risk levels of the evaluation module, the clinical cases and treatment plans in the medical database are deeply mined. Suppose the case set stored in the medical database is , using the nearest neighbor algorithm to calculate the current evaluation score The distance from the evaluation scores of each case in the database , select the closest The recommended treatment plan is determined by weighted voting. ; ； ； in, For the most recent A collection of cases, is the indicator function, represents the treatment regimen variable, Indicates the first Treatment options for each case; Based on different risk levels and changes in the condition, emergency treatment measures are recommended for high-risk patients, suggestions for adjusting existing treatment plans are provided for medium-risk patients, and maintenance treatment or gradual reduction of treatment plans are given for low-risk patients.

7. A deep learning-based pediatric pneumonia lung image analysis and evaluation system according to claim 6, characterized in that: The data recording module stores alveolar density, clarity, quantity and echo intensity data and corresponding image images, records the patient's treatment status in detail in a structured table form, and stores the evaluation results of the evaluation module, risk level and treatment suggestions generated by the treatment processing module.

8. An evaluation method applied to the deep learning-based pediatric pneumonia lung image analysis and evaluation system as claimed in claim 7, characterized in that: The following steps are involved: S1, image acquisition; Capture lung images through the lung shadow detection module; S2, image processing and data calculation; The acquired images are transmitted to the data calculation module, and the alveolar density, clarity, quantity and echo intensity data are calculated; S3, indicator analysis and disease diagnosis; Receive data on alveolar density, clarity, quantity and echo intensity, and use cosine similarity, Mahalanobis distance and comprehensive distance algorithms to determine the progression of the disease; S4. Disease assessment and risk stratification; Based on the results of the image contrast module, the assessment score is calculated to evaluate the worsening or improvement of pneumonia and determine the risk level; S5. Imaging evaluation and treatment decision-making; Based on the results of the evaluation module, a decision-making algorithm is used to generate treatment recommendations.

Images