Bone tumor treatment scheme acquisition method and system, terminal and storage medium
By preprocessing and feature analysis of image images of knee bone tumor patients, combined with cyclic convolutional neural network and multi-model fusion method, adaptively adjusting the treatment plan, the problem of insufficient accuracy of treatment plan in the existing technology is solved, and the problem of More efficient and personalized therapeutic effects.
Patent Information
- Application Number
- CN202410988091.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-15
AI Technical Summary
The prior art relies on doctors' experience in the treatment of knee joint bone tumors, which makes it difficult to guarantee the accuracy and effect of the treatment plan, and there is great subjectivity.
By preprocessing the patient's image images, preliminary features are extracted and importance judged, tumor image feature analysis is performed by combining cyclic convolutional neural network and multi-model fusion method, tumor image classification results are obtained, and adaptive adjustments are made based on historical patient cases. , determine the final treatment plan.
It improves the accuracy and personalization of knee bone tumor treatment plans, reduces the subjectivity of the treatment plans, and provides a more objective and scientific treatment plan.
Smart Images

Figure CN118969172A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of medical data processing technology, and in particular to a method, system, terminal and storage medium for obtaining a bone tumor treatment plan. Background Art
[0002] Knee bone tumor is a disease that seriously affects the quality of life of patients. With the change of lifestyle and the aging of the population, the incidence of bone tumors has increased year by year. According to statistics, bone tumors are more common in adolescents and the elderly, especially tumors in the knee joint, including osteosarcoma and osteochondroma, which have a high degree of malignancy and a poor prognosis. At present, the treatment of knee bone tumors mainly relies on doctors to judge through imaging images combined with experience and professional knowledge. This method is very subjective, and the accuracy of the treatment plan cannot be guaranteed, which in turn affects the treatment of patients. Summary of the invention
[0003] In view of this, the embodiments of the present application provide a method, system, terminal and storage medium for obtaining a bone tumor treatment plan to improve the accuracy and treatment effect of clinical treatment.
[0004] In a first aspect, an embodiment of the present application provides a method for obtaining a treatment plan for a knee bone tumor, comprising:
[0005] Performing preliminary feature extraction on the preprocessed image of the patient to obtain preliminary features of the preprocessed image, performing importance judgment on the preliminary features to obtain an important feature set, and analyzing the image features in the important feature set to obtain tumor image features; wherein the preliminary features include edge features, texture features, shape features, contrast features, and density features;
[0006] Performing feature analysis on the tumor image features by using a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result;
[0007] According to the tumor image classification result, a treatment plan matching the current patient is obtained from a medical database in combination with a preset algorithm, and the matching treatment plan is adaptively adjusted in combination with the information of the current patient to obtain a preliminary treatment plan, and a simulation evaluation is performed on the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response;
[0008] According to the information of the current patient, the predicted treatment effect and the predicted treatment response, a decision tree algorithm is determined through a pre-trained scheme to perform analysis to obtain a final treatment plan.
[0009] In some embodiments, the method further includes: acquiring the patient's rehabilitation data in real time, and generating a follow-up adjustment plan based on the rehabilitation data; wherein the follow-up adjustment plan includes follow-up frequency and follow-up content.
[0010] In some embodiments, before extracting preliminary features from the preprocessed image of the patient to obtain preliminary features of the preprocessed image, the method further includes:
[0011] The acquired image of the patient is preprocessed to obtain an image with high recognition.
[0012] In some embodiments, the step of performing preliminary feature extraction on the pre-processed image of the patient to obtain preliminary features of the pre-processed image, and performing importance judgment on the preliminary features to obtain an important feature set, and then analyzing the image features in the important feature set by a binary large object analysis method to obtain tumor image features, includes:
[0013] Performing preliminary feature extraction on the preprocessed image through a pretrained ResNet-50 network; wherein the pretrained ResNet-50 network is obtained by adjusting the last residual block and the fully connected layer of the ResNet-50 network and performing knee bone tumor image feature extraction training;
[0014] The importance of the preliminary features is scored according to LASSO regression and principal component analysis, and then important features are obtained according to the scoring results and preset criteria to form the important feature set;
[0015] Morphological operations are applied to the image corresponding to the important feature set to obtain a highly recognizable image, and the highly recognizable image is converted into a binary image to obtain a potential tumor area, and then possible tumor areas are extracted from the potential tumor areas through a connected component analysis method, and then non-tumor areas are excluded through preset rules to obtain a target tumor area.
[0016] In some embodiments, excluding non-tumor areas by a preset rule to obtain a target tumor area includes:
[0017] The complexity of the edge of the possible tumor region is judged by an edge monitoring algorithm, so as to perform a first mark on the possible tumor region that meets the preset tumor edge complexity, and perform a second mark on the remaining possible tumor regions;
[0018] Analyzing the texture features of the possible tumor regions to obtain possible tumor regions that meet preset texture features, performing a third mark on the possible tumor regions that meet the preset texture features, and performing a fourth mark on the remaining possible tumor regions;
[0019] Analyzing the regional features of the possible tumor regions to obtain possible tumor regions that meet the preset regional features, performing a fifth mark on the possible tumor regions that meet the preset regional features, and performing a sixth mark on the remaining possible tumor regions; wherein the regional features include area features, perimeter features, and grayscale value features;
[0020] Determine whether each of the possible tumor regions is marked with a preset number of marks among the first mark, the third mark, and the fifth mark; wherein the preset number is greater than or equal to 2;
[0021] If so, the current possible tumor region is determined to be the target tumor region.
