Tumor recognition detection method, device and equipment based on large model and medium
Through the tumor recognition detection method based on large models, image data augmentation technology is used to generate synthetic tumor images, expand the data set and train the model, which solves the problem of scarcity of data in tumor detection and improves the accuracy and efficiency of detection.
Patent Information
- Application Number
- CN202510223638.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-27
- Publication Date
- 2025-06-06
AI Technical Summary
The prior art is limited by the scarcity of tumor image data in tumor detection and recognition, resulting in insufficient generalization ability and accuracy of deep learning models.
Using a tumor recognition detection method based on a large model, the multimodal tumor image data is obtained, and the image data is enhanced by using the target big model to generate synthetic tumor images, expand the data set, and the target tumor recognition detection model is trained through this expanded data set.
It effectively expands the tumor image data set, improves the diversity of data and the generalization ability of models, significantly improves the accuracy and efficiency of tumor detection and identification, and solves the problem of data scarcity.
Smart Images
Figure CN120107722A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical image analysis, and in particular to a tumor recognition and detection method, device, equipment and medium based on a large model. Background Art
[0002] Accurate tumor detection and identification are crucial for the diagnosis and treatment of diseases. Traditional tumor detection and identification methods mainly rely on manual film reading, which is inefficient, easily affected by subjective factors, and difficult to achieve accurate quantitative analysis of tumors. In recent years, with the rapid development of artificial intelligence technology, deep learning has played an increasingly important role in medical image analysis. In the field of medical image analysis, especially in tumor detection and identification tasks, the quality and quantity of image data have a decisive impact on the performance of the model. However, due to privacy protection, ethical review and difficulty in obtaining, high-quality tumor image data is often scarce. The lack of sufficient annotated data has greatly limited the training and performance improvement of deep learning models. In addition, the morphological diversity and pathological complexity of tumors also pose great challenges to automatic recognition algorithms. Different tumors may have similar imaging features, while the same tumor may show completely different appearances in different patients or under different scanning conditions.
[0003] It can be seen that due to the scarcity of tumor image data, traditional training methods are difficult to achieve ideal results, resulting in deficiencies in the generalization ability and accuracy of the model. Therefore, how to provide a solution to the above technical problems is a problem that technicians in this field currently need to solve. Summary of the invention
[0004] In view of this, the purpose of the present invention is to provide a tumor recognition and detection method, device, equipment and medium based on a large model, which can generate high-quality tumor image data, effectively expand the data set, provide sufficient data support for model training, and improve the accuracy of tumor detection and recognition to meet the needs of clinical applications. The specific scheme is as follows:
[0005] In a first aspect, the present application discloses a tumor recognition and detection method based on a large model, comprising:
[0006] Acquire multimodal tumor image data;
[0007] Determining a target macromodel for medical image processing, and performing image data enhancement processing on the multimodal medical image data using the target macromodel to obtain a plurality of synthetic tumor images;
[0008] Generate a target data set using the multimodal tumor image data and the synthetic tumor image, and train a target tumor recognition and detection model using the target data set;
[0009] The trained target tumor recognition and detection model is integrated into the hospital information system to provide real-time tumor recognition and detection results.
[0010] Optionally, after acquiring the multimodal tumor image data, the method further includes:
[0011] Preprocessing the multimodal tumor image data; the preprocessing process includes: data standardization, denoising, normalization, segmentation, contrast adjustment and image registration;
[0012] Accordingly, the multimodal medical image data is subjected to image data enhancement processing using the target large model to obtain a plurality of synthetic tumor images, including:
[0013] The target large model is used to perform image data enhancement processing on the preprocessed multimodal medical image data to obtain a plurality of synthetic tumor images.
[0014] Optionally, determining a target macromodel for medical image processing, and performing image data enhancement processing on the multimodal medical image data using the target macromodel to obtain a plurality of synthetic tumor images includes:
[0015] Determine an industry-wide large model for medical image processing, and use transfer learning technology to fine-tune the industry-wide large model to learn feature representations of tumor images to obtain the target large model;
[0016] The data enhancement pipeline is implemented through the data enhancement tool in the deep learning framework, and the target large model is used to perform image data enhancement processing on each batch of the multimodal medical image data in real time based on the data enhancement pipeline to obtain multiple synthetic tumor images.
