Re-parameterization-based medical image data intelligent auxiliary interpretation method and system, medium, program product and terminal

Abstract

According to the medical image data intelligent auxiliary interpretation method and system based on re-parameterization, the medium, the program product and the terminal provided by the invention, intelligent auxiliary interpretation of the medical image is realized by combining neural network design, training strategy optimization and incremental learning strategy optimization. According to the invention, through the design of the basic convolution operator and the optimization of the training strategy, the model is more suitable for the intelligent recognition of the medical image, the characteristics of the medical image data set are well utilized, and a large amount of label-free data and a small amount of accurately labeled data are fully utilized to improve the recognition accuracy and robustness. According to the method, the intelligent detection and recognition model is updated regularly by utilizing an incremental learning strategy, and the knowledge of the model is continuously increased by learning new data under the condition that the whole model is not retrained by combining the thought of'playback 'data and regularization, so that the long-term practicability and stability of the detection and recognition model are improved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent assisted interpretation of medical imaging data, and in particular to a method, system, medium, program product and terminal for intelligent assisted interpretation of medical imaging data based on re-parameterization. Background Art

[0002] The construction of intelligent auxiliary interpretation system for medical images is of great significance and has broad application prospects. Intelligent auxiliary interpretation methods can automatically detect and mark lesions, provide quantitative analysis results, and assist doctors in making decisions by using artificial intelligence algorithms to analyze medical images. This can not only improve the accuracy of diagnosis, but also reduce the workload of doctors and improve work efficiency. In addition, intelligent auxiliary interpretation technology can help solve the problem of insufficient medical resources, especially in primary medical institutions, where doctors' imaging diagnosis capabilities are relatively weak. By using auxiliary interpretation systems, their diagnostic level can be improved, so that more patients can receive timely and accurate diagnosis and treatment. By analyzing a large amount of medical imaging data, artificial intelligence models can discover new disease characteristics, treatment effect indicators, etc., providing valuable data resources for medical research. The same set of technologies can be deployed in multiple locations to expand the scope of medical services and benefit more patients. For example, by using remote imaging diagnosis systems, high-quality medical services can be provided to patients in remote areas to solve the problem of "difficulty in seeing a doctor."

[0003] At present, the research status of intelligent recognition technology based on medical images shows a booming trend. With the continuous progress of computer vision, artificial intelligence and imaging technology, intelligent recognition technology of medical images has achieved remarkable research results and has been widely used in multiple medical imaging fields. At the algorithm level, artificial intelligence algorithms are developing rapidly. Based on the optimization of the backbone network structure, from the linear convolutional neural network VGG to the residual convolutional neural network ResNet, to the network introducing the attention mechanism (such as Swin Transformer), and RepLKNet based on reparameterization and large convolution kernels, the backbone network's ability to represent features is gradually enhanced; the module optimization of detection and recognition has developed a large number of Neck, Head module designs and label assignment methods, such as FPN, BiFPN, PAN and CSP-PAN, which have become the mainstream method in the field of general image recognition. At the application level, the current status of intelligent assisted interpretation of medical images is: 1) a small amount of precisely labeled data and a large amount of unusable unlabeled data, 2) unsatisfactory interpretation accuracy, both of which are problems that need to be solved in the actual application of intelligent assisted interpretation.

[0004] Image recognition methods based on universal detection and recognition networks are widely used in natural image recognition and have been verified on public data sets. However, their application in medical imaging is limited by problems such as large differences in lesions and small sample sizes. The practicality and versatility of the recognition methods are unsatisfactory. For example, in X-ray images, taking fractures of the extremities and pneumonia on chest radiographs as examples, the former has an unfixed position, diverse morphology, low resolution (small target area), and a small sample size, while the latter has a relatively fixed position, diverse lesion morphology, high resolution (large target area), and a large sample size. As mentioned above, targets with large spans of appearance and statistical features will produce suppression of the majority class against the minority class and suppression of the large target against the small target in the same detection and recognition framework, affecting the recognition performance of the model. On the other hand, from the perspective of the long-term practicality of intelligent auxiliary interpretation methods, the interpretation basis will produce statistical deviations with scientific development and regional characteristics, and the model needs to be iterated in a timely manner with changes to adapt to the latest interpretation standards. Summary of the invention

[0005] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method, system, medium, program product and terminal for intelligent assisted interpretation of medical image data based on re-parameterization, which are used to solve the problems of a small amount of precisely labeled data and a large amount of unusable unlabeled data, unsatisfactory interpretation accuracy, and the inability to update and iterate the model in real time in practical applications of intelligent assisted interpretation.

[0006] To achieve the above-mentioned purpose and other related purposes, the first aspect of the present application provides a method for intelligent assisted interpretation of medical image data based on re-parameterization, including: obtaining a sample set of unlabeled medical image data and a sample set of accurately labeled medical image data; using re-parameterization technology to construct a re-parameterized convolution module, replacing the basic module in the deep learning neural network with the re-parameterized convolution module to obtain an initial medical image detection and recognition model; using the sample set of unlabeled medical image data to perform self-supervised pre-training on the initial medical image detection and recognition model; using the sample set of accurately labeled medical image data to perform fully supervised fine-tuning on the initial medical image detection and recognition model after self-supervised pre-training to obtain a medical image detection and recognition model; constructing an incremental learning sample set, and in response to a preset trigger condition, inputting the incremental learning sample set into the medical image detection and recognition model for incremental learning to update the training of the medical image detection and recognition model.