[0022] In some embodiments, the analyzing the tumor image features by a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result includes:
[0023] Extracting spatial features of the target tumor area in the image at each time point in time sequence by a recurrent convolutional neural network, and processing the spatial features to obtain interdependencies of the spatial features;
[0024] Combining the interdependence of the spatial features at each time point, the spatial features at each time point are fused to generate a comprehensive feature vector;
[0025] Inputting the comprehensive feature vector into a plurality of preset models respectively to obtain a preliminary prediction result of each model;
[0026] The multiple preliminary prediction results are combined according to preset rules to obtain a final prediction result.
[0027] In some embodiments, the method of obtaining the treatment plan closest to the current patient from a medical database based on the tumor image classification result in combination with a preset algorithm, and adaptively adjusting the closest treatment plan in combination with the information of the current patient to obtain a preliminary treatment plan, and performing a simulation evaluation on the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response, includes:
[0028] Calculating the similarity between the patient's medical feature vector and the medical feature vector of each historical patient stored in the medical library by using the Euclidean algorithm; wherein the medical feature vector is a vector converted according to the patient's medical parameters; wherein the medical parameters include age, gender, pathological characteristics, gene information and the tumor image classification result;
[0029] Determine the closest treatment plan according to the similarity with the medical feature vector of each historical patient, and use the closest treatment plan as the preliminary treatment plan;
[0030] According to the relevant information of the current patient, the preliminary treatment plan is adjusted by a particle swarm optimization algorithm to obtain an adjusted treatment plan, and a simulation evaluation is performed on the adjusted treatment plan to obtain a predicted treatment effect and a predicted treatment response.
[0031] In a second aspect, an embodiment of the present application provides a knee bone tumor treatment plan acquisition system, including: a tumor image feature extraction module, a tumor classification module and a plan decision module;
[0032] The tumor image feature extraction module is used to perform preliminary feature extraction on the patient's preprocessed image to obtain preliminary features of the preprocessed image, and to perform importance judgment on the preliminary features to obtain an important feature set, and to analyze the image features in the important feature set to obtain tumor image features; wherein the preliminary features include edge features, texture features, shape features, contrast features and density features;
[0033] The tumor classification module is used to perform feature analysis on the tumor image features through a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result;
[0034] The treatment plan acquisition module is used to obtain a treatment plan matching the current patient from a medical database according to the tumor image classification result in combination with a preset algorithm, and adaptively adjust the matching treatment plan in combination with the information of the current patient to obtain a preliminary treatment plan, and simulate and evaluate the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response;
[0035] The treatment plan acquisition module is also used to determine the decision tree algorithm for analysis through the pre-trained plan based on the current patient information, the predicted treatment effect and the predicted treatment response to obtain a final treatment plan.
[0036] In a third aspect, an embodiment of the present application provides a terminal device, which includes a processor and a memory, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the above-mentioned method for obtaining a treatment plan for a knee bone tumor.
[0037] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when executed on a processor, implements the above-mentioned method for obtaining a treatment plan for a knee bone tumor.
[0038] The embodiments of the present application have the following beneficial effects: the present application first extracts preliminary features from the preprocessed image, then obtains important features from the preliminary features, and then analyzes the important features to obtain tumor image features; then analyzes tumor image features through a recurrent convolutional neural network and a preset multi-model fusion method to obtain tumor image classification results; according to the type classification results, historical patient cases matching the patient are obtained in combination with a preset algorithm, and the matching historical patient cases are adaptively adjusted according to the specific circumstances of the patient to obtain a treatment plan that suits the patient himself. It can be seen that the knee bone tumor treatment plan obtained by the method of the present application has a high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0040] Figure 1 A first flow chart of the method for obtaining a treatment plan for a knee bone tumor according to an embodiment of the present application is shown;
[0041] Figure 2 A second flow chart of the method for obtaining a treatment plan for a knee bone tumor according to an embodiment of the present application is shown;
[0042] Figure 3 A third flow chart of the method for obtaining a treatment plan for a knee bone tumor according to an embodiment of the present application is shown;
[0043] Figure 4 A fourth process diagram of the method for obtaining a treatment plan for a knee bone tumor according to an embodiment of the present application is shown;
[0044] Figure 5 A fifth process diagram of the method for obtaining a treatment plan for a knee bone tumor according to an embodiment of the present application is shown;
[0045] Figure 6 A sixth flow chart of the method for obtaining a treatment plan for a knee bone tumor according to an embodiment of the present application is shown;
[0046] Figure 7 A structural schematic diagram of a system for acquiring a treatment plan for a knee bone tumor according to an embodiment of the present application is shown. DETAILED DESCRIPTION
[0047] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.
[0048] The components of the embodiments of the present application generally described and shown in the drawings herein may be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.
[0049] Hereinafter, the terms "including", "having" and their cognates that can be used in various embodiments of the present application are intended only to indicate specific features, numbers, steps, operations, elements, components or a combination of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or a combination of the foregoing items or increasing the possibility of one or more features, numbers, steps, operations, elements, components or a combination of the foregoing items. In addition, the terms "first", "second", "third" and the like are only used to distinguish descriptions and cannot be understood as indicating or implying relative importance.
[0050] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present application belong. Terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present application.
[0051] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.
[0052] At present, the treatment of knee bone tumors mainly relies on doctors to make judgments based on images, combined with experience and professional knowledge. This method is very subjective, and the accuracy of the treatment plan given cannot be guaranteed, which in turn affects the treatment of patients. Based on this, the present application proposes a method, system, terminal and storage medium for obtaining a treatment plan for knee bone tumors. First, the acquired patient images are preprocessed to obtain highly recognizable images; then, feature extraction is performed based on the preprocessed images to obtain tumor image features, and then the tumor image features are classified by a preset method to obtain tumor image classification results. Finally, based on the tumor image classification results and combined with historical patient cases, a final treatment plan is given. The method of the embodiment of the present application can formulate an objective, accurate and more suitable treatment plan for patients.
