Liver fibrosis detection grading model construction method applied to schistosomiasis

Abstract

The invention discloses a liver fibrosis detection grading model construction method applied to schistosomiasis, and relates to the technical field of medical detection, multi-modal liver sample data is collected and subjected to standardization processing to generate standard multi-modal data, different standard multi-modal data is subjected to feature processing by adopting corresponding processing methods, and the classification model construction method is applied to the schistosomiasis. Liver image features, serum biochemical marker features and pathological section features representing the influence of the schistosomiasis on liver fibrosis are obtained, a multi-mode combined grading diagnosis model is constructed based on the features to conduct liver fibrosis detection on patients related to the schistosomiasis, and the detection result of liver fibrosis detection is verified. And performing dynamic optimization on the multi-modal joint grading diagnosis model based on a detection result, evaluating a model indication of the multi-modal joint grading diagnosis model after dynamic optimization is completed, and determining whether to adjust optimization parameters for performing dynamic optimization on the multi-modal joint grading diagnosis model or not based on the model indication.

Description

Technical Field

[0001] The present invention relates to the technical field of medical detection, and specifically to a method for constructing a liver fibrosis detection and grading model applied to schistosomiasis. Background Art

[0002] Schistosomiasis is a parasitic disease that seriously endangers human health and is widely prevalent in many regions of the world, especially in some areas with relatively poor sanitary conditions and easily polluted water sources. After long-term infection with schistosomes, the liver is extremely vulnerable to damage, leading to liver fibrosis. Different degrees of liver fibrosis correspond to completely different treatment strategies and prognosis.

[0003] In the past, the detection methods for liver fibrosis caused by schistosomiasis were relatively single and scattered, mainly including traditional liver imaging examinations such as ultrasound and CT, serum biochemical marker detections. For example, specific protein and enzyme indicators can be detected to indirectly reflect liver damage, and pathological section examinations are also included. Different detection methods are isolated from each other, lacking effective integration. The data of each modality have not been fully mined and synergistically utilized, and cannot provide comprehensive, accurate, and dynamic liver fibrosis grading information for clinical practice, resulting in low diagnostic efficiency and poor accuracy of schistosomiasis liver fibrosis, delaying the best treatment time for patients, and increasing medical costs and patient suffering. Summary of the Invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a method for constructing a liver fibrosis detection and grading model applied to schistosomiasis.

[0005] The purpose of the present invention can be achieved by the following technical solutions: A method for constructing a liver fibrosis detection and grading model applied to schistosomiasis, including the following steps: Step S1: Collect multi-modal liver sample data related to schistosomiasis, and perform standardization processing on the multi-modal liver sample data to generate corresponding standard multi-modal data respectively; Step S2: Adopt corresponding processing methods for different standard multi-modal data for feature processing to obtain liver imaging features, serum biochemical marker features, and pathological section features that characterize the impact of schistosomiasis on liver fibrosis; Step S3: Construct a multi-modal joint grading diagnosis model based on the liver imaging features, serum biochemical marker features, and pathological section features, and use the multi-modal joint grading diagnosis model to detect liver fibrosis in related patients with schistosomiasis; Step S4: Verify the detection results of liver fibrosis detection, dynamically optimize the multi-modal joint grading diagnosis model based on the detection results, and evaluate the model indicators of the multi-modal joint grading diagnosis model after dynamic optimization; Step S5: Based on the model indicators, decide whether to adjust the optimization parameters for dynamically optimizing the multi-modal joint grading diagnosis model; The process of collecting multi-modal liver sample data related to schistosomiasis and performing standardization processing on the multi-modal liver sample data, and then generating their respective corresponding standard multi-modal data includes: Select a number of confirmed patients and a number of healthy control populations in the schistosomiasis epidemic area, and set the selection ratio of confirmed patients to healthy control populations; Configure an ultrasonic diagnostic instrument, set the gain values corresponding to the near field, middle field, and far field of the TGC curve respectively, collect multi-sectional dynamic images of the liver, and perform two-dimensional wavelet transform filtering, adaptive histogram equalization, and cubic spline interpolation processing on the multi-sectional dynamic images of the liver; Scan the multi-sectional dynamic images of the liver of schistosomiasis confirmed patients through a spiral CT device, and perform image gray normalization and rigid registration processing after scanning. Use a superconducting magnetic resonance imaging system to perform MRI scanning according to the set parameters, and perform gray normalization and rigid registration after scanning; Furthermore, obtain the standardized ultrasonic images corresponding to the confirmed patients. The standardized ultrasonic images include MRI images and CT images; Collect several milliliters of venous blood from the confirmed patients, centrifuge and separate the serum, detect specific biochemical markers and perform standardization conversion to obtain the corresponding serum biochemical marker data. Puncture and take liver pathological section samples under ultrasonic guidance, and perform non-local means filtering, color calibration, and brightness contrast adjustment after staining and digitization.

