Method and system for interpreting medical image report based on multilayer deep learning model

By preprocessing and training data based on a multi-layer deep learning model, medical imaging reports can be automatically interpreted, solving the problem of poor report readability, achieving fast and convenient report interpretation, and improving patients' comprehension.

CN119028507APending Publication Date: 2024-11-26CHONGQING JUNZI WUBEN TECHNOLOGY CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410966597.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The poor readability of existing medical imaging reports makes it difficult for patients to accurately understand their health conditions, increasing the reporting burden on radiologists.

Method used

A multi-layer deep learning model-based approach is adopted to interpret medical image reports through data preprocessing, construction of a multi-layer deep learning model, and training of the model. The model includes an input layer, an embedding layer, a bidirectional long short-term memory network, a transformer module, a global max pooling layer, a full-density layer, and a regularization layer, and outputs the probability of a binary classification task.

Benefits of technology

It enables automatic interpretation of medical imaging reports, improving report readability, reducing the workload of radiologists, and providing rapid and easy-to-understand interpretation for non-professional users, with a feedback time of no more than 1 minute.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028507A_ABST
    Figure CN119028507A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses a method and system for interpreting a medical image report based on a multilayer deep learning model, and the method comprises the steps: collecting a certain number of historical medical image reports, and carrying out the data preprocessing of the historical medical image reports, pre-processed data are obtained; constructing a multi-layer deep learning model, and inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model; and inputting preprocessed data corresponding to a to-be-explained medical image report into the optimal multilayer deep learning model to obtain a medical image report explanation result corresponding to the to-be-explained medical image report. The problem that in the prior art, medical image reports are poor in readability is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence generation technology, and specifically to a system, method, electronic device and storage medium for interpreting medical imaging reports based on a multi-layer deep learning model. Background Art

[0002] Medical imaging is the science of obtaining images of the internal structures of the human body using non-invasive technology. It includes technologies such as X-ray, CT, MRI, ultrasound, and nuclear medicine, which aim to reveal invisible anatomical and physiological conditions. Medical imaging reports include detailed descriptions of various imaging results, technical specifications of diagnostic tools, measurement results, and clinical recommendations. It is a complex communication tool mainly used for communication between doctors and consists of professional medical terms and non-grammatical fragments. However, because medical imaging reports are full of professional terms and complex medical descriptions, patients cannot accurately grasp their health status based on medical imaging reports, which may affect treatment effects and prognosis.

[0003] In recent years, various medical institutions have been trying to develop patient-centered annotated imaging reports using simplified terms and infographics. For example, patient-friendly reports are provided, which use simplified language and illustrations so that non-professional patients can understand complex medical terms and results; or interactive interpretation tools are provided, through the use of online platforms and applications, patients can enter their report data and obtain personalized interpretations and guidance. However, these strategies will increase the reporting burden of radiologists, and the feasibility of these methods is questionable in an environment with tight medical resources.

[0004] Therefore, there is an urgent need for a method to improve the readability of medical imaging reports. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a system, method, electronic device and storage medium for interpreting medical imaging reports based on a multi-layer deep learning model, so as to solve the problem of poor readability of medical imaging reports in the prior art.

[0006] To achieve the above objectives, an embodiment of the present invention provides a method for interpreting medical imaging reports based on a multi-layer deep learning model, and the method specifically includes:

[0007] Collecting a certain number of historical medical imaging reports, and performing data preprocessing on the historical medical imaging reports to obtain preprocessed data;

[0008] Constructing a multi-layer deep learning model, and inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model;

[0009] The preprocessed data corresponding to the medical image report to be interpreted is input into the optimal multi-layer deep learning model to obtain the medical image report interpretation result corresponding to the medical image report to be interpreted.

[0010] On the basis of the above technical solution, the present invention can also be improved as follows:

[0011] Furthermore, the collecting of a certain number of historical medical imaging reports and performing data preprocessing on the historical medical imaging reports to obtain preprocessed data include:

[0012] Determine whether the text data in the historical medical imaging report will affect the performance of the multi-layer deep learning model. When the text data will not affect the performance of the multi-layer deep learning model, retain the text data; when the text data will affect the performance of the multi-layer deep learning model, remove the text data.

