Medical image analysis method based on deep learning

Abstract

The invention discloses a medical image analysis method based on deep learning, and belongs to the field of medical image processing and artificial intelligence. According to the method, firstly, a pre-trained OfficientNet model is used as a feature extractor, and feature extraction is carried out on a medical image; through a self-defined classifier, a network structure is adjusted according to different types of medical images so as to adapt to detection requirements of different diseases. In the training stage, a data enhancement technology is adopted to process training data, the generalization ability of the model is improved, and meanwhile, a learning rate scheduler is introduced to optimize the training process. In a test stage, the model performs disease detection on an input medical image, outputs the probability of disease occurrence, and stores a result in a CSV file. In addition, the method is also combined with a large model, and corresponding suggestions are obtained after a detection result is given. In addition, the invention further provides a detection process, the operation mode of the model is flexibly configured through command line parameters, and training, testing and image analysis of the model are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of medical image processing and artificial intelligence, and particularly relates to a medical image analysis method based on deep learning. Background Art

[0002] The analysis of medical images (such as CT, MRI, PET, etc.) is crucial for the early detection and treatment of diseases. Traditional medical image analysis methods mainly rely on doctors' professional knowledge and experience. Although this method can provide accurate diagnostic results to a certain extent, there are also some limitations. First, doctors may experience fatigue after long hours of work, resulting in a decline in the accuracy and efficiency of diagnosis. Second, for some complex diseases, especially early lesions, even experienced doctors may miss or misdiagnose. In addition, with the continuous increase in the amount of medical image data, the efficiency of manual analysis is difficult to meet clinical needs.

[0003] In recent years, with the rapid development of artificial intelligence technology, deep learning has been widely applied in the field of medical image analysis. Deep learning models can automatically learn the features in images, thereby realizing the automatic detection and diagnosis of diseases. However, there are still some problems with existing deep learning methods in medical image disease detection. For example, some methods need to manually adjust the model structure when processing different types of medical images, which increases the complexity and training difficulty of the model. In addition, for some rare diseases or early lesions, due to the lack of training data, the generalization ability of the model may be affected, resulting in a decline in the detection accuracy.

[0004] To address the above problems, the present invention proposes a medical image analysis method based on deep learning, aiming to improve the accuracy and efficiency of medical image analysis while reducing the dependence on doctors' professional knowledge. Summary of the Invention

[0005] The purpose of the present invention is to provide a medical image analysis method based on deep learning for the deficiencies existing in the existing medical image analysis technology, such as high dependence on doctors' professional knowledge, low detection efficiency, and low detection accuracy for rare diseases or early lesions.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A medical image analysis method based on deep learning, comprising:

[0008] S1. Obtain a medical image sample set, where the medical images include but are not limited to CT images, MRI images, PET images, X-Ray images, and slice images;

[0009] S2. Preprocess the medical image sample set, including operations such as resizing the images and normalizing them;

[0010] S3. Use the pre-trained EfficientNet model to extract features from the preprocessed medical images;

[0011] S4. According to the type of medical images, classify the extracted features through a custom classifier and output the analysis results;

[0012] S5. Combine the analysis results with a large model to obtain life suggestions given by the large model;

[0013] S6. Save the analysis results to a specified file.

[0014] Through the use of deep learning techniques, especially the pre-trained EfficientNet model and custom classifier, the present invention can automatically extract features from medical images and perform image analysis, improving the accuracy and efficiency of the analysis and reducing the dependence on doctors' professional knowledge. At the same time, by combining the life suggestions of the large model, more comprehensive health guidance is provided for patients.

[0015] Further, the preprocessing in S2 also includes resizing the images: Resize all images to 224x224 pixels; Normalization: Use the ImageOps.fit method for normalization processing; Data augmentation operations, which include random horizontal flipping, random vertical flipping, random rotation, random cropping, Gaussian blur, color jitter, and random erasing. These data augmentation operations can increase the generalization ability of the model and improve the adaptability of the model to different image variations.

[0016] Further, the EfficientNet model in S3 is initialized with pre-trained weights. The pre-trained weights can help the model converge faster, improve the training efficiency, and also contribute to improving the detection performance of the model.

[0017] Further, the custom classifier in S4 includes multiple fully connected layers and activation functions, and adjusts the classification head according to the type of medical images. There are different classification heads for different medical images, and each classification head has different processing for different medical images. This design can perform targeted image analysis according to the characteristics of different types of medical images and improve the accuracy of image analysis.

[0018] Further, the image analysis results in S4 include the probability of the occurrence of a disease. By outputting the probability of the occurrence of a disease, more detailed information can be provided for doctors to help doctors make more accurate judgments.

[0019] Furthermore, the life advice given by the large model in S5. Combining the life advice of the large model can provide personalized health guidance for patients and help patients better manage their health.