[0006] Furthermore, the process of performing corresponding processing methods on different standard multi-modal data for feature processing, and then obtaining liver image features, serum biochemical marker features, and pathological section features that characterize the impact of schistosomiasis on liver fibrosis includes: Construct an improved convolutional neural network, divide the standardized ultrasonic images into a training set and a test set according to a ratio of 7:3, and use the Adam optimizer and data augmentation technology to extract ultrasonic image features in the standardized ultrasonic images; Adopt the Otsu threshold segmentation and region growing algorithm to extract the liver region of the CT image in the standardized ultrasonic image, calculate CT values, morphological, and vascular-related features. Use the 3D-U-Net model to divide the MRI images into a training set and a test set according to a ratio of 6:4, and use the Adagrad optimizer to train and normalize the images to extract features; Construct a multi-modal feature vector corresponding to the serum biochemical marker data, and with the help of pathological image analysis software, identify fibrous tissue, hepatocytes, and inflammatory cells in Masson and HE stained sections respectively, and calculate the fiber ratio, cell morphological parameters, and the number of inflammatory cells; Summarize the liver imaging features, serum biochemical marker features, and pathological section features that characterize the impact of schistosomiasis on liver fibrosis.

[0007] Furthermore, the process of constructing a multimodal joint grading diagnosis model based on liver imaging features, serum biochemical marker features, and pathological section features includes: Concatenate the liver imaging features, serum biochemical marker features, and pathological section feature vectors in a ratio of 1:1:1 to form a comprehensive feature vector, and select the support vector machine algorithm and radial basis kernel function to construct an initial multimodal joint grading diagnosis model; Obtain sample data and divide the sample data into a training set and a test set in a ratio of 8:2; Set the penalty parameter and kernel function parameter of the initial multimodal joint grading diagnosis model; Use the grid search algorithm combined with 5-fold cross-validation to assign values to the penalty parameter and kernel function parameter respectively, optimize the parameters within the value range until the training set accuracy reaches over 95% and the test set accuracy stabilizes above 90%, and apply the initial multimodal joint grading diagnosis model to detect liver fibrosis in relevant patients with schistosomiasis and output the grading results.

[0008] Furthermore, the process of detecting liver fibrosis in relevant patients with schistosomiasis through the multimodal joint grading diagnosis model includes: Collect relevant data of different modalities corresponding to schistosomiasis patients who need liver fibrosis detection, and perform standardization processing on all relevant data to obtain different multimodal standard data of the patients, as well as the liver imaging features, serum biochemical marker features, and pathological section features obtained after feature processing of different multimodal standard data; Extract the multimodal feature vectors corresponding to the liver imaging features, serum biochemical marker features, and pathological section features respectively, and input all the multimodal feature vectors into the trained multimodal joint grading diagnosis model to output the grading results of liver fibrosis detection for relevant patients, and the grading results include grades F0 - F4.

[0009] Furthermore, the process of verifying the detection results of liver fibrosis detection, dynamically optimizing the multimodal joint grading diagnosis model based on the detection results, and evaluating the model indicators of the multimodal joint grading diagnosis model after dynamic optimization includes: Summarize the liver fibrosis grading results output by the multi-modal combined grading diagnosis model, compare and analyze the liver fibrosis grading results with the patient's clinical symptoms, signs and other examination results, and then verify the test results. Use the sliding window method to dynamically optimize the multi-modal combined grading diagnosis model, and then monitor the model accuracy and sensitivity of the multi-modal combined grading diagnosis model, calculate the mean and standard deviation of the last 5 results, and evaluate the model indicators of the multi-modal combined grading diagnosis model.