[0013] Further, the multi-layer deep learning model includes an input layer, an embedding layer, a bidirectional long short-term memory network, a transformer module, a conversion layer, a global maximum pooling layer, a full density layer, a regularization layer, and a full density layer with an activation function, which are connected in sequence;

[0014] Receiving the preprocessed data through the input layer;

[0015] Converting the text sequence of the preprocessed data into a vector representation through the embedding layer;

[0016] Extracting sequence features of the text sequence through the bidirectional long short-term memory network;

[0017] Extracting complex sequence features of the text sequence through the transformer module;

[0018] Extracting local features of the text sequence and shortening the length of the text sequence through the conversion layer;

[0019] The features at different positions are fused into a vector representation of fixed length through the global maximum pooling layer;

[0020] Learn higher-level feature representations through full density layers and regularization layers;

[0021] Pass through a full density layer with an activation function to output the probability of a binary classification task.

[0022] Further, the transformer module includes a multi-head attention mechanism and a feed-forward neural network;

[0023] Capturing global dependencies through multi-head attention mechanism;

[0024] Feature transformation and nonlinear mapping are performed via a feed-forward neural network.

[0025] Furthermore, the constructing of a multi-layer deep learning model, inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model, includes:

[0026] Dividing the preprocessed data into a training set, a validation set, and a test set;

[0027] Training the multi-layer deep learning model based on the training set;

[0028] Performing a performance evaluation on the trained multi-layer deep learning model based on the validation set to obtain a multi-layer deep learning model that meets performance requirements;

[0029] The prediction result of the multi-layer deep learning model that meets the performance condition is evaluated based on the test set to obtain an evaluation index corresponding to the multi-layer deep learning model.

[0030] Furthermore, the constructing of a multi-layer deep learning model, inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model, includes:

[0031] Based on the nature of the task, choose an appropriate loss function to evaluate the difference between the model's prediction and the true label to obtain the loss value;

[0032] Use gradient descent or its variants to perform parameter updates based on the calculated gradients;

[0033] Repeat the steps of forward propagation, loss calculation, backpropagation, and parameter update until the performance of the multi-layer deep learning model on the validation set is no longer significantly improved or the preset number of iterations is reached.

[0034] Furthermore, the constructing of a multi-layer deep learning model, inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model, includes:

[0035] Determine whether the multi-layer deep learning model needs fine-tuning, and if so, fine-tune the multi-layer deep learning model based on the training results.

[0036] A system for interpreting medical imaging reports based on a multi-layer deep learning model, comprising:

[0037] The collection module is used to collect a certain number of historical medical imaging reports;

[0038] A preprocessing module, used for performing data preprocessing on the historical medical imaging report to obtain preprocessed data;

[0039] A construction module, used to construct a multi-layer deep learning model, input the preprocessed data into the multi-layer deep learning model for training, so as to obtain an optimal multi-layer deep learning model;

[0040] The interpretation module is used to input the preprocessed data corresponding to the medical image report to be interpreted into the optimal multi-layer deep learning model to obtain the medical image report interpretation result corresponding to the medical image report to be interpreted.

[0041] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the method are implemented when the processor executes the computer program.

[0042] A non-transitory computer-readable storage medium stores a computer program, which implements the steps of the method when executed by a processor.

[0043] The embodiments of the present invention have the following advantages:

[0044] The method for interpreting medical imaging reports based on a multi-layer deep learning model in the present invention collects a certain number of historical medical imaging reports, performs data preprocessing on the historical medical imaging reports to obtain preprocessed data; constructs a multi-layer deep learning model, and inputs the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model; inputs the preprocessed data corresponding to the medical imaging report to be interpreted into the optimal multi-layer deep learning model to obtain a medical imaging report interpretation result corresponding to the medical imaging report to be interpreted, thereby solving the problem of poor readability of medical imaging reports in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0046] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0047] Figure 1A flowchart of a method for interpreting medical imaging reports based on a multi-layer deep learning model according to the present invention;

[0048] Figure 2 The present invention is an architecture diagram of a system for interpreting medical imaging reports based on a multi-layer deep learning model;

[0049] Figure 3 This is an architecture diagram of the multi-layer deep learning model of the present invention;

[0050] Figure 4 A schematic diagram of specific parameters of the multi-layer deep learning model of the present invention;

[0051] Figure 5 This is a schematic diagram of the physical structure of the electronic device provided by the present invention.