[0007] In some embodiments of the first aspect of the present application, the process of constructing a re-parameterized convolution module using re-parameterization technology includes: in the training stage, dividing the input channel into multiple branches, processing each branch separately, adding the processing results of each branch, and inputting the added result into the activation function layer; in the inference stage after the training stage is completed, the convolution structure of multiple branches is equivalently replaced by a single-branch convolution structure.

[0008] In some embodiments of the first aspect of the present application, the multiple branches are respectively a first branch, a second branch, a third branch, a fourth branch and a fifth branch; wherein: the first branch is a residual branch; the second branch includes a 1×1 convolutional layer and a first BN layer connected in sequence; the third branch includes a 3×1 convolutional layer, a 1×3 convolutional layer and a second BN layer connected in sequence; the fourth branch includes a 5×1 convolutional layer, a 1×5 convolutional layer and a third BN layer connected in sequence; the fifth branch includes a 7×1 convolutional layer, a 1×7 convolutional layer and a fourth BN layer connected in sequence.

[0009] In some embodiments of the first aspect of the present application, the process of self-supervised pre-training of an initial medical image detection and recognition model using an unlabeled medical image data sample set includes: constructing a positive sample pair set and a negative sample pair set based on the unlabeled medical image data sample set; using the positive sample pair set and the negative sample pair set to perform self-supervised comparative learning, and performing repeated iterative training based on a loss function until the training is terminated when a preset condition is met, so as to obtain an initial medical image detection and recognition model after self-supervised pre-training.

[0010] In some embodiments of the first aspect of the present application, the process of constructing a positive sample pair set based on the unlabeled medical imaging data sample set includes: using two different preprocessing methods on any unlabeled medical imaging data sample in the unlabeled medical imaging data sample set to obtain a pair of positive sample inputs; performing the same feature mapping processing on the pair of positive sample inputs to obtain a pair of high-dimensional feature vectors; performing dimensionality reduction mapping processing on the pair of high-dimensional feature vectors to obtain a pair of low-dimensional feature vectors, and the pair of low-dimensional feature vectors serve as the positive sample pair of any unlabeled medical imaging data sample.

[0011] In some embodiments of the first aspect of the present application, the process of inputting the incremental learning sample set into the medical image detection and recognition model for incremental learning to update the training of the medical image detection and recognition model includes: calculating the weight value corresponding to each neuron in the medical image detection and recognition model, and dividing each neuron into a baseline neuron and a target neuron through preset screening conditions and the weight value corresponding to each neuron; using the incremental learning sample set to fine-tune the medical image detection and recognition model, and setting the regularization loss function conditions so that when the parameters of the target neuron are updated during the fine-tuning training process, the parameter changes of the baseline neuron are minimized.

[0012] To achieve the above-mentioned purpose and other related purposes, the second aspect of the present application provides a medical imaging data intelligent auxiliary interpretation system based on re-parameterization, including: a data sample set acquisition module, used to acquire an unlabeled medical imaging data sample set and an accurately labeled medical imaging data sample set; a re-parameterized convolution module replacement module, used to construct a re-parameterized convolution module using re-parameterization technology, and replace the basic module in the deep learning neural network with the re-parameterized convolution module to obtain an initial medical imaging detection and recognition model; a model generation module, used to perform self-supervised pre-training on the initial medical imaging detection and recognition model using the unlabeled medical imaging data sample set; perform fully supervised fine-tuning on the initial medical imaging detection and recognition model after the self-supervised pre-training using the accurately labeled medical imaging data sample set to obtain a medical imaging detection and recognition model; a model update training module, used to construct an incremental learning sample set, and input the incremental learning sample set into the medical imaging detection and recognition model for incremental learning in response to a preset trigger condition, so as to update and train the medical imaging detection and recognition model.

[0013] To achieve the above-mentioned purpose and other related purposes, the third aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent assisted interpretation method of medical imaging data based on re-parameterization.

[0014] To achieve the above-mentioned purpose and other related purposes, the fourth aspect of the present application provides a computer program product, which includes a computer program code. When the computer program code runs on a computer, the computer implements the intelligent assisted interpretation method of medical image data based on re-parameterization.

[0015] To achieve the above-mentioned purpose and other related purposes, the fifth aspect of the present application provides an electronic terminal, including a memory, a processor and a computer program stored in the memory; the processor executes the computer program to implement the intelligent assisted interpretation method of medical imaging data based on re-parameterization.

[0016] As described above, the medical imaging data intelligent assisted interpretation method, system, medium, program product and terminal based on re-parameterization provided by the present application have the following beneficial effects:

[0017] This application realizes intelligent assisted interpretation of medical images by combining neural network design, training strategy optimization and incremental learning strategy optimization. In this application, by designing the basic convolution operator and optimizing the training strategy, the model is made more suitable for intelligent recognition of medical images, making good use of the characteristics of medical image data sets, and making full use of a large amount of unlabeled data and a small amount of accurately labeled data to improve the accuracy and robustness of recognition. This application also uses an incremental learning strategy to regularly update the intelligent detection and recognition model. By combining the idea of ​​"playback" data with regularization, the model is allowed to continuously increase its knowledge by learning new data without retraining the entire model, thereby improving the long-term practicality and stability of the detection and recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Shown is a flow chart of a method for intelligent assisted interpretation of medical imaging data based on re-parameterization in one embodiment of the present application.

[0019] Figure 2 Shown is a structural schematic diagram of a deep learning neural network in one embodiment of the present application.

[0020] Figure 3 Shown is a structural schematic diagram of a heavy parameter convolution module in one embodiment of the present application.

[0021] Figure 4 Shown is a specific embodiment diagram of constructing a positive sample pair set in one embodiment of the present application.

