A medical image recognition method and system integrating deep learning and big data algorithms

Abstract

The present invention discloses a medical image recognition method and system that integrates deep learning and big data algorithms, aiming to improve the efficiency and accuracy of medical image recognition. The method includes the following steps: collecting medical image data from multiple medical institutions and building a large-scale medical image database; performing standardization processing on the collected medical images, including image enhancement, denoising, and resizing; based on the processed data, using a convolutional neural network to build a deep learning model, and combining it with a big data algorithm to extract features; using an integrated learning method to train the deep learning model and optimize the model parameters; after the model training is completed, the input medical image is recognized in real time, and the recognition results are fed back to the doctor; the recognition results and their related image data are stored in the cloud to provide support for subsequent analysis. The method of the present invention can effectively improve the efficiency of medical image recognition and provide more reliable auxiliary support for clinical practice.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a medical image recognition method and system integrating deep learning and big data algorithms. Background Art

[0002] With the rapid development of medical imaging technology, the scale and complexity of medical imaging data have increased significantly. The widespread use of various medical imaging technologies, such as CT, MRI, and X-rays, has provided an important basis for the early diagnosis and treatment of diseases. However, traditional medical imaging analysis methods often rely on the experience of professional radiologists and face the following major challenges: Traditional imaging analysis methods are typically manually designed based on image processing and feature extraction, resulting in slow processing speed and strong reliance on physician expertise, making it difficult to meet clinical needs for real-time diagnosis. Due to the diversity and complexity of imaging features, manual analysis can lead to decreased diagnostic accuracy, especially when processing large amounts of imaging data, which can easily lead to fatigue and omissions, affecting the final diagnostic results. The lack of effective data sharing and integration mechanisms among medical institutions has resulted in medical imaging data being scattered across different platforms, making it impossible to fully utilize rich historical data for model training and optimization. Although deep learning has demonstrated excellent performance in image recognition, its training process requires a large amount of high-quality labeled data, which traditional image acquisition and annotation methods cannot meet.

[0003] To address these issues, researchers have recently begun exploring the integration of deep learning and big data technologies to improve the efficiency and accuracy of medical image analysis. Deep learning models can automatically learn features from images, reducing reliance on handcrafted features. Meanwhile, big data technologies can help collect, process, and store large amounts of imaging data, supporting model training.

[0004] Although some deep learning-based medical image recognition methods have been proposed, they still suffer from issues such as insufficient model generalization, imbalanced training data, and high computing resource requirements. Therefore, a novel medical image recognition method that integrates deep learning and big data algorithms is urgently needed to achieve more efficient and accurate medical image analysis and provide more reliable support for clinical decision-making. Summary of the Invention

[0005] The present invention provides a medical image recognition method integrating deep learning and big data algorithms, which includes the following steps:

[0006] Collect medical imaging data from multiple medical institutions and build a large-scale medical imaging database;

[0007] Perform standardized processing on collected medical images, including image enhancement, denoising, and resizing;

[0008] Based on the processed data, a deep learning model is constructed using convolutional neural networks and combined with big data algorithms for feature extraction;

[0009] Use ensemble learning methods to train deep learning models and optimize model parameters;

[0010] After the model training is completed, the input medical images are recognized in real time and the recognition results are fed back to the doctor;

[0011] The recognition results and their related image data are stored in the cloud to provide support for subsequent analysis.

[0012] As described above, a medical image recognition method that integrates deep learning and big data algorithms, wherein the deep learning model adopts a convolutional neural network structure, and the model can be adaptively adjusted according to different types of medical images.

[0013] As described above, a medical image recognition method that integrates deep learning and big data algorithms, wherein the data preprocessing steps include image enhancement, denoising and standardization processing, etc., to ensure that the model can handle diverse data inputs during training.

[0014] As described above, a medical image recognition method that integrates deep learning and big data algorithms, wherein, during the model training process, the deep learning model is trained by adopting an integrated learning method to optimize the model parameters.

[0015] As described above, a medical image recognition method integrating deep learning and big data algorithms, wherein the real-time recognition also includes providing the confidence value of the recognition result to the doctor to assist in clinical judgment.

[0016] As described above, a medical image recognition method that integrates deep learning and big data algorithms is described, in which the system is able to generate recognition reports for different disease categories and analysis charts of model performance to help doctors understand the recognition capabilities of the model and its effectiveness in practical applications.

[0017] As described above, a medical image recognition method that integrates deep learning and big data algorithms, wherein the cloud storage has high security and data encryption mechanisms to protect the privacy of medical image data, and supports regular data backup and recovery functions to ensure the security and integrity of the data.

[0018] The present invention also provides a medical image recognition system integrating deep learning and big data algorithms, wherein the system comprises:

[0019] A data acquisition module for collecting medical imaging data from multiple medical institutions;

[0020] Data preprocessing module, used to standardize image data;

[0021] Model training module, used to build and train deep learning models;

[0022] Real-time recognition module, used for real-time recognition of input medical images;

[0023] The data storage module is used to store recognition results and image data in the cloud.

