Focus image recognition method and system based on deep learning model

Abstract

The invention discloses a focus image recognition method and system based on a deep learning model, and relates to the technical field of medical equipment. Comprising the following steps: acquiring medical images of various lesion types, and preprocessing the medical images to generate sample images; marking a focus area, extracting image feature parameters, and constructing a sample training data set; and based on the data set, training a segmentation identification model, preprocessing a medical image to be identified, identifying and segmenting a focus area by using the trained segmentation identification model, inputting a feature extraction model to obtain an image feature parameter prediction value, and calculating a biochemical discrimination influence index in combination with biochemical index data of a patient. And generating a comprehensive lesion identification coefficient according to the lesion texture complexity index, the edge influence index and the biochemical discrimination influence index, and comparing the comprehensive lesion identification coefficient with a set threshold to complete lesion image identification. And the accuracy and reliability of focus identification are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical equipment, in particular to a lesion image recognition method and system based on a deep learning model. BACKGROUND

[0002] With the development of medical imaging technology, medical images play an increasingly important role in disease diagnosis, monitoring and treatment. In recent years, computer vision and deep learning technology have made significant progress in medical image analysis, especially in automatic identification and segmentation of lesions. However, existing medical image recognition technology still faces many challenges.

[0003] First, the diversity and complexity of medical images make the task of lesion recognition difficult. Different types of lesions (such as tumors, inflammation and other lesions) vary in shape, size, color and texture, and are often accompanied by changes in surrounding tissues. This diversity makes it difficult for traditional image processing methods to effectively identify and segment lesion areas, affecting clinical treatment decisions.

[0004] Second, the correlation between biochemical indicators and tumor lesions is also an important research area. Biochemical indicators (such as erythrocyte sedimentation rate, white blood cell count, carcinoembryonic antigen level and alpha-fetoprotein level) have an important influence on the characteristics of lesions and the survival prognosis of patients. However, current technology mostly processes image data and biochemical data separately, failing to fully utilize the complementarity between the two. Lack of comprehensive analysis leads to insufficient comprehensive and accurate discrimination and evaluation of lesions.

[0005] Therefore, in view of the above problems, a medical image recognition method based on deep learning is developed, which can effectively improve the recognition accuracy and reliability of lesions through image preprocessing, feature extraction, model training and comprehensive analysis, and has important clinical significance.

[0006] In the prior art, the authorization announcement No. CN110232383B discloses a lesion image recognition method and system based on a deep learning model. By sequentially performing segmentation of the smallest organ tissue image, image recognition of the corresponding organ tissue, deep learning model prediction and marking on the prediction result image, the potential disease in the medical image can be found and the lesion tissue in the medical image can be automatically recognized and marked, so as to remind the doctor to make further diagnosis, timely diagnose the disease, and thus reduce the work intensity of the doctor, timely diagnose whether the lesion occurs, avoid delaying the treatment opportunity of the disease, and especially help to find the early lesion. However, in this scheme, although the system can assist the doctor to identify the potential lesion, the final clinical judgment still depends on the professional knowledge and experience of the doctor. At the same time, the performance of the deep learning model is highly dependent on the quality and quantity of the training data. If the training data is insufficient or not diversified enough, the model may not be able to effectively identify all types of lesions. Therefore, relying only on the model for judgment will reduce the accuracy and effectiveness of the recognition system.

[0007] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0008] The purpose of the present application is to provide a lesion image recognition method and system based on a deep learning model to solve the problems raised in the background.

[0009] To achieve the above purpose, the present application provides the following technical solutions:

[0010] A lesion image recognition method based on a deep learning model, the specific steps comprising:

[0011] Obtain a plurality of medical images of different lesion types, and perform image preprocessing to obtain sample images. The lesion area in the sample image is marked by manual marking, and the lesion marking area image feature parameters are obtained to generate a sample training data set;

[0012] Based on the sample training data set, a segmentation and recognition model is established. The sample image is used as input, and the lesion area marking and image feature parameters in the image are used as labels to train the segmentation and recognition model;

[0013] Obtain the lesion medical image to be recognized, input the trained segmentation and recognition model, obtain the image feature parameter prediction value of the lesion medical image to be recognized, and obtain the biochemical index data of the patient corresponding to the lesion medical image to be recognized;

[0014] Based on the obtained image feature parameter prediction value, the lesion texture complexity and the lesion edge influence are characterized, and the biochemical influence is evaluated based on the biochemical index data;

[0015] According to the lesion texture complexity and the lesion edge influence, combined with the biochemical influence, a comprehensive lesion recognition coefficient is generated, the comprehensive lesion recognition coefficient is compared with the set lesion judgment threshold, different comparison results are generated, corresponding lesion recognition judgment results are issued, and lesion image recognition is completed.