[0022] Figure 5 Shown is a specific embodiment diagram of a medical image data intelligent auxiliary interpretation system based on re-parameterization in one embodiment of the present application.

[0023] Figure 6 Shown is a structural schematic diagram of a medical image data intelligent auxiliary interpretation system based on re-parameterization in one embodiment of the present application.

[0024] Figure 7 Shown is a schematic diagram of the structure of an electronic terminal in one embodiment of the present application. DETAILED DESCRIPTION

[0025] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0026] The embodiment of the present application provides a method for intelligent assisted interpretation of medical image data based on re-parameterization. This method realizes intelligent assisted interpretation of medical images by combining neural network design, training strategy optimization and incremental learning strategy optimization. In the present application, the model is more suitable for intelligent recognition of medical images through the design of basic convolution operators and the optimization of training strategies. It makes good use of the characteristics of medical image data sets and makes full use of a large amount of unlabeled data and a small amount of accurately labeled data to improve the accuracy and robustness of recognition. The present application also uses an incremental learning strategy to regularly update the intelligent detection and recognition model. By combining the idea of ​​"playback" data with regularization, the model is allowed to continuously increase its knowledge by learning new data without retraining the entire model, thereby improving the long-term practicality and stability of the detection and recognition model.

[0027] To facilitate understanding of the embodiments of the present application, first Figure 1 Detailed description. Figure 1 A method for intelligent assisted interpretation of medical image data based on re-parameterization in an embodiment of the present invention is presented; the method comprises:

[0028] Step S11: Obtain an unlabeled medical imaging data sample set and an accurately labeled medical imaging data sample set.

[0029] It should be noted that medical imaging data are medical photos taken by X-ray, ultrasound, CT, MRI, etc., which are used to assist in disease diagnosis. The unlabeled medical imaging data in this embodiment refers to data that does not contain any form of annotation information, that is, there is no description of the image content, classification labels or any form of metadata. They are original, unprocessed image data. Accurately labeled medical imaging data contains detailed annotation information and is usually annotated by professional doctors or annotators. The annotation can be a mark of a specific area in the image (such as a tumor, lesion, etc.), or a classification label for the image content (such as normal, abnormal, etc.).

[0030] Step S12: Use re-parameterization technology to construct a re-parameterized convolution module, and replace the basic module in the deep learning neural network with the re-parameterized convolution module to obtain an initial medical image detection and recognition model.

[0031] It should be noted that this embodiment constructs a medical image detection and recognition model through a deep learning neural network. Figure 2As shown, the structure of the deep learning neural network includes three parts: feature extraction network (Backbone network) 110, feature fusion network (Neck network) 120 and detection head network (Head network) 130. The Backbone network is responsible for feature extraction, the Neck network strengthens features, and the Head network is responsible for classification and positioning. The detection head network 130 can include a positioning network 131, a classification network 132 or a segmentation network 133 corresponding to different task requirements.

[0032] In this embodiment, a re-parameterized convolution module is constructed based on the re-parameterization technology to replace the basic modules in the feature extraction network (Backbone) 110, the feature fusion network (Neck) 120 and the detection head network (Head) 130 in the deep learning neural network, that is, the re-parameterized convolution module can be applied to the basic modules of the feature extraction network, the feature fusion network, and the positioning / classification / segmentation network to achieve efficient feature modeling. The use of re-parameterization technology can increase the width and depth of the model without additional burden, and improve the model's feature extraction and characterization capabilities.

[0033] In one embodiment, the process of constructing a re-parameterized convolution module using re-parameterization technology includes: in the training stage, dividing the input channel into multiple branches, processing each branch separately, adding the processing results of each branch, and inputting the added result into the activation function layer; in the inference stage after the training stage is completed, the convolution structure of multiple branches is equivalently replaced by the convolution structure of a single branch.

[0034] It should be explained that the process of building a re-parameterized convolution module consists of two stages: training and inference. Figure 3 Note that during the training phase, the input features are copied into multiple copies so that different operations can be performed on these input features at the same time in subsequent processing to extract richer feature information. The input channel is then divided into multiple branch inputs. This structure allows the network to process the same input in parallel on different paths, which helps the model capture different features in the input data. The copied input features are input into the above different branches respectively, and each branch is processed separately.

[0035] Specifically, the multiple branches are respectively a first branch, a second branch, a third branch, a fourth branch and a fifth branch; wherein: the first branch is a residual branch; the second branch includes a 1×1 convolutional layer and a first BN layer connected in sequence; the third branch includes a 3×1 convolutional layer, a 1×3 convolutional layer and a second BN layer connected in sequence; the fourth branch includes a 5×1 convolutional layer, a 1×5 convolutional layer and a third BN layer connected in sequence; the fifth branch includes a 7×1 convolutional layer, a 1×7 convolutional layer and a fourth BN layer connected in sequence.

[0036] like Figure 3 As shown, the multiple branches in this embodiment are preferably five branches. The first branch is a residual branch, and the processing process is to input the copied input features and perform self-mapping processing, that is, directly pass the input features to the next stage without any transformation, which helps to form a residual connection, thereby alleviating the gradient problem in the training process. Residual connection is a common network design technique. By adding the input directly to the output of the subsequent layer without being affected by the intermediate layer, it can effectively alleviate the gradient vanishing or gradient exploding problems in deep network training and help maintain the stability of the gradient. The use of a residual branch structure allows the network to learn the identity mapping, that is, directly pass the input information, thereby making the training of deep networks more efficient and stable.