[0024] The present invention also provides a computer storage medium, comprising: at least one memory and at least one processor;

[0025] The memory is used to store one or more program instructions;

[0026] A processor is used to run one or more program instructions to execute a medical image recognition method that integrates deep learning and big data algorithms.

[0027] The present invention achieves the following beneficial effects: It uses a deep learning algorithm to automatically extract key features from medical images, significantly improving the accuracy of disease identification. By integrating deep learning with big data technologies, it enables rapid processing and real-time recognition of medical images, shortening the time it takes doctors to analyze images. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0029] Figure 1 This is a flow chart of a medical image recognition method that integrates deep learning and big data algorithms, as provided in Example 1 of the present application;

[0030] Figure 2 This is a schematic diagram of a medical image recognition system that integrates deep learning and big data algorithms, provided in Example 2 of the present application. DETAILED DESCRIPTION

[0031] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0032] Example 1

[0033] like Figure 1 As shown, the first embodiment of the present application provides a medical image recognition method integrating deep learning and big data algorithms, including the following steps:

[0034] Step S10: Collect medical imaging data from multiple medical institutions and build a large-scale medical imaging database;

[0035] Data collection is the first step in the medical image recognition method, which aims to systematically collect medical image data from multiple medical institutions to build a large-scale medical image database. The specific implementation steps are as follows:

[0036] Selecting Partner Medical Institutions: Establish partnerships with multiple medical institutions with extensive imaging resources, including large general hospitals, specialized hospitals, and research institutions. Ensure these institutions have a strong reputation and strong technical support capabilities in the field of medical imaging. When selecting partner institutions, consider the diversity of imaging types (e.g., CT, MRI, X-ray, etc.) and disease types (e.g., cardiovascular disease, tumors, orthopedic diseases, etc.) to cover different clinical scenarios and patient populations.

[0037] Develop a data collection protocol: Develop a detailed data collection protocol that clearly defines the scope, format, and requirements of data collection, ensuring that the collected data complies with relevant laws, regulations, and ethical standards. All data should be formatted in a standardized medical imaging format, such as DICOM (Digital Imaging and Communications in Medicine), to facilitate subsequent processing.

[0038] Data collection process: By connecting to the hospital's information system, the medical images of patients are automatically collected. This enables real-time data collection and ensures timely data updates. Each image data collection requires the recording of relevant metadata, including patient information (such as age, gender, medical history), image type, acquisition date, device model, etc. This metadata is crucial for subsequent data analysis and model training. To ensure data quality, the following formula is used to assess the diversity and completeness of the collected image data: Among them, Q represents the data quality index; M is the number of medical institutions from which the data comes; N j is the number of images provided by the jth institution; N total is the total number of images provided by all institutions; Var(C j ) is the variance of the disease category distribution of the jth institution's images. By calculating data quality indicators, we ensure the balance and representativeness of the data from each institution.

[0039] Data Annotation and Storage: We collaborate with radiologists to professionally annotate images, including features, disease diagnoses, and treatment plans. This annotated data serves as a crucial foundation for model training. Collected images and their annotated information are stored in a cloud database, ensuring data security and accessibility. Efficient storage solutions are employed to support rapid reading and writing of large amounts of data.

[0040] Real-time update mechanism: A real-time data update mechanism is established to regularly obtain new data from partner hospitals to maintain the currentness and integrity of the imaging database. Each data update requires a quality assessment to ensure that the new data meets the collection standards. Existing data is reviewed and unqualified data is eliminated.

[0041] Step S20: performing standardization processing on the collected medical images, including image enhancement, denoising, and size adjustment;

[0042] Data preprocessing is a critical step in ensuring the quality and usability of medical images. This step includes image enhancement, denoising, and normalization, aiming to improve the quality of image data and prepare it for subsequent deep learning model training.

[0043] In order to improve the visualization effect of the image, a comprehensive image enhancement function E is used to make the image details clearer by adjusting the contrast, brightness and saturation. The enhancement function is defined as follows: Where I(x,y) represents the pixel value of the original image; α is the contrast adjustment parameter that controls the enhancement strength; μ is the mean value of the image; β is the brightness adjustment parameter; k is the domain range, which is used to calculate the local average; γ is the saturation adjustment parameter, which is used to dynamically adjust different parameters and enhance according to the overall characteristics of the image to ensure the visualization of details.

[0044] In order to reduce the noise in the image, a denoising function D based on wavelet transform is used. This function combines local features and global structure and is defined as follows: Among them, W j =(x0,y0) represents the jth wavelet coefficient, representing information at different frequencies; (x0,y0) represents the center coordinates of the image; σ is a parameter that controls the denoising strength; smaller values ​​result in stronger denoising effects; and n represents the number of wavelet transform layers. This formula effectively preserves image edge features while denoising.