[0017] Optionally, the generating a target data set using the multimodal tumor image data and the synthetic tumor image, and training a target tumor recognition and detection model using the target data set, includes:
[0018] Merging the multimodal tumor image data and the synthetic tumor image to generate a target data set;
[0019] The target data set is balanced by using an oversampling technique or an undersampling technique, and a target tumor recognition and detection model is trained by using the balanced target data set.
[0020] Optionally, the using the target data set to train a target tumor recognition and detection model includes:
[0021] Dividing the target data set into a training set, a validation set, and a test set;
[0022] Selecting a target loss function and a target optimization algorithm required for training the target tumor recognition and detection model, and determining a strategy to prevent overfitting when obtaining the target tumor recognition and detection model;
[0023] The initial tumor recognition and detection model is trained using the training set according to the target loss function and the target optimization algorithm, and the learning rate of the initial tumor recognition and detection model is dynamically adjusted using the validation set based on the overfitting prevention strategy to obtain the target tumor recognition and detection model.
[0024] Optionally, after using the target data set to train the target tumor recognition and detection model, the method further includes:
[0025] A cross-validation method is adopted to evaluate the performance of the target tumor recognition detection model using at least one evaluation indicator on the test set.
[0026] Optionally, the process of integrating the trained target tumor recognition and detection model into a hospital information system to provide real-time tumor recognition and detection results further includes:
[0027] The target tumor recognition and detection model is updated online through online learning, and a user feedback mechanism is established to collect user feedback information;
[0028] The feedback information is used to optimize the performance of the target tumor recognition and detection model.
[0029] In a second aspect, the present application discloses a tumor recognition and detection device based on a large model, comprising:
[0030] A data acquisition module, used for acquiring multimodal tumor image data;
[0031] A data enhancement module, used to determine a target macromodel for medical image processing, and perform image data enhancement processing on the multimodal medical image data using the target macromodel to obtain a plurality of synthetic tumor images;
[0032] A model training module, used to generate a target data set using the multimodal tumor image data and the synthetic tumor image, and train a target tumor recognition and detection model using the target data set;
[0033] The model integration module is used to integrate the trained target tumor recognition and detection model into the hospital information system to provide real-time tumor recognition and detection results.
[0034] In a third aspect, the present application discloses an electronic device, comprising a processor and a memory; wherein the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the large model-based tumor recognition and detection method as described above.
[0035] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the large model-based tumor recognition and detection method as described above.
[0036] The present application provides a tumor recognition and detection method based on a large model, comprising: acquiring multimodal tumor image data; determining a target large model for medical image processing, and using the target large model to perform image data enhancement processing on the multimodal medical image data to obtain multiple synthetic tumor images; generating a target data set using the multimodal tumor image data and the synthetic tumor images, and training a target tumor recognition and detection model using the target data set; integrating the trained target tumor recognition and detection model into a hospital information system to provide real-time tumor recognition and detection results.
[0037] The beneficial technical effects of this application are:
[0038] 1. Improve data diversity: This invention effectively expands the data set of tumor images by using synthetic tumor images generated by industry large models, increases data diversity, covers a wider range of tumor morphology and pathological characteristics, and solves the problem of data scarcity;
[0039] 2. Improve model performance: Due to the expansion and diversification of the data set, the tumor detection and recognition model trained based on this method shows stronger generalization ability, can stably detect and identify tumors in different clinical environments, and reduce overfitting of specific data sets;
[0040] 3. Improve model training efficiency: The image generation process significantly reduces the time for data preparation, allowing researchers and medical experts to obtain the data sets required for training more quickly, speeding up model development and iteration;
[0041] 4. Improve diagnostic accuracy: The introduction of synthetic tumor images provides the model with more subtle and difficult-to-observe tumor features, enabling the model to more accurately identify the edges and internal structures of the tumor, thereby improving the accuracy of clinical diagnosis;
[0042] 5. Enhance model interpretability: Through images generated by industry-leading models, researchers can gain a deeper understanding of how the models learn and identify tumor features, thereby improving the interpretability of the models and helping medical experts better trust and adopt AI technology.