[0053] The method for obtaining the treatment plan for knee bone tumors is described below with reference to some specific embodiments.
[0054] Figure 1 A schematic flow chart of a method for obtaining a treatment plan for a knee joint bone tumor according to an embodiment of the present application is shown. Exemplarily, the method for obtaining a treatment plan for a knee joint bone tumor comprises the following steps:
[0055] Step S100, performing preliminary feature extraction on the patient's preprocessed image to obtain preliminary features of the preprocessed image, performing importance judgment on the preliminary features to obtain an important feature set, and analyzing the image features in the important feature set to obtain tumor image features.
[0056] In some implementations, before step S100, the acquired image needs to be preprocessed to obtain an image with high recognition.
[0057] The image is usually obtained through medical equipment such as CT, PET, X-ray, magnetic resonance or ultrasound imaging. The preprocessing of the image in this step is mainly to improve the visualization quality of the image and provide a clear image basis for subsequent feature extraction and feature classification.
[0058] like Figure 2 As shown, the acquired image is preprocessed to obtain a highly recognizable image, specifically including:
[0059] Step S10, continue to denoise the image using Gaussian filtering denoising technology.
[0060] Step S20, enhancing the contrast of the image after denoising by using an adaptive histogram equalization technique.
[0061] In some embodiments, step S10 is mainly to remove random noise caused by mechanical vibration or electronic equipment during the image acquisition process of medical equipment such as CT, PET, X-ray, magnetic resonance or ultrasound imaging, so as to improve the quality of the image.
[0062] When preprocessing an image using Gaussian filtering denoising technology, first select an appropriate Gaussian kernel size. The selected Gaussian kernel size must ensure that it can cover the direct neighborhood around the key pixels, effectively smooth small noise fluctuations, and retain important details. In this application, the Gaussian kernel size selects a 5x5 kernel.
[0063] In addition, the standard deviation of the Gaussian filter function of the filter in the Gaussian filter denoising technology needs to be determined, and the standard deviation determines the blur degree of the filter. In this application, the standard deviation can be set to 1.5 to achieve a balance between the denoising kernel and the image details.
[0064] After setting the Gaussian kernel size and kernel standard deviation, the Gaussian filtering algorithm is used to process the image. After preprocessing the image, the present application also needs to compare the image before and after the preprocessing to ensure that key tumor identification details are not lost; in addition, the present application also needs to score the quality of the preprocessed image. When scoring the image quality, the image quality can be quantified by calculating the signal-to-noise ratio (SNR) or structural similarity (SSIM) index of the image to ensure that the quality of the denoised image meets the predetermined standard; wherein, SNR measures the ratio of signal intensity to noise, and SSIM measures the similarity of the image before and after denoising. If the denoising does not achieve the preset effect and / or the quality of the preprocessed image does not meet the preset requirements, the image is re-preprocessed until the preprocessed image reaches the preset denoising effect and meets the quality requirements.
[0065] Step S20, enhancing the contrast of the image after denoising by using an adaptive histogram equalization technique.
[0066] In step S20, the contrast of the denoised image is enhanced by using the adaptive histogram equalization technique (CLAHE), so that the details of the tumor area can be more obvious.
[0067] Before contrast enhancement, it is necessary to first set the CLAHE cropping limit and grid size; wherein, the cropping limit means that the CLAHE algorithm will limit the maximum value of the histogram during local contrast enhancement during the processing process to prevent artifacts caused by excessive contrast enhancement; the present application can set the CLAHE cropping limit to 2.0 to avoid image distortion caused by excessive enhancement; in addition, the present application can set the grid size to 8x8 to adjust the contrast of small areas of the image and enhance the local details of the image; after setting the cropping limit and grid size, the CLAHE algorithm is applied to the denoised image to enhance the contrast. In order to avoid the image after contrast enhancement not meeting the requirements, the image after contrast enhancement can also be evaluated. During the evaluation, it can be evaluated by a professional radiologist, or other automated evaluation methods can be used. The specific evaluation method is not specifically limited here.
[0068] In some embodiments, Figure 3 As shown, step S100 specifically includes:
[0069] Step S110, performing preliminary feature extraction on the preprocessed image through a pre-trained ResNet-50 network.
[0070] In this step, the pre-trained ResNet-50 network is pre-trained on a large-scale image dataset and can capture complex features in the image. The pre-trained ResNet-50 network in this application mainly extracts preliminary features including edges, textures, and shapes from the pre-processed image, which are important for distinguishing benign tumors from malignant tumors. The dataset used for training the pre-trained ResNet-50 network of this application is a knee MRI dataset annotated by professional radiologists, which includes most types of knee bone tumors, such as osteosarcoma and osteochondroma.
[0071] Among them, the pre-trained ResNet-50 network is obtained by adjusting the last residual block and fully connected layer of the ResNet-50 network and performing knee bone tumor image feature extraction training.
[0072] The adjustment of the pre-trained ResNet-50 network in this application specifically includes: first, freezing the previous residual block layer, and only training a few pre-set layers, so that the efficiency of model training can be improved on the basis of achieving the purpose of extracting preliminary features in this application, for example, only layer4 and fc layers can be trained. In addition, the last fully connected layer can be replaced with a new layer suitable for knee bone tumor classification. Then, when training the ResNet-50 network, it is necessary to set the optimizer and learning rate, and optimize the performance of the model by adjusting the parameters; when setting the learning rate, a lower learning rate is usually used (for example, the initial learning rate can be 0.1, and the learning rate set in this step can be 1 / 10 of the original learning rate) to ensure the stability of the model during fine-tuning. At the same time, select an appropriate batch size (for example, 32 or 64) to balance training efficiency and memory usage; finally, use the training data to train the model, and adjust the number of iterations according to the performance on the validation set to ensure that the model converges within an appropriate time; wherein the number of iterations is usually tens to hundreds of epochs, which can be adjusted specifically according to the training effect; in this application, the speed of model update can be controlled by adjusting the learning rate to avoid overfitting or underfitting; adjusting the batch size can balance computing resources and training effect; according to the training effect, adjust the number of iterations to ensure that the model converges within an appropriate time.