[0045] To ensure the uniformity of the image, the image pixel values ​​are normalized to between 0 and 1 using the following normalization formula: Where S represents the normalized pixel value; I(x,y) represents the pixel value of the original image; μ local Represents the mean value of the local area of ​​the image, which can improve the retention of local features; σ globalRepresents the global standard deviation of the image, which is used to ensure the overall consistency of the data; ∈ is a small constant to prevent division by zero; ∑ i,j I(x+i,y+j) represents the sum of all pixels in the local window centered at pixel (x,y); N local Represents the number of local pixels, used to calculate the local mean.

[0046] Step S30: Based on the processed data, a deep learning model is constructed using a convolutional neural network, and feature extraction is performed in combination with a big data algorithm;

[0047] In this step, a deep learning model is constructed based on a convolutional neural network and adaptive feature extraction is performed in conjunction with a big data algorithm. Traditional convolutional neural networks extract image features through fixed convolution kernels and pooling operations. However, in practical applications, the diversity and complexity of image features require the model to automatically adapt and optimize the feature extraction process. This paper proposes an adaptive feature extraction mechanism that combines adaptive convolution kernels, dynamic feature selection, and a group learning strategy to improve the performance of the model in medical image recognition. The specific steps are as follows:

[0048] Adaptive convolution kernel adjustment: In traditional convolutional neural networks, the size and shape of the convolution kernel are usually fixed. However, due to the large differences in features between different regions in medical images, a fixed convolution kernel often cannot fully capture all important information. Therefore, this paper proposes an adaptive convolution kernel adjustment mechanism that enables the convolution kernel to automatically adjust its size and shape according to the local features of the input image. Specifically, the size and shape of the convolution kernel S L It is dynamically generated based on the local statistical characteristics of the image data. For example, image segmentation or cluster analysis can be used to determine the different regions and features of the image. The output F of the convolutional layer is L (x,y) is represented as: Among them, N L is the number of convolution kernels used in the Lth convolution layer; F L (x,y) represents the output feature map of the Lth layer; W L,i (I(x,y),S L ) represents the weight of the i-th convolution kernel in the L-th layer, which is based on the current image I(x,y) and the feature S of the specific layer L L Dynamically generated; S L Represents the shape and size of the convolution kernel, which is adjusted according to the local features of the image; F L-1 (x,y) represents the output feature map of the previous layer. In this way, the convolution kernel can be adaptively adjusted to extract local features in different regions, enhancing the ability to capture image details.

[0049] Dynamic feature selection and weight update: In image data, not all features contribute equally to the classification task. Traditional convolutional neural networks process features uniformly after extracting them at each layer. However, in a big data environment, the number and dimensions of features are often very large. Therefore, this paper introduces a dynamic feature selection and weighting mechanism, which selects and updates features based on their classification contribution in a weighted manner at each layer. Each feature F i The weight α i It will be dynamically adjusted according to its contribution to the classification task. Specifically, the final optimized feature can be expressed as: Among them, M represents the total number of extracted features; F optimized is the final feature representation after weighting; F i represents the i-th extracted feature; α i Represents the weight coefficient of the i-th feature, which is dynamically adjusted according to its importance during the training process. i The update is achieved through the back-propagation algorithm, which optimizes each feature based on its impact on model performance. This mechanism enables the model to automatically learn which features are more important for the classification task, thereby improving the accuracy of the model.

[0050] Group learning feature fusion: In medical image recognition, image data usually has a complex structure, and a single feature extraction strategy may not be able to fully capture all information. Therefore, this step combines the group learning strategy in big data algorithms, using distributed computing and cluster learning to extract features from multiple sub-networks and fuse these features into a unified feature representation. Assume that the network is divided into multiple sub-networks N1, N2, ..., N k , each sub-network independently extracts features from different subsets of the data. Each sub-network N j The extracted features are F j , the final fused feature F final It can be expressed as: Among them, F final represents the final fused feature representation; K represents the number of sub-networks; F j represents the features extracted by the jth sub-network; β j Represents the weight coefficient of the jth sub-network, indicating the importance of the sub-network in the final feature fusion. These weight coefficients β j It is dynamically updated through group learning, ensuring that the final fused features effectively represent the common characteristics of all image data while reducing possible redundant information. Through adaptive convolution kernels, dynamic feature selection, and group learning fusion, this model can continuously optimize the feature extraction process based on the characteristics of the image data.