[0016] Further, the image feature parameters include color contrast, texture entropy, lesion edge strength, lesion contour perimeter and lesion contour area; the lesion area in the sample image is marked in an artificial marking manner, specifically by marking tools in the image, and the different lesion types specifically include hemorrhage, inflammation and tumor;

[0017] The image preprocessing specifically includes image enhancement and denoising preprocessing, wherein a wavelet transform denoising method is used to denoise each lesion medical image, and a bilateral filter is used to perform image enhancement preprocessing on each lesion medical image.

[0018] The training sample data set is generated by one-to-one mapping the marked sample image and the corresponding image feature parameters to form a corresponding grid, and the formed grid is recorded as the training sample data set.

[0019] Further, a segmentation recognition model is established based on the sample training data set, wherein the segmentation recognition model is constructed by an LSTM model, an activation function and an optimization algorithm are selected, wherein a Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm of the LSTM model; the formula of the Tanh function is:

[0020]

[0021] In the formula, Tanh function, independent variable Indicates the input weighted sum of neurons, that is, the result of the weighted sum of the inputs received by the neurons from the previous layer;

[0022] Meanwhile, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing quantity and the number of hidden layer neurons.

[0023] The number of network layers is set to 5-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch processing quantity is set to 128, and the number of hidden layer neurons is set to 32.

[0024] The segmented and recognized model after training inputs a medical image of a lesion, and outputs an image with a marked lesion region and corresponding predicted values of image feature parameters, including predicted values of color contrast, texture entropy, lesion edge strength, lesion contour perimeter, and lesion contour area.

[0025] Further, based on the obtained predicted values of image feature parameters, a lesion texture complexity index and a lesion edge influence index are calculated, wherein a formula for calculating the lesion texture complexity index is specifically as follows:

[0026]

[0027] In the formula, is the lesion texture complexity index, is a normalized value of the predicted value of the color contrast of the image of the marked region, is the predicted value of the texture entropy of the image of the marked region, is the maximum predicted value of the texture entropy of all images of the marked region;

[0028] wherein a formula for calculating the lesion edge influence index is as follows:

[0029]

[0030] In the formula, is the lesion edge influence index, is the predicted value of the lesion contour perimeter of the image of the marked region, is the predicted value of the lesion contour area of the image of the marked region, is the lesion edge strength of the image of the marked region, is the circularity of the lesion contour of the image of the marked region.

[0031] Further, the circularity of the lesion contour of the image of the marked region is specifically calculated according to the formula:

[0032]

[0033] The biochemical index data includes erythrocyte sedimentation rate, white blood cell count, carcinoembryonic antigen level, and alpha-fetal protein level, and a biochemical discrimination influence index is calculated based on the biochemical index data, wherein a formula for calculating the biochemical discrimination influence index is specifically as follows:

[0034]

[0035] In the formula, is the biochemical discrimination influence index, is a normalized value of the erythrocyte sedimentation rate of the corresponding patient, is a normalized value of the white blood cell count of the corresponding patient, and respectively are the normalized values of carcinoembryonic antigen level and alpha-fetoprotein level of the corresponding patient.

[0036] Further, according to the obtained lesion texture complexity index and lesion edge influence index, a comprehensive lesion recognition coefficient is generated in combination with a biochemical discrimination influence index, wherein a formula for calculating the comprehensive lesion recognition coefficient is as follows:

[0037]

[0038] In the formula, is a comprehensive lesion recognition coefficient, , and respectively are weight coefficients of the biochemical discrimination influence index, the lesion edge influence index and the lesion texture complexity index, wherein and , and are all greater than 0.

[0039] Further, the comprehensive lesion recognition coefficient is compared with a set lesion judgment threshold value to generate different comparison results, and a corresponding lesion recognition judgment result is issued, wherein the specific judgment logic is as follows:

[0040] when , it is judged that the lesion in the to-be-recognized lesion medical image is a tumor, and other means should be taken for further diagnosis and treatment;

[0041] when , it is judged that the lesion in the to-be-recognized lesion medical image is inflammation, and the inflammation should be treated;

[0042] when , it is judged that the lesion in the to-be-recognized lesion medical image is hemorrhage;

[0043] wherein is a lesion judgment threshold value.