[0043] In addition, the present application provides a large model-based tumor recognition and detection device, equipment and storage medium, which correspond to the above-mentioned large model-based tumor recognition and detection method, and have the same effect as above. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0045] Figure 1 A flow chart of a tumor recognition and detection method based on a large model disclosed in this application;
[0046] Figure 2 A flowchart of a specific tumor identification and detection method disclosed in this application;
[0047] Figure 3 This is a schematic diagram of the structure of a tumor recognition and detection device based on a large model disclosed in this application;
[0048] Figure 4 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION
[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0050] Currently, with the rise of deep learning technology, especially the breakthrough progress of Convolutional Neural Networks (CNN) in the field of image recognition, medical image analysis has begun to try to use these advanced algorithms to improve the accuracy of tumor detection and recognition. Image detection and recognition models, with their excellent feature extraction and generalization capabilities, occupy an important position in this field. The reason why these models stand out is that they have made breakthrough progress in deep representation learning of images and complex pattern recognition, and have demonstrated unprecedented accuracy and efficiency in the classification and location of benign and malignant tumors, as well as in the quantitative analysis of pathological characteristics.
[0051] However, despite the great success of these technologies in image recognition, existing models still face some challenges in practical applications, especially in medical scenarios that require precise positioning and identification of tumors. Specifically, due to the scarcity of high-quality tumor image data, deep learning models often find it difficult to obtain sufficient training samples, resulting in deficiencies in the generalization ability and accuracy of the models. In addition, the morphological diversity and pathological complexity of tumors also pose great challenges to automatic recognition algorithms. Different tumors may have similar imaging features, while the same tumor may show completely different appearances in different patients or under different scanning conditions, all of which increase the difficulty of tumor detection and identification.
[0052] Image enhancement can play a role in expanding data. Although traditional image enhancement techniques, such as rotation, flipping, and scaling, can increase the diversity and quantity of training data to a certain extent by performing various transformations on the original images, they cannot create new tumor samples and cannot fundamentally solve the problem of insufficient data. Generative adversarial nets (GAN), as a powerful image generation tool, are used to generate realistic images to expand training data sets.
[0053] However, existing GAN-based methods often require a lot of computing resources and have high requirements for network structure and parameter adjustment. In addition, the generated images need to be consistent with the real data distribution, otherwise additional noise may be introduced, affecting the generalization ability of the model.
[0054] To overcome these challenges, this application provides a large model-based tumor recognition and detection solution that can solve the problem of poor training effects of deep learning models caused by the scarcity of tumor image data in the medical industry, improve the accuracy of tumor detection and recognition, and meet the needs of clinical applications.
[0055] The present invention discloses a tumor recognition and detection method based on a large model. Figure 1 As shown, the method includes:
[0056] Step S11: Acquire multimodal tumor image data.
[0057] First, tumor image data from different sources are collected from the medical field to obtain multimodal tumor image data. Multimodal tumor image data refers to tumor-related data that combines multiple imaging technologies, such as CT (Computed tomography), MRI (Magic Resonance Imaging), PET (Positron Emission Tomography) and other scanning images and different data sources (such as pathology, genomics, clinical data, etc.). By integrating information from multiple modalities, these data can more comprehensively describe the biological characteristics, tissue structure and metabolic activities of tumors, thereby improving the accuracy of diagnosis, treatment and prognosis assessment.
[0058] In the embodiment of the present application, after the multimodal tumor image data is obtained, these data need to undergo strict preprocessing. Specifically, the preprocessing process includes: data standardization, denoising, normalization, segmentation, contrast adjustment and image registration. Standardization can ensure the quality and consistency of the data; preprocessing techniques such as denoising, normalization and segmentation can improve the availability of the data; contrast adjustment and image registration can enhance the recognizability of tumor features in the image. Through these steps, the diversity, consistency and representativeness of the data set are ensured, the image quality can be improved, and the alignment between multimodal images is ensured, providing a high-quality data foundation for subsequent model training.
[0059] Step S12: determining a target large model for medical image processing, and using the target large model to perform image data enhancement processing on the multimodal medical image data to obtain a plurality of synthetic tumor images.
[0060] In actual clinical applications, there are many limitations in obtaining a large amount of high-quality tumor image data for training and verifying deep learning models. These limitations include but are not limited to patient privacy protection, data acquisition costs, ethical review, and the complexity and diversity of tumor images themselves. Due to these factors, current models often perform poorly in terms of generalization ability, accuracy, and robustness. In the face of these challenges, the application of data enhancement technology has become the key to improving the performance of small sample image detection and recognition models. In the embodiment of the present application, the industry large model is used for image data enhancement to improve the accuracy and efficiency of tumor detection and recognition.
