Medical image similarity detection method adopting neural network feature matching, electronic equipment, computer readable storage medium and computer software product

Abstract

The invention discloses a medical image similarity detection method adopting neural network feature matching, electronic equipment, a computer readable storage medium and a computer software product. The method comprises the following steps: performing type identification on an input medical paper image by using a trained convolutional neural network model, and dividing the input medical paper image into four types: a streaming image, a fluorescence image, a particle fluorescence image and other images; respectively carrying out image preprocessing on the four types of images; performing feature extraction on the preprocessed different types of images, and matching the extracted features; judging whether the two images are similar or not according to the matched feature points and the proportion of the matched points in the total points; if yes, further analyzing transformation parameters; and for the detected similar image pairs, feature matching points are drawn, similar areas are marked, and then a result image is visually output and output.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to a medical image similarity detection method using neural network feature matching, as well as electronic equipment, computer-readable storage medium, and computer software product. Background Art

[0002] In modern medical research and clinical practice, image data, as an important data carrier, is widely used in disease diagnosis, efficacy evaluation, scientific research paper publication, and other aspects. In medical research papers in particular, images are not only used to visually demonstrate research results, but also often serve as key evidence to support experimental conclusions. However, during the paper writing and submission process, the reuse, misuse, and even falsification of images are common, seriously undermining academic integrity and the quality of scientific research. Therefore, establishing an efficient, accurate, and intelligent image similarity detection mechanism is of great significance for enhancing the authenticity and credibility of scientific research results.

[0003] Currently, most medical institutions and scientific research institutions rely mainly on manual review to check the duplication and similarity of images in papers. This method is not only time-consuming and labor-intensive, but also easily influenced by subjective judgment, resulting in a high rate of missed detection or misjudgment. In addition, some hospitals have tried to use image duplication detection services provided by third parties. Although this has improved detection efficiency to a certain extent, it has also brought many problems: such as expensive service fees, long response cycles, inability to meet large-scale concurrent processing needs, and the most core issue of data privacy and security. Since medical images often contain sensitive patient information, if they are handed over to external institutions for processing, they are very likely to lead to the risk of data leakage. This potential threat is particularly prominent in unpublished research results.

[0004] Existing image similarity detection technologies mainly include traditional image processing algorithms, such as histogram comparison, structural similarity index (SSIM), hash algorithm, SIFT (Scale Invariant Feature Transform) and SURF (Speeded Robust Features) and other feature point matching methods. Figure 2 These methods have achieved some success in general image processing, but they face significant limitations in the specialized application of medical images. For example, medical images typically have high resolution and complex internal structures, making it difficult for traditional methods to effectively capture their deep semantic information. Furthermore, during the acquisition process, medical images often experience scale changes, rotational offsets, and brightness differences due to factors such as device parameters, shooting angles, and imaging conditions. These issues can lead to significant appearance differences between images, thus affecting the accuracy of similarity judgments.

[0005] Furthermore, current technologies generally lack the ability to automatically identify and classify image types, making it impossible to differentiate between different modalities (such as X-ray, CT, MRI, and ultrasound), further reducing the system's applicability and generalization capabilities. Faced with the ever-increasing volume of medical image data and the diverse range of image types, there is an urgent need for an intelligent image similarity detection system capable of adaptive learning and high-precision recognition.

[0006] More importantly, the present invention can be deployed in the hospital's local server environment, completely avoiding the privacy risks associated with uploading sensitive data to the cloud or a third-party platform. Local deployment not only ensures the security and controllability of data, but also significantly reduces long-term operating costs and enhances the hospital's ability to independently manage the image duplication checking system. By introducing artificial intelligence technology, the present invention is expected to significantly improve the automation level of image review, reduce manual intervention, improve overall work efficiency, and provide strong support for quality control in medical research.

[0007] In summary, the present invention solves the problems of low image recognition accuracy, poor adaptability, insufficient privacy protection, etc. in the prior art, and proposes an efficient, intelligent and secure solution with good application prospects and promotion value. Summary of the Invention

[0008] The purpose of the embodiments of the present application is to provide a medical image similarity detection method using neural network feature matching, as well as an electronic device, a computer-readable storage medium, and a computer software product.

[0009] A medical image similarity detection method using neural network feature matching, comprising: The trained convolutional neural network model is used to identify the types of input medical paper images and classify them into four categories: flow images, fluorescence images, particle fluorescence images, and other images; Image preprocessing is performed on each of the four image types. Preprocessing of flow images includes: cutting edges, removing coordinate axes, improving resolution, and grayscale; preprocessing of fluorescence images / particle fluorescence images and other images includes: reducing resolution, grayscale, and balancing brightness. Perform feature extraction on different types of pre-processed images and match the extracted features. For streaming images, SIFT algorithm is used for feature extraction and matching; for fluorescence images / particle fluorescence images and other images, AKAZE algorithm is used for feature extraction and matching. Determine whether two images are similar based on the number of matched feature points and the ratio of matched points to the total number of points; if they are similar, further analyze their transformation parameters; For the detected similar image pairs, feature matching points are drawn, similar areas are marked, and the resulting images are visualized and output.

[0010] An electronic device includes a processor, a memory, and a computer program that can be run on the processor, and the processor implements the above method when executing the program.

