Video intelligent analysis system and method for skin infected by monkey pox based on deep learning
Through the spatial and temporal feature extraction and timing analysis based on three-dimensional convolutional neural network, the accuracy and efficiency of monkeypox infection video diagnosis in the existing technology are solved, and accurate identification and dynamic tracking of monkeypox infection areas are realized, which is suitable for a variety of clinical environments.
Patent Information
- Application Number
- CN202510400990.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-01
- Publication Date
- 2025-07-18
AI Technical Summary
The existing medical image analysis methods lack spatial and temporal feature integration, dynamic evolution modeling and noise processing capabilities when processing monkeypox infected videos, resulting in poor diagnostic accuracy and efficiency, especially in complex backgrounds and high noise environments.
Using three-dimensional convolutional neural network (3D CNN) based on the combination of timing information, intelligent identification and analysis of monkey pox skin lesions is achieved through preprocessing, spatiotemporal feature extraction, timing analysis and post-processing.
It significantly improves the accuracy and efficiency of monkeypox infection detection, can automatically process video streams, reduce the risk of misdiagnosis and misdiagnosis, is suitable for different types of infection symptoms, has real-time analysis capabilities and high ease of use.
Smart Images
Figure CN120339909A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image processing, and particularly to an intelligent analysis system and method for monkey pox-infected skin videos based on deep learning. Background Art
[0002] As the spread of the monkey pox virus has received increasing attention, medical assisted diagnosis based on image and video analysis has become an important direction to improve the efficiency and accuracy of clinical diagnosis. In this regard, traditional medical image analysis methods mainly rely on the processing and analysis of two-dimensional images, such as static images collected through microscopes, skin imaging, or other traditional devices. However, these methods face many challenges when dealing with dynamic monkey pox-infected video sequences, especially in the case of multi-temporal, complex backgrounds, and high noise interference, where the spatio-temporal features of the infected area cannot be fully extracted, resulting in low diagnostic accuracy and efficiency.
[0003] Existing medical image analysis methods usually adopt techniques such as image classification and segmentation based on two-dimensional convolutional neural networks (CNNs). Two-dimensional convolutional neural networks have achieved certain success in the processing of single-frame images, but they have the following deficiencies when dealing with continuous video sequences:
[0004] 1. Lack of integration of spatio-temporal features: Traditional two-dimensional convolutional neural networks only process the spatial features of images, ignoring the temporal correlation between frames in the video. Therefore, the dynamic changes of monkey pox virus-infected skin cannot be effectively captured, resulting in inaccurate recognition of the lesion area.
[0005] 2. Insufficient modeling of dynamic evolution: Monkey pox skin lesions have obvious time-evolution characteristics, and the lesion area changes continuously over time. Existing technologies have not fully utilized the temporal information in the video sequence, resulting in the inability to accurately capture and analyze the change trend of the lesion.
[0006] 3. Noise and background interference: In actual monkey pox skin videos, there are often varying degrees of noise and complex backgrounds, which may affect the performance and accuracy of deep learning models based on two-dimensional images, especially in low-light or blurred environments.
[0007] 4. Lack of optimized models for specific diseases: Existing video analysis methods are mostly general models and have not been optimized for the characteristics of monkey pox virus-infected skin, resulting in limited application in the professional field, and the accuracy and robustness of the models are difficult to meet clinical needs.
[0008] The main causes of the above problems are as follows:
[0009] 1. Limitations of 2D convolutional models: Traditional 2D convolutional neural networks only perform feature extraction from a spatial perspective and lack the ability to model time series features, resulting in an inability to fully understand the information brought about by temporal changes when processing dynamic videos.
[0010] 2. Loss of temporal features: Video data is a temporal signal, while traditional methods mainly rely on single-frame analysis of static images, ignoring the spatio-temporal correlation between frames in video data, leading to the inability to capture the temporal evolution and dynamic information of the lesion area.