[0051] Step S40: Based on the processed data, a deep learning model is constructed using a convolutional neural network, and feature extraction is performed in combination with a big data algorithm;

[0052] The convolutional neural network is initialized and trained using the training set. During the training process, the gradient descent method is used to optimize the loss function and gradually update it in combination with the adaptive learning rate. Assume that the loss function is The model parameter is θ, and the parameter is optimized by gradient descent. The formula is as follows: Among them, θ t is the model parameter at the current moment; τ t is the learning rate at the current moment, which is dynamically adjusted according to the training progress; is the gradient of the loss function with respect to the parameters. In order to effectively avoid the local minimum problem, stochastic gradient descent is used in this step, which randomly extracts part of the data for gradient calculation at each update, thereby accelerating the training process and avoiding falling into the local minimum. In order to accelerate convergence and improve training results, the learning rate τ t It will be adaptively adjusted with the number of training rounds t. The following adaptive learning rate adjustment formula is used to ensure that the model is optimized with different learning rates at different stages of training: Where τ0 is the initial learning rate; γ is the decay factor that controls the decay rate of the learning rate; and t is the current training round. The purpose of this formula is to gradually reduce the learning rate as training progresses, ensuring that the model can make more precise adjustments when approaching the optimal solution and avoid over-adjustment. To avoid overtraining of the model, a dynamic early stopping mechanism is introduced, which determines when to stop training by calculating the gain function during the training process. The gain function is defined as: in, and are the loss function values ​​of the current and previous rounds of training, G t It measures the rate of decrease of the loss function during training. If the value of the gain function is less than a preset threshold in N consecutive rounds of training, training is stopped to prevent overfitting.

[0053] After model training and optimization are completed, the model performance is further optimized through the integration method. The results of multiple optimizations are fused through the weighted average method to obtain the final prediction output. The fusion formula is: in, is the final prediction result; K represents the number of sub-models involved in the fusion; is the predicted value of the i-th model; α i The weights of the model are dynamically adjusted based on the training performance of the model; is the average of all predicted values; β is the adjustment factor to control the impact of prediction deviation.

[0054] Step S50: After the model training is completed, the input medical image is recognized in real time, and the recognition result is fed back to the doctor;

[0055] In this step, the input medical images are processed and recognized in real time, generating corresponding diagnostic results. These results are then promptly fed back to the physician through a feedback mechanism. This process ensures that medical decisions can be made in the shortest possible time and provides effective support to improve the efficiency and accuracy of diagnosis and treatment.

[0056] When medical images are input into the system, they are first preprocessed to ensure that the input data meets the model's requirements. After preprocessing, the images are fed into a trained deep learning model for real-time feature extraction and prediction. Based on a convolutional neural network architecture, the model extracts key features from the images through multiple convolutional, pooling, and fully connected layers. Once the model completes its inference of the image, the output diagnosis is transmitted to the physician via a real-time feedback mechanism. To enhance the effectiveness of this feedback, the model not only provides a simple classification result but also a corresponding confidence score. Based on this feedback, physicians can quickly understand the image's diagnostic results and their credibility, enabling them to make further clinical decisions. To enhance physicians' understanding of diagnostic results and improve decision-making efficiency, the system also visualizes the recognition results and uses heatmaps to annotate key areas in the image with the predicted results.

[0057] To further improve diagnostic accuracy and physician decision support, the system will combine multimodal data (such as medical imaging, medical records, etc.) to generate a comprehensive feedback report. The comprehensive report includes the following:

[0058] Diagnostic results: The system's analysis results of medical images, including lesion type and location.

[0059] Confidence score: The confidence level of each diagnosis result.

[0060] Multimodal data: Combined with relevant medical records data to provide reference opinions.

[0061] The feedback report will be sent to doctors in real time through system interfaces (such as hospital information systems HIS) to help them make more accurate decisions.

[0062] Step S60: store the recognition results and related image data in the cloud to provide support for subsequent analysis.

[0063] In this step, the system stores the identified medical imaging diagnostic results and related imaging data for subsequent analysis, retrieval, and clinical decision support. Cloud storage not only ensures efficient data management and security but also enables data sharing and collaborative analysis across different medical institutions, providing valuable data support for future diagnoses.

[0064] In the previous steps, the system has completed real-time recognition of medical images and obtained diagnostic results. These recognition results, including the diagnosis category, confidence level, and heat map data for each case, need to be systematically stored. All diagnostic results and related imaging data will be encrypted and uploaded to the cloud storage system. Cloud storage has high availability and redundancy mechanisms to ensure the security and consistency of data when used across devices and platforms. The upload process includes the following steps:

[0065] Data encryption and transmission: All data are encrypted before transmission to ensure that sensitive information (such as patient privacy) is effectively protected.

[0066] Data storage: Data storage adopts a distributed storage architecture to ensure reliable storage and fast access to large-scale data.