[0011] A computer-readable storage medium stores a computer program, which implements the above method when executed by a processor.

[0012] A computer software product comprises a computer program / instruction, which implements the above method when executed by a processor.

[0013] The above embodiment of the present application proposes a medical image similarity detection method based on the combination of deep learning and feature matching, which has the following significant technical effects: 1. Highly accurate intelligent classification and similarity detection Deep learning technology is used to automatically classify medical images, and targeted preprocessing and feature matching strategies are adopted according to different categories of images, which effectively improves the accuracy of image similarity detection.

[0014] In actual tests, the overall accuracy of this method in image classification and similarity judgment can reach more than 90%, and it has good recognition capabilities.

[0015] 2. Efficient batch processing capability significantly improves detection efficiency The system supports automated and batch processing of large amounts of medical paper image data without the need for human intervention.

[0016] The image processing speed can reach 20 images per minute, which greatly shortens the time cycle of image review and significantly improves the efficiency of duplicate checking and quality control.

[0017] 3. Strong robustness, adaptable to complex image changes Through systematic image preprocessing (such as scale normalization, grayscale conversion, brightness balance, etc.) and post-processing (such as feature point annotation and region visualization), we can effectively address problems such as multi-scale, rotation, and noise interference in medical images.

[0018] The algorithm's adaptability to different imaging conditions and image transformations has been improved, and the stability and robustness of the system have been enhanced.

[0019] 4. Intuitive visual output to assist manual review For the detected similar image pairs, the system can automatically generate a visual result map containing feature matching points and corresponding area annotations.

[0020] The visual information is clear and intuitive, making it easy for researchers or reviewers to quickly locate similar content, providing strong support for the authenticity review of medical papers.

[0021] 5. Low cost, high security, and data privacy protection Compared with the traditional image duplication checking method that relies on third-party services, the present invention can be deployed on the hospital's local server to avoid uploading sensitive images to external platforms.

[0022] It significantly reduces the hospital's image duplication checking costs, while greatly improving the privacy protection and security of medical data and preventing the risk of data leakage.

[0023] 6. Modular design, easy to expand and maintain The system adopts a modular architecture design with clear responsibilities and low coupling between functional modules, which facilitates flexible deployment, functional expansion and system upgrades based on the actual needs of the hospital.

[0024] It supports high availability and long-term maintainability, and is suitable for promotion and application in various medical institutions.

[0025] In summary, the present invention not only excels in image recognition accuracy, processing efficiency, and robustness, but also takes into account data security, cost control, and system scalability, and has broad application prospects and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0027] Figure 1 This is a flow chart of a method in one embodiment of the present application; Figure 2 This is a schematic diagram of a histogram comparison in one embodiment of the present application; Figure 3 This is a schematic diagram showing matching points of two images in one embodiment of the present application; Figure 4 A schematic diagram of similar parts in the two figures in one embodiment of the present application; Figure 5 It is a “matching line” between two images in one embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0029] This application provides a medical image similarity detection method based on neural network feature matching, aiming to address existing issues such as the difficulty in effectively identifying duplicate or highly similar medical images, weak image classification capabilities, low duplication detection accuracy, and insufficient data privacy protection. This method combines the advantages of deep convolutional neural networks with traditional image feature extraction algorithms to achieve automatic classification and refined similarity detection of medical images, and supports differential processing of multiple image types.

[0030] This application provides a medical image similarity detection method using neural network feature matching, such as Figure 1 Shown, including: S110: Use the trained convolutional neural network model to identify the type of input medical paper images and divide them into four categories: flow images, fluorescence images, particle fluorescence images, and other images.

[0031] First, image classification is performed. This stage uses a trained convolutional neural network (CNN) model to identify the type of input medical paper images, classifying them into four categories: streaming images, fluorescent images, granular fluorescent images, and other images. This allows for differentiated processing strategies based on these image categories. During the image classification stage, a trained convolutional neural network (CNN) model is used to identify the type of input medical paper images. This process covers not only the basic principles and applications of CNNs, but also model design and selection, training procedures, and applicability analysis.

[0032] Convolutional neural networks are deep learning models specifically designed for processing data with grid-like structures, such as images. CNNs automatically learn and extract feature representations from input data by mimicking the workings of the human visual system. They have achieved remarkable success in computer vision tasks, including but not limited to image classification, object detection, and image segmentation. They automatically learn feature representations of input data through a series of convolutional, pooling, and fully connected layers, ultimately achieving tasks such as classification or regression. The primary advantage of CNNs is their ability to effectively capture the spatial hierarchical structure of input data while reducing the number of parameters, thus avoiding overfitting.

[0033] The basic components of CNN include: 1. Convolutional Layer Function: The convolution layer is one of the core components of CNN. It uses a set of learnable filters (also called kernels or convolution kernels) to scan the input data to generate a feature map. Each filter focuses on a specific type of feature, such as edges or textures.

[0034] Connection method: Each neuron in the convolutional layer is connected only to a local region of the input data. This local connectivity reduces the number of parameters, helps improve computational efficiency and reduces the risk of overfitting.

[0035] 2. Activation Function Layer Function: Usually follows the convolutional layer, using nonlinear activation functions (such as ReLU, Sigmoid, etc.) to increase the nonlinear expression ability of the network, enabling the model to learn more complex patterns.