[0010] Further, the process of determining whether to adjust the optimization parameters for dynamically optimizing the multi-modal combined grading diagnosis model includes: Monitor and record the model accuracy and sensitivity of the multi-modal combined grading diagnosis model, set the indicator threshold, compare the current model indicator with the indicator threshold. If the accuracy improves by at least 2% for 10 consecutive times, the difference between the training set and test set accuracy is within 10%, the sensitivity exceeds 85% and the specificity exceeds 90%, decide to stop adjusting the optimization parameters; if the threshold is not met, such as slow improvement of indicators, overfitting or underfitting, continue to adjust the optimization parameters.

[0011] Further, for the multi-modal combined grading diagnosis model trained by the gradient descent algorithm, increase the learning rate when the training is slow, and adjust the number of cross-validation folds when the oscillation decreases and the cross-validation generalization ability decreases, and retrain and verify the multi-modal combined grading diagnosis model.

[0012] Compared with the prior art, the beneficial effects of the present invention are: 1. The present invention comprehensively collects multi-modal liver sample data of liver images, serum biochemical markers and pathological sections, and organically integrates information in different dimensions. Among them, liver images visually show the macroscopic morphological changes of the liver, serum biochemical markers reflect the microscopic changes at the liver cell metabolism level, and pathological sections provide accurate histological basis. The three complement each other, can comprehensively capture the details of the impact of schistosomiasis on liver fibrosis, overcome the limitations of single-modal detection, and significantly improve the diagnostic accuracy.

[0013] 2. After constructing the multi-modal combined grading diagnosis model, by verifying the test results, continuously dynamically optimize the model, continuously adapt to differences in different patient groups, schistosome strains in different regions and disease conditions, and flexibly adjust the optimization parameters according to the model indicators to ensure that the model is always in the best diagnostic efficiency state, and provide reliable liver fibrosis grading for clinical practice in the long term and stably, and help precision medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] Such as Figure 1As shown in the figure, a method for constructing a liver fibrosis detection and grading model for schistosomiasis includes the following steps: Step S1: Collect multi-modal liver sample data related to schistosomiasis, and perform standardization processing on the multi-modal liver sample data to generate respective corresponding standard multi-modal data; Step S2: Adopt corresponding processing methods for different standard multi-modal data for feature processing to obtain liver image features, serum biochemical marker features, and pathological section features that characterize the impact of schistosomiasis on liver fibrosis; Step S3: Construct a multi-modal joint grading diagnosis model based on the liver image features, serum biochemical marker features, and pathological section features, and use the multi-modal joint grading diagnosis model to detect liver fibrosis in patients related to schistosomiasis; Step S4: Verify the detection results of liver fibrosis detection, dynamically optimize the multi-modal joint grading diagnosis model based on the detection results, and evaluate the model indicators of the multi-modal joint grading diagnosis model after dynamic optimization; Step S5: Based on the model indicators, decide whether to adjust the optimization parameters for dynamically optimizing the multi-modal joint grading diagnosis model.

[0016] It should be further noted that in the specific implementation process, the process of collecting multi-modal liver sample data related to schistosomiasis and performing standardization processing on the multi-modal liver sample data to generate respective corresponding standard multi-modal data includes: First, perform sample selection. In the schistosomiasis epidemic area, 500 patients diagnosed with schistosomiasis are selected to form a patient group. The selected patients need to cover different infection stages, different age groups, and different genders, aiming to ensure the diversity of the samples; At the same time, 100 healthy control subjects confirmed by examination are selected. The selected healthy control subjects are matched with the patient group in terms of age group and gender distribution to reduce the interference of individual basic differences on subsequent detection and model construction; Select an ultrasonic diagnostic instrument equipped with a 5MHz probe to obtain the internal liver structure image of the patient. Set the gain to 50dB, and set the time gain compensation curve to 20dB, 30dB, and 40dB in the near field, middle field, and far field respectively. Through this setting, it is possible to effectively compensate for the signal attenuation caused by the increase in the ultrasonic propagation distance and ensure the consistency of the image quality in different depth regions of the liver; When performing liver ultrasound examinations on each patient, collect dynamic images of the sagittal section of the left lobe of the liver, the oblique section of the right lobe, and the transverse section under the rib. The collection duration for each section is set to 10 seconds to ensure sufficient liver structure information is obtained, covering the morphological changes of the liver at different angles and positions; After the acquisition, the two-dimensional wavelet transform filtering algorithm is used to remove noise from the ultrasonic images. The two-dimensional wavelet transform can decompose the images at different scales and directions, effectively suppressing noise while retaining the edge and texture information of the images, avoiding the influence of noise interference on subsequent feature extraction; The contrast is enhanced by the adaptive histogram equalization method. This method adaptively adjusts the histogram according to the gray distribution of the local area of the image, making the details of the liver tissue clearer, such as the course of the blood vessels in the liver and the texture of the liver parenchyma, etc., facilitating subsequent analysis; The cubic spline interpolation method is used to uniformly resample the image resolution to 512×512 pixels, ensuring the consistency of the images in subsequent processing and providing a standardized image data basis for feature extraction; A 128-slice spiral CT device is used for scanning. Before scanning, the patient needs to fast for more than 6 hours to reduce the interference of gastrointestinal contents on liver imaging, ensure the clarity of liver images, and take an appropriate amount of contrast agent orally. The contrast agent can enhance the contrast between the liver tissue and the surrounding tissues, making the boundary and internal structure of the liver clearer and distinguishable; After the scanning is completed, the gray normalization process is performed on the CT images, mapping the CT values to the interval [0,1], eliminating the differences brought by different devices and scanning parameters, making the CT images of different patients comparable. The rigid registration algorithm based on mutual information is used for spatial standardization registration to find the optimal spatial transformation parameters, so that all images have the same position and direction in the spatial coordinate system.