[0067] The goal of cloud storage is to ensure the persistence and efficiency of data, while facilitating access, sharing, and analysis at any time. Data stored in the cloud can not only be queried by doctors, but can also be used for subsequent statistical analysis, model training, and performance optimization. For subsequent analysis, the system will retrieve stored case data based on specific query conditions (such as disease type, imaging characteristics, doctor feedback, etc.) and generate corresponding analysis reports. Through data analysis, the system can generate trend forecasts, clinical recommendations, and early diagnosis models for diseases based on big data. As the system continues to be used and improved, the data in the cloud storage will be continuously updated as new data is added. The system will continuously optimize existing models and analysis algorithms based on new diagnostic results and feedback, thereby continuously improving diagnostic accuracy and efficiency.

[0068] Example 2

[0069] like Figure 2 As shown, the second embodiment of the present application provides a medical image recognition system integrating deep learning and big data algorithms, including:

[0070] Data collection module 21: used to collect medical imaging data from multiple medical institutions;

[0071] The data acquisition module is the first step in the entire medical image recognition system. Its primary function is to collect medical imaging data from multiple medical institutions, ensuring that the acquired data is diverse and comprehensive. The basic function of the data acquisition module is to acquire imaging data from the medical institution's data system. This data includes, but is not limited to, X-rays, CT scans, MRIs, ultrasound images, and other types. Furthermore, the acquired imaging data includes not only the image itself but also image-related metadata, such as the time the image was taken, the device model, basic patient information (such as age, gender, and medical history), and image annotation information. This metadata is crucial for subsequent data processing and model training.

[0072] This system supports access to multiple data sources, ensuring that imaging data can be collected from different types of medical institutions and equipment, including hospital information systems (HIS), picture archiving and communication systems (PACS), and electronic health record systems (EHR). Medical imaging data comes in a variety of different formats, and the data acquisition module needs to be able to support image files in these different formats, including DICOM format (Digital Imaging and Communications in Medicine), JPEG, PNG and other common image formats. Some non-standard image files (such as the standard image format used by hospitals) also need to be recognized and processed by the system.

[0073] To ensure the efficiency and security of data when it arrives at the system from multiple medical institutions and equipment, the data acquisition module supports multiple data transmission protocols. Since medical imaging data usually involves patients' personal privacy information, the data acquisition module strictly abides by relevant laws and regulations to ensure the security and privacy protection of patient data. All data during data transmission and storage must be encrypted using a strong encryption algorithm (such as AES-256) to ensure the confidentiality of the data; only authorized personnel can access the data acquisition system, and the system should have strict user authentication and access control mechanisms; during the processing process, the patient's private information (such as name, ID number, etc.) should be desensitized to ensure that the data is not leaked.

[0074] The data acquisition module also supports two methods of automatic and manual acquisition: automatic acquisition, for routine imaging data, the system can automatically acquire the data according to the set rules. For example, by connecting with the PACS system, it can automatically obtain the latest imaging data from the hospital's imaging library; manual acquisition, for special cases or emergency situations, users can manually upload imaging data to the system. Manually uploaded imaging data also needs to undergo system security verification and format conversion.

[0075] To ensure data quality, the data acquisition module performs a preliminary quality check on the collected image data, eliminating images with serious problems such as blurry, damaged, or incomplete images. After data acquisition is completed, all image data will be stored in a local database or cloud storage system.

[0076] Data pre-processing module 22: used for standardizing image data;

[0077] The data preprocessing module is responsible for performing a series of operations to standardize medical imaging data. These operations primarily include image enhancement, denoising, and image normalization. The goal of these operations is to optimize the quality of the imaging data, making it suitable for training and prediction using deep learning models.

[0078] Image enhancement is the key to data preprocessing. It aims to enhance the key information in the image by transforming and modifying the image, making the image features more prominent and facilitating subsequent feature extraction and model training. To improve the visualization of the image, a comprehensive image enhancement function E is used to make the image details clearer by adjusting the contrast, brightness, and saturation. The enhancement function is defined as follows: Where I(x,y) represents the pixel value of the original image; α is the contrast adjustment parameter, controlling the enhancement strength; μ is the image mean; β is the brightness adjustment parameter; k is the range, used to calculate the local average; and γ is the saturation adjustment parameter. This is used to dynamically adjust different parameters and enhance the image based on its overall characteristics to ensure detailed visualization.

[0079] Medical imaging data is often subject to interference from noise, which can originate from the image acquisition device, the transmission process, or the external environment. Denoising can effectively reduce the impact of noise on image quality, thereby preventing the learning of irrelevant information during model training. To reduce noise in images, a denoising function D based on the wavelet transform is used. This function combines local features and global structure and is defined as follows: Where n represents the number of layers of wavelet transform; W j represents the jth layer wavelet coefficient, representing information at different frequencies; (x0, y0) represents the center coordinates of the image; and σ represents the parameter that controls the denoising strength; smaller values ​​result in stronger denoising effects. This formula effectively preserves the edge features of the image while denoising.