[0036] 3. Pooling Layer Purpose: Pooling layers are used to reduce the spatial size of feature maps, reduce computational complexity, and control overfitting. The most commonly used pooling operation is max pooling, which selects the maximum value within each sliding window as the output, thereby retaining the most important information.

[0037] 4. Normalization Layer Function: Like Batch Normalization, it is used to accelerate the training process, stabilize gradient descent, and allow higher learning rates, thereby improving model performance.

[0038] 5. Fully Connected Layer Function: In the final stage of a CNN, the fully connected layer maps all features to a fixed-length vector and then makes a final prediction based on this vector. This is a key step in achieving classification or regression tasks.

[0039] 6. Dropout layer Function: To prevent overfitting, the Dropout layer randomly "discards" some neurons, forcing the network to learn more robust feature representations.

[0040] In a typical CNN architecture, input data first passes through a series of convolutional and pooling layers. Each layer takes input from the previous layer and produces a new feature map as output to the next layer. Convolutional layers extract features, while pooling layers reduce the size of feature maps. Alternating between these two layers effectively captures information at multiple scales. Finally, after several rounds of convolution and pooling, the feature map is flattened and passed through one or more fully connected layers to make the final classification decision.

[0041] Although the general structure of CNN is mentioned above, for a specific problem to be solved, the structure of CNN needs to be designed according to the characteristics of the problem, mainly including determining the number and configuration of convolutional layers, pooling layers, and fully connected layers.

[0042] In this application, the convolutional neural network model preferably comprises three convolutional layers, three pooling layers, and two fully connected layers. The convolutional layers are used to extract local texture and structural features of the image, the pooling layers are used to reduce image dimensionality and enhance scale invariance, and the fully connected layers are responsible for the final image classification decision. The model is trained and optimized using a large amount of annotated data, achieving high accuracy in medical image classification.

[0043] The three convolutional layers allow the model to gradually extract features from low-level to high-level. For example, each convolutional layer is equipped with a certain number of filters that gradually extract features from low-level to high-level. The first layer may learn basic features such as edges and colors, while subsequent layers can capture more complex shapes and patterns. The three pooling layers help further compress the amount of information while retaining the most important features.

[0044] The three pooling layers mentioned above can be followed by a pooling layer after each convolutional layer. This helps further compress the spatial size of the feature map while retaining the most important feature information. The presence of pooling layers also increases the model's invariance to input transformations (such as translation and rotation).

[0045] A two-layer fully connected layer design, where the final set of convolutional and pooling layers can be followed by two fully connected layers. The first fully connected layer can map all learned features into a fixed-length vector, while the second fully connected layer makes the final classification decision based on this vector. This design enables the model to perform complex nonlinear transformations in high-dimensional space, thereby improving classification accuracy.

[0046] In a preferred embodiment of the present application, a CNN architecture comprising three convolutional layers, three pooling layers, and two fully connected layers is adopted. This design is intended to effectively extract features from medical images and adapt to the unique challenges of medical images, such as high resolution and complex internal structures. Once the network structure is determined, the next step is the training process, which is to use a large amount of labeled data to learn the CNN of the structure, and adjust the network weights through the back-propagation algorithm to minimize the error between the predicted results and the actual labels. In short, a specific CNN structure is designed first, and then this structure is trained using a large-scale labeled data set in a specific field, and the model parameters are gradually optimized to finally obtain a model that can make accurate predictions for specific problems. This process ensures that the model can not only learn the basic features of the input data, but also show efficiency and accuracy when facing specific applications. The summary is as follows: Efficient feature extraction: The design of three convolutional layers plus three pooling layers is very suitable for processing detailed features at different scales, which is particularly important for medical images because medical images often contain rich detailed information.

[0047] Reducing overfitting: Designing an appropriate number of layers can help prevent the model from overfitting the training data while maintaining sufficient expressive power to distinguish different image categories.

[0048] Flexibility and scalability: This architecture can be easily adjusted according to specific needs, such as adding more convolutional layers or changing the number of units in the fully connected layer to suit specific application scenarios.

[0049] In summary, the design of three convolutional layers, three pooling layers, and two fully connected layers is based on a deep understanding of the characteristics of medical images and a comprehensive consideration of model performance. This design ensures that the model can effectively learn the key features of the input image while balancing computational efficiency and generalization capabilities, making it an ideal choice for medical image classification tasks.

[0050] The training process of the CNN model includes: 1. Data Preparation: First, you need to collect a large number of labeled medical images as a training set. These images should cover all target categories (streaming images, fluorescent images, granular fluorescent images, and other types of medical images) and be as diverse as possible to ensure good generalization of the model.

[0051] Furthermore, obtaining sufficient labeled data for training machine learning models in the medical field is often a challenge. Because the labeling process requires specialized knowledge and data privacy and security issues must be properly addressed, obtaining large amounts of high-quality training data is often difficult. Transfer learning offers an effective solution to this problem. Transfer learning is a machine learning method that applies a model previously trained on one task to a related but different task. This approach leverages the feature representations learned by the pre-trained model to accelerate learning for the new task and reduce the need for large-scale labeled data. For medical image analysis, this means using deep learning models trained on larger, more diverse image datasets, such as ImageNet, as a starting point and then fine-tuning them for specific medical image classification tasks. In summary, when faced with the challenge of insufficient medical image data, transfer learning provides a powerful and flexible framework, enabling the development of high-performance image classification systems even with limited resources. This approach not only saves significant time and costs but also contributes to advances in the field of medical image analysis.