[0017] It should be further noted that in the specific implementation process, the process of taking corresponding processing methods for different standard multi-modal data for feature processing, and then obtaining the liver image features, serum biochemical marker features, and pathological section features that characterize the impact of schistosomiasis on liver fibrosis includes: Construct a convolutional neural network model improved based on ResNet-50, retain the first four convolutional modules of the original ResNet-50, remove the global average pooling layer and the fully connected layer, add two convolutional layers after the fourth convolutional module, where the convolutional kernel size of one convolutional layer is set to 3×3, and the other is set to 1×1, the stride of both is 1, select ReLU as the activation function, set the pooling layer as average pooling, the pooling kernel size is 2×2, and the stride is 2; The standardized ultrasonic image data is divided into a training set and a test set according to the ratio of 7:3. The Adam optimizer is used, and the initial learning rate of the Adam optimizer is set to 0.0001. The data augmentation technology is used to perform random rotation, horizontal flipping, and vertical flipping operations on the ultrasonic images within the range of -15° to 15°; Through the above operations, the convolutional neural network model learns the texture features of the patient's liver, and the texture features include morphological features such as the uniformity of echo, thickness, and the running and branching of blood vessels; Feature mining is carried out through a 3D-U-Net deep learning model. The 3D-U-Net model includes an encoder, a decoder, and a skip connection structure. The encoder part consists of multiple 3D convolutional layers and 3D pooling layers, which are used to extract deep features of the image; The decoder part consists of multiple 3D transposed convolutional layers and 3D convolutional layers, which are used to restore the spatial resolution of the image. The skip connection structure fuses the feature maps of different levels in the encoder with the corresponding level feature maps in the decoder to improve the utilization rate of features; The MRI image data is divided into a training set and a test set according to a ratio of 6:4. The Adagrad optimizer is used, and the learning rate of the Adagrad optimizer is set to 0.001. The MRI images are normalized, and the image intensity values are normalized to the interval [0,1] to learn the signal intensity differences and tissue contrasts of the liver tissue under different sequences.

[0018] A dimensionality reduction algorithm combining principal component analysis and factor analysis is adopted to standardize the data of biochemical markers such as hyaluronic acid, laminin, type III procollagen, and type IV collagen in serum, so that the data mean is 0 and the standard deviation is 1; Calculate the covariance matrix of the standardized data. Through eigenvalue decomposition, eigenvalues and eigenvectors are obtained. Select the principal components with a cumulative contribution rate of more than 90% to construct a preliminary eigenvector. Factor analysis is used to further interpret and reconstruct the principal components, and factors with practical biological significance are extracted. Finally, a comprehensive serum biochemical marker eigenvector is constructed, and the serum biochemical marker eigenvector is used to characterize the serum biochemical marker features; With the help of pathological image analysis software, feature analysis is carried out on the pathological sections of liver tissue. In the stained sections, the blue fibrous tissue is identified using a color threshold segmentation algorithm, and noise and small fragments are removed through morphological operations. Calculate the percentage of the fibrous tissue area in the entire liver tissue area; By setting the color features and morphological parameters of the cell nucleus, and the morphological parameters include roundness, area, and perimeter, a cell detection algorithm is used to identify hepatocytes. Randomly select 10 high-power fields of view, and measure the morphological parameters such as the diameter, nucleus area, and nuclear-cytoplasmic ratio of 50 hepatocytes in each field of view, and calculate the average value. A cell counting algorithm is used to count the inflammatory cells, and the number of inflammatory cells per unit area is calculated, and then the pathological section features corresponding to the pathological section are obtained.