[0061] It should be pointed out that the target large model is a large model obtained after a series of processing of the industry large model. The industry large model in the present invention refers to a large model that is deeply customized and optimized for medical image processing. The model is usually further trained and adjusted based on the general large model to better adapt to and understand the language, knowledge, data and other characteristics of a specific industry, so as to more effectively perform more professional tasks and be more adaptable to specific medical image data. Through advanced image generation technology, combined with deep feature learning of large models, realistic and clinically valuable synthetic tumor images are created. Synthetic tumor images are highly similar to real data, which can solve the data scarcity problem of tumor detection and identification in medical image analysis, improve the performance of the model, and significantly improve the accuracy of tumor detection and identification, providing clinicians with more reliable auxiliary diagnostic tools and supporting more accurate clinical decisions.
[0062] It is understandable that, since the acquired multimodal medical image data has undergone strict preprocessing, when image data enhancement processing is performed based on the target large model, multiple synthetic tumor images are generated using the preprocessed multimodal medical image data.
[0063] Step S13: Generate a target data set using the multimodal tumor image data and the synthetic tumor image, and train a target tumor recognition and detection model using the target data set.
[0064] In the embodiment of the present application, after generating high-quality synthetic tumor image data, the original multimodal tumor image data is combined to form a complete target data set. The target data set is used to train the target tumor recognition and detection model.
[0065] It should be noted that the target data set needs to be balanced to deal with the class imbalance problem and improve the model's recognition ability for minority classes. Specifically, the target data set is balanced using oversampling technology or undersampling technology to ensure the model's recognition ability for various types of tumors. The target tumor recognition and detection model is then trained using the balanced target data set.
[0066] It can be seen that the generated synthetic tumor images can increase the diversity of data, create new tumor samples, expand the original data set, and fundamentally solve the problem of insufficient data. When the target data set is obtained by combining the original multimodal tumor image data and the generated synthetic tumor images, the target tumor recognition detection model trained with the target data set can significantly improve the accuracy of tumor detection and recognition, and effectively solve the problem of poor training effect of deep learning models in the medical industry due to the scarcity of tumor image data.
[0067] Step S14: Integrate the trained target tumor recognition and detection model into the hospital information system to provide real-time tumor recognition and detection results.
[0068] In the embodiment of the present application, the trained target tumor recognition and detection model is integrated into the clinical workflow, and a seamless connection with the hospital information system is achieved by developing an API (Application Program Interface). System integration integrates the model into the existing medical information system to achieve seamless data flow and workflow. It provides real-time tumor detection and recognition results, supports doctors in early diagnosis and decision-making of tumors, and assists doctors in rapid diagnosis.
[0069] The present application provides a tumor recognition and detection method based on a large model, comprising: acquiring multimodal tumor image data; determining a target large model for medical image processing, and using the target large model to perform image data enhancement processing on the multimodal medical image data to obtain multiple synthetic tumor images; generating a target data set using the multimodal tumor image data and the synthetic tumor images, and training a target tumor recognition and detection model using the target data set; integrating the trained target tumor recognition and detection model into a hospital information system to provide real-time tumor recognition and detection results.
[0070] The beneficial technical effects of this application are:
[0071] 1. Improve data diversity: This invention effectively expands the data set of tumor images by using synthetic tumor images generated by industry large models, increases data diversity, covers a wider range of tumor morphology and pathological characteristics, and solves the problem of data scarcity;
[0072] 2. Improve model performance: Due to the expansion and diversification of the data set, the tumor detection and recognition model trained based on this method shows stronger generalization ability, can stably detect and identify tumors in different clinical environments, and reduce overfitting of specific data sets;
[0073] 3. Improve model training efficiency: The image generation process significantly reduces the time for data preparation, allowing researchers and medical experts to obtain the data sets required for training more quickly, speeding up model development and iteration;
[0074] 4. Improve diagnostic accuracy: The introduction of synthetic tumor images provides the model with more subtle and difficult-to-observe tumor features, enabling the model to more accurately identify the edges and internal structures of the tumor, thereby improving the accuracy of clinical diagnosis;
[0075] 5. Enhance model interpretability: Through images generated by industry-leading models, researchers can gain a deeper understanding of how the models learn and identify tumor features, thereby improving the interpretability of the models and helping medical experts better trust and adopt AI technology.
[0076] In a specific implementation, data enhancement is used in the field of image detection and recognition to solve the problem of data scarcity, so as to improve the model's ability to recognize the diversity and complexity of object morphology in images. Data enhancement provides the model with richer training samples by simulating different image conditions and morphologies, thereby effectively alleviating the problem of scarcity of high-quality data. Through data enhancement, the model can learn more features and patterns, enhancing its ability to understand the diversity and complexity of data.