[0052] 2. Initialize weights: Initialize the weights in all convolutional layers and fully connected layers with random values.

[0053] Forward propagation: Input an image and output the predicted category probability distribution after processing through each layer.

[0054] Loss calculation: The cross entropy loss function is used to measure the difference between the predicted results and the true labels.

[0055] Backpropagation: Adjust the weights in the network based on the loss gradient to minimize the loss function.

[0056] Optimization algorithm: Commonly used algorithms such as Adam (adaptive gradient algorithm) or SGD (Stochastic Gradient Descent) can be used for optimization. These algorithms can dynamically adjust the learning rate to help the model converge faster.

[0057] 3. Hyperparameter tuning: To achieve optimal performance, you may need to adjust hyperparameters such as learning rate, batch size, convolution kernel size, etc. This is usually done through experimental verification.

[0058] The convolutional neural network (CNN) model trained using the above scheme can identify the type of input medical paper images and classify them into four categories: flow images, fluorescence images, particle fluorescence images, and other images. This can be achieved by following the following steps: 1. Data Preprocessing Before inputting images into a trained CNN model, data preprocessing is required. This includes, but is not limited to, resizing the images to fit the model's input dimensions and normalizing pixel values. For certain types of medical images, additional preprocessing steps, such as grayscale conversion and brightness adjustment, may be required to better extract features.

[0059] 2. Input to CNN model The preprocessed image is fed into the CNN model as input. The first part of the model is a feature extractor consisting of multiple convolutional and pooling layers. These layers automatically learn local features in the input image and, by stacking multiple layers, gradually build up more complex and abstract feature representations.

[0060] Convolutional layer: used to detect various patterns or features in the input image, such as edges, textures, etc.

[0061] Pooling layer: reduces the spatial dimension of the feature map while enhancing the model's invariance to input changes (such as translation and rotation).

[0062] 3. Feature extraction and mapping As image data passes through a series of convolutional and pooling layers, the model gradually extracts the key features of the image and maps these features into a low-dimensional space. This process converts the original high-dimensional image data into a feature vector that contains rich semantic information.

[0063] 4. Fully Connected Layer and Classification Decision The feature vector output from the last pooling layer then enters a fully connected layer. The first fully connected layer integrates all previously learned features, while the second fully connected layer makes the final classification decision based on these features. Specifically, the last fully connected layer is usually followed by a Softmax activation function, which converts the scores corresponding to each category into a probability value, thereby determining the most likely category to which the image belongs.

[0064] 5. Output classification results Based on the probability distribution given by the Softmax layer, the category with the highest probability value is selected as the prediction result. This means that each input medical paper image is assigned to one of four predetermined categories: flow cytometry image, fluorescence image, particle fluorescence image, or other image. The entire process not only relies on the carefully designed internal architecture of the model but also benefits from previous training using a large amount of annotated datasets, which ensures that the model can accurately identify different types of medical images.

[0065] 6. Post-processing (optional) In some application scenarios, to further improve classification accuracy or meet specific requirements, the output of the CNN model can be post-processed. For example, you can set a confidence threshold to filter out samples with uncertain classifications, or use non-maximum suppression (NMS) technology to address overlapping areas.

[0066] In this way, the pre-trained CNN model can effectively identify the type of new medical paper images and correctly classify them into one of the four types mentioned above, providing basic support for subsequent similarity detection.

[0067] S120: performing image preprocessing on the four types of images respectively.

[0068] Image preprocessing is crucial to ensuring the effectiveness of subsequent steps. Different types of medical images require different preprocessing strategies to optimize feature extraction and matching, due to their differing imaging principles, application scenarios, and visual characteristics. The following describes specific preprocessing methods and their rationale for streaming images, fluorescent images / granular fluorescent images, and other image types.

[0069] The preprocessing of streaming images includes: cutting edges, removing coordinate axes, improving resolution, and grayscale. Cutting edges and removing coordinate axes: Streaming images often contain additional non-image information, such as coordinate axes or legends, which are not helpful for understanding the image content and may interfere with the feature extraction process.

[0070] Improving resolution: To more accurately capture detailed features in an image, it is sometimes necessary to increase the image resolution. This is particularly important for streaming images, as these images often contain subtle structural changes, and high resolution helps accurately identify these changes.

[0071] Grayscale conversion: Converting color images to grayscale can reduce data dimensionality while highlighting the texture and shape information of the image. After grayscale conversion, the model is more likely to focus on the content of the image rather than the color information, which is particularly useful when analyzing streaming images.

[0072] Preprocessing of fluorescent images / granular fluorescent images includes: reducing resolution, grayscale, and balancing brightness. Reduce resolution: Although increasing resolution helps capture details, for fluorescence images, too high a resolution may increase computational complexity without providing additional information gain. Appropriately reducing resolution can help speed up processing while maintaining sufficient details for feature extraction.