[0019] It should be further noted that in the specific implementation process, the process of constructing a multi-modal combined grading diagnosis model based on liver imaging features, serum biochemical marker features, and pathological section features includes: Merge the liver imaging features, serum biochemical marker features, and pathological section features obtained through the previous steps. Connect the feature vectors corresponding to the liver imaging features, serum biochemical marker features, and pathological section features end to end in a ratio of 1:1:1 to construct a comprehensive feature vector, and use the comprehensive feature vector as the input data for subsequent model construction. This comprehensive feature vector can comprehensively reflect various aspects of information on the impact of schistosomiasis on liver fibrosis; Select the support vector machine algorithm to construct a multi-modal combined grading diagnosis model. Divide all sample data into a training set and a test set in a ratio of 8:2. The training set is used for parameter learning and optimization of the model, and the test set is used to evaluate the generalization ability and prediction performance of the model; It should be noted that ensure that the proportion of various samples in the training set and the test set is the same as that in the original dataset to ensure the balance and representativeness of data division; Set the penalty parameter and kernel function parameter of the multi-modal combined grading diagnosis model; Among them, denote the penalty parameter and the kernel function parameter as C and D respectively, C = 2, D = 0.1; The penalty parameter C controls the tolerance of the multi-modal combined grading diagnosis model to classification errors during the training process. The larger the value of C, the lower the penalty of the multi-modal combined grading diagnosis model for classification errors, which may lead to overfitting; the smaller the value of C, the higher the tolerance of the multi-modal combined grading diagnosis model to classification errors, which may lead to underfitting. The kernel function parameter D determines the width of the radial basis kernel function and affects the complexity and generalization ability of the model; Use the grid search algorithm combined with 5-fold cross-validation to optimize the penalty parameter C. The grid search algorithm performs an exhaustive search within the pre-set parameter range. The value range of C is set as follows: C = [0.1, 1, 2, 5, 10]; The value range of the kernel function parameter D is set as [0.01, 0.05, 0.1, 0.5, 1]; Each time, select 4 pieces of data from the training set as the training subset, and select 1 piece of data as the validation subset, and calculate the accuracy of the multi-modal combined grading diagnosis model on the validation subset under different parameter combinations; Select the parameter combination with the highest accuracy as the optimal parameter, and continuously iterate and optimize until the accuracy of the multi-modal combined grading diagnosis model on the training set reaches more than 95%, and the accuracy on the test set is stable at more than 90%, and construct the final multi-modal combined grading diagnosis model.