[0073] Step S120, scoring the importance of the preliminary features according to LASSO regression and principal component analysis, and then obtaining important features according to the scoring results and preset criteria to form an important feature set.
[0074] The preliminary features extracted through step S110 include features such as edges, textures, shapes, contrast, and density. In this step, the importance of the extracted preliminary features is scored through LASSO regression and principal component analysis, and important features are obtained based on the scoring results and preset criteria, where the preset criteria can be a score range. If the score of a feature is higher than the preset criteria, it can be classified into the important feature set. By selecting the important feature set, the present application can improve the accuracy and stability of the classification model.
[0075] In step S130, morphological operations are applied to the image corresponding to the important feature set to obtain a high-recognition image, and the high-recognition image is converted into a binary image to obtain a potential tumor area, and then the possible tumor area is extracted from the potential tumor area by a connected component analysis method, and then the non-tumor area is excluded by preset rules to obtain the target tumor area.
[0076] In this step, the morphological operations applied can be dilation and erosion; the dilation operation can increase the pixels at the edge of the image, fill the internal holes, and make the tumor area more prominent; the erosion operation can remove the noise at the edge of the image, making the tumor boundary clearer. By applying morphological operations to the image corresponding to the important feature set, the recognition of the tumor area can be enhanced and the impact of noise on the image can be reduced.
[0077] After acquiring a highly recognizable image, the present application converts the image into a binary image by using global or adaptive threshold technology to distinguish potential tumor areas, and then uses connected component analysis to identify and calibrate all connected pixel blocks, and extract possible tumor areas from potential tumor areas. This process ensures that the tumor area is accurately identified and extracted, facilitating subsequent tumor classification.
[0078] After obtaining the possible tumor area, in order to ensure that the selected tumor area is more accurate, this application also needs to verify each possible tumor area to exclude non-tumor areas and retain only the target tumor area, which can further improve the accuracy of subsequent tumor classification.
[0079] like Figure 4 As shown, in this application, non-tumor areas are excluded by preset rules and algorithms to obtain the target tumor area, which specifically includes:
[0080] Step S131, the complexity of the edge of the possible tumor region is determined by an edge detection algorithm, so as to perform a first mark on the possible tumor region that meets the preset tumor edge complexity, and perform a second mark on the remaining possible tumor regions.
[0081] Since tumor areas usually have specific shape characteristics, such as irregular shapes, blurred edges, etc. In this application, the shape factors of possible tumor areas, such as circularity, aspect ratio, etc., are analyzed by edge detection algorithms, so that possible tumor areas whose shape characteristics obviously do not conform to the characteristics of tumor images can be found, and possible tumor areas that obviously do not conform to the characteristics of tumor images are marked as second, and possible tumor areas that meet the preset tumor edge complexity are marked as first. Among them, the edge detection algorithm of this application can be a Canny edge detection algorithm.
[0082] Step S132, analyzing the texture features of the possible tumor regions to obtain possible tumor regions that meet the preset texture features, performing a third mark on the possible tumor regions that meet the preset texture features, and performing a fourth mark on the remaining possible tumor regions.
[0083] Since tumor areas usually have specific texture patterns, such as high contrast, heterogeneity, etc. In this application, a texture analysis method can be used to evaluate the texture features of each possible tumor area. By calculating texture features (such as contrast, correlation, entropy, etc.), it is possible to find areas where the texture pattern is obviously inconsistent with the tumor image features, and the areas where the texture pattern is obviously inconsistent with the tumor image features are marked fourth, and the remaining possible tumor areas are marked third. Among them, the texture analysis method used in this application can be a gray level co-occurrence matrix (GLCM).
[0084] Step S133, analyzing the regional features of the possible tumor regions to obtain possible tumor regions that meet the preset regional features, and performing a fifth marking on the possible tumor regions that meet the preset regional features, and performing a sixth marking on the remaining possible tumor regions.
[0085] Among them, regional features include area features, perimeter features and gray value features.
[0086] Since the regional features of the tumor region usually also have specific features, in this step, the possible tumor region that obviously does not meet the area and perimeter can be determined by calculating the area and perimeter of the possible tumor region; and the grayscale histogram of the possible tumor region can be analyzed to further determine the possible tumor region that obviously does not meet the grayscale value distribution. The possible tumor region that does not meet the area and perimeter and / or the possible tumor region that obviously does not meet the grayscale value distribution is marked sixthly, and the remaining possible tumor regions are marked fifthly.
[0087] Step S134 , determining whether each possible tumor region is marked with a preset number of marks among the first mark, the third mark, and the fifth mark.
[0088] Among them, the preset number is greater than or equal to 2.
[0089] If it is determined in step S134 that the possible tumor region contains two or more of the first mark, the third mark and the fifth mark, step S135 is executed to determine that the current possible tumor region is the target tumor region.
[0090] If it is determined in step S134 that there is no first mark, third mark or fifth mark in the possible tumor area, or there is only one mark among the first mark, third mark and fifth mark, step S136 is executed to determine that the current possible tumor area is not the target tumor area.
[0091] Based on the above-mentioned morphological feature analysis, texture feature analysis and regional feature analysis, it is determined whether each possible tumor area is marked with at least two of the first mark, the third mark and the fifth mark. If so, the possible tumor area is determined as the target tumor area.