[0073] Grayscale conversion: Similar to flow cytometry images, fluorescence images are often converted to grayscale images to simplify feature representation and focus on brightness information. This is because the main information of fluorescence images is usually expressed through brightness changes rather than color.

[0074] Brightness Balancing: Fluorescence images can be affected by shooting conditions, resulting in uneven brightness. Brightness balancing eliminates this difference, making images acquired under different conditions consistent and improving the accuracy of feature extraction and matching.

[0075] The preprocessing of other images includes: reducing resolution, grayscale, and balancing brightness. Reduce resolution: Similar to fluorescence images, other types of medical images may also benefit from appropriate resolution adjustment. Reducing the resolution can reduce the computational burden without compromising the core information.

[0076] Grayscale: Converting color images to grayscale images helps simplify feature representation, making it easier for the model to learn useful patterns.

[0077] Balancing brightness: Images captured with different imaging devices or in different environmental conditions may have different brightness levels. By standardizing brightness, we can reduce the impact of these external factors on the feature extraction process and improve the consistency and reliability of the algorithm.

[0078] Each of the different types of images mentioned above has its own unique characteristics and challenges: Flow images: Usually involve complex dynamic processes, so higher resolution is required to capture subtle changes and remove unnecessary background elements to focus on the main content.

[0079] Fluorescence images and particle fluorescence images rely primarily on brightness changes to reveal structure, so grayscale conversion and brightness balance are particularly important. In addition, considering that fluorescence images may have a high native resolution, moderately reducing the resolution can help simplify processing without losing important information.

[0080] Other images: This category covers a variety of medical images, each with different characteristics, but all require standardized preprocessing steps to enhance feature extraction, such as grayscale and brightness adjustment.

[0081] In summary, different preprocessing strategies are employed for different types of medical images to better adapt to their respective characteristics and challenges, thereby improving the efficiency and accuracy of subsequent feature extraction and matching steps. These preprocessing steps not only consider the physical properties of the image itself but also incorporate practical challenges encountered in real-world applications, aiming to provide an optimized foundation for the effective execution of subsequent analysis tasks.

[0082] S130: Extract features from the pre-processed images of different types and match the extracted features.

[0083] Feature extraction and matching are core steps in image similarity detection. By extracting unique features from different image types and finding corresponding relationships between these features, similar images can be effectively identified. However, because different image types have different visual characteristics and complexities, selecting appropriate algorithms and parameters is crucial for improving detection accuracy. The following are specific feature extraction and matching strategies and their rationale for streaming images, fluorescent images / granular fluorescent images, and other image types.

[0084] 1. Streaming Images Algorithm of choice: SIFT Reasoning: Streaming images often contain complex dynamic structures and rich details. The SIFT (Scale-Invariant Feature Transform) algorithm excels in processing such images due to its excellent scale and rotation invariance. It accurately locates and describes key points, maintaining high matching accuracy even in the presence of scale changes, rotation, or partial occlusion.

[0085] Specific operation: Use the SIFT algorithm to extract the key points of the image and calculate the descriptor of each key point. Then, based on the distance between the descriptors (such as Euclidean distance), find similarities and corresponding relationships between images.

[0086] 2. Fluorescent Images / Granular Fluorescent Images Algorithm selection: AKAZE Reasoning: Fluorescence images and particle fluorescence images typically have high contrast and rich texture information, but they can also be affected by imaging conditions, such as uneven brightness or noise. Compared to SIFT, the Accelerated-KAZE algorithm maintains good scale and rotation invariance while providing faster speed and higher accuracy, especially when processing small-scale details.

[0087] Threshold setting: Set the matching threshold to 20. A lower threshold means stricter requirements for feature point matching, which helps improve matching accuracy and is especially suitable for analyzing demanding fluorescence images, ensuring that similar points can be accurately identified even with subtle differences.

[0088] Optimizing matching: RANSAC Reasoning: To further improve matching quality, the RANSAC (Random Sample Consensus) algorithm is used to eliminate mismatched points. RANSAC uses iterative random sampling and consistency checks to find the largest set of consistent matching points, resulting in more reliable matching results.

[0089] Specific operation: First, use the AKAZE algorithm to extract the key feature points of the image, then apply the RANSAC algorithm to optimize the matching, and finally obtain a set of high-quality key point pairs.

[0090] 3. Other Images Algorithm selection: AKAZE Rationale: While other types of medical images are diverse, scale and rotation invariance are still important in most cases. The AKAZE algorithm is an ideal choice due to its efficiency and robustness. Compared to fluorescence images, these images can be more diverse, allowing for slightly looser matching criteria.

[0091] Threshold setting: Set the matching threshold to 40. A higher threshold allows the algorithm to tolerate a certain degree of error when facing more complex image content and adapt to more diverse application scenarios, although this means that a certain degree of accuracy may be sacrificed in some cases.

[0092] Specific operation: Similar to the processing flow of fluorescence images, first use the AKAZE algorithm to extract feature points, then perform preliminary matching based on the set threshold, and finally use the RANSAC algorithm to further optimize the matching results as needed.

[0093] The key to choosing different algorithms and thresholds for different types of images is: Streaming images are more suitable for using SIFT algorithm to ensure high matching accuracy due to their complex structure and detail requirements.