[0020] Furthermore, the specified file in S6 is a CSV format file, and the file contains image IDs and corresponding image analysis results.

[0021] {q i F j}

[0022] Among them, q i represents a certain query, and F j represents the prediction result of sample j and the advice of the large model. This format facilitates data storage and subsequent analysis.

[0023] Furthermore, it also includes a model training step, and the model training step includes:

[0024] S1. Construct a training data set and a validation data set, and the training data set and the validation data set include medical images and their corresponding labels;

[0025] S2. Define a loss function and an optimizer, the loss function is a cross-entropy loss function, and the optimizer is an Adam optimizer;

[0026] S3. Use the training data set to train the model, and at the same time use the validation data set for validation, and adjust the model parameters according to the validation results;

[0027] S4. Save the trained model weights.

[0028] Furthermore, during the training process in S3, it also includes using the Adam optimizer with a learning rate of 0.001 and a weight decay of 1e-4. Learning rate scheduling, and the learning rate scheduling adopts the StepLR strategy, multiplying the learning rate by 0.1 every 10 epochs. Learning rate scheduling can help the model better adjust the learning rate during the training process and improve the training effect.

[0029] Furthermore, during the training process in S3, it also includes performing gradient clipping (nn.utils.clip_grad_norm_) after each epoch to prevent gradient explosion. Gradient clipping can prevent the problem of gradient explosion during the training process and improve the stability and training effect of the model.

[0030] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are:

[0031] By leveraging deep learning techniques, particularly the pre-trained EfficientNet model and a custom classifier, the present invention can automatically extract features from medical images for disease detection, improving the accuracy and efficiency of detection and reducing the dependence on doctors' professional knowledge.

[0032] Through data augmentation operations and the use of pre-trained weights, the generalization ability and training efficiency of the model are improved, enabling the model to better adapt to different types of medical images.

[0033] Through the design of a custom classifier, targeted disease detection is performed based on different types of medical images, further improving the accuracy of detection and ensuring the effectiveness of the model in different application scenarios.

[0034] By outputting the probability of disease occurrence and combining with life suggestions from large models, more comprehensive information is provided for doctors and patients, helping with better diagnosis and health management and enhancing the scientific nature and practicality of medical decision-making.

[0035] Through the optimization of the model training steps, including learning rate scheduling and gradient clipping, the training effect and stability of the model are improved, ensuring the reliability and accuracy of the model in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a flowchart of the deep learning-based medical image analysis method described in Embodiment 1;

[0038] Figure 2 It is a reference schematic diagram of the data file types supported by the deep learning-based medical image analysis described in Embodiment 1;

[0039] Figure 3 It is a schematic diagram of the model architecture;

[0040] Figure 4 It is a schematic diagram of the structure of the custom classifier;

[0041] Figure 5 It is a schematic diagram of the output format of the analysis results in Embodiment 1;

[0042] Figure 6 It is a graph showing the changes in loss and accuracy during the model training process in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention usually described and illustrated in the accompanying drawings herein can be arranged and designed in various different configurations.

[0044] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0045] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, article or device including the said element.

[0046] The features and performance of the present invention will be further described in detail below in conjunction with the embodiments.

[0047] Refer to Figure 1 , a preferred embodiment of the present invention provides a medical image analysis method based on deep learning, including:

[0048] S1. Obtain a medical image sample set, where the medical images include but are not limited to CT images, MRI images, and PET images;

[0049] S2. Preprocess the medical image sample set, including operations such as adjusting the image size, normalizing, and data augmentation operations, such as random horizontal flipping, random vertical flipping, random rotation, random cropping, Gaussian blur, color jitter, and random erasing;

[0050] S3. Use a pre-trained EfficientNet model to extract features from the preprocessed medical images, and the architecture of the model is as Figure 4as shown;

[0051] S4. Classify the extracted features through a custom classifier according to the type of medical image, and output the analysis result. The structure of the custom classifier is as Figure 4 shown;

[0052] S5. Combine the analysis result with the large model to obtain life suggestions given by the large model;

[0053] S6. Save the analysis result to a specified file, and the file format is as Figure 5 shown.

[0054] Among them, the reference schematic diagram of the data file type of the medical image sample set is as Figure 2 shown.

[0055] In step S3 of this embodiment, the EfficientNet model is initialized with pre-trained weights. The pre-trained weights can help the model converge faster, improve the training efficiency, and also help improve the detection performance of the model.