[0092] Step S200, analyzing the tumor image features through a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result.
[0093] In this step, a recurrent network is used to process the time-dependent features in the image sequence, and the features of each time point are fused to generate a comprehensive feature vector that can reflect the dynamic characteristics of the tumor. The neutralized feature vector is then input into different prediction models to obtain multiple prediction results, which are then comprehensively analyzed to obtain the tumor image classification results.
[0094] like Figure 5 As shown, step S200 specifically includes:
[0095] Step S210, extracting the spatial features of the target tumor area in the image at each time point in time sequence through a recurrent convolutional neural network, and processing the spatial features to obtain the interdependence of the spatial features.
[0096] Step S220 , combining the interdependence of the spatial features at each time point, and fusing the spatial features at each time point to generate a comprehensive feature vector.
[0097] In some embodiments, for each time point of the image, the spatial features of the target tumor area are extracted by the convolution layer of the recurrent convolutional neural network, wherein the spatial features include, but are not limited to, features such as edges, textures, and shapes. The extracted spatial features are then input into a gated recurrent unit (GRU) so that the GRU processes the data obtained in time series and captures the dynamic relationship of spatial features over time. At each time point, the hidden state output by the GRU is fused with the spatial features of the current time point. Specifically, the hidden state can be combined with the spatial features by concatenation or weighted sum to generate a new feature vector, and then the features of each time point are fused by the above method to generate a comprehensive feature vector that can reflect the dynamic characteristics of the tumor over time.
[0098] Step S230, inputting the comprehensive feature vector into a plurality of preset models respectively to obtain a preliminary prediction result of each model.
[0099] Step S240, multiple preliminary prediction results are merged according to preset rules to obtain a final prediction result.
[0100] In some embodiments, the preset models include but are not limited to random forests, decision trees, and deep learning models. The comprehensive feature vectors obtained above are respectively input into random forests, decision trees, and deep learning models, and each model makes predictions independently to generate preliminary prediction results for tumor classification of each model. Among them, random forests make predictions by integrating multiple decision trees, decision trees perform classification through tree structures, and deep learning models classify complex features through neural networks. According to the performance of each model on the training set, different weights can be assigned to each model, and then the application combines the prediction results of each model with the multiple preliminary prediction results through a weighted voting mechanism, and finally outputs the final prediction results.
[0101] It should be noted that in order to ensure the accuracy of the classification results, the random forest, decision tree and deep learning models used in this application must be clinically verified. During the clinical verification, the results of each classification technology are compared with the actual pathological results to verify the accuracy of the model; and a retrospective analysis is performed to evaluate the classification accuracy and reliability of each technology on different tumor types. Based on clinical feedback and the results of retrospective analysis, the classification model is continuously optimized to ensure the high accuracy and reliability of the classification results in actual clinical applications.
[0102] The above scheme can not only improve the accuracy and efficiency of tumor classification, but also provide an objective and scientific scheme for subsequent treatment.
[0103] Step S300, based on the tumor image classification result, combined with a preset algorithm, a treatment plan matching the current patient is obtained from the medical database, and the matching treatment plan is adaptively adjusted in combination with the current patient's information to obtain a preliminary treatment plan, and a simulation evaluation is performed on the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response.
[0104] Among them, the medical database in this step is a medical database related to the knee bone tumor of this application. The medical database can be obtained by collecting and organizing historical patient data, where the patient data includes but is not limited to medical history information, treatment methods, treatment results and patient feedback, etc. The medical database must ensure that each case is recorded in detail, including but not limited to the patient's treatment process, follow-up data, and final health status.
[0105] In some embodiments, Figure 6 As shown, step S300 specifically includes:
[0106] Step S310, calculating the similarity between the patient's medical feature vector and the medical feature vector of each historical patient stored in the medical database by using the Euclidean algorithm.
[0107] The medical feature vector is a vector converted according to the patient's medical parameters, including age, gender, pathological characteristics, gene information, and tumor image classification results.
[0108] In this step, the medical parameters of the patient are converted into feature vectors, and then the distance between the feature vectors of the patient and the patients in the medical database is calculated by the Euclidean algorithm. The smaller the distance, the higher the similarity. Furthermore, different weight values can be assigned to different medical parameters, and then the weighted Euclidean distance is calculated to reflect the different importance of each medical parameter in the similarity assessment.
[0109] Step S320, determining the closest treatment plan based on the similarity with the medical feature vector of each historical patient, and using the closest treatment plan as the preliminary treatment plan.
[0110] In step S310, if a historical patient with high similarity is obtained, the treatment plan of the historical patient can be used as the closest treatment plan, and then the closest treatment plan can be used as the preliminary treatment plan.
[0111] Step S330, according to the relevant information of the current patient, the preliminary treatment plan is adjusted by a particle swarm optimization algorithm to obtain an adjusted treatment plan, and a simulation evaluation is performed on the adjusted treatment plan to obtain a predicted treatment effect and a predicted treatment response.
[0112] In this step, the preliminary treatment plan is adjusted according to the individual conditions of the patient, such as age, drug sensitivity, complications, etc. The specific contents of the adjustment include but are not limited to the treatment method, dosage, duration of treatment, etc. For example, if the patient has an allergic reaction to a certain drug, the system will automatically exclude the drug and select an alternative drug. In this step, each content in the preliminary treatment plan is adjusted. The particle swarm optimization algorithm simulates the search behavior of the particle swarm in the solution space to find the optimal solution. Each particle represents an adjusted content. After adjusting one of the contents, the algorithm evaluates the effect of the plan (such as the expected treatment results), and then continuously updates the position and speed of the particles, gradually approaches the optimal solution, and then obtains the adjusted treatment plan.