[0037] The second branch includes a 1×1 convolutional layer and the first BN (Batch Normalization) layer connected in sequence. After the input feature is input, a convolution operation is performed by a single-layer k=1 convolution operator in the 1×1 convolutional layer. The convolution operator performs a linear mapping between channels without changing the spatial dimension. Then, the output of the 1×1 convolutional layer is normalized by the batch normalization operator in the first BN layer. Batch normalization can reduce internal covariate shift and accelerate the network training process.

[0038] The third branch includes a 3×1 convolution layer, a 1×3 convolution layer, and a second BN layer connected in sequence. The processing process is that the input features are processed in sequence by two layers of asymmetric k=3 convolution operators. The two convolution layers perform spatial modeling. The second BN layer normalizes the output results of the sequentially connected 3×1 convolution layer and 1×3 convolution layer.

[0039] The fourth branch includes a 5×1 convolution layer, a 1×5 convolution layer, and a third BN layer connected in sequence. The processing process is that the input features are processed in sequence by two layers of asymmetric k=5 convolution operators. The convolution operators of the two convolution layers perform spatial modeling, which can capture feature information in a larger spatial range than the third branch. The third BN layer normalizes the output results of the sequentially connected 5×1 convolution layer and 1×5 convolution layer.

[0040] The fifth branch includes a 7×1 convolutional layer, a 1×7 convolutional layer, and a fourth BN layer connected in sequence. The processing process is that the input features are processed in sequence by two layers of asymmetric k=7 convolution operators. The convolution operators of the two convolutional layers perform spatial modeling, which further captures feature information in a larger spatial range than the fourth branch. The fourth BN layer normalizes the output results of the sequentially connected 7×1 convolutional layer and the 1×7 convolutional layer.

[0041] The final output results of the above five branches are added together, and the added result is sent to the activation function layer for nonlinear mapping processing, which can increase the nonlinear expression ability of the model. The activation function layer can be a ReLU activation function.

[0042] After the training phase, in the inference phase, if Figure 3 As shown in Figure 2, the convolutional structures in each branch can be fused into a single branch convolutional structure. Specifically, Figure 3 The 1×1 convolution layer, 3×1 convolution layer, 1×3 convolution layer, 5×1 convolution layer, 1×5 convolution layer, 7×1 convolution layer, and 1×7 convolution layer in the multiple branches on the left are fused into Figure 3 The 7×7 convolution layer on the right. By replacing each branch with a single 7×7 convolution operation, the network inference speed can be improved during the inference phase, reducing the model's demand for video memory and the amount of computation during each forward propagation, allowing the model to efficiently process input data and meet the requirements of real-time applications.

[0043] Step S13: Use the unlabeled medical imaging data sample set to perform self-supervised pre-training on the initial medical imaging detection and recognition model; use the precise labeled medical imaging data sample set to perform fully supervised fine-tuning on the initial medical imaging detection and recognition model after the self-supervised pre-training to obtain the medical imaging detection and recognition model.

[0044] After completing the construction of the initial medical image detection and recognition model through the above steps, the model needs to be further trained. The training strategy adopted in this embodiment is self-supervised pre-training and fully supervised fine-tuning. Self-supervised pre-training is a deep learning training strategy that does not rely on external labels, but uses the structure and characteristics of the data itself to train the model. Fully supervised fine-tuning is performed on the basis of self-supervised pre-training, and the pre-trained model is further fine-tuned. Through the combination of self-supervised pre-training and fully supervised fine-tuning, the model can better adapt to the needs of specific tasks while maintaining generalization capabilities, and improve performance on target tasks.

[0045] In one embodiment, the process of using an unlabeled medical imaging data sample set to perform self-supervised pre-training on an initial medical imaging detection and recognition model includes: constructing a positive sample pair set and a negative sample pair set based on the unlabeled medical imaging data sample set; using the positive sample pair set and the negative sample pair set to perform self-supervised comparative learning, and performing repeated iterative training based on a loss function until the training is terminated when a preset condition is met, so as to obtain an initial medical imaging detection and recognition model after self-supervised pre-training.

[0046] It should be noted that there are a large number of unlabeled medical imaging data in the dataset. Self-supervised contrastive learning can make full use of unlabeled samples and mine the characteristic information of the data itself. In this embodiment, a method for contrastive learning based on a constructed positive and negative sample pair set is proposed, so that the model can distinguish between positive sample pairs and negative sample pairs, thereby learning the representation information of the image.

[0047] Specifically, the process of constructing a positive sample pair set based on the unlabeled medical imaging data sample set includes: using two different preprocessing methods on any unlabeled medical imaging data sample in the unlabeled medical imaging data sample set to obtain a pair of positive sample inputs; performing the same feature mapping processing on the pair of positive sample inputs to obtain a pair of high-dimensional feature vectors; performing dimensionality reduction mapping processing on the pair of high-dimensional feature vectors to obtain a pair of low-dimensional feature vectors, and the pair of low-dimensional feature vectors serve as the positive sample pair of any unlabeled medical imaging data sample.

[0048] For example, Figure 4 As shown in Figure 2, the specific steps for constructing a positive sample pair set are as follows:

[0049] Step 1: Select any unlabeled medical image data in the unlabeled medical image data sample set, that is, select an input image I, and perform different random preprocessing (such as rotation, translation, noise addition, masking) on ​​the input image I. Specifically, two different preprocessing methods are used, and different preprocessed input images are output corresponding to different preprocessing methods to obtain a pair of positive sample inputs. and i and j represent different preprocessing methods.

[0050] Step 2: Input this pair of positive samples and After the same feature mapping f(·) is processed, a corresponding pair of high-dimensional feature vectors h is obtained. i and h j .