[0052] The accompanying drawings are denoted as follows:

[0053] Acquisition module 10 , preprocessing module 20 , construction module 30 , interpretation module 40 , electronic device 50 , processor 501 , memory 502 , bus 503 . DETAILED DESCRIPTION

[0054] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are 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.

[0055] Figure 1 This is a flow chart of an embodiment of a method for interpreting medical imaging reports based on a multi-layer deep learning model according to the present invention. Figure 1 As shown, a method for interpreting medical imaging reports based on a multi-layer deep learning model provided by an embodiment of the present invention includes the following steps:

[0056] S101, collecting a certain number of historical medical imaging reports, performing data preprocessing on the historical medical imaging reports, and obtaining preprocessed data;

[0057] Specifically: determine whether the text data in the historical medical imaging report will affect the performance of the multi-layer deep learning model. When the text data will not affect the performance of the multi-layer deep learning model, retain the text data; when the text data will affect the performance of the multi-layer deep learning model, remove the text data.

[0058] Some text content in historical medical imaging reports that is unnecessary for the target and may affect the performance of the model should be removed. For example: (1) Non-Chinese text: The imaging reports in the database come from multiple databases. This study only focuses on Chinese reports and removes all other text. (2) Other invalid characters and text: such as non-ASCII (American Standard Code for Information Interchange) characters and empty strings. Invalid values ​​are computationally unacceptable to the model.

[0059] S102, constructing a multi-layer deep learning model, and inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model;

[0060] Specifically, Figure 3 and Figure 4 As shown, the multi-layer deep learning model includes an input layer, an embedding layer, a bidirectional long short-term memory network, a transformer module, a conversion layer, a global maximum pooling layer, a full density layer, a regularization layer, and a full density layer with an activation function, which are connected in sequence;

[0061] Receiving the preprocessed data through the input layer;

[0062] Converting the text sequence of the preprocessed data into a vector representation through the embedding layer;

[0063] Extracting sequence features of the text sequence through the bidirectional long short-term memory network;

[0064] The transformer module is used to extract the complex sequence features of the text sequence; in the multi-layer deep learning model structure, the transformer module is applied to the sequence representation output by the long short-term memory network to improve the modeling ability of the multi-layer deep learning model for long-distance dependencies. The stacking and residual connection mechanism of the transformer blocks helps to alleviate the gradient vanishing problem and allows the multi-layer deep learning model to better capture the complex features of the input sequence.

[0065] The transformer module includes a multi-head attention mechanism and a feedforward neural network;

[0066] Capturing global dependencies through multi-head attention mechanism;

[0067] Feature transformation and nonlinear mapping are performed via a feed-forward neural network.

[0068] Applying a one-dimensional transformation layer to the output of the transformer module, extracting local features of the text sequence and shortening the text sequence length through the transformation layer;

[0069] The features at different positions are fused into a vector representation of fixed length through the global maximum pooling layer;

[0070] Learn higher-level feature representations and prevent overfitting through full density layers and regularization layers;

[0071] Pass through a full density layer with an activation function to output the probability of a binary classification task.

[0072] The entire multi-layer deep learning model is built using the functional network interface, where inputs and outputs are defined in the model class, forming an end-to-end trainable multi-layer deep learning model. The architecture of this multi-layer deep learning model incorporates different types of neural network layers, and fully utilizes the respective advantages of structures such as long short-term memory networks, transformers, and convolutional neural networks when processing sequence data, aiming to improve the performance of the model when classifying or labeling text data. At the same time, the use of technical means such as residual connections, multi-head attention mechanisms, and convolution operations enables the multi-layer deep learning model to have better modeling and generalization capabilities, and performs well when processing tasks such as text interpretation.

[0073] Dividing the preprocessed data into a training set, a validation set, and a test set;

[0074] Training the multi-layer deep learning model based on the training set;

[0075] Performing a performance evaluation on the trained multi-layer deep learning model based on the validation set to obtain a multi-layer deep learning model that meets performance requirements;

[0076] The prediction result of the multi-layer deep learning model that meets the performance condition is evaluated based on the test set to obtain an evaluation index corresponding to the multi-layer deep learning model.