[0113] After obtaining the adjusted treatment plan, this application will conduct a simulation evaluation of the adjusted treatment plan to ensure the feasibility of the adjusted treatment plan in terms of safety and effectiveness. The simulation evaluation includes predicting the treatment effect, evaluating possible side effects, etc.
[0114] It should be noted that after treating the patient, the present application will also collect the treatment situation, including but not limited to treatment response and health results, and feed this information back to the above-mentioned model and case library, and continuously update the above-mentioned model and case library to continuously optimize the model and algorithm used in the present application. Step S400, according to the current patient information, predicted treatment effect and predicted treatment response, the decision tree algorithm is determined through the pre-trained scheme for analysis to obtain the final treatment plan.
[0115] In this step, the patient's tumor image features, individual condition features, predicted treatment effects, and predicted treatment responses are input into the trained solution determination decision tree model, which is classified according to the preset decision rules and provides optional treatment options. Each treatment option is accompanied by a success rate prediction, so that the treatment option with the highest success rate is used as the final treatment plan. It should be noted that when determining the final treatment plan, the treatment option with the highest success rate may not be selected, but the doctor may select it and then determine the final treatment plan.
[0116] The training process of the decision tree model determined in the scheme of this step is: first, key medical features are selected from the medical database. These features have a high predictive value for the treatment of bone tumors. The selected medical features include but are not limited to pathological type, tumor image features, tumor location, patient age, treatment effect and treatment screening. A training set can be established based on the medical features of historical patients obtained from the medical database; then a decision tree learning algorithm (such as ID3, C4.5 or CART) is applied to train the data in the training set, and the model is constructed. The constructed decision tree model recursively divides the data and gradually constructs decision rules according to the values of different features. For example, the decision tree model constructed using the CART algorithm forms a tree structure through binary segmentation, each node represents the judgment of a feature, and the leaf node represents the final classification result. After the model is constructed, the effect and accuracy of the model must be verified by cross-validation technology to ensure the generalization ability of the model on different data sets.
[0117] In this application, CBR technology and decision tree algorithm are used simultaneously. These two methods complement each other and together provide patients with more accurate and personalized treatment recommendations. Among them, CBR technology generates personalized treatment recommendations by finding similar historical cases and based on these cases, while the decision tree algorithm further refines and diversifies treatment options by analyzing and classifying patient data. This combined approach takes advantage of the advantages of both technologies and ensures the reliability and personalization of treatment recommendations.
[0118] In some embodiments, the method of the present application further includes: acquiring the patient's rehabilitation data in real time, and generating a follow-up adjustment plan based on the rehabilitation data; wherein the follow-up adjustment plan includes the follow-up frequency and the follow-up content.
[0119] It can be understood that in the present application, the rehabilitation data of the patient is obtained in real time, and the rehabilitation data is input into a preset follow-up model. The follow-up model outputs an adjusted follow-up plan based on the input rehabilitation data, and then the follow-up frequency and content can be adjusted in real time according to the actual situation of the patient.
[0120] The construction of the follow-up model includes: firstly, determining the rehabilitation data that has an impact on the patient's condition, and the rehabilitation data includes health indicators and clinical parameters, wherein the health indicators include but are not limited to pain level, treatment response and activity ability, etc.; clinical parameters include but are not limited to biochemical markers, tumor marker levels and post-treatment review results, etc. Then, a training data set is established, and the rehabilitation data of each historical patient is obtained in time series, and the obtained rehabilitation data is cleaned and preprocessed to ensure the quality and consistency of the input data, wherein the data cleaning and preprocessing can be processing missing values, outliers, standardizing data, etc. Then, a suitable model for processing medical data is selected, such as one of random forests, support vector machines (SVM) and neural networks. When training the selected model, the data set is first divided into a training set, a validation set and a test set, usually divided in a ratio of 8:1:1, and then the preprocessed data features are input into the model, and finally the model is trained using the training set, and the model parameters are optimized through repeated iterations to make it perform best on the training set. During the follow-up model construction process, parameters need to be adjusted, mainly using grid search or random search methods for hyperparameter optimization; for example, for random forests, adjust the number of trees and the maximum depth; for SVM, adjust the kernel function type and regularization parameters; for neural networks, adjust the number of layers, the number of neurons in each layer, the learning rate, etc. In addition, after each parameter adjustment, use the validation set to evaluate the model performance, and reduce the risk of overfitting through cross-validation. When evaluating the constructed follow-up model, the generalization ability of the model can be evaluated through k-fold cross-validation, dividing the data set into k subsets, using k-1 subsets for training each time, and the remaining 1 subset for validation, and repeating k times; among them, the evaluation indicators include but are not limited to accuracy, precision, recall and F1 score; among them, accuracy is the proportion of samples predicted correctly to the total samples; precision is the proportion of true positive examples in samples predicted as positive examples; recall is the proportion of true positive examples to all positive examples; F1 score is the harmonic mean of precision and recall, which comprehensively evaluates the classification ability of the model. After evaluating the constructed model, it is necessary to validate it on the test set to assess its performance on unseen data.
[0121] In some embodiments, the rehabilitation data can be obtained through a wearable device deployed on the patient, or the patient can enter the data through a mobile device. In the present application, the patient's rehabilitation data can also be displayed in real time. The display form is not specifically limited. It can be in a tabular form or a graphical trend chart, such as a time series chart of pain level, daily changes in activity level, etc. In the present application, thresholds or conditions can also be set for the patient's rehabilitation data. If the patient's rehabilitation data is no longer at the threshold or does not meet the set conditions, an alarm can be issued to remind the doctor and / or patient so that the doctor and / or patient can intervene in time.