[0044] The application also provides a lesion image recognition system based on a deep learning model, which is used to execute the above-mentioned lesion image recognition method based on a deep learning model, and comprises:

[0045] a sample data acquisition module, which is used to acquire medical images of several different lesion types, and perform image preprocessing to obtain sample images, mark the lesion area in the sample images in an artificial marking manner, acquire lesion marking area image feature parameters, and generate a sample training data set;

[0046] The prediction model training module is configured to establish a segmentation identification model based on a sample training data set, take a sample image as input, and take a label of a lesion region in the image and an image feature parameter as a label, and train the segmentation identification model;

[0047] The biochemical index collection module is configured to obtain a to-be-identified lesion medical image, input the trained segmentation identification model, and obtain an image feature parameter prediction value of the to-be-identified lesion medical image, and biochemical index data corresponding to the to-be-identified lesion medical image of a patient;

[0048] The influence parameter analysis module is configured to characterize lesion texture complexity and lesion edge influence based on the obtained image feature parameter prediction value, and evaluate biochemical influence based on the biochemical index data.

[0049] The lesion image identification module is configured to generate a comprehensive lesion identification coefficient according to the lesion texture complexity and the lesion edge influence, combine the biochemical influence, compare the comprehensive lesion identification coefficient with a set lesion judgment threshold, generate different comparison results, issue a corresponding lesion identification judgment result, and complete lesion image identification.

[0050] Compared with the prior art, the present application has the following advantages:

[0051] Firstly, the image preprocessing and manual labeling method is adopted to comprehensively analyze the collected different types of medical images, so that the lesion region labeling is more accurate, thereby providing high-quality training data for the deep learning model and improving the training efficiency and accuracy of the model.

[0052] Secondly, the established segmentation identification model can fully capture the shape, texture and structure information of the lesion through accurate segmentation of the lesion region and comprehensive feature extraction. By extracting multi-dimensional features such as color contrast, texture entropy, edge intensity, contour perimeter and contour area, a comprehensive description of the lesion is provided, which helps to improve the recognition rate of the lesion. In addition, combined with the biochemical index data of the to-be-identified medical image, multi-dimensional analysis of the lesion can be realized.

[0053] In addition, the combination of biochemical indicators and lesion features forms comprehensive indicators such as lesion texture complexity index and biochemical discrimination influence index. This multi-source data-based analysis method not only improves the accuracy of lesion identification, but also provides more rich information support for clinicians, which can better assist decision-making. Finally, by generating a comprehensive lesion identification coefficient and comparing it with a set judgment threshold, intelligent and automatic lesion identification can be realized. This process simplifies the workflow of doctors, improves the diagnosis efficiency, and provides more timely medical intervention for patients. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1The whole method flowchart of the application is shown in the figure;

[0055] Figure 2 The erythrocyte sedimentation rate normalized value-white blood cell count normalized value fitting curve is shown in the figure;

[0056] Figure 3 The erythrocyte sedimentation rate normalized value-carcinoembryonic antigen level normalized value fitting curve is shown in the figure;

[0057] Figure 4 The white blood cell count normalized value-alpha-fetoprotein level normalized value fitting curve is shown in the figure;

[0058] Figure 5 The white blood cell count normalized value-biochemical discriminant influence index color mapping is shown in the figure;

[0059] Figure 6 The whole system structure diagram of the application is shown in the figure. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical scheme and advantages of the application more clear and understandable, the application will be further described in detail below in combination with specific embodiments.

[0061] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the application should be understood as the general meaning understood by those skilled in the art to which the application belongs. The "first", "second" and similar words used in the application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent the relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0062] EMBODIMENT

[0063] Please refer to Figures 1-5 The application provides a technical scheme:

[0064] A lesion image recognition method based on a deep learning model, the specific steps comprising:

[0065] Step 1: Obtain a plurality of medical images of different lesion types, and perform image preprocessing to obtain sample images. The lesion area in the sample image is marked by manual marking, and the lesion marking area image feature parameters are obtained to generate a sample training data set.

[0066] The lesion area in the sample image is marked by artificial marking, specifically by marking tool in the image, specifically using LabelImg marking tool for marking, selecting irregular curve to mark the lesion contour in the image, after completing all marking, ensuring that the marking result is saved in a proper format (such as XML, JSON, CSV, etc.), so as to facilitate subsequent model training and analysis. The different lesion types specifically include hemorrhage, inflammation and tumor;

[0067] The image preprocessing specifically includes image enhancement and denoising preprocessing, wherein a wavelet transform denoising method is used to denoise each lesion medical image, and a bilateral filter is used to perform image enhancement preprocessing on each lesion medical image;