[0094] Fluorescence images and particle fluorescence images benefit from the efficient processing and detailed feature capture capabilities provided by the AKAZE algorithm, while the low matching threshold ensures high accuracy during analysis.

[0095] For other types of images, considering their diversity and potential complex backgrounds, the AKAZE algorithm is also applicable, but a relatively loose matching threshold is adopted to adapt to a wider range of application scenarios without unduly affecting the accuracy.

[0096] This hierarchical and targeted approach not only improves the effect of feature extraction and matching, but also lays a solid foundation for subsequent similarity analysis.

[0097] S140: Determine whether the two images are similar based on the number of matched feature points and the ratio of the number of matched points to the total number of points; if they are similar, further analyze their transformation parameters.

[0098] After feature extraction and matching, the next step is similarity analysis. The goal of this stage is to determine whether two images are similar based on the matching results and further analyze the geometric transformation relationships between them (such as scale, rotation, and flipping). The following is a detailed description: S141: Preliminary similarity judgment.

[0099] First, a preliminary similarity assessment is performed based on the feature point matching results obtained in the previous step. This is mainly based on two key indicators: Number of matched feature points: Counts the number of successfully matched feature points between two images. Generally, if two images have a high degree of similarity, there will be more matching points between them. Figure 3 As shown, it represents the matching points of the two images.

[0100] Ratio of matching points to total points: In addition to the number of matching points, we also need to consider the ratio of these matching points to the total number of detected keypoints in each image. This ratio reflects the distribution of matching points across the image and their importance. For example, even if there are many matching points, if the ratio is low relative to the total number of keypoints, it may mean that the two images are not truly similar.

[0101] By setting appropriate thresholds, we can determine which image pairs are considered “similar.” For example, if the number of matching points exceeds a certain preset value and the proportion of matching points to the total number of points is higher than a certain percentage, then the two images are considered similar.

[0102] S142: Further similarity analysis.

[0103] Once we determine that two images are similar, we need to analyze the specific transformation parameters between them. This includes but is not limited to the following common geometric transformations: Scale Variation: Analyzes the relative size difference between two images. Even with the same content, the actual size of an image may vary due to differences in shooting distance or device settings.

[0104] Rotation Angle: Identifying whether one image is rotated relative to another. This is crucial for accurately understanding the relationship between images, especially in medical image analysis, where even small angular differences can affect diagnostic results.

[0105] Flip / Mirror Transformation: Checks for horizontal or vertical flipping. This is common in real-world applications, especially when processing images from different sources or after different editing processes.

[0106] To perform the above analysis, we can obtain these transformation parameters by calculating the homography matrix between matching point pairs. A homography is a two-dimensional projective transformation that describes the mapping relationship between planes. It is suitable for analyzing affine transformations (such as translation, rotation, and scaling) between images, as well as more complex perspective transformations.

[0107] In an example, suppose there is a pair of medical images A and B that are considered similar. A series of corresponding feature point pairs are obtained through feature matching. The following operations are performed: Calculate the homography transformation matrix H between these matching point pairs, which contains all the transformation information from image A to image B (including scale, rotation, and translation, etc.).

[0108] According to the H matrix, the specific transformation parameters can be solved, such as: Scale factor: Indicates how many times image A is enlarged or reduced relative to image B.

[0109] Rotation Angle: Gives how many degrees image A is rotated relative to image B.

[0110] Translation vector: Displays the displacement of image A relative to image B in the x-axis and y-axis directions.

[0111] These transformation parameters not only help verify the true similarity between two images, but also provide valuable information for subsequent possible applications, such as image alignment, stitching, or tampering detection.

[0112] like Figure 4 As shown, the similar parts in the two figures are marked with black boxes.

[0113] In short, similarity analysis involves more than simply counting the number and ratio of matching points. It also requires a deeper exploration of the various transformation relationships between images to comprehensively and accurately assess the degree of similarity between them. This process is crucial for ensuring the accuracy and reliability of similarity detection in medical images.

[0114] S150: For the detected similar image pairs, feature matching points are drawn, similar regions are marked, and then the result images are visualized and output.

[0115] After feature extraction, matching, and similarity analysis are completed, in order to facilitate users to intuitively understand the detection results and support further manual verification or report generation, the detected similar image pairs need to be post-processed and visualized. This process mainly includes the following steps: S151: Draw feature matching points.

[0116] Define matching points: Based on the matching feature point pairs calculated in the previous step (found by SIFT or AKAZE algorithm), determine the specific location of each pair of matching points in the two images.

[0117] Plotting matching points: Mark the locations of these matching points on the original image with obvious markers (such as circles or crosses). Typically, different colors can be used to distinguish different types of matching points (for example, correctly matched points can be represented in green, and incorrectly matched points can be represented in red). This step helps to visually indicate which parts are considered to be similar key areas.

[0118] S152: Mark similar areas.

[0119] Connect matching points: For each pair of matching points, draw a line between the two images to connect them, forming a so-called "matching line", such as Figure 5This connection method not only clearly shows which points correspond to each other, but also helps the observer quickly identify the corresponding relationship between images and possible transformations (such as rotation, scaling, etc.).