[0080] To ensure the uniformity of the image, the image pixel values ​​are normalized to between 0 and 1 using the following normalization formula: Where S represents the normalized pixel value; I(x,y) represents the pixel value of the original image; μ local Represents the mean value of the local area of ​​the image, which can improve the retention of local features; σglobal Represents the global standard deviation of the image, which is used to ensure the overall consistency of the data; ∈ is a small constant to prevent division by zero; ∑ i,j I(x+i,y+j) represents the sum of all pixels in the local window centered at pixel (x,y); N local Represents the number of local pixels, used to calculate the local mean.

[0081] Model training module 23: used to build and train deep learning models;

[0082] The model training module is the core component of the entire medical image recognition system. It is responsible for building and training deep learning models, extracting effective features from image data, and optimizing model parameters through large-scale data training.

[0083] In this module, a deep learning model is constructed using convolutional neural networks and combined with big data algorithms for adaptive feature extraction. Traditional convolutional neural networks extract image features through fixed convolution kernels and pooling operations. However, in practical applications, the diversity and complexity of image features require the model to automatically adapt and optimize the feature extraction process. This paper proposes an adaptive feature extraction mechanism that combines adaptive convolution kernels, dynamic feature selection, and a group learning strategy to improve the model's performance in medical image recognition. Specific adjustments are as follows:

[0084] Adaptive convolution kernel adjustment: In traditional convolutional neural networks, the size and shape of the convolution kernel are usually fixed. However, due to the large differences in features between different regions of medical images, a fixed convolution kernel often cannot fully capture all important information. Therefore, this module proposes an adaptive convolution kernel adjustment mechanism that enables the convolution kernel to automatically adjust its size and shape according to the local features of the input image. Specifically, the size and shape of the convolution kernel S L It is dynamically generated based on the local statistical characteristics of the image data. For example, image segmentation or cluster analysis can be used to determine the different regions and features of the image. The output F of the convolutional layer is L (x,y) is represented as: Among them, N L is the number of convolution kernels used in the Lth convolution layer; F L (x,y) represents the output feature map of the Lth layer; W L,i (I(x,y),S L ) represents the weight of the i-th convolution kernel in the L-th layer, which is based on the current image I(x,y) and the feature S of the specific layer L L Dynamically generated; S L Represents the shape and size of the convolution kernel, which is adjusted according to the local features of the image; F L-1(x,y) represents the output feature map of the previous layer. In this way, the convolution kernel can be adaptively adjusted to extract local features in different regions, enhancing the ability to capture image details.

[0085] Dynamic feature selection and weight update: In image data, not all features contribute equally to the classification task. Traditional convolutional neural networks extract features at each layer and process them uniformly. However, in a big data environment, the number and dimensions of features are often very large. Therefore, this module introduces a dynamic feature selection and weighting mechanism, which selects and updates features based on their classification contribution in a weighted manner at each layer. Each feature F i The weight α i It will be dynamically adjusted according to its contribution to the classification task. Specifically, the final optimized feature can be expressed as: Among them, M is the total number of features extracted; F optimized is the final feature representation after weighting; F i represents the i-th extracted feature; α i Represents the weight coefficient of the i-th feature, which is dynamically adjusted according to its importance in the training process. The weight α i The update is achieved through the back-propagation algorithm, which optimizes each feature based on its impact on model performance. This mechanism enables the model to automatically learn which features are more important for the classification task, thereby improving the accuracy of the model.

[0086] Group learning feature fusion: In medical image recognition, image data usually has a complex structure, and a single feature extraction strategy may not be able to fully capture all the information. Therefore, this module combines the group learning strategy in big data algorithms, using distributed computing and cluster learning to extract features from multiple sub-networks and fuse these features into a unified feature representation. Assume that the network is divided into multiple sub-networks N1, N2, ..., N k , each sub-network independently extracts features from different subsets of the data. Each sub-network N j The extracted features are F j , the final fused feature F final It can be expressed as: Where K represents the number of subnetworks; F final represents the final fused feature representation; F j represents the features extracted by the jth sub-network; β j Represents the weighted coefficient of the jth sub-network, indicating the importance of the sub-network in the final feature fusion. These weighted coefficients β jIt is dynamically updated through group learning, ensuring that the final fused features effectively represent the common characteristics of all image data while reducing possible redundant information. Through adaptive convolution kernels, dynamic feature selection, and group learning fusion, this model can continuously optimize the feature extraction process based on the characteristics of the image data.