[0068] The method for denoising and enhancing the lesion medical image is: a wavelet transform denoising method is used to denoise the distortion-corrected image, and the specific steps of the wavelet transform denoising method include: the distortion-corrected image is decomposed by wavelet transform to obtain wavelet coefficients of the image in different scales and directions; the wavelet coefficients are threshold processed, the low-amplitude wavelet coefficients are set to zero, and the high-amplitude wavelet coefficients are retained; the wavelet coefficients after threshold processing are inverse transformed to reconstruct the processed coefficients into an image, and the image denoising processing is completed;

[0069] The lesion medical image is enhanced by a bilateral filter, and the formula on which the filter transform is based is:

[0070]

[0071] In the formula, is a coordinate vector in the image coordinate system, is a coordinate vector is a gray value at the coordinate, is a gray value is a gray value after bilateral filter transform, are all Gaussian functions, wherein The formula on which the filter transform is based is:

[0072]

[0073]

[0074] In the formula, is a coordinate vector in the image coordinate system, is a coordinate vector is a gray value at the coordinate, and are standard deviations of

[0075] ​Color contrast generally refers to the difference degree of color distribution in the image. It can be obtained by the following steps: converting the image from the RGB color space to the HSV or Lab color space for better analysis of the color, calculating the histogram using the function in OpenCV, and then calculating the color contrast using the same.

[0076] Texture entropy is a parameter used to describe the complexity of image texture. The gray level co-occurrence matrix of the image is calculated to describe the spatial relationship of different gray level pixels. Then the probability distribution of each gray level is calculated through the gray level co-occurrence matrix, and the texture entropy is calculated using the entropy formula.

[0077] Edge strength describes the saliency of edges in the image. It can be obtained by using edge detection algorithms such as Canny edge detection or Sobel operator. The edge strength can be obtained by calculating the gradient amplitude of the edge image or using the output result of the edge detection algorithm, which is usually the mean value of the intensity values of the edge pixels.

[0078] Contour perimeter refers to the total length of the boundary of an object in an image. The edge image is obtained using an edge detection algorithm (such as Canny), and then the contour detection function is used. Similarly, the contour area is obtained using the edge detection and contour detection method.

[0079] The method for generating the training sample data set is: mapping the labeled sample image and the corresponding image feature parameter one by one to form a corresponding grid, and recording the formed grid as the training sample data set.

[0080] Step 2: Establish a segmentation and recognition model based on the sample training data set, take the sample image as the input, and take the label of the lesion area in the image and the image feature parameter as the label, and train the segmentation and recognition model.

[0081] The segmentation and recognition model is established based on the sample training data set, wherein the segmentation and recognition model is constructed by an LSTM model, an activation function and an optimization algorithm are selected, wherein a Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm of the LSTM model; the formula of the Tanh function is:

[0082]

[0083] In the formula, Tanh function, independent variable represents the input weighted sum of neurons, that is, the result of the weighted sum of the inputs received by the neurons from the previous layer;

[0084] Meanwhile, the hyperparameters of the LSTM model are set, including: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the number of batches, and the number of hidden layer neurons.

[0085] The network layer number is set to a 5-layer network structure, the iteration number is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the training number is set to 100, the batch processing number is set to 128, and the number of hidden layer neurons is 32.

[0086] The trained segmentation recognition model takes the lesion medical image as input and outputs the image with the marked lesion region and the corresponding image feature parameter prediction values, including the prediction values of color contrast, texture entropy, lesion edge strength, lesion contour perimeter, and lesion contour area.

[0087] LSTM can capture the nonlinear relationship of data through its complex network structure. This is particularly important for feature extraction such as color contrast and texture entropy, as the relationship between these features is often not linear. Feature parameters such as color contrast and edge strength usually change over time in image sequences. LSTM can adapt to input sequences of different lengths, so it performs particularly well when dealing with irregular or variable-length time series. LSTM can automatically learn features without the need for manually designed complex feature extraction algorithms. Through training, LSTM can learn an effective way to extract color contrast, edge strength, and other features, thereby improving the accuracy and reliability of feature extraction.

[0088] Step 3: Obtain the lesion medical image to be identified, input it into the trained segmentation recognition model, and obtain the image feature parameter prediction values of the lesion medical image to be identified. At the same time, obtain the biochemical indicator data of the patient corresponding to the lesion medical image to be identified.

[0089] Obtain the lesion medical image to be identified, and perform image preprocessing on it. The preprocessing steps are consistent with the above and will not be repeated here.

[0090] The biochemical indicator data includes erythrocyte sedimentation rate, white blood cell count, carcinoembryonic antigen level, and alpha-fetoprotein level.