[0120] Highlighting Similar Areas: In addition to individually marking each matching point, you can also highlight the area covered by a series of closely related matching points as a whole. For example, you can highlight these similar areas by drawing a rectangular box or filling them with a transparent color. This makes it easier for users to notice local areas with high similarity.

[0121] S153: Visually output the result image, and output the matching result of each type of similar image pair to a result file.

[0122] Integration: The matching points, lines, and highlighted areas drawn above are combined to form a comprehensive result image. Ideally, this result image should contain both the input images and their corresponding annotation information for direct comparison and evaluation.

[0123] Format selection: Select the appropriate output format based on actual needs. Common practices include: Still Image: Save the processed image in formats such as PNG and JPEG, suitable for report generation or archiving.

[0124] Dynamic interactive interface: If implemented in a software system, an interactive interface can be designed that allows users to zoom in to view specific areas, switch display modes (such as showing only matching points without connecting lines), adjust transparency, etc., to obtain a more flexible and detailed analysis experience.

[0125] Additional information: In addition to basic visual annotation, some auxiliary information can be added next to the image, such as matching scores, transformation parameters (such as scale change rate, rotation angle, etc.), confidence levels, etc., to provide users with more background knowledge and data to support decision-making.

[0126] In a specific application example, assume that the analysis of two suspected similar medical images A and B has been completed, and several reliable matching point pairs have been determined. The following steps will be performed: Mark the locations of all matching points with red circles on images A and B respectively; Use blue lines to connect each pair of matching points, forming a line spanning the two images; For those matching point groups that are clustered together and show strong similarity, they are surrounded by yellow semi-transparent rectangular frames to emphasize the importance of these areas; Finally, images A and B with all the annotation information are displayed side by side on a PDF report page, with a brief text description explaining the main findings and significance of the matching.

[0127] This meticulous and comprehensive post-processing and visualization not only greatly improves the readability and usability of the results, but also provides a powerful support tool for subsequent manual review, which is crucial for ensuring the accuracy and reliability of medical image similarity detection systems.

[0128] Furthermore, when processing sensitive information such as medical images, ensuring data security and privacy is crucial. This is especially true when the system consists of multiple modules that need to communicate and exchange data with each other, so data transmission security becomes a key consideration. To this end, using encrypted channels to protect data transmission within and outside the system is an effective strategy. The following details how this measure is implemented: 1. The importance of data encryption Privacy protection: Medical images and other health-related information are highly sensitive data types, and leakage can have serious consequences. Transmitting data through encrypted channels can effectively prevent unauthorized access.

[0129] Integrity assurance: In addition to confidentiality, encryption ensures that data is not tampered with or corrupted during transmission, which is critical for maintaining the accuracy of medical diagnosis and research.

[0130] Compliance requirements: Many countries and regions have strict laws and regulations regarding the processing of medical data (such as GDPR and HIPAA), requiring appropriate technical measures to protect patient information. Using encrypted channels can help meet these legal requirements.

[0131] 2. Methods for implementing encrypted channels The main methods for implementing encrypted channels include but are not limited to the following: SSL / TLS protocol: This is one of the most commonly used network layer encryption methods, providing end-to-end secure communications between clients and servers. All modern browsers support HTTPS, the HTTP protocol over SSL / TLS, making it an ideal choice for protecting data transmission in web applications.

[0132] VPN (Virtual Private Network): For enterprise applications or cross-organizational data sharing, establishing a secure VPN connection ensures data security throughout the entire transmission path. This approach is particularly suitable for communication between different departments within a hospital or between the hospital and other medical institutions.

[0133] Message queue and API gateway encryption: If your system architecture relies on a microservices architecture, the messages passed between services may require additional security measures. You can enable TLS encryption by configuring your message queue (such as RabbitMQ, Kafka) or API gateway (such as AWS API Gateway) to protect inter-service communications.

[0134] End-to-end encryption: For specific application scenarios, such as frequent and large-scale data exchange between two fixed terminals, end-to-end encryption solutions can be directly implemented. This approach usually involves implementing encryption algorithms (such as AES and RSA) directly at the application level to ensure that each transmission is secure.

[0135] 3. Configure and manage encrypted channels Key management: Effective key management is fundamental to the successful deployment of encrypted channels. Strict policies must be in place for the generation, storage, distribution, and rotation of keys, and key usage must be regularly audited to prevent key leaks.

[0136] Certificate Authority (CA): For SSL / TLS-based solutions, you need to obtain a digital certificate from a trusted certificate authority. This not only enhances user trust in the system but also simplifies the certificate verification process.

[0137] Monitoring and Logging: Continuously monitor the status of encrypted channels to promptly detect and respond to any unusual activity. Detailed logs are also maintained for subsequent analysis and tracking of potential security incidents.

[0138] 4. Application Examples Consider a large hospital system where multiple departments, such as the Imaging Department and the Pathology Department, need to share patient medical image data. To ensure the security of this highly sensitive data during transmission, the following steps can be taken: An SSL / TLS-based encrypted connection is established between the workstations and servers of each department to ensure that every image request and transmission is completed within the encrypted channel.

[0139] For remote consultations or cross-hospital collaborations, VPN technology is used to create a secure private network environment, allowing participants to efficiently view and discuss cases in a safe and secure manner.

[0140] If the system adopts a microservice architecture, it is necessary to ensure that the data interaction between each microservice is encrypted, for example, by configuring the API gateway to enforce the TLS protocol.