[0020] It should be further noted that in the specific implementation process, the process of detecting liver fibrosis in schistosomiasis-related patients through the multi-modal joint grading diagnosis model includes: For schistosomiasis-related patients who need to undergo liver fibrosis detection, multi-modal data collection is performed using the same equipment and parameters as those used when constructing the multi-modal joint grading diagnosis model; That is, an ultrasonic diagnostic instrument equipped with a 5MHz probe, a gain of 50dB, and a TGC curve of 20dB, 30dB, and 40dB in the near field, middle field, and far field respectively is used to collect multi-sectional dynamic images of the liver; That is, a 128-slice spiral CT device is used to perform scans with parameters of tube voltage 120kV, tube current 200mA, slice thickness 1mm, and pitch 1.25. The patient fasts for more than 6 hours and takes an oral contrast agent before the scan; a 3.0T superconducting magnetic resonance imaging system is used to perform MRI scans according to the parameters of the T1WI sequence: TR = 500ms, TE15ms, and the T2WI sequence: TR = 4000ms, TE = 100ms; 8ml of venous blood is collected from the patient on an empty stomach, centrifuged at 3000r / min for 15 minutes to separate the serum, and enzyme-linked immunosorbent assay is used to detect serum biochemical markers such as hyaluronic acid, laminin, type III procollagen, and type IV collagen; Under ultrasonic guidance, an 18G puncture needle is used to obtain a liver tissue sample of about 1.5 - 2 cm from the edge of the right lobe of the liver. After being fixed, dehydrated, cleared, infiltrated with wax, and embedded, etc., a pathological section with a thickness of 4μm is made, and hematoxylin-eosin (HE) staining and Masson staining are performed; The ultrasonic image is processed by two-dimensional wavelet transform filtering, adaptive histogram equalization, and cubic spline interpolation, the image resolution is resampled to 512×512 pixels, the CT and MRI images are subjected to gray normalization processing, the CT value or MRI signal intensity is mapped to the [0,1] interval, and a rigid registration algorithm based on mutual information is used for spatial standardization registration; The serum biochemical marker data is standardized and converted based on the mean and standard deviation of this marker in the healthy population according to the characteristics of the detection method. After the pathological section is digitized, non-local mean filtering algorithm is used for noise reduction, color calibration is performed using a color calibration card, and the brightness is adjusted to 80 and the contrast is adjusted to 1.2; According to the feature extraction method used when constructing a multimodal joint hierarchical diagnosis model, feature extraction was performed on the standardized new data, i.e., ultrasound image features were extracted using an improved ResNet-50 convolutional neural network, CT image features were extracted based on the region growing and morphological analysis algorithm, MRI image features were extracted using the 3D-U-Net deep learning model, serum biochemical marker features were extracted using a dimensionality reduction algorithm combining principal component analysis with factor analysis, and pathological section features were extracted using professional pathological image analysis software; The multimodal feature vectors corresponding to the extracted liver image features, serum biochemical marker features and pathological section features of the new patient are input into the trained multimodal joint grading diagnosis model, and then the grading results of liver fibrosis detection are output, including grades F0-F4.

[0021] It should be further explained that, in the specific implementation process, the process of verifying the test results of liver fibrosis detection, dynamically optimizing the multimodal joint hierarchical diagnosis model based on the test results, and evaluating the model indication of the multimodal joint hierarchical diagnosis model after the dynamic optimization is completed includes: Summarize the liver fibrosis grading results output by the multimodal joint grading diagnosis model, covering the F0-F4 grading results corresponding to each patient, and simultaneously collect other relevant clinical data of the patient, including symptoms, such as fatigue, abdominal distension, etc., physical signs, such as liver and spleen size, and other medical examination results, such as liver function indicators and pathogenic examination indicators; The grading results of the multimodal joint grading diagnosis model are compared in detail with the patient's clinical symptoms, signs, and other medical examination results. If the patient diagnosed with F3 grade fibrosis by the multimodal joint grading diagnosis model has obvious liver function abnormalities and hepatosplenomegaly, which is consistent with the model grading results, it is determined to have passed the verification; if there is a large deviation, such as the multimodal joint grading diagnosis model diagnoses early fibrosis F1 grade, but the patient has severe liver function damage, it is marked as a result to be verified; The sliding window method is used to continuously monitor the key performance indicators of the model, including model accuracy, sensitivity, and specificity. A sliding window containing the results of the last five optimizations is set, and the average and standard deviation of each key performance indicator in the sliding window are calculated; If the average increase in the model accuracy, sensitivity, and specificity within the sliding window is less than 1%, and the standard deviation is greater than 5%, it is determined that the model optimization effect is poor and the multimodal joint hierarchical diagnosis model falls into a local optimum; When the accuracy of the corresponding model in the training set is higher than 95%, but the accuracy of the corresponding model in the test set is lower than 80%, and the difference in the loss function value between the training set and the test set is greater than 0.5, it is determined that the multimodal joint hierarchical diagnosis model is overfitting; If the accuracy rates of both the training set and the test set are lower than 70% and the loss function value is greater than 0.75, it indicates that there is an underfitting problem with the model; If the model training speed is too slow, increase the learning rate; If the model shows oscillations, decrease the learning rate; If the generalization ability of the model does not meet the standard when using 5-fold cross-validation currently, adjust the number of cross-validation folds to 7-fold or 3-fold, and re-train and validate the model; After each dynamic optimization is completed, evaluate the key performance indicators of the model; If the model accuracy improves by at least 2% in each of the consecutive 10 optimizations, and the difference in accuracy rates between the training set and the test set always remains within 10%, while the sensitivity is greater than 85% and the specificity is greater than 90%, it is determined that the model performance is in a stable improvement state; otherwise, no determination is made. When the model performance is in a stable improvement state, stop the dynamic optimization and determine the final model parameters.