[0091] Meanwhile, the biochemical index data of the patient corresponding to the to-be-identified lesion medical image is obtained, and the specific data acquisition method is as follows: the biochemical index is usually determined through a blood sample. Medical staff will use aseptic technique to extract a certain amount of blood sample from the patient's body, and ESR is usually determined by using the Westergren method or the Capillary method. The anticoagulated blood is placed in a special test tube, and the sedimentation speed of red blood cells in a certain time is measured, and the number of white blood cells can be counted by using an automatic blood analysis instrument (such as a full-automatic blood cell counter). The detection method of the carcinoembryonic antigen level is: the immunological detection method (such as enzyme-linked immunosorbent assay ELISA or chemiluminescence immunoassay) is used for detection. The detection method of the alpha-fetal protein level is: the enzyme-linked immunosorbent assay (ELISA), the chemiluminescence method or the radioimmunoassay method can also be used for detection.

[0092] Step 4: Based on the obtained image feature parameter prediction value, the lesion texture complexity and the lesion edge influence are characterized, and the biochemical influence is evaluated based on the biochemical index data.

[0093] Based on the obtained image feature parameter prediction value, the lesion texture complexity index and the lesion edge influence index are calculated, and the formula for calculating the lesion texture complexity index is specifically as follows:

[0094]

[0095] In the formula, the lesion texture complexity index, the color contrast prediction value normalization value of the marked region image, the texture entropy prediction value of the marked region image, the maximum texture entropy prediction value of all the marked region images;

[0096] It should be noted that the lesion texture complexity index is used to represent the texture complexity of the lesion region, and the higher the value, the more complex the texture.

[0097] In the formula, the color contrast is used to reflect the diversity of colors in the image, and a higher color contrast means that the lesion region contains more color changes. High color contrast is usually associated with complex structure and morphology, and the probability of the presence of a lesion increases. Therefore, the color contrast prediction value normalization value is directly proportional to the lesion texture complexity index , and the influence of the change in color contrast can be smoothly processed by using a logarithmic function . The specific method for normalization is as follows: the original value of the color contrast is extracted from all the to-be-analyzed images, and the maximum value and the minimum value in the data set are calculated, and the maximum value is set as , and the minimum value is Using a linear normalization formula, the predicted color contrast values ​​of the labeled region image are converted into normalized predicted color contrast values ​​of the labeled region image. The linear normalization formula is a conventional existing technology and will not be elaborated here.

[0098] Texture entropy is an important parameter describing texture complexity. A higher entropy value indicates a more complex and irregular texture, and the more complex and irregular the texture, the greater the probability of the presence of lesions. Therefore, the predicted texture entropy value of the labeled region image... With lesion texture complexity index Proportional, through an exponential function In the denominator, the higher the entropy value, the greater the texture complexity of the lesion. Used to normalize the predicted texture entropy values ​​of the labeled region image.

[0099] The formula used to calculate the lesion margin impact index is as follows:

[0100]

[0101] In the formula, The lesion margin impact index, This is the predicted perimeter of the lesion outline in the labeled region image. The predicted area of ​​the lesion contour in the labeled region image. The intensity of the lesion edge in the labeled region image. The circularity of the lesion outline in the labeled region image.

[0102] It should be noted that the lesion margin impact index It indicates the complexity of the lesion's edge and the degree of its impact on surrounding tissues. The higher the value, the more complex the structure of the lesion area and the more severe the lesion.

[0103] Among these factors, the perimeter of the lesion is a crucial influence on its marginal characteristics. A longer perimeter often indicates a more complex and tortuous edge, potentially leading to a greater impact on surrounding tissues. The ratio of perimeter to area (i.e., the relationship between perimeter and area) provides a quantitative assessment of the impact of the lesion's margins, while edge intensity reflects the clarity of the lesion's edges. Clearer edges may indicate a more significant impact on surrounding tissues. Edge intensity and perimeter together constitute the molecular component; taking the square root of the sum of the squares of perimeter and edge intensity enhances the reflection of their combined characteristics, ultimately leading to a more comprehensive assessment. It reflects the overall characteristics of the edge.

[0104] The concept of roundness is introduced to adjust for the edge influence index. A value closer to 1 indicates a shape closer to a circle, meaning relatively regular edges. Lesions, however, are often irregular in shape, potentially exhibiting serrated or protruding features, or lobulated, spiculated, or infiltrative edges. Therefore, roundness... Influence index of lesion margin Inversely proportional, through It emphasizes the non-linear relationship between shape and edge complexity, while also indicating that when... When reduced, the lesion margin impact index Significantly increased.