[0051] Step 3: For this pair of high-dimensional feature vectors h i and h j Perform a simple linear mapping to reduce the dimension, thus obtaining a pair of low-dimensional feature vectors z i and z j . Low-dimensional feature vector z i and z j As the positive sample pairs of the currently selected unlabeled medical imaging data.

[0052] Step 4: By performing the above processing on each sample in the unlabeled medical image data sample set, the positive sample pairs corresponding to each unlabeled medical image data sample set can be obtained to form a positive sample pair set. A positive sample pair refers to the feature vectors obtained by different processing of the same image, and the similarity between the feature vectors is relatively high.

[0053] Furthermore, the method for constructing a negative sample pair set is as follows: select two different unlabeled medical image data samples in the unlabeled medical image data sample set, the two image data samples have a low similarity, and may come from different patients, or even if they come from the same patient, they are completely different anatomical parts or pathological conditions, that is, select two different image inputs I1 and I2; perform high-dimensional feature mapping on the image inputs I1 and I2 to obtain a high-dimensional feature vector h1 of the image input I1 and a high-dimensional feature vector h2 of the image input I2; perform simple linear mapping on the high-dimensional feature vector h1 of the image input I1 and the high-dimensional feature vector h2 of the image input I2 to reduce the dimension, and obtain a low-dimensional feature vector z1 of the high-dimensional feature vector h1 and a low-dimensional feature vector z2 of the high-dimensional feature vector h2, and the low-dimensional feature vectors z1 and z2 are used as a pair of negative sample pairs. The feature vectors of the negative sample pairs need to have a low similarity, and multiple pairs of negative sample pairs are obtained to form a negative sample pair set.

[0054] After the construction of the positive sample pair set and the negative sample pair set is completed, self-supervised comparative learning is performed based on the positive sample pair set and the negative sample pair set. In the process of self-supervised pre-training using the positive sample pair set, a loss function is set to guide the training process of the initial medical image detection and recognition model to improve the low-dimensional feature vector z of the positive sample pair. i and z j The cosine similarity of the two vectors converges in the direction of cosine similarity, which is an indicator to measure the directional similarity of two vectors, to ensure that the feature vectors of the positive sample pair are closer in the vector space, that is, to bring the feature representation of the positive sample pair closer. In the process of self-supervised pre-training using the negative sample pair set, the loss function guides the model to push the feature representation of the negative sample pair further away. In each iterative training process, the model weights are updated according to the gradient of the loss function.

[0055] The initial medical image detection and recognition model obtained after the self-supervised pre-training process can capture the characteristics of medical image data and distinguish different image samples. In the above pre-training step, a large amount of unlabeled medical image data can be used to complete comparative learning to obtain an initial encoder, which grasps some characteristics of medical data sets and sample semantics. The initial encoder is the initial medical image detection and recognition model after self-supervised pre-training, which can lay the foundation for subsequent fully supervised fine-tuning.

[0056] Through this self-supervised pre-training method, the intrinsic structural information of a large amount of unlabeled medical imaging data can be used to train the model, laying the foundation for subsequent medical imaging analysis tasks. This method is particularly suitable for the field of medical imaging, because high-quality labeled data is often difficult to obtain, while unlabeled data is relatively abundant. Furthermore, using a small number of accurately labeled medical imaging data sample sets to perform fully supervised fine-tuning on the model can improve the performance of the model on specific tasks, achieve higher accuracy and practicality, and ultimately obtain a practical medical imaging detection and recognition model.

[0057] It should be emphasized that in this embodiment, a better medical image detection and recognition model is constructed by using the heavy parameter convolution module based on the design of deep learning neural networks. At the same time, the use of self-supervised pre-training improves the utilization rate of a large amount of unlabeled medical image data, enhances the model's semantic understanding and feature representation of images, and then fine-tunes with a small amount of accurately labeled medical image data to obtain a medical image detection and recognition model with strong practicality and good robustness, forming an important cornerstone of the intelligent auxiliary interpretation method. The medical image detection and recognition model plays an important role in medical image analysis. It can not only improve the accuracy and efficiency of diagnosis, but also assist doctors in making treatment decisions, thereby improving the treatment effect of patients.

[0058] Step S14: construct an incremental learning sample set, and input the incremental learning sample set into the medical image detection and recognition model for incremental learning in response to a preset trigger condition, so as to update and train the medical image detection and recognition model.

[0059] It should be explained that the incremental learning sample set refers to the sample data corrected and newly annotated by the imaging doctor as new sample data, which is centrally organized into an incremental learning sample set for incremental learning. The preset trigger condition is the trigger condition for the medical image detection and recognition model to update the training based on the incremental learning sample set. It can be a threshold parameter for triggering the model update based on time or the number of new samples. This threshold parameter (preset trigger condition) determines when to start fine-tuning training to ensure that the model can adapt to new data and annotations in a timely manner.

[0060] In one embodiment, the process of inputting the incremental learning sample set into the medical image detection and recognition model for incremental learning to update the training of the medical image detection and recognition model includes: calculating the weight value corresponding to each neuron in the medical image detection and recognition model, and dividing each neuron into a baseline neuron and a target neuron through preset screening conditions and the weight value corresponding to each neuron; using the incremental learning sample set to fine-tune the medical image detection and recognition model, and setting the regularization loss function conditions so that when the parameters of the target neuron are updated during the fine-tuning training process, the parameter changes of the baseline neuron are minimized.