[0141] Through the above measures, not only can the ability of medical information systems to resist external attacks be significantly improved, but the security of internal data circulation can also be enhanced, ultimately providing patients with more reliable service guarantees.

[0142] This paper proposes a medical image similarity detection method based on the combination of deep learning and feature matching, which has the following significant technical effects: 1. Highly accurate intelligent classification and similarity detection Deep learning technology is used to automatically classify medical images, and targeted preprocessing and feature matching strategies are adopted according to different categories of images, which effectively improves the accuracy of image similarity detection.

[0143] In actual tests, the overall accuracy of this method in image classification and similarity judgment can reach more than 90%, and it has good recognition capabilities.

[0144] 2. Efficient batch processing capability significantly improves detection efficiency The system supports automated and batch processing of large amounts of medical paper image data without the need for human intervention.

[0145] The image processing speed can reach 20 images per minute, which greatly shortens the time cycle of image review and significantly improves the efficiency of duplicate checking and quality control.

[0146] 3. Strong robustness, adaptable to complex image changes Through systematic image preprocessing (such as scale normalization, grayscale conversion, brightness balance, etc.) and post-processing (such as feature point annotation and region visualization), we can effectively address problems such as multi-scale, rotation, and noise interference in medical images.

[0147] The algorithm's adaptability to different imaging conditions and image transformations has been improved, and the stability and robustness of the system have been enhanced.

[0148] 4. Intuitive visual output to assist manual review For the detected similar image pairs, the system can automatically generate a visual result map containing feature matching points and corresponding area annotations.

[0149] The visual information is clear and intuitive, making it easy for researchers or reviewers to quickly locate similar content, providing strong support for the authenticity review of medical papers.

[0150] 5. Low cost, high security, and data privacy protection Compared with the traditional image duplication checking method that relies on third-party services, the present invention can be deployed on the hospital's local server to avoid uploading sensitive images to external platforms.

[0151] It significantly reduces the hospital's image duplication checking costs, while greatly improving the privacy protection and security of medical data and preventing the risk of data leakage.

[0152] 6. Modular design, easy to expand and maintain The system adopts a modular architecture design with clear responsibilities and low coupling between functional modules, which facilitates flexible deployment, functional expansion and system upgrades based on the actual needs of the hospital.

[0153] It supports high availability and long-term maintainability, and is suitable for promotion and application in various medical institutions.

[0154] In summary, the present invention not only excels in image recognition accuracy, processing efficiency, and robustness, but also takes into account data security, cost control, and system scalability, and has broad application prospects and promotion value.

[0155] An embodiment of the present application further provides an electronic device, comprising a processor, a memory, and a computer program that can be run on the processor, wherein the processor implements the above-described method when executing the program.

[0156] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the above-described method when executed by a processor.

[0157] An embodiment of the present application also provides a computer product, including a computer program / instruction, which implements the above-mentioned method when executed by a processor.

[0158] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0159] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0160] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0161] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0162] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0163] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0164] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0165] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0166] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0167] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A medical image similarity detection method using neural network feature matching, comprising: The trained convolutional neural network model is used to identify the types of input medical paper images and classify them into four categories: flow images, fluorescence images, particle fluorescence images, and other images; Image preprocessing is performed on each of the four image types. Preprocessing of flow images includes: cutting edges, removing coordinate axes, improving resolution, and grayscale; preprocessing of fluorescence images / particle fluorescence images and other images includes: reducing resolution, grayscale, and balancing brightness. Perform feature extraction on different types of pre-processed images and match the extracted features. For streaming images, SIFT algorithm is used for feature extraction and matching; for fluorescence images / particle fluorescence images and other images, AKAZE algorithm is used for feature extraction and matching. Determine whether two images are similar based on the number of matched feature points and the ratio of matched points to the total number of points; if they are similar, further analyze their transformation parameters; For the detected similar image pairs, feature matching points are drawn, similar areas are marked, and the resulting images are visualized and output.

2. The method according to claim 1, wherein the trained convolutional neural network model structure includes three convolutional layers, three pooling layers, and two fully connected layers, wherein: The convolutional layer is used to extract local texture and structural features of the image, the pooling layer is used to reduce the image dimension and enhance scale invariance, and the fully connected layer is responsible for the final image classification decision.

3. The method according to claim 1, wherein the AKAZE algorithm is used to extract and match features of the fluorescence image / particle fluorescence image, and the matching threshold is set to 20.

4. The method according to claim 1, wherein the AKAZE algorithm is used to extract features and match other images, and the matching threshold is set to 40.

5. The method according to claim 1, after extracting and matching features of the fluorescence image / particle fluorescence image and other images using the AKAZE algorithm, further comprising: The machine sampling consensus algorithm is used to eliminate false matching points. The method according to claim 1 , wherein the transformation parameters include scale, rotation, and flipping.

7. The method according to claim 1, wherein when the system implementing the method is composed of multiple modules, data transmission between the modules adopts an encrypted channel.

8. An electronic device comprising a processor, a memory, and a computer program that can be run on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the program.

9. A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method according to any one of claims 1 to 7 when executed by a processor.

10. A computer software product comprising a computer program / instructions, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images