[0077] When expanding a data set through data enhancement, first of all, you need to select an appropriate enhancement strategy based on the type of data. For image data, traditional enhancement methods include rotation, flipping, cropping, scaling, color adjustment, etc. There are also methods that use generative models to synthesize images to expand data. Choosing an appropriate strategy is the basis of data enhancement, ensuring that the enhanced data can effectively improve the diversity and robustness of the model. In an embodiment of the present application, an industry large model is used to enhance image data to improve the accuracy and efficiency of tumor detection and identification. In the context of industry large models, using their powerful feature extraction and learning capabilities, combined with tumor imaging data for fine-tuning, has become an effective way to improve model performance. Specifically, the use of large models for image data enhancement specifically includes the following steps:
[0078] Determine an industry-wide large model for medical image processing, and use transfer learning technology to fine-tune the industry-wide large model to learn feature representations of tumor images to obtain the target large model;
[0079] The data enhancement pipeline is implemented through the data enhancement tool in the deep learning framework, and the target large model is used to perform image data enhancement processing on each batch of the multimodal medical image data in real time based on the data enhancement pipeline to obtain multiple synthetic tumor images.
[0080] Once the enhancement strategy is selected, the specific parameter range of each operation needs to be determined. For example, based on the traditional enhancement method, the rotation operation may need to specify the range of rotation angles, and the cropping operation needs to set the cropping ratio. In addition, the application probability of each operation needs to be determined to control the strength and frequency of the enhancement, so as to strike a balance between diversity and data authenticity.
[0081] In the embodiment of the present application, a large industry model suitable for medical image processing and excellent in image recognition tasks, such as LLM (Large Language Model), is selected. A customized model for medical images can also be used, which is not limited here. Then, the large industry model is fine-tuned to adapt to specific data characteristics. By utilizing transfer learning technology, the model weights pre-trained on a large-scale data set are migrated to a specific tumor detection task, and the high-level features of the model are fine-tuned through a small amount of labeled data to learn the feature representation of tumor images.
[0082] It is understandable that in a deep learning model, the high-level features of the model refer to the features with higher abstraction levels and semantic information extracted by the model in the deeper network layers. In this way, the model is fine-tuned and trained, that is, through transfer learning technology, it can reduce the dependence on large-scale annotated data, ensure the authenticity of the synthesized images, and improve the adaptability of the model in specific fields.
[0083] Furthermore, the data enhancement pipeline is implemented in the data enhancement process: the data enhancement pipeline is implemented using the data enhancement tools in the deep learning framework, such as the predefined enhancement functions in TensorFlow or PyTorch (open source frameworks for machine learning and deep learning); custom enhancement logic can also be written according to needs. Implementing an efficient enhancement pipeline helps to dynamically process data during training and improve efficiency. Next, apply data enhancement: using the target large model, based on the data enhancement pipeline, each batch of data is enhanced in real time to avoid storing a large amount of enhanced data. In other words, by applying data enhancement technology to the input data in real time during the data preparation stage of each batch, instead of enhancing and storing the entire data set in advance. This real-time enhancement method can ensure data diversity while avoiding large amounts of storage, and the enhancement strategy can be adjusted dynamically as needed. It should be pointed out that it is necessary to ensure that the enhanced data labels remain correct to avoid affecting the accuracy of model training. In addition, during this process, the model effect will be verified: the model is trained using the enhanced data set and compared with the unenhanced data set. The impact of data enhancement on model performance is evaluated by monitoring training and validation losses.
[0084] Finally, the fine-tuned target large model is used to enhance the image data and generate multiple synthetic tumor images, ensuring that there is no significant visual difference from the real data and avoiding the introduction of noise. The generated synthetic tumor images are used to expand the original dataset and increase the sample diversity of model training.
[0085] It can be seen that high-quality tumor image data can be generated by enhancing image data through large models. Compared with traditional image enhancement technology, the industry large model used in this method can more deeply understand the morphological characteristics and pathological complexity of tumors, generate more realistic and clinically valuable image data, and effectively expand the data set. In this embodiment, through data enhancement, combined with the industry large model, sufficient data support is provided for model training, which significantly improves the recognition ability and recognition accuracy of tumor images, not only improves the efficiency of tumor detection, but also reduces the subjective errors caused by manual film reading, and provides doctors with a more reliable basis for diagnosis, which is of great significance to improving the accuracy of early diagnosis of tumors, meets the needs of clinical applications, and helps promote the development of medical image detection and recognition technology.