[0087] Furthermore, an ensemble learning method is used to train the deep learning model and optimize the model parameters. The convolutional neural network is initialized and trained using the training set. During the training process, the gradient descent method is used to optimize the loss function and gradually update it in combination with the adaptive learning rate. Assume that the loss function is The model parameter is θ, and the parameter is optimized by gradient descent. The formula is as follows: Among them, θ t is the model parameter at the current moment; τ t is the learning rate at the current moment, which is dynamically adjusted according to the training progress; is the gradient of the loss function with respect to the parameters. In order to effectively avoid the local minimum problem, stochastic gradient descent is used in this module, which randomly extracts part of the data for gradient calculation at each update, thereby accelerating the training process and avoiding falling into the local minimum. In order to accelerate convergence and improve training results, the learning rate τ t It will be adaptively adjusted with the number of training rounds t. The following adaptive learning rate adjustment formula is used to ensure that the model is optimized with different learning rates at different stages of training: Where τ0 is the initial learning rate; γ is the decay factor that controls the decay rate of the learning rate; and t is the current training round. The purpose of this formula is to gradually reduce the learning rate as training progresses, ensuring that the model can make more precise adjustments when approaching the optimal solution and avoid over-adjustment. To avoid overtraining of the model, a dynamic early stopping mechanism is introduced, which determines when to stop training by calculating the gain function during the training process. The gain function is defined as: in, and are the loss function values ​​of the current and previous rounds of training, G t It measures the rate of decrease of the loss function during the training process. If the value of the gain function is less than a preset threshold in N consecutive rounds of training, the training is stopped to prevent overfitting.

[0088] After model training and optimization are completed, in order to further optimize the model performance, multiple optimization results are fused through weighted averaging to obtain the final prediction output. The fusion formula is: in, is the final prediction result; K represents the number of sub-models involved in the fusion; is the predicted value of the i-th model; αi The weights of the model are dynamically adjusted based on the training performance of the model; is the average of all predicted values; β is the adjustment factor to control the impact of prediction deviation.

[0089] Real-time recognition module 24: used for real-time recognition of input medical images;

[0090] In this module, input medical images are processed and recognized in real time, generating corresponding diagnostic results. These results are then promptly fed back to doctors through a feedback mechanism. This process ensures that medical decisions can be made in the shortest possible time and provides effective support to improve the efficiency and accuracy of diagnosis and treatment.

[0091] When medical images are input into the system, they are first preprocessed to ensure that the input data meets the model's requirements. After preprocessing, the images are fed into a trained deep learning model for real-time feature extraction and prediction. Based on a convolutional neural network architecture, the model extracts key features from the images through multiple convolutional, pooling, and fully connected layers. Once the model completes its inference of the image, the output diagnosis is transmitted to the physician via a real-time feedback mechanism. To enhance the effectiveness of this feedback, the model not only provides a simple classification result but also a corresponding confidence score. Based on this feedback, physicians can quickly understand the image's diagnostic results and their credibility, enabling them to make further clinical decisions. To enhance physicians' understanding of diagnostic results and improve decision-making efficiency, the system also visualizes the recognition results and uses heatmaps to annotate key areas in the image with the predicted results.

[0092] To further improve diagnostic accuracy and physician decision support, the system will combine multimodal data (such as medical imaging, medical records, etc.) to generate a comprehensive feedback report. The comprehensive report includes the following:

[0093] Diagnostic results: The system's analysis results of medical images, including lesion type and location.

[0094] Confidence score: The confidence level of each diagnosis result.

[0095] Multimodal data: Combined with relevant medical records data to provide reference opinions.

[0096] The feedback report will be sent to doctors in real time through system interfaces (such as hospital information systems HIS) to help them make more accurate decisions.

[0097] Data storage module 25: used to store recognition results and image data in the cloud.

[0098] In this module, the system stores identified medical imaging diagnostic results and related imaging data for subsequent analysis, retrieval, and clinical decision support. Cloud storage not only ensures efficient data management and security but also enables data sharing and collaborative analysis across different medical institutions, providing valuable data support for future diagnoses.

[0099] The goal of cloud storage is to ensure the persistence and efficiency of data, while facilitating access, sharing, and analysis at any time. Data stored in the cloud can not only be queried by doctors, but can also be used for subsequent statistical analysis, model training, and performance optimization. For subsequent analysis, the system will retrieve stored case data based on specific query conditions (such as disease type, imaging characteristics, doctor feedback, etc.) and generate corresponding analysis reports. Through data analysis, the system can generate trend forecasts, clinical recommendations, and early diagnosis models for diseases based on big data. As the system is continuously used and improved, the data in the cloud storage will be continuously updated as new data is added. The system will continuously optimize existing models and analysis algorithms based on new diagnostic results and feedback, thereby continuously improving diagnostic accuracy and efficiency.

[0100] Corresponding to the above embodiment, an embodiment of the present invention provides a computer storage medium, comprising: at least one memory and at least one processor;

[0101] The memory is used to store one or more program instructions;

[0102] A processor is used to run one or more program instructions to execute a medical image recognition method that integrates deep learning and big data algorithms.

[0103] Corresponding to the above embodiments, an embodiment of the present invention provides a computer-readable storage medium, which contains one or more program instructions, and the one or more program instructions are used by a processor to execute a medical image recognition method that integrates deep learning and big data algorithms.