[0022] It should be further noted that in the specific implementation process, the process of determining whether to adjust the optimization parameters for dynamically optimizing the multi-modal joint grading diagnosis model based on the model indicators includes: According to the historical performance of the multi-modal joint grading diagnosis model in the training and validation processes and the actual requirements of clinical applications, set corresponding thresholds for each model indicator of the multi-modal joint grading diagnosis model; For example, set the threshold for the improvement of the model accuracy to at least 2% in each of the consecutive multiple optimizations, the threshold for the difference in accuracy rates between the training set and the test set to not exceed 10%, the sensitivity threshold to be greater than 85%, the specificity threshold to be greater than 90%, and for the difference in loss function values, set an acceptable range, such as not exceeding 0.5; Compare and analyze the monitored model indicators with the corresponding set thresholds, as follows: When the model accuracy of the multi-modal joint grading diagnosis model improves by at least 2% in each of the consecutive 10 optimizations, and the difference in accuracy rates between the training set and the test set always remains within 10%, while the sensitivity is greater than 85% and the specificity is greater than 90%, it is decided not to adjust the optimization parameters when dynamically optimizing the current multi-modal joint grading diagnosis model; When the multi-modal joint grading diagnosis model falls into a local optimum, shows overfitting or has an underfitting problem, adjust the optimization parameters when dynamically optimizing the current multi-modal joint grading diagnosis model.

[0023] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a liver fibrosis detection and grading model applied to schistosomiasis, characterized in that, It includes the following steps: Step S1: Collect multi-modal liver sample data related to schistosomiasis, perform standardization processing on the multi-modal liver sample data, and then generate corresponding standard multi-modal data respectively; Step S2: Adopt corresponding processing methods for different standard multi-modal data for feature processing, and then obtain liver image features, serum biochemical marker features, and pathological section features that characterize the impact of schistosomiasis on liver fibrosis; Step S3: Construct a multi-modal joint grading diagnosis model based on liver image features, serum biochemical marker features, and pathological section features, and detect liver fibrosis in related patients with schistosomiasis through the multi-modal joint grading diagnosis model; Step S4: Verify the detection results of liver fibrosis detection, dynamically optimize the multi-modal joint grading diagnosis model based on the detection results, and evaluate the model indicators of the multi-modal joint grading diagnosis model after dynamic optimization; Step S5: Based on the model indicators, decide whether to adjust the optimization parameters for dynamically optimizing the multi-modal joint grading diagnosis model; The process of collecting multi-modal liver sample data related to schistosomiasis and performing standardization processing on the multi-modal liver sample data to generate corresponding standard multi-modal data respectively includes: Select a number of confirmed patients and a number of healthy control populations in schistosomiasis epidemic areas, and set the selection ratio of confirmed patients to healthy control populations; Configure an ultrasonic diagnostic instrument, set the gain values corresponding to the near field, middle field, and far field of the TGC curve respectively, collect dynamic images of multiple liver sections, and perform two-dimensional wavelet transform filtering, adaptive histogram equalization, and cubic spline interpolation processing on the dynamic images of multiple liver sections; Scan the dynamic images of multiple liver sections of schistosomiasis confirmed patients through a spiral CT device, perform image gray normalization and rigid registration processing after scanning, and perform MRI scanning according to the set parameters using a superconducting magnetic resonance imaging system, and perform gray normalization and rigid registration after scanning; Furthermore, obtain the standardized ultrasonic images corresponding to the confirmed patients, and the standardized ultrasonic images include MRI images and CT images; Collect several milliliters of venous blood from the confirmed patients, centrifuge to separate the serum, detect specific biochemical markers and perform standardization conversion to obtain corresponding serum biochemical marker data, puncture to obtain liver pathological section samples under ultrasonic guidance, and perform non-local means filtering, color calibration, and brightness contrast adjustment after staining and digitization.

2. The method for constructing a liver fibrosis detection and grading model applied to schistosomiasis according to claim 1, wherein, The process of adopting corresponding processing methods for different standard multi-modal data for feature processing, and then obtaining liver image features, serum biochemical marker features, and pathological section features that characterize the impact of schistosomiasis on liver fibrosis includes: Construct an improved convolutional neural network, divide the standardized ultrasonic images into a training set and a test set according to a ratio of 7:3, and use the Adam optimizer and data augmentation technology to extract ultrasonic image features from the standardized ultrasonic images; The Otsu threshold segmentation and region growing algorithm are used to extract the liver region of CT images in standardized ultrasound images, calculate CT values, morphological and vascular-related features. The 3D-U-Net model is adopted to divide the MRI images into a training set and a test set in a ratio of 6:4, and the Adagrad optimizer is used to train and normalize the images to extract features; Construct a multi-modal feature vector corresponding to the serum biochemical marker data. With the help of pathological image analysis software, fibrous tissue, hepatocytes and inflammatory cells are respectively identified in Masson and HE staining sections, and the fiber proportion, cell morphological parameters and the number of inflammatory cells are calculated; Summarize to obtain the liver image features, serum biochemical marker features and pathological section features that characterize the impact of schistosomiasis on liver fibrosis.