[0122] In addition, adjustment strategies can be preset in this application, so that after obtaining the patient's rehabilitation data, the follow-up plan can be adjusted in a targeted manner according to the rehabilitation data. If the patient's condition is stable according to the rehabilitation data analysis, it may be manifested as pain levels and mobility continuing to remain within the normal range. In this case, the follow-up frequency can be reduced, such as adjusting to a follow-up every 6 months to check basic health indicators and imaging examinations; if the patient's condition is a slight symptom change according to the rehabilitation data analysis, it may be manifested as a slight increase in pain or a slight decrease in mobility. In this case, close monitoring is required. In this case, the follow-up can be adjusted to every 3 months, and the frequency of blood tests and imaging examinations can be increased; if the patient's condition is a significant symptom change according to the rehabilitation data analysis, it may be manifested as a significant increase in pain levels or a significant decrease in mobility. In this case, timely intervention is required. This In this case, follow-up can be conducted every month, with additional detailed blood tests, imaging tests, and possible further treatment tests; if the patient's condition is found to be at increased risk of recurrence based on recovery data analysis, preventive measures should be taken in a timely manner. In this case, follow-up can be conducted every 1-2 months, with additional advanced imaging tests (such as MRI or CT scans) and tumor marker testing, and preventive treatment measures may need to be initiated; in addition, the applicant can also adjust the follow-up plan as needed, such as personalized adjustments based on individual differences of patients (such as age, gender, comorbidities, etc.), and personalized follow-up plans, which may include specific examination items and consultation content to meet the special needs of patients.
[0123] The present application pre-processes images through an automated image processing method, thereby achieving the effect of improving image recognition, and can improve the accuracy of subsequent tumor image feature extraction and classification. By pre-processing images in an automated manner, the efficiency of image processing can also be improved, and the workload of doctors can be reduced. In addition, in the process of tumor image feature extraction, tumor division and treatment plan determination, the present application integrates multiple decision-making algorithms, combines imaging data and clinical information, and can provide personalized treatment recommendations to assist doctors in formulating the best treatment plan. Furthermore, the present application adopts deep learning technology and diversified data training, and therefore has strong generalization ability and robustness.
[0124] Figure 7 A schematic diagram of a knee joint bone tumor treatment plan acquisition system according to an embodiment of the present application is shown. Exemplarily, the knee joint bone tumor treatment plan acquisition system includes: a tumor image feature extraction module 100, a tumor classification module 200 and a treatment plan acquisition module 300;
[0125] The tumor image feature extraction module 100 is used to perform preliminary feature extraction on the pre-processed image of the patient to obtain preliminary features of the pre-processed image, and to judge the importance of the preliminary features to obtain an important feature set, and to analyze the image features in the important feature set to obtain tumor image features; wherein the preliminary features include edge features, texture features, shape features, contrast features and density features;
[0126] The tumor classification module 200 is used to perform feature analysis on tumor image features through a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result;
[0127] The treatment plan acquisition module 300 is used to obtain a treatment plan matching the current patient from a medical database according to the tumor image classification result and in combination with a preset algorithm, and to adaptively adjust the matching treatment plan in combination with the current patient's information to obtain a preliminary treatment plan, and to simulate and evaluate the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response;
[0128] The treatment plan acquisition module 300 is also used to determine the decision tree algorithm for analysis based on the current patient information, predicted treatment effect and predicted treatment response through the pre-trained plan to obtain the final treatment plan.
[0129] It can be understood that the device of this embodiment corresponds to the method for obtaining a treatment plan for a knee bone tumor in the above embodiment, and the optional items in the above embodiment are also applicable to this embodiment, so they will not be described again here.
[0130] The present application also provides a terminal device. Exemplarily, the terminal device includes a processor and a memory, wherein the memory stores a computer program, and the processor runs the computer program to enable the terminal device to execute the functions of each module in the above-mentioned method for acquiring a knee bone tumor treatment plan or the above-mentioned device for acquiring a knee bone tumor treatment plan.
[0131] Among them, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU) and a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or at least one of other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc., which can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application.
[0132] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. Among them, the memory is used to store a computer program, and the processor can execute the computer program accordingly after receiving an execution instruction.
[0133] The present application also provides a computer-readable storage medium for storing a computer program used in the above-mentioned terminal device. For example, the computer-readable storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0134] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or the flow diagram, and the combination of boxes in the structure diagram and / or the flow diagram, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.
[0135] In addition, the functional modules or units in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
[0136] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a smart phone, a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application.
[0137] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.
Claims
1. A method for obtaining a treatment plan for a knee bone tumor, characterized in that: include: performing preliminary feature extraction on the patient's preprocessed image to obtain preliminary features of the preprocessed image, performing importance judgment on the preliminary features to obtain an important feature set, and analyzing image features in the important feature set to obtain tumor image features; wherein the preliminary features include edge features, texture features, shape features, contrast features, and density features; Performing feature analysis on the tumor image features using a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result; Based on the tumor image classification results, a treatment plan matching the current patient is obtained from a medical database in combination with a preset algorithm, and the matching treatment plan is adaptively adjusted in combination with the current patient's information to obtain a preliminary treatment plan, and a simulation evaluation is performed on the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response; A decision tree algorithm is determined based on the current patient information, the predicted treatment effect and the predicted treatment response through a pre-trained scheme to perform analysis to obtain a final treatment plan.
2. The method for obtaining a knee bone tumor treatment plan according to claim 1, characterized in that: The method further comprises: Acquire the patient's rehabilitation data in real time, and generate a follow-up adjustment plan based on the rehabilitation data; wherein the follow-up adjustment plan includes follow-up frequency and follow-up content.
3. The method for obtaining a knee joint bone tumor treatment plan according to claim 1, characterized in that: Before extracting preliminary features from the pre-processed image of the patient to obtain preliminary features of the pre-processed image, the method further includes: The acquired image of the patient is preprocessed to obtain a highly recognizable image.