[0077] Training a multi-layer deep learning network model is a complex process involving multiple steps aimed at optimizing the parameters of the network to improve its performance on a specific task. The first step is to initialize the network parameters. Before training begins, the weights and biases in the network are usually randomly initialized, which provides a starting point for the learning process. Choose a loss function. Depending on the nature of the task, choose an appropriate loss function to evaluate the difference between the model's predictions and the true labels. In the forward propagation, the input data is fed into the network and processed by weighted summation and activation functions layer by layer until the last layer, where the predicted output is obtained. Calculate the loss. Use the loss function to calculate the difference between the model output and the true label to obtain the loss value. The smaller this value, the more accurate the model's prediction. Backward propagation. Use the chain rule to calculate the gradient of the loss function with respect to each parameter. This process starts from the output layer and goes backwards through each layer of the network. Parameter update: Use gradient descent or its variants to update the parameters of the network according to the calculated gradients. The learning rate is a key hyperparameter in this step and determines the magnitude of the parameter update. Iterative training: Repeat the forward propagation, loss calculation, back propagation, and parameter update steps until the performance of the model on the validation set is no longer significantly improved, or the preset number of iterations is reached. Regularization: In order to prevent overfitting, regularization techniques such as regularization, L1 / L2 regularization, etc. may be applied during the training process. Hyperparameter adjustment: During or before training, it is necessary to select appropriate hyperparameters such as learning rate, batch size, number of network layers, and number of neurons. Model fine-tuning: According to the training results, the model may need to be fine-tuned, such as adjusting the network structure or hyperparameters to further improve performance. The training of multi-layer deep learning network models is a trial-and-error and iterative process that requires careful monitoring and adjustment to obtain the best results. As the training progresses, the model should gradually learn the complex features and patterns in the data, and eventually be able to make accurate analytical references on unseen data.

[0078] S103, inputting the preprocessed data corresponding to the medical image report to be interpreted into the optimal multi-layer deep learning model to obtain a medical image report interpretation result corresponding to the medical image report to be interpreted;

[0079] Specifically, automated measurements and human evaluation are used to track the model loss during the training process, which helps determine whether the training is complete, and to compare the quality of interpretations provided in manual interpretations and the quality of interpretations in automatic replies generated by multi-layer deep learning models.

[0080] The method for interpreting medical image reports based on a multi-layer deep learning model collects a certain number of historical medical image reports, performs data preprocessing on the historical medical image reports to obtain preprocessed data; constructs a multi-layer deep learning model, inputs the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model; inputs the preprocessed data corresponding to the medical image report to be interpreted into the optimal multi-layer deep learning model to obtain a medical image report interpretation result corresponding to the medical image report to be interpreted. The problem of poor readability of medical image reports in the prior art is solved.

[0081] This method of interpreting medical imaging reports based on a multi-layer deep learning model automatically interprets imaging diagnostic reports through artificial intelligence algorithms, without the need for manual intervention by imaging doctors, saving the workload of imaging doctors. The report interpretation is fast, with feedback time of no more than 1 minute. It is convenient and easy to use, with strong readability, making it easy for ordinary users without professional background to understand complex diagnostic reports.

[0082] Figure 2 Flow chart of a system embodiment of the present invention for interpreting medical imaging reports based on a multi-layer deep learning model; Figure 2 As shown, an embodiment of the present invention provides a system for interpreting medical imaging reports based on a multi-layer deep learning model, comprising the following steps:

[0083] The collection module 10 is used to collect a certain number of historical medical imaging reports;

[0084] A preprocessing module 20, used for performing data preprocessing on the historical medical imaging report to obtain preprocessed data;

[0085] The pre-processing module 20 is also used for:

[0086] Determine whether the text data in the historical medical imaging report will affect the performance of the multi-layer deep learning model. When the text data will not affect the performance of the multi-layer deep learning model, retain the text data; when the text data will affect the performance of the multi-layer deep learning model, remove the text data.

[0087] A construction module 30 is used to construct a multi-layer deep learning model, and input the pre-processed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model;

[0088] The multi-layer deep learning model includes an input layer, an embedding layer, a bidirectional long short-term memory network, a transformer module, a conversion layer, a global maximum pooling layer, a full density layer, a regularization layer, and a full density layer with an activation function connected in sequence;

[0089] Receiving the preprocessed data through the input layer;

[0090] Converting the text sequence of the preprocessed data into a vector representation through the embedding layer;

[0091] Extracting sequence features of the text sequence through the bidirectional long short-term memory network;

[0092] Extracting complex sequence features of the text sequence through the transformer module;

[0093] The transformer module includes a multi-head attention mechanism and a feedforward neural network;

[0094] Capturing global dependencies through multi-head attention mechanism;

[0095] Feature transformation and nonlinear mapping are performed via a feed-forward neural network.