[0104] The embodiments disclosed in the present invention provide a computer-readable storage medium, which stores computer program instructions. When the computer program instructions are executed on a computer, the computer executes the above-mentioned medical image recognition method that integrates deep learning and big data algorithms.

[0105] In the embodiments of the present invention, the processor may be an integrated circuit chip having signal processing capabilities. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0106] The methods, steps, and logic diagrams disclosed in the embodiments of the present invention can be implemented or executed. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules within the decoding processor. The software modules can be located in a storage medium well-established in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. The processor reads the information from the storage medium and, in conjunction with its hardware, completes the steps of the aforementioned methods.

[0107] The storage medium may be a memory and may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0108] Among them, the non-volatile memory can be 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), or a flash memory.

[0109] Volatile memory may be random access memory (RAM), which is used as an external cache memory. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM).

[0110] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0111] Those skilled in the art will appreciate that in one or more of the above examples, the functions described herein can be implemented using a combination of hardware and software. When software is used, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media includes any medium that facilitates the transmission of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0112] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A medical image recognition method integrating deep learning and big data algorithms, characterized in that: The following steps are involved: Collect medical imaging data from multiple medical institutions and build a large-scale medical imaging database; Perform standardized processing on collected medical images, including image enhancement, denoising, and resizing; Based on the processed data, a deep learning model is constructed using convolutional neural networks and combined with big data algorithms for feature extraction; Use ensemble learning methods to train deep learning models and optimize model parameters; After the model training is completed, the input medical images are recognized in real time and the recognition results are fed back to the doctor; Store the recognition results and related image data in the cloud to support subsequent analysis; A deep learning model is built based on a convolutional neural network and combined with a big data algorithm for adaptive feature extraction. This combines adaptive convolution kernels, dynamic feature selection, and group learning strategies to improve the model's performance in medical image recognition. The specific steps are as follows: Adaptive convolution kernel adjustment: enables the convolution kernel to automatically adjust its size and shape according to the local features of the input image; specifically, the size and shape of the convolution kernel It is dynamically generated based on the local statistical characteristics of the image data, and the different areas and features of the image are determined through image segmentation or cluster analysis; the output of the convolution layer is represented as: ,in, is the number of convolution kernels used in the Lth convolution layer; Represents the output feature map of the Lth layer; Represents the weight of the i-th convolution kernel in the L-th layer, based on the current image and the features of a specific layer L Dynamic generation; Indicates the shape and size of the convolution kernel, which is adjusted according to the local features of the image; Represents the output feature map of the previous layer; Dynamic feature selection and weight update: Introduce a dynamic feature selection and weighting mechanism, select and update the features according to their classification contribution in a weighted manner at each layer; each feature Weight It will be dynamically adjusted according to its contribution to the classification task; the final optimized feature is expressed as: , where M represents the total number of features extracted; is the final feature representation after weighting; represents the i-th extracted feature; Represents the weight coefficient of the i-th feature, which is dynamically adjusted according to its importance in the training process; weight The update is achieved through the back-propagation algorithm, which optimizes the performance of each feature according to its impact on the model; Group learning feature fusion: Combining the group learning strategy in big data algorithms, using distributed computing and cluster learning to extract features from multiple sub-networks and fuse these features into a unified feature representation; assuming that the network is divided into multiple sub-networks Each sub-network independently extracts features from a different subset of the data. The extracted features are , the final fused features Expressed as: ,in, represents the final fused feature representation; K represents the number of sub-networks; Represents the features extracted by the j-th sub-network; Represents the weighted coefficient of the jth sub-network, indicating the importance of the sub-network in the final feature fusion; weighted coefficient Dynamic updates are made through group learning to ensure that the final fused features can effectively represent the common features of all image data.

2. The medical image recognition method integrating deep learning and big data algorithms according to claim 1, characterized in that: The real-time recognition also includes providing the confidence value of the recognition result to the doctor to assist in clinical judgment.

3. The medical image recognition method integrating deep learning and big data algorithms according to claim 1, characterized in that: The system can generate recognition reports for different disease categories and analytical charts of model performance to help doctors understand the model's recognition capabilities and its effectiveness in practical applications.

4. A medical image recognition system integrating deep learning and big data algorithms, which executes a medical image recognition method integrating deep learning and big data algorithms as described in any one of claims 1 to 3, characterized in that: The system comprises: A data acquisition module for collecting medical imaging data from multiple medical institutions; Data preprocessing module, used to standardize image data; Model training module, used to build and train deep learning models; Real-time recognition module, used for real-time recognition of input medical images; The data storage module is used to store recognition results and image data in the cloud.

5. A computer storage medium comprising: at least one memory and at least one processor; The memory is used to store one or more program instructions; A processor for running one or more program instructions to execute a medical image recognition method integrating deep learning and big data algorithms as described in any one of claims 1 to 3.

Images