3. The method for constructing a liver fibrosis detection and grading model applied to schistosomiasis according to claim 2, wherein The process of constructing a multi-modal joint grading diagnosis model based on liver image features, serum biochemical marker features and pathological section features includes: The liver image features, serum biochemical marker features and pathological section feature vectors are spliced according to a ratio of 1:1:1 to form a comprehensive feature vector, and the support vector machine algorithm and the radial basis kernel function are selected to construct an initial multi-modal joint grading diagnosis model; Obtain sample data and divide the sample data into a training set and a test set in a ratio of 8:2; Set the penalty parameter and kernel function parameter of the initial multi-modal joint grading diagnosis model; Use the grid search algorithm combined with 5-fold cross-validation to respectively obtain values for the penalty parameter and the kernel function parameter, optimize the parameters within the value range until the accuracy of the training set reaches more than 95% and the accuracy of the test set stabilizes above 90%. Apply the initial multi-modal joint grading diagnosis model to detect liver fibrosis in patients with schistosomiasis and output the grading results.

4. The method for constructing a liver fibrosis detection and grading model applied to schistosomiasis according to claim 3, wherein The process of detecting liver fibrosis in patients with schistosomiasis through the multi-modal joint grading diagnosis model includes: Collect relevant data of different modalities corresponding to schistosomiasis patients who need liver fibrosis detection, and perform standardized processing on all relevant data to obtain different multi-modal standard data of the patients, as well as the liver image features, serum biochemical marker features and pathological section features obtained after feature processing of different multi-modal standard data; Extract the multi-modal feature vectors corresponding to the liver image features, serum biochemical marker features and pathological section features respectively, and input all the multi-modal feature vectors into the trained multi-modal joint grading diagnosis model to output the grading results of liver fibrosis detection for relevant patients. The grading results include grades F0 - F4.

5. The method for constructing a liver fibrosis detection and grading model applied to schistosomiasis according to claim 4, wherein Verify the detection results of liver fibrosis detection, dynamically optimize the multi-modal joint grading diagnosis model based on the detection results, and evaluate the model indicators of the multi-modal joint grading diagnosis model after dynamic optimization, the process includes: Summarize the liver fibrosis grading results output by the multi-modal joint grading diagnosis model, compare and analyze the liver fibrosis grading results with the patient's clinical symptoms, signs and other examination results, and then verify the test results. Use the sliding window method to dynamically optimize the multi-modal joint grading diagnosis model, and then monitor the model accuracy and sensitivity of the multi-modal joint grading diagnosis model. Calculate the mean and standard deviation of the last 5 results, and evaluate the model indicators of the multi-modal joint grading diagnosis model.

6. The method for constructing a liver fibrosis detection and grading model applied to schistosomiasis according to claim 5, wherein, The process of determining whether to adjust the optimization parameters for the dynamic optimization of the multi-modal joint grading diagnosis model includes: Monitor and record the model accuracy and sensitivity of the multi-modal joint grading diagnosis model, set the indicator threshold, and compare the current model indicator with the indicator threshold. If the accuracy improves by at least 2% for 10 consecutive times, the difference between the training set and test set accuracy is within 10%, the sensitivity exceeds 85%, and the specificity exceeds 90%, decide to stop adjusting the optimization parameters; if the threshold is not met, such as slow indicator improvement, overfitting or underfitting, continue to adjust the optimization parameters.

7. The method for constructing a liver fibrosis detection and grading model applied to schistosomiasis according to claim 6, wherein For the multi-modal joint grading diagnosis model trained by the gradient descent algorithm, increase the learning rate when the training is slow, and adjust the number of cross-validation folds when the oscillation decreases and the cross-validation generalization ability decreases, and retrain and validate the multi-modal joint grading diagnosis model.

Images