[0096] Extracting local features of the text sequence and shortening the length of the text sequence through the conversion layer;

[0097] The features at different positions are fused into a vector representation of fixed length through the global maximum pooling layer;

[0098] Learn higher-level feature representations through full density layers and regularization layers;

[0099] Pass through a full density layer with an activation function to output the probability of a binary classification task.

[0100] The building block 30 is also used for:

[0101] Dividing the preprocessed data into a training set, a validation set, and a test set;

[0102] Training the multi-layer deep learning model based on the training set;

[0103] Performing a performance evaluation on the trained multi-layer deep learning model based on the validation set to obtain a multi-layer deep learning model that meets performance requirements;

[0104] The prediction result of the multi-layer deep learning model that meets the performance condition is evaluated based on the test set to obtain an evaluation index corresponding to the multi-layer deep learning model.

[0105] Based on the nature of the task, choose an appropriate loss function to evaluate the difference between the model's prediction and the true label to obtain the loss value;

[0106] Use gradient descent or its variants to perform parameter updates based on the calculated gradients;

[0107] Repeat the steps of forward propagation, loss calculation, backpropagation, and parameter update until the performance of the multi-layer deep learning model on the validation set is no longer significantly improved or the preset number of iterations is reached.

[0108] Determine whether the multi-layer deep learning model needs fine-tuning, and if so, fine-tune the multi-layer deep learning model based on the training results.

[0109] The interpretation module 40 is used to input the preprocessed data corresponding to the medical image report to be interpreted into the optimal multi-layer deep learning model to obtain the medical image report interpretation result corresponding to the medical image report to be interpreted.

[0110] The system for interpreting medical image reports based on a multi-layer deep learning model of the present invention collects a certain number of historical medical image reports through an acquisition module 10; performs data preprocessing on the historical medical image reports through a preprocessing module 20 to obtain preprocessed data; constructs a multi-layer deep learning model through a construction module 30, inputs the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model; and inputs the preprocessed data corresponding to the medical image report to be interpreted into the optimal multi-layer deep learning model through an interpretation module 40 to obtain a medical image report interpretation result corresponding to the medical image report to be interpreted. The method for interpreting medical image reports based on a multi-layer deep learning model solves the problem of poor readability of medical image reports in the prior art.

[0111] Figure 5 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as Figure 5 As shown, the electronic device 50 includes: a processor 501 (processor), a memory 502 (memory) and a bus 503;

[0112] The processor 501 and the memory 502 communicate with each other via the bus 503;

[0113] The processor 501 is used to call the program instructions in the memory 502 to execute the methods provided by the above-mentioned method embodiments, for example, including: collecting a certain number of historical medical imaging reports, performing data preprocessing on the historical medical imaging reports to obtain preprocessed data; constructing a multi-layer deep learning model, and inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model; inputting the preprocessed data corresponding to the medical imaging report to be interpreted into the optimal multi-layer deep learning model to obtain a medical imaging report interpretation result corresponding to the medical imaging report to be interpreted.

[0114] The present embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions. The computer instructions enable the computer to execute the methods provided by the above-mentioned method embodiments, for example, including: collecting a certain number of historical medical imaging reports, performing data preprocessing on the historical medical imaging reports, and obtaining preprocessed data; constructing a multi-layer deep learning model, inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model; inputting the preprocessed data corresponding to the medical imaging report to be interpreted into the optimal multi-layer deep learning model, and obtaining the medical imaging report interpretation result corresponding to the medical imaging report to be interpreted. A person of ordinary skill in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other storage media that can store program codes.

[0115] The device embodiments described above are merely illustrative, wherein 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, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0116] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0117] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for interpreting medical imaging reports based on a multi-layer deep learning model, characterized in that: The method for interpreting medical imaging reports based on a multi-layer deep learning model specifically includes: Collecting a certain number of historical medical imaging reports, and performing data preprocessing on the historical medical imaging reports to obtain preprocessed data; Constructing a multi-layer deep learning model, and inputting the preprocessed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model; The preprocessed data corresponding to the medical image report to be interpreted is input into the optimal multi-layer deep learning model to obtain the medical image report interpretation result corresponding to the medical image report to be interpreted.