[0061] It should be explained that this embodiment provides an incremental learning method that combines the ideas of "replay" and regularization. The incremental learning method can allow the model to continuously increase its knowledge by learning new data without retraining the entire model. The introduction of neuron regularization and incremental learning methods in this embodiment can fully explore the continuous learning ability of the model. The specific steps of updating and training the medical image detection and recognition model based on incremental learning include the following:

[0062] Step 1: Calculate the weight value (Importance value) of each neuron in the medical image detection and recognition model, and select the important neurons in the model as benchmark neurons according to the preset screening conditions. For example, sort each neuron from high to low by Importance value, and the preset screening condition is to select the Top 10% of neurons as benchmark neurons. According to this preset screening condition, the neurons in the Top 10% of the weight value can be used as benchmark neurons, and the rest as target neurons. Among them, the benchmark neurons represent the most critical part of the model and have a significant impact on the decision-making of the model. Therefore, try to avoid updating its parameters or making large adjustments during this update training process, and the target neural unit is the target that needs to update parameters during this update training process.

[0063] Step 2: Filter out representative samples from the precisely labeled medical imaging data sample set (10 to 20 samples of each type) to ensure that the representative samples can cover the responses of all the above-mentioned benchmark neurons. The representative sample data set is {X′, Y′}; at the same time, store the image doctor-corrected and newly labeled sample data as an incremental learning sample set, recorded as {X″, Y″}.

[0064] Step three: In response to preset trigger conditions (i.e., updating threshold parameters), use representative sample data sets and incremental learning sample sets to perform fine-tuning training of the medical image detection and recognition model. At this time, it is necessary to set the regularized loss function conditions. During the model fine-tuning training process, the regularized loss function is calculated. The regularized loss function meets the regularized loss function conditions so that when the parameters of the target neuron are updated, the parameter changes of the benchmark neuron are minimized. That is, the parameter update of the benchmark neuron needs to be kept within a certain range, and the parameter update of the benchmark neuron should be avoided as much as possible, which helps to maintain the stability of the model on key features. During the forward propagation of each training process, the representative sample data set {X′, Y′} is used to calculate the distance of the benchmark neuron as part of the regularization term to measure and control parameter changes.

[0065] The above steps complete the update training of the medical image detection and recognition model. It should be understood that the parameter values ​​of the model will be adjusted according to the incremental learning sample set to improve the performance of the model on new data. At the same time, according to the set regularized loss function conditions, it can be ensured that the model will not deviate too much from the original key features while adapting to the new data.

[0066] It should be emphasized that in this embodiment, by collecting samples that have been corrected and newly labeled by imaging doctors and combining them with an incremental learning strategy based on the ideas of "replay" and regularization, the medical imaging detection and recognition model is regularly updated, so that the model can gradually adapt to new data while retaining previously learned knowledge, thereby maintaining better robustness and practicality in long-term applications.

[0067] In order to facilitate the demonstration of a method for intelligent assisted interpretation of medical image data based on re-parameterization provided by the present application, the following specific embodiments are provided for illustration.

[0068] like Figure 5 As shown in the figure, a specific implementation process of a medical image data intelligent auxiliary interpretation system based on reparameterization applicable to real-time X-ray images is demonstrated. The system includes an intelligent detection and recognition module applicable to medical images and an incremental learning module based on user interaction correction labels. Among them, the intelligent detection and recognition module introduces a reparameterization strategy in structural design, designs a multi-branch convolution structure, and merges the features of each branch for forward inference, which enhances the feature fusion and expression ability of the model; in terms of training strategy, through self-supervised training, a large amount of unlabeled X-ray image data is used to learn the shallow texture features and local features of the image, and then a small amount of accurate label data is used for fine-tuning to obtain intelligent model parameters suitable for X-ray image interpretation. The incremental learning module based on interaction takes into account the practical application of intelligent auxiliary interpretation methods in actual applications, uses the interactive function to store the sample data corrected and newly labeled by radiologists, and performs incremental fine-tuning on the model, so that the model can quickly adapt to the regional characteristics of new image data and changes in interpretation basis during long-term use.

[0069] The specific implementation process includes: inputting new images to be diagnosed into the intelligent detection and recognition module for intelligent auxiliary interpretation of image data; sending the auxiliary interpretation results to the doctor for judgment; the doctor performs visual interpretation based on the auxiliary interpretation results and the images to be diagnosed; using the interactive function, the doctor's multiple re-calibration data or newly labeled sample data is sent to the incremental learning module; the incremental learning module regularly updates and trains the intelligent detection and recognition module.

[0070] It should be noted that the construction of the medical image detection and recognition model in this embodiment includes deep learning network design and training strategy optimization, which can make full use of existing data to achieve recognition performance that meets user needs and different application scenarios. At the same time, the incremental learning method based on user interactive correction labels uses interactive functions to save correction labels and add new data, and regularly adjusts the medical image detection and recognition model.

[0071] The intelligent auxiliary interpretation method of medical imaging data based on reparameterization provided in this application enables the deep learning neural network model to adapt to the characteristics of medical imaging, and can make full use of a large amount of unlabeled data and a small amount of accurately annotated data. At the same time, the introduction of the reparameterization strategy improves the model's feature fusion and representation of cross-scale targets, and maintains the model's practicality and generalization for actual application scenarios through subsequent incremental learning. This application can greatly reduce the workload of medical imaging-related medical personnel, expand the scope of medical services, benefit more patients, and provide strong technical support for application fields such as smart healthcare.

[0072] In the embodiments of the present application, words such as "first" and "second" are used to distinguish the same or similar items with substantially the same functions and effects, and do not limit their order. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit them to be different.