[0105] The circularity of the lesion outline in the labeled region image The specific formula used for the calculation is as follows:

[0106]

[0107] The biochemical discriminant influence index is calculated based on biochemical index data. The specific formula for the biochemical discriminant influence index is as follows:

[0108]

[0109] In the formula, To influence the biochemical discrimination index, The normalized value of erythrocyte sedimentation rate for the corresponding patient. This represents the normalized value of the white blood cell count for the corresponding patient. and These are the normalized values ​​of the carcinoembryonic antigen (CEA) and alpha-fetoprotein (AFP) levels for the corresponding patients.

[0110] The specific method used for normalization preprocessing is as follows: data can be normalized by subtracting the maximum and minimum values. This method compresses all data values ​​into a fixed range (usually 0 to 1), eliminating the influence of different units or dimensions between indicators, and making each indicator have the same dimension, which is convenient for subsequent calculations.

[0111] It should be noted that the biochemical discrimination index These are used to assess the importance of certain biochemical indicators in determining the impact of lesions; higher values ​​may indicate more severe or complex lesions.

[0112] Among them, the normalized value of erythrocyte sedimentation rate corresponding to the patient It is an indicator reflecting the state of inflammation or infection, and is usually used clinically to determine the degree of inflammatory response. The higher the value, the greater the probability that the patient has inflammation. Therefore, it corresponds to the patient's normalized erythrocyte sedimentation rate (ESR). Impact of biochemical discrimination index Proportional, through the square The increase in the degree of inflammation has a non-linearly enhanced effect on reflecting that high values can be associated with more severe pathological states.

[0113] The normalized value of the white blood cell count of the patient, which is usually used to assess the immune response or the infection situation, is an important component of the immune system responsible for resisting infection and other diseases. When the body is infected or inflamed, the white blood cell count usually increases, and a higher white blood cell count often indicates the presence of infection, inflammation, or other pathological states. Therefore, the normalized value of the white blood cell count of the patient is directly proportional to the biochemical discriminant impact index , which is prevented from having too much impact on the result by the logarithmic function to prevent excessive volatility from having too much impact on the result and to ensure stability.

[0114] The level of carcinoembryonic antigen (CEA), which is usually used to monitor the presence of tumors and their activity. An increase in the level of carcinoembryonic antigen is usually associated with the presence or development of tumors, so the normalized value of the level of carcinoembryonic antigen of the patient is higher, indicating that the patient's lesions are more severe. The level of alpha-fetoprotein is similar to that of carcinoembryonic antigen, and an increase in the normalized value of the level of alpha-fetoprotein is usually also associated with the presence of tumors, so the normalized value of the level of alpha-fetoprotein is directly proportional to the biochemical discriminant impact index , which is represented in the denominator, and the exponential relationship indicates that when the level of alpha-fetoprotein and the level of carcinoembryonic antigen increase, the biochemical discriminant impact index increases significantly, indicating a significant impact on the judgment of the severity of the lesions.

[0115] Step 5: According to the lesion texture complexity and the lesion edge impact, combined with the biochemical impact, a comprehensive lesion recognition coefficient is generated, the comprehensive lesion recognition coefficient is compared with the set lesion judgment threshold value to generate different comparison results, and the corresponding lesion recognition judgment result is issued to complete the lesion image recognition.

[0116] According to the obtained lesion texture complexity index and lesion edge impact index, combined with the biochemical discriminant impact index, a comprehensive lesion recognition coefficient is generated, wherein the formula for calculating the comprehensive lesion recognition coefficient is:

[0117]

[0118] In the formula, is the comprehensive lesion recognition coefficient, , and are the weight coefficients of the biochemical discriminant impact index, the lesion edge impact index, and the lesion texture complexity index, respectively, wherein and , and are both greater than 0.

[0119] It should be noted that the comprehensive lesion recognition coefficient is used to represent the judgment of the lesion, and the greater the value, the more serious the existing lesion. Since the above-mentioned biochemical discrimination influence index, lesion edge influence index and lesion texture complexity index have been explained relative to the lesion recognition, they will not be repeated here.

[0120] where the exponential function represents the significant influence of the biochemical discrimination influence index on the lesion recognition, and the logarithmic function prevents excessive fluctuations from having too much impact on the results and ensures the stability of the model.

[0121] wherein in clinical practice, biochemical indicators (such as tumor markers) are important basis for judging the condition and risk. They can directly reflect the physiological state and pathological changes of patients, and the lesion edge influence index reflects the lesion structure and the degree of influence on the surrounding tissue, which is also very important in the evaluation of the lesion. Although the edge influence is a key factor, its importance is generally slightly lower than that of biochemical indicators, and the weight of the lesion texture complexity index, although the shape and texture features of the lesion help to evaluate the characteristics of the lesion, its clinical value is relatively low compared to biochemical indicators and edge influence, so and , and are both greater than 0.