2. The method for interpreting medical imaging reports based on a multi-layer deep learning model according to claim 1, characterized in that: The collecting of a certain number of historical medical imaging reports and performing data preprocessing on the historical medical imaging reports to obtain preprocessed data include: Determine whether the text data in the historical medical imaging report will affect the performance of the multi-layer deep learning model. When the text data will not affect the performance of the multi-layer deep learning model, retain the text data; when the text data will affect the performance of the multi-layer deep learning model, remove the text data.

3. The method for interpreting medical imaging reports based on a multi-layer deep learning model according to claim 1, characterized in that: The multi-layer deep learning model includes an input layer, an embedding layer, a bidirectional long short-term memory network, a transformer module, a conversion layer, a global maximum pooling layer, a full density layer, a regularization layer, and a full density layer with an activation function connected in sequence; Receiving the preprocessed data through the input layer; Converting the text sequence of the preprocessed data into a vector representation through the embedding layer; Extracting sequence features of the text sequence through the bidirectional long short-term memory network; Extracting complex sequence features of the text sequence through the transformer module; Extracting local features of the text sequence and shortening the length of the text sequence through the conversion layer; The features at different positions are fused into a vector representation of fixed length through the global maximum pooling layer; Learn higher-level feature representations through full density layers and regularization layers; Pass through a full density layer with an activation function to output the probability of a binary classification task.

4. The method for interpreting medical imaging reports based on a multi-layer deep learning model according to claim 3, characterized in that: The transformer module includes a multi-head attention mechanism and a feedforward neural network; Capturing global dependencies through multi-head attention mechanism; Feature transformation and nonlinear mapping are performed via a feed-forward neural network.

5. The method for interpreting medical imaging reports based on a multi-layer deep learning model according to claim 1, characterized in that: The constructing of a multi-layer deep learning model, inputting the pre-processed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model, includes: Dividing the preprocessed data into a training set, a validation set, and a test set; Training the multi-layer deep learning model based on the training set; Performing a performance evaluation on the trained multi-layer deep learning model based on the validation set to obtain a multi-layer deep learning model that meets performance requirements; The prediction result of the multi-layer deep learning model that meets the performance condition is evaluated based on the test set to obtain an evaluation index corresponding to the multi-layer deep learning model.

6. The method for interpreting medical imaging reports based on a multi-layer deep learning model according to claim 1, characterized in that: The constructing of a multi-layer deep learning model, inputting the pre-processed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model, includes: Based on the nature of the task, choose an appropriate loss function to evaluate the difference between the model's prediction and the true label to obtain the loss value; Use gradient descent or its variants to perform parameter updates based on the calculated gradients; Repeat the steps of forward propagation, loss calculation, backpropagation, and parameter update until the performance of the multi-layer deep learning model on the validation set is no longer significantly improved or the preset number of iterations is reached.

7. The method for interpreting medical imaging reports based on a multi-layer deep learning model according to claim 1, characterized in that: The constructing of a multi-layer deep learning model, inputting the pre-processed data into the multi-layer deep learning model for training to obtain an optimal multi-layer deep learning model, includes: Determine whether the multi-layer deep learning model needs fine-tuning, and if so, fine-tune the multi-layer deep learning model based on the training results.

8. A system for interpreting medical imaging reports based on a multi-layer deep learning model, characterized in that: include: The collection module is used to collect a certain number of historical medical imaging reports; A preprocessing module, used for performing data preprocessing on the historical medical imaging report to obtain preprocessed data; A construction module, used to construct a multi-layer deep learning model, input the preprocessed data into the multi-layer deep learning model for training, so as to obtain an optimal multi-layer deep learning model; The interpretation module is used to input the preprocessed data corresponding to the medical image report to be interpreted into the optimal multi-layer deep learning model to obtain the medical image report interpretation result corresponding to the medical image report to be interpreted.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Citation Information

Cited By

  • Analysis method, system and equipment for nuclear medicine image radiation report and medium

    CN119851962A

  • Analysis Method, System, Device and Medium for Nuclear Medicine Imaging Radiation Report

    CN119851962B