[0073] It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" represent examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0074] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can represent: a, b, c, ab, ac, bc or abc, where a, b, c can be single or multiple.

[0075] Figure 61 is a schematic block diagram of a medical image data intelligent auxiliary interpretation system 100 based on re-parameterization provided in an embodiment of the present application. The system includes:

[0076] A data sample set acquisition module 110, used to acquire an unlabeled medical imaging data sample set and an accurately labeled medical imaging data sample set;

[0077] A re-parameterized convolution module replacement module 120, used to construct a re-parameterized convolution module using a re-parameterization technique, and replace the basic module in the deep learning neural network with the re-parameterized convolution module to obtain an initial medical image detection and recognition model;

[0078] The model generation module 130 is used to perform self-supervised pre-training on the initial medical image detection and recognition model using the unlabeled medical image data sample set; and perform fully supervised fine-tuning on the initial medical image detection and recognition model after the self-supervised pre-training using the accurate labeled medical image data sample set to obtain the medical image detection and recognition model;

[0079] The model update training module 140 is used to construct an incremental learning sample set, and input the incremental learning sample set into the medical image detection and recognition model for incremental learning in response to a preset trigger condition, so as to update the training of the medical image detection and recognition model.

[0080] It should be understood that the specific process of each module executing the above corresponding steps has been described in detail in the above system embodiment, and for the sake of brevity, it will not be repeated here.

[0081] It should also be understood that the division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present application may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0082] Figure 7 is a schematic block diagram of an electronic terminal provided in an embodiment of the present application. Figure 7 As shown, the electronic terminal includes: at least one processor 701, a memory 702, at least one network interface 703 and a user interface 705. The various components in the device are coupled together through a bus system 704. It can be understood that the bus system 704 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 704 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 7 In the specification, various buses are labeled as bus systems.

[0083] The user interface 705 may include a display, a keyboard, a mouse, a trackball, a click gun, keys, buttons, a touch pad or a touch screen.

[0084] It is understood that the memory 702 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), which is used as an external cache. By way of exemplary but not limiting explanation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include but is not limited to these and any other suitable categories of memory.

[0085] The memory 702 in the embodiment of the present invention is used to store various categories of data to support the operation of the electronic terminal 700. Examples of these data include: any executable program for operating on the electronic terminal 700, such as an operating system 7021 and an application 7022; the operating system 7021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 7022 can include various applications, such as a media player (Media Player), a browser (Browser), etc., for implementing various application services. The intelligent auxiliary interpretation method of medical imaging data based on re-parameterization provided in the embodiment of the present invention can be included in the application 7022.

[0086] The method disclosed in the above embodiment of the present invention can be applied to the processor 701, or implemented by the processor 701. The processor 701 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 701 or the instruction in the form of software. The above processor 701 may be a general processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 701 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor 701 may be a microprocessor or any conventional processor, etc. In combination with the steps of the accessory optimization method provided in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0087] In an exemplary embodiment, the electronic terminal 700 may be implemented by one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD) to execute the aforementioned method.

[0088] According to the method provided in the embodiments of the present application, the present application also provides a computer program product, which includes: computer program code, when the computer program code is run on a computer, the computer executes the intelligent assisted interpretation method of medical image data based on re-parameterization in any of the embodiments shown.

[0089] According to the method provided in the embodiments of the present application, the present application also provides a computer-readable storage medium, which stores a program code. When the program code runs on a computer, the computer executes the intelligent assisted interpretation method of medical imaging data based on re-parameterization in any of the embodiments shown.

[0090] The terms "component", "module", "system", etc. used in this specification are used to represent computer-related entities, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program and / or a computer. By way of illustration, both applications running on a computing device and a computing device can be components. One or more components may reside in a process and / or an execution thread, and a component may be located on a computer and / or distributed between two or more computers. In addition, these components may be executed from various computer-readable media having various data structures stored thereon. Components may, for example, communicate through local and / or remote processes according to signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system and / or a network, such as the Internet interacting with other systems through signals).

[0091] Those of ordinary skill in the art will appreciate that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0092] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0093] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0094] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0095] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0096] In the above embodiments, the functions of each functional unit can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions (programs). When loading and executing computer program instructions (programs) on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. Computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media integrations. Available media may be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., high-density digital video discs (DVDs), or semiconductor media (e.g., solid state disks (SSDs), etc.).

[0097] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0098] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0099] In summary, the medical imaging data intelligent assisted interpretation method, system, medium, program product and terminal based on re-parameterization provided by the present application include: obtaining an unlabeled medical imaging data sample set and an accurately labeled medical imaging data sample set; using re-parameterization technology to construct a re-parameterized convolution module, replacing the basic module in the deep learning neural network with the re-parameterized convolution module to obtain an initial medical imaging detection and recognition model; using the unlabeled medical imaging data sample set to perform self-supervised pre-training on the initial medical imaging detection and recognition model; using the accurately labeled medical imaging data sample set to perform fully supervised fine-tuning on the initial medical imaging detection and recognition model after self-supervised pre-training to obtain a medical imaging detection and recognition model; constructing an incremental learning sample set, and in response to a preset trigger condition, inputting the incremental learning sample set into the medical imaging detection and recognition model for incremental learning to update the training of the medical imaging detection and recognition model.