[0086] In a specific implementation, the process of training and optimizing, validating and evaluating the model using the synthetic target data set may include the following steps:
[0087] Dividing the target data set into a training set, a validation set, and a test set;
[0088] Selecting a target loss function and a target optimization algorithm required for training the target tumor recognition and detection model, and determining a strategy to prevent overfitting when obtaining the target tumor recognition and detection model;
[0089] Using the training set, training the initial tumor recognition and detection model according to the target loss function and the target optimization algorithm, and using the validation set, dynamically adjusting the learning rate of the initial tumor recognition and detection model based on the overfitting prevention strategy to obtain the target tumor recognition and detection model;
[0090] A cross-validation method is adopted to evaluate the performance of the target tumor recognition detection model using at least one evaluation indicator on the test set.
[0091] In the embodiment of the present application, the data is divided into a training set, a validation set, and a test set during model training. The enhanced training set is used to train the initial tumor recognition and detection model, and a variety of techniques are used to optimize the model performance. For example, a suitable loss function, such as cross entropy loss or Focal Loss, is selected to solve the problem of class imbalance; the model weights are adjusted using optimization algorithms such as Adam or SGD (Stochastic Gradient Descent). Furthermore, after the model is trained with the training set, the model is verified with the validation set. Since the best model has not been selected at this time, the model is continuously adjusted according to the situation, and the best model is selected as the final target tumor recognition and detection model. In this process, techniques such as Dropout and L2 regularization are applied to prevent overfitting, and the learning rate is dynamically adjusted according to the progress of model training to speed up convergence. At the same time, an early stopping strategy is used to monitor the performance on the validation set to avoid overfitting.
[0092] Furthermore, the accuracy and reliability of the model are evaluated on an independent test set. In order to more accurately evaluate the performance of the model, a cross-validation method can be used to divide the training data into multiple subsets, which are used as validation sets or test sets in turn to reduce bias. Then, the performance of the target tumor recognition detection model is evaluated using at least one evaluation metric on the test set. Exemplarily, indicators such as sensitivity, specificity, accuracy, F1 score, and AUC (Area Under ROC Curve) can be used to comprehensively evaluate the performance of the model to ensure its reliability and effectiveness in clinical applications. Analyze the model prediction error and identify the model performance bottleneck.
[0093] It can be seen that in this embodiment, the enhanced data set is used to train the target tumor recognition detection model, and adaptive learning rate adjustment, early stopping strategy and cross-validation technology are used to improve the convergence speed and stability of the model and avoid overfitting. The model validation and evaluation steps include comprehensive evaluation of multiple indicators such as sensitivity, specificity and AUC to ensure the reliability of the model in different clinical scenarios.
[0094] In a feasible implementation, the target tumor recognition and detection model supports real-time data updates and model feedback, has continuous learning and updating capabilities, provides timely diagnostic support for doctors, and adapts to changing clinical needs. Specifically, the process of integrating the trained target tumor recognition and detection model into the hospital information system to provide real-time tumor recognition and detection results also includes the following steps:
[0095] The target tumor recognition and detection model is updated online through online learning, and a user feedback mechanism is established to collect user feedback information;
[0096] The feedback information is used to optimize the performance of the target tumor recognition and detection model.
[0097] In the embodiments of this application, online learning enables the model to be updated online according to new data, adaptively adjust model parameters according to different tumor types and patient characteristics, continuously optimize performance, improve personalized diagnostic capabilities, and enhance the flexibility of clinical applications. At the same time, a feedback loop from clinical application to model training is established, and feedback from doctors and technicians is collected through a user feedback mechanism to continuously optimize model performance and achieve iterative improvement of the model.
[0098] It can be seen that the target tumor recognition and detection model in this embodiment has the ability to continuously learn and update, and through the user feedback mechanism, it continuously optimizes data enhancement and model training strategies to maintain the advancement and competitiveness of the technology.