4. The method for obtaining a knee joint bone tumor treatment plan according to claim 1, characterized in that: The performing preliminary feature extraction on the preprocessed image of the patient to obtain preliminary features of the preprocessed image, performing importance judgment on the preliminary features to obtain an important feature set, and analyzing image features in the important feature set to obtain tumor image features, includes: Performing preliminary feature extraction on the preprocessed image using a pretrained ResNet-50 network; wherein the pretrained ResNet-50 network is obtained by adjusting the last residual block and fully connected layer of the ResNet-50 network and performing knee bone tumor image feature extraction training; The importance of the preliminary features is scored according to LASSO regression and principal component analysis, and then important features are obtained according to the scoring results and preset criteria to form the important feature set; Morphological operations are applied to the image corresponding to the important feature set to obtain a highly recognizable image, and the highly recognizable image is converted into a binary image to obtain a potential tumor area. Then, a connected component analysis method is used to extract possible tumor areas from the potential tumor areas, and then non-tumor areas are excluded according to preset rules to obtain a target tumor area.
5. The method for obtaining a knee joint bone tumor treatment plan according to claim 4, characterized in that: The method of excluding non-tumor areas by a preset rule to obtain a target tumor area includes: Determining the complexity of the edge of the possible tumor region using an edge monitoring algorithm to perform a first marking on the possible tumor region that meets a preset tumor edge complexity, and performing a second marking on the remaining possible tumor regions; Analyzing the texture features of the possible tumor regions to obtain possible tumor regions that meet preset texture features, performing a third marking on the possible tumor regions that meet the preset texture features, and performing a fourth marking on the remaining possible tumor regions; Analyzing regional features of the possible tumor regions to obtain possible tumor regions that meet preset regional features, performing a fifth marking on the possible tumor regions that meet the preset regional features, and performing a sixth marking on the remaining possible tumor regions; wherein the regional features include area features, perimeter features, and grayscale value features; Determine whether each of the possible tumor regions is marked with a preset number of the first marker, the third marker, and the fifth marker; wherein the preset number is greater than or equal to 2; If so, the current possible tumor region is determined to be the target tumor region.
6. The method for obtaining a knee joint bone tumor treatment plan according to claim 1, characterized in that: The analyzing the tumor image features by using a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result includes: extracting spatial features of the target tumor region in the image at each time point in chronological order using a recurrent convolutional neural network, and processing the spatial features to obtain interdependencies among the spatial features; Combining the interdependence of the spatial features at each time point, the spatial features at each time point are fused to generate a comprehensive feature vector; Inputting the comprehensive feature vector into a plurality of preset models respectively to obtain preliminary prediction results of each model; The multiple preliminary prediction results are combined according to preset rules to obtain the final prediction result.
7. The method for obtaining a knee joint bone tumor treatment plan according to claim 1, characterized in that: The method comprises: obtaining a treatment plan closest to the current patient from a medical database based on the tumor image classification result and a preset algorithm, adaptively adjusting the closest treatment plan based on the current patient information to obtain a preliminary treatment plan, and performing a simulation evaluation on the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response, including: Calculating the similarity between the patient's medical feature vector and the medical feature vector of each historical patient stored in the medical database using the Euclidean algorithm; wherein the medical feature vector is a vector converted according to the patient's medical parameters; wherein the medical parameters include age, gender, pathological characteristics, genetic information, and the tumor image classification result; Determining the closest treatment plan based on the similarity with the medical feature vector of each historical patient, and using the closest treatment plan as the preliminary treatment plan; According to the relevant information of the current patient, the preliminary treatment plan is adjusted by a particle swarm optimization algorithm to obtain an adjusted treatment plan, and the adjusted treatment plan is simulated and evaluated to obtain a predicted treatment effect and a predicted treatment response.
8. A knee joint bone tumor treatment plan acquisition system, characterized in that: include: Tumor image feature extraction module, tumor classification module and solution decision module; The tumor image feature extraction module is configured to perform preliminary feature extraction on the patient's preprocessed image to obtain preliminary features of the preprocessed image, perform importance judgment on the preliminary features to obtain an important feature set, and analyze image features in the important feature set to obtain tumor image features; wherein the preliminary features include edge features, texture features, shape features, contrast features, and density features; The tumor classification module is configured to perform feature analysis on the tumor image features using a recurrent convolutional neural network and a preset multi-model fusion method to output a tumor image classification result; The treatment plan acquisition module is used to obtain a treatment plan that matches the current patient from a medical database based on the tumor image classification result and a preset algorithm, and to adaptively adjust the matched treatment plan based on the current patient's information to obtain a preliminary treatment plan, and to perform a simulation evaluation on the preliminary treatment plan to obtain a predicted treatment effect and a predicted treatment response; The treatment plan acquisition module is further used to determine a decision tree algorithm for analysis based on the information of the current patient, the predicted treatment effect and the predicted treatment response through a pre-trained plan to obtain a final treatment plan.
9. A terminal device, characterized in that: The terminal device includes a processor and a memory, the memory stores a computer program, and the processor is used to execute the computer program to implement the method for obtaining a knee bone tumor treatment plan according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that It stores a computer program, which, when executed on a processor, implements the method for obtaining a knee joint bone tumor treatment plan according to any one of claims 1-7.
Citation Information
Cited By
Schistosomiasis B ultrasonic image fibrosis grading data dynamic analysis method
CN119850696A
Desktop article rapid positioning method based on smart home
CN119963644A
A smart home-based desktop item rapid positioning method
CN119963644B
Clinical treatment scheme recommendation method based on brain diseases
CN120108698A
Prediction system for chronic obstructive pulmonary disease and storage medium
CN120299689A