[0122] The comprehensive lesion recognition coefficient is compared with the set lesion judgment threshold to generate different comparison results, and the corresponding lesion recognition judgment results are issued, wherein the specific judgment logic is:

[0123] When , it is judged that the lesion in the lesion medical image to be recognized is a tumor, and other means should be taken for further diagnosis and treatment;

[0124] When , it is judged that the lesion in the lesion medical image to be recognized is inflammation, and it should be treated for inflammation;

[0125] When , it is judged that the lesion in the lesion medical image to be recognized is bleeding;

[0126] wherein is the lesion judgment threshold.

[0127] Please refer to Figure 6The application further provides a lesion image recognition system based on a deep learning model, which is used to execute the above-mentioned lesion image recognition method based on a deep learning model and comprises the following modules.

[0128] A sample data acquisition module is configured to acquire medical images of different lesion types, perform image preprocessing to obtain sample images, mark the lesion regions in the sample images in an artificial marking manner, acquire image feature parameters of the lesion marking regions, and generate a sample training data set.

[0129] A prediction model training module is configured to establish a segmentation and recognition model based on the sample training data set, take the sample images as input, take the marking of the lesion regions in the images and the image feature parameters as labels, and train the segmentation and recognition model.

[0130] A biochemical index acquisition module is configured to acquire a lesion medical image to be recognized, input the lesion medical image to be recognized into the trained segmentation and recognition model, obtain image feature parameter prediction values of the lesion medical image to be recognized, and acquire biochemical index data of a patient corresponding to the lesion medical image to be recognized.

[0131] An influence parameter analysis module is configured to characterize lesion texture complexity and lesion edge influence based on the obtained image feature parameter prediction values, and evaluate biochemical influence based on the biochemical index data.

[0132] A lesion image recognition module is configured to generate a comprehensive lesion recognition coefficient according to the lesion texture complexity and the lesion edge influence and in combination with the biochemical influence, compare the comprehensive lesion recognition coefficient with a set lesion judgment threshold, generate different comparison results, issue a corresponding lesion recognition judgment result, and complete lesion image recognition.

[0133] The above formulas are all dimensionless values calculated, the formulas are obtained by software simulation of a large amount of data to obtain a formula of the nearest real situation, and preset parameters in the formulas are set by a person skilled in the art according to actual conditions.

[0134] The above embodiments can be realized by software, hardware, firmware or any combination thereof in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part. A person skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0135] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, and may be located in one place, or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment of the present application according to actual needs.

[0136] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for identifying a lesion image based on a deep learning model, the method comprising: The specific steps include: Obtain medical images of several different lesion types, and perform image preprocessing to obtain sample images, mark the lesion regions in the sample images by manual marking, obtain lesion marking region image feature parameters, and generate a sample training data set; Based on the sample training data set, a segmentation and recognition model is established, the sample images are taken as input, and the lesion region marking and image feature parameters in the images are taken as labels to train the segmentation and recognition model; Obtain the lesion medical image to be recognized, input it into the trained segmentation and recognition model, obtain the image feature parameter prediction value of the lesion medical image to be recognized, and obtain the biochemical index data of the patient corresponding to the lesion medical image to be recognized; Based on the obtained image feature parameter prediction value, the lesion texture complexity and lesion edge influence are characterized, and the biochemical influence is evaluated based on the biochemical index data; According to the lesion texture complexity and lesion edge influence, combined with the biochemical influence, a comprehensive lesion recognition coefficient is generated, the comprehensive lesion recognition coefficient is compared with the set lesion judgment threshold, different comparison results are generated, the corresponding lesion recognition judgment result is issued, and the lesion image recognition is completed. 2.The method of claim 1, wherein the method further comprises: The image feature parameters include color contrast, texture entropy, lesion edge strength, lesion contour perimeter, and lesion contour area; the lesion regions in the sample images are marked by manual marking, specifically by marking tools in the images, and the different lesion types specifically include bleeding, inflammation, and tumor; The image preprocessing specifically includes image enhancement and denoising preprocessing, wherein each lesion medical image is denoised by a wavelet transform denoising method, and each lesion medical image is image enhanced by a bilateral filter; The training sample data set is generated by one-to-one mapping of the marked sample images and the corresponding image feature parameters to form a corresponding grid, and the formed grid is recorded as the training sample data set. 3.The method of claim 2, wherein the method further comprises: Based on the sample training data set, a segmentation and recognition model is established, wherein the segmentation and recognition model is constructed by an LSTM model, an activation function and an optimization algorithm are selected, wherein a Tanh function is selected as the activation function, and Adam is selected as the optimization algorithm of the LSTM model; the formula of the Tanh function is: wherein denotes the Tanh function, the argument denotes the input weighted sum of a neuron, i.e. the result of the weighted sum of the inputs received by the neuron from the previous layer; Meanwhile, the hyperparameters of the LSTM model are set, including the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the batch processing quantity, and the number of hidden layer neurons; The number of network layers is set to 5-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the batch processing quantity is set to 128, and the number of hidden layer neurons is set to 32; The input of the trained segmentation and recognition model is the lesion medical image, the output is the image with completed lesion region marking, and the corresponding image feature parameter prediction value, including the prediction value of color contrast, texture entropy, lesion edge strength, lesion contour perimeter, and lesion contour area.