[0099] like Figure 2 The figure shows a specific schematic diagram of the tumor recognition and detection process provided based on the content in the above-mentioned embodiment. First, tumor image data from different sources, including multimodal images such as CT, MRI, and PET, are collected and standardized to ensure the quality and consistency of the data. This step also involves preprocessing techniques such as denoising, normalization, and segmentation to improve the availability of the data. Then, a large industry model suitable for tumor detection is selected and fine-tuned through transfer learning technology to adapt to specific medical image data. Further, the fine-tuned large industry model is used to generate high-quality synthetic tumor images, expand the original data set, and ensure that the generated image is visually not significantly different from the real data. The original data and the generated data are combined to form a complete data set. Secondly, the tumor detection and recognition model is trained using the enhanced data set, and adaptive learning rate adjustment, early stopping strategy, and cross-validation technology are used to improve the convergence speed and stability of the model. The detection and recognition accuracy of the model is evaluated on an independent test set, and a variety of indicators such as sensitivity, specificity, and AUC are used for comprehensive evaluation to ensure its reliability and effectiveness in clinical applications. Finally, the optimized model is integrated into the clinical workflow, and the API interface is developed to connect with the hospital information system to support doctors in early diagnosis and decision-making of tumors. At the same time, a user feedback mechanism is established to collect feedback from doctors and technicians and continuously optimize model performance.
[0100] Correspondingly, the present application also discloses a tumor recognition and detection device based on a large model, see Figure 3 As shown, the device comprises:
[0101] A data acquisition module 11, used to acquire multimodal tumor image data;
[0102] A data enhancement module 12, used to determine a target macromodel for medical image processing, and perform image data enhancement processing on the multimodal medical image data using the target macromodel to obtain a plurality of synthetic tumor images;
[0103] A model training module 13, configured to generate a target data set using the multimodal tumor image data and the synthetic tumor image, and to train a target tumor recognition and detection model using the target data set;
[0104] The model integration module 14 is used to integrate the trained target tumor recognition and detection model into the hospital information system to provide real-time tumor recognition and detection results.
[0105] Among them, for more specific working processes of the above-mentioned modules, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.
[0106] It can be seen that the above scheme of this embodiment includes: acquiring multimodal tumor image data; determining a target large model for medical image processing, and using the target large model to perform image data enhancement processing on the multimodal medical image data to obtain multiple synthetic tumor images; generating a target data set using the multimodal tumor image data and the synthetic tumor images, and training a target tumor recognition and detection model using the target data set; integrating the trained target tumor recognition and detection model into a hospital information system to provide real-time tumor recognition and detection results.
[0107] The beneficial technical effects of this application are:
[0108] 1. Improve data diversity: This invention effectively expands the data set of tumor images by using synthetic tumor images generated by industry large models, increases data diversity, covers a wider range of tumor morphology and pathological characteristics, and solves the problem of data scarcity;
[0109] 2. Improve model performance: Due to the expansion and diversification of the data set, the tumor detection and recognition model trained based on this method shows stronger generalization ability, can stably detect and identify tumors in different clinical environments, and reduce overfitting of specific data sets;
[0110] 3. Improve model training efficiency: The image generation process significantly reduces the time for data preparation, allowing researchers and medical experts to obtain the data sets required for training more quickly, speeding up model development and iteration;
[0111] 4. Improve diagnostic accuracy: The introduction of synthetic tumor images provides the model with more subtle and difficult-to-observe tumor features, enabling the model to more accurately identify the edges and internal structures of the tumor, thereby improving the accuracy of clinical diagnosis;
[0112] 5. Enhance model interpretability: Through images generated by industry-leading models, researchers can gain a deeper understanding of how the models learn and identify tumor features, thereby improving the interpretability of the models and helping medical experts better trust and adopt AI technology.
[0113] Furthermore, the present application also discloses an electronic device. Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment, and the content in the diagram cannot be considered as any limitation on the scope of use of the present application.
[0114] Figure 4 A schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the tumor recognition and detection method based on a large model disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be a computer.
[0115] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.
[0116] In addition, the memory 22 as a carrier for resource storage may be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon may include an operating system 221, a computer program 222 and data 223, etc. The data 223 may include various data. The storage method may be temporary storage or permanent storage.
[0117] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, and can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the large model-based tumor recognition and detection method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks.
[0118] Furthermore, the embodiment of the present application also discloses a computer-readable storage medium, wherein the computer-readable storage medium mentioned here includes random access memory (Random Access Memory, RAM), memory, read-only memory (Read-Only Memory, ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, magnetic disk or optical disk or any other form of storage medium known in the technical field. Wherein, when the computer program is executed by the processor, the aforementioned tumor recognition and detection method based on the large model is implemented. For the specific steps of the method, reference can be made to the corresponding contents disclosed in the aforementioned embodiment, which will not be repeated here.