[0056] In step S4 of this embodiment, the custom classifier includes multiple fully connected layers and activation functions. The classification head is adjusted according to the type of medical image, and there are different classification heads for different medical images. Each classification head has different processing for different medical images. For CT images: 4 categories are output (no cancer, squamous cell carcinoma, large cell carcinoma, adenocarcinoma), for MRI images: 2 categories are output (no cancer, cancer), for PET images: 3 categories are output (no abnormality, COVID-19, other viral pneumonia), for X-Ray images: 2 categories are output (no abnormality, COVID-19), for slice images: 3 categories are output (no cancer, lung adenocarcinoma, lung squamous cell carcinoma). This design can perform targeted disease detection according to the characteristics of different types of medical images and improve the detection accuracy.

[0057] In step S5 of this embodiment, the life suggestions combined with the large model can provide personalized health guidance for patients and help patients better manage their health.

[0058] In step S6 of this embodiment, the disease detection result includes the probability of the occurrence of the disease. By outputting the probability of the occurrence of the disease, more detailed information can be provided for doctors to help doctors make more accurate diagnoses.

[0059] In addition, the present invention also includes a model training step, and the model training step includes:

[0060] S1. Construct a training data set and a validation data set, and the training data set and the validation data set include medical images and their corresponding labels;

[0061] S2. Define the loss function and the optimizer. The loss function is the cross-entropy loss function, and the optimizer is the Adam optimizer;

[0062] S3. Use the training dataset to train the model and at the same time use the validation dataset for validation. Adjust the model parameters according to the validation results. The training process also includes learning rate scheduling, which adopts the StepLR strategy, and gradient clipping to prevent gradient explosion;

[0063] S4. Save the weights of the trained model.

[0064] In step S3 of this embodiment, the graphs of the loss and accuracy changes during the training process are as Figure 6 shown. Through these curves, the performance changes of the model during the training process can be intuitively observed.

[0065] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A medical image analysis method based on deep learning, characterized in that: include: S1. Acquire a sample set of medical images, including but not limited to CT images, MRI images, PET images, X-ray images, and slice images; S2. Preprocessing the medical image sample set, including adjusting image size and normalization; S3. Use the pre-trained EfficientNet model to extract features from the pre-processed medical images; S4. According to the type of medical image, the extracted features are classified by a custom classifier and the analysis results are output; S5. Combine the final analysis results with the big model to finally get the life suggestions given by the big model; S6. Save the analysis results to a specified file.

2. The medical image analysis method based on deep learning according to claim 1, characterized in that: The medical image sample set includes but is not limited to the following types: MRI images: label 0 indicates no cancer, label 1 indicates cancer; CT images: label 0 indicates no cancer, label 1 indicates squamous cell carcinoma, label 2 indicates large cell carcinoma, and label 3 indicates adenocarcinoma; PET images: label 0 indicates no abnormality, label 1 indicates COVID-19, and label 2 indicates other viral pneumonia; -X-Ray images: label 0 indicates no abnormality, label 1 indicates COVID-19; slice images: label 0 indicates no cancer, label 1 indicates lung adenocarcinoma, and label 2 indicates lung squamous cell carcinoma.

3. The medical image analysis method based on deep learning according to claim 1, characterized in that: The preprocessing in S2 includes the following operations: increasing the robustness of the model to left-right flipped images through random horizontal flipping; increasing the robustness of the model to up-down flipped images through random vertical flipping; simulating shooting at different angles through random rotation; simulating images with different fields of view through random cropping; simulating image blur through Gaussian blur; simulating different lighting conditions through color jittering; simulating local occlusion of the image through random erasing.

4. The medical image analysis method based on deep learning according to claim 1, characterized in that: The EfficientNet model in S3 is initialized using pre-trained weights.

5. The medical image analysis method based on deep learning according to claim 1, characterized in that: The custom classifier in S4 includes multiple fully connected layers and activation functions. The classification head is adjusted according to the type of medical image. Different classification heads are used for different medical images, and each classification head processes different medical images differently.

6. The medical image analysis method based on deep learning according to claim 1, characterized in that: The analysis results in S4 include the probability of disease occurrence.

7. The medical image analysis method based on deep learning according to claim 1, characterized in that: The file specified in S6 is a CSV format file, which contains the image ID and the corresponding analysis results. {q i F j } Among them, q i Represents a query, F j Represents the prediction result and of sample j.

8. The medical image analysis method based on deep learning according to claim 1, characterized in that: The training steps of the model include: S1. Constructing a training data set and a validation data set, wherein the training data set and the validation data set include medical images and their corresponding labels; S2. define a loss function and an optimizer, wherein the loss function is a cross entropy loss function and the optimizer is an Adam optimizer; S3. Use the training data set to train the model, use the validation data set to validate it, and adjust the model parameters according to the validation results; S4. Save the trained model weights.

9. The medical image analysis method based on deep learning according to claim 7, characterized in that: The training process in S3 also includes learning rate scheduling, and the learning rate scheduling adopts the StepLR strategy.

10. The medical image analysis method based on deep learning according to claim 7, characterized in that: The training process in S3 also includes gradient clipping.

Images