4. The method of claim 3, wherein the method is based on a deep learning model. Based on the obtained image feature parameter prediction value, a lesion texture complexity index and a lesion edge influence index are calculated, wherein the formula for calculating the lesion texture complexity index is specifically: wherein is a lesion texture complexity index, is a color contrast prediction value normalized value of the marked region image, is a texture entropy prediction value of the marked region image, is a maximum texture entropy prediction value of all marked region images; The formula for calculating the lesion edge influence index is: wherein is a lesion margin influence index, is a lesion contour perimeter prediction value for the marked region image, is a lesion contour area prediction value for the marked region image, is a lesion margin intensity for the marked region image, is a lesion contour circularity for the marked region image. 5.The method of claim 4, wherein the method further comprises: The circularity of the lesion outline of the marked region image The formula used for the calculation is: The biochemical index data includes erythrocyte sedimentation rate, white blood cell count, carcinoembryonic antigen level and alpha-fetal protein level, and a biochemical discrimination influence index is calculated based on the biochemical index data, wherein the formula for calculating the biochemical discrimination influence index is specifically: wherein is a biochemical discriminating influence index, is a normalized value of erythrocyte sedimentation rate corresponding to the patient, is a normalized value of white blood cell count corresponding to the patient, and are normalized values of carcinoembryonic antigen level and alpha-fetoprotein level, respectively, corresponding to the patient. 6.The method of claim 5, wherein the method further comprises: According to the obtained lesion texture complexity index and lesion edge influence index, combined with the biochemical discrimination influence index, a comprehensive lesion recognition coefficient is generated, wherein the formula for calculating the comprehensive lesion recognition coefficient is: In the formula, is a comprehensive lesion recognition coefficient, , and are weight coefficients of biochemical discrimination influence index, lesion edge influence index and lesion texture complexity index respectively, wherein and , and are all greater than 0. 7.The method of claim 1, wherein the method further comprises: determining a first feature of the first image based on a first deep learning model; determining a second feature of the second image based on a second deep learning model; and determining the first feature and the second feature based on a third deep learning model. The comprehensive lesion recognition coefficient is compared with the set lesion judgment threshold to generate different comparison results, and the corresponding lesion recognition judgment result is issued, wherein the specific judgment logic is: When the lesion in the medical image to be identified is determined as a tumor, other means should be taken for further diagnosis and treatment. When the lesion in the medical image to be identified is determined to be inflammation, the inflammation should be treated. When the lesion in the medical image to be identified is determined to be a hemorrhage; wherein is a lesion judgment threshold value. 8.A lesion image recognition system based on a deep learning model, characterized in that: The lesion image recognition system based on the deep learning model is used to execute the lesion image recognition method based on the deep learning model in any one of claims 1-7, comprising: A sample data acquisition module is configured to obtain medical images of different lesion types, perform image preprocessing to obtain sample images, mark the lesion regions in the sample images by using an artificial marking method, obtain image feature parameters of the lesion marked regions, and generate a sample training data set; A prediction model training module is configured to establish a segmentation and recognition model based on the sample training data set, take the sample images as input, take the labels of the lesion regions in the images and the image feature parameters as labels, and train the segmentation and recognition model; A biochemical index acquisition module is configured to obtain a medical image of a lesion to be recognized, input the medical image of the lesion to be recognized into the trained segmentation and recognition model to obtain image feature parameter prediction values of the medical image of the lesion to be recognized, and obtain biochemical index data of a patient corresponding to the medical image of the lesion to be recognized; An influence parameter analysis module is configured to characterize lesion texture complexity and lesion edge influence based on the obtained image feature parameter prediction values, and evaluate biochemical influence based on the biochemical index data; A lesion image recognition module is configured to generate a comprehensive lesion recognition coefficient according to the lesion texture complexity and the lesion edge influence, combined with the biochemical influence, compare the comprehensive lesion recognition coefficient with a set lesion judgment threshold to generate different comparison results, issue a corresponding lesion recognition judgment result, and complete lesion image recognition.

Images