[0119] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0120] The steps of the large model-based tumor recognition detection method or algorithm described in conjunction with the embodiments disclosed herein can be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the technical field.
[0121] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0122] The above is a detailed introduction to a large model-based tumor recognition and detection method, device, equipment and medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for a person skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
Claims
1. A tumor recognition and detection method based on a large model, characterized in that: include: Acquire multimodal tumor image data; Determining a target macromodel for medical image processing, and performing image data enhancement processing on the multimodal medical image data using the target macromodel to obtain a plurality of synthetic tumor images; Generate a target data set using the multimodal tumor image data and the synthetic tumor image, and train a target tumor recognition and detection model using the target data set; The trained target tumor recognition and detection model is integrated into the hospital information system to provide real-time tumor recognition and detection results.
2. The tumor recognition and detection method based on a large model according to claim 1, characterized in that: After acquiring the multimodal tumor image data, the method further includes: Preprocessing the multimodal tumor image data; the preprocessing process includes: data standardization, denoising, normalization, segmentation, contrast adjustment and image registration; Accordingly, the multimodal medical image data is subjected to image data enhancement processing using the target large model to obtain a plurality of synthetic tumor images, including: The target large model is used to perform image data enhancement processing on the preprocessed multimodal medical image data to obtain a plurality of synthetic tumor images.
3. The tumor recognition and detection method based on a large model according to claim 1, characterized in that: The step of determining a target macromodel for performing medical image processing, and performing image data enhancement processing on the multimodal medical image data using the target macromodel to obtain a plurality of synthetic tumor images includes: Determine an industry-wide large model for medical image processing, and use transfer learning technology to fine-tune the industry-wide large model to learn feature representations of tumor images to obtain the target large model; The data enhancement pipeline is implemented through the data enhancement tool in the deep learning framework, and the target large model is used to perform image data enhancement processing on each batch of the multimodal medical image data in real time based on the data enhancement pipeline to obtain multiple synthetic tumor images.
4. The tumor recognition and detection method based on a large model according to claim 1, characterized in that: The step of generating a target data set by using the multimodal tumor image data and the synthetic tumor image, and training a target tumor recognition and detection model by using the target data set, includes: Merging the multimodal tumor image data and the synthetic tumor image to generate a target data set; The target data set is balanced by using an oversampling technique or an undersampling technique, and a target tumor recognition and detection model is trained by using the balanced target data set.
5. The tumor recognition and detection method based on a large model according to claim 1, characterized in that: The using the target data set to train a target tumor recognition and detection model comprises: Dividing the target data set into a training set, a validation set, and a test set; Selecting a target loss function and a target optimization algorithm required for training the target tumor recognition and detection model, and determining a strategy to prevent overfitting when obtaining the target tumor recognition and detection model; The initial tumor recognition and detection model is trained using the training set according to the target loss function and the target optimization algorithm, and the learning rate of the initial tumor recognition and detection model is dynamically adjusted using the validation set based on the overfitting prevention strategy to obtain the target tumor recognition and detection model.
6. The tumor recognition and detection method based on a large model according to claim 5, characterized in that: After using the target data set to train the target tumor recognition and detection model, the method further includes: A cross-validation method is adopted to evaluate the performance of the target tumor recognition detection model using at least one evaluation indicator on the test set.
7. The tumor recognition and detection method based on a large model according to any one of claims 1 to 6, characterized in that: The process of integrating the trained target tumor recognition and detection model into the hospital information system to provide real-time tumor recognition and detection results also includes: The target tumor recognition and detection model is updated online through online learning, and a user feedback mechanism is established to collect user feedback information; The feedback information is used to optimize the performance of the target tumor recognition and detection model.
8. A tumor recognition and detection device based on a large model, characterized in that: include: A data acquisition module, used for acquiring multimodal tumor image data; A data enhancement module, used to determine a target macromodel for medical image processing, and perform image data enhancement processing on the multimodal medical image data using the target macromodel to obtain a plurality of synthetic tumor images; A model training module, used to generate a target data set using the multimodal tumor image data and the synthetic tumor image, and train a target tumor recognition and detection model using the target data set; The model integration module is used to integrate the trained target tumor recognition and detection model into the hospital information system to provide real-time tumor recognition and detection results.
9. An electronic device, characterized in that: The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the large model-based tumor recognition and detection method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: Used to store computer programs; wherein when the computer program is executed by a processor, the large model-based tumor recognition and detection method as described in any one of claims 1 to 7 is implemented.