[0100] This application realizes intelligent auxiliary interpretation of medical images by combining neural network design, training strategy optimization and incremental learning strategy optimization. In this application, by designing the basic convolution operator and optimizing the training strategy, the model is more suitable for intelligent recognition of medical images, making good use of the characteristics of the medical image data set, and making full use of a large amount of unlabeled data and a small amount of accurately labeled data to improve the accuracy and robustness of recognition. This application also uses an incremental learning strategy to regularly update the intelligent detection and recognition model. By combining the idea of ​​"playback" data with regularization, the model is allowed to continuously increase its knowledge by learning new data without retraining the entire model, thereby improving the long-term practicality and stability of the detection and recognition model. Therefore, this application effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0101] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. A method for intelligent assisted interpretation of medical image data based on re-parameterization, characterized in that: include: Obtaining an unlabeled medical imaging data sample set and an accurately labeled medical imaging data sample set; A re-parameterization technique is used to construct a re-parameterized convolution module, and the basic module in the deep learning neural network is replaced with the re-parameterized convolution module to obtain an initial medical image detection and recognition model; The initial medical image detection and recognition model is self-supervised pre-trained using the unlabeled medical image data sample set; the initial medical image detection and recognition model after self-supervised pre-training is fully supervised fine-tuned using the precise labeled medical image data sample set to obtain the medical image detection and recognition model; An incremental learning sample set is constructed, and in response to a preset trigger condition, the incremental learning sample set is input into the medical image detection and recognition model for incremental learning, so as to update and train the medical image detection and recognition model.

2. The intelligent assisted interpretation method of medical image data based on re-parameterization according to claim 1 is characterized in that: The process of constructing a reparameterized convolution module using reparameterization technology includes: In the training phase, the input channel is divided into multiple branches, each branch is processed separately, the processing results of each branch are added together, and the added result is input to the activation function layer; In the inference phase after the training phase is completed, the convolutional structure of multiple branches is equivalently replaced by a single-branch convolutional structure.

3. The intelligent assisted interpretation method of medical image data based on re-parameterization according to claim 1 is characterized in that: The multiple branches are respectively a first branch, a second branch, a third branch, a fourth branch and a fifth branch; wherein: The first branch is a residual branch; The second branch includes a 1×1 convolutional layer and a first BN layer connected in sequence; The third branch includes a 3×1 convolutional layer, a 1×3 convolutional layer and a second BN layer connected in sequence; The fourth branch includes a 5×1 convolutional layer, a 1×5 convolutional layer and a third BN layer connected in sequence; The fifth branch includes a 7×1 convolutional layer, a 1×7 convolutional layer and a fourth BN layer which are connected in sequence.

4. The intelligent assisted interpretation method of medical image data based on re-parameterization according to claim 1, characterized in that: The process of self-supervising pre-training the initial medical image detection and recognition model using the unlabeled medical image data sample set includes: Constructing a positive sample pair set and a negative sample pair set based on the unlabeled medical imaging data sample set; The positive sample pair set and the negative sample pair set are used for self-supervised comparative learning, and repeated iterative training is performed based on the loss function until the training is terminated when the preset conditions are met, so as to obtain an initial medical image detection and recognition model after self-supervised pre-training.

5. The intelligent auxiliary interpretation method of medical image data based on re-parameterization according to claim 4 is characterized in that: The process of constructing a positive sample pair set based on the unlabeled medical imaging data sample set includes: Using two different preprocessing methods on any unlabeled medical image data sample in the unlabeled medical image data sample set to obtain a pair of positive sample inputs; Performing the same feature mapping process on the pair of positive sample inputs to obtain a pair of high-dimensional feature vectors; A dimensionality reduction mapping process is performed on the pair of high-dimensional feature vectors to obtain a pair of low-dimensional feature vectors, and the pair of low-dimensional feature vectors serve as a positive sample pair of any unlabeled medical imaging data sample.

6. The intelligent auxiliary interpretation method of medical image data based on re-parameterization according to claim 1 is characterized in that: The process of inputting the incremental learning sample set into the medical image detection and recognition model for incremental learning to update and train the medical image detection and recognition model includes: Calculating the weight value corresponding to each neuron in the medical image detection and recognition model, and dividing each neuron into a reference neuron and a target neuron according to a preset screening condition and the weight value corresponding to each neuron; The medical image detection and recognition model is fine-tuned and trained using an incremental learning sample set, and a regularized loss function condition is set so that the parameter change of the benchmark neuron is minimized when the parameters of the target neuron are updated during the fine-tuning training process.

7. An intelligent auxiliary interpretation system for medical image data based on re-parameterization, characterized in that: include: A data sample set acquisition module, used to acquire an unlabeled medical imaging data sample set and an accurately labeled medical imaging data sample set; A re-parameterized convolution module replacement module, used to construct a re-parameterized convolution module using a re-parameterization technique, and replace the basic module in the deep learning neural network with the re-parameterized convolution module to obtain an initial medical image detection and recognition model; A model generation module is used to perform self-supervised pre-training on an initial medical image detection and recognition model using an unlabeled medical image data sample set; and to perform fully supervised fine-tuning on the initial medical image detection and recognition model after self-supervised pre-training using an accurate labeled medical image data sample set to obtain a medical image detection and recognition model; The model update training module is used to construct an incremental learning sample set, and in response to a preset trigger condition, input the incremental learning sample set into the medical image detection and recognition model for incremental learning to update the medical image detection and recognition model.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for intelligent assisted interpretation of medical imaging data based on re-parameterization as described in any one of claims 1 to 6 is implemented.

9. A computer program product, characterized in that The computer program product includes computer program codes, and when the computer program codes are executed on a computer, the computer is enabled to implement the intelligent auxiliary interpretation method for medical image data based on re-parameterization according to any one of claims 1 to 6.

10. An electronic terminal comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the intelligent assisted interpretation method of medical imaging data based on re-parameterization as described in any one of claims 1 to 6.

Images