[0011] 3. Noise and blurring problems: Medical images are usually affected by shooting conditions (such as low light, out-of-focus blurring, etc.), and existing technologies lack means of noise reduction and image enhancement for such special scenarios, resulting in unsatisfactory performance of the diagnostic model in complex environments. Summary of the Invention
[0012] To solve the above technical problems existing in the prior art, the present invention proposes an intelligent analysis system and method for monkey pox-infected skin videos based on deep learning. By combining a three-dimensional convolutional neural network (3D CNN) with temporal information, it aims to achieve intelligent identification of monkey pox skin lesions.
[0013] On the one hand, to achieve the above object, the present invention provides an intelligent analysis system for monkey pox-infected skin videos based on deep learning, including:
[0014] Input end: Used to receive video sequence data to be detected;
[0015] Processing unit: Used to input the video sequence data to be detected into an improved three-dimensional convolutional neural network for processing to obtain a video sequence analysis result;
[0016] Output end: Used to display the dynamic evolution process, infection location and scope of the monkey pox-infected area, and generate clinical analysis suggestions.
[0017] Preferably, the input end is a video input module, and the video sequence data to be detected is a grayscale or color image of a fixed size.
[0018] Preferably, the processing unit includes:
[0019] Preprocessing module: Used to preprocess the video sequence data, including video image denoising, frame alignment, illumination equalization, and video image enhancement operations;
[0020] Three-dimensional convolutional neural network module: Used to perform joint convolution in space and time on the preprocessed video images through convolutional kernels, extract spatio-temporal features, and generate spatio-temporal expressions of the monkey pox-infected area.
[0021] Temporal analysis module: used to further extract the characteristics of the evolution of the monkeypox infection area over time, establish a temporal dynamic model of monkeypox infection, and identify the evolution law of the monkeypox infection area;
[0022] Post-processing module: used to receive the output result of the temporal analysis module, and perform detail restoration, noise removal, and image smoothing operations.
[0023] Preferably, the temporal analysis module includes:
[0024] Temporal modeling sub-unit: used to model the characteristics of consecutive time steps through a temporal model;
[0025] Dynamic evolution analysis sub-unit: used to analyze the dynamic evolution trend of the monkeypox infection area through the hidden state of the temporal model, and predict the development direction and speed of monkeypox virus infection; wherein, the hidden state of the temporal model refers to the potential representation of the development process of the monkeypox infection area inside the temporal model, and is used to capture complex patterns and change trends during the infection process;
[0026] Stage determination sub-unit: used to judge the change stage of the monkeypox infection area through a determination mechanism; wherein, the determination mechanism determines the specific change stage where the infection area is located by analyzing the dynamic characteristics of the infection area, combining historical data and statistical methods.
[0027] Preferably, the post-processing module uses the non-local means denoising method to remove random noise in the image, uses bilateral filtering to smooth the image, removes noise in the flat area while retaining edge information, and applies the Laplace operator to enhance the details and edge sharpness of the image.
[0028] On the other hand, to achieve the above object, the present invention also provides an intelligent analysis method for monkeypox-infected skin videos based on deep learning, including:
[0029] Preprocess the input monkeypox-infected skin video sequence;
[0030] Extract and analyze the features of the preprocessed video frame data through an improved three-dimensional convolutional neural network, and simultaneously learn the spatial features and time series features of the video sequence to generate a spatio-temporal expression of the monkeypox infection area;
[0031] Based on the output result of the improved three-dimensional convolutional neural network, use the temporal analysis module to analyze the dynamic evolution of the monkeypox infection area, and establish a temporal dynamic model of monkeypox infection;
[0032] Post-process the output result of the temporal analysis module, including noise removal, image smoothing, and detail enhancement;
[0033] Output the dynamic evolution process, infection location and scope of the monkeypox infection area, as well as clinical analysis suggestions. Preferably, the processing process of the improved three-dimensional convolutional neural network module is as follows:
[0034] y = Conv3D(V, K);
[0035] In the formula, V is the input three-dimensional video sequence, K is the convolutional kernel, and y is the output feature map after the convolutional operation.
[0036] Preferably, the time series analysis module uses a time series model to model the features of consecutive time steps, capture the non-linear changes of the lesion area over time, and retain long-term dependencies.
[0037] Compared with the prior art, the present invention has the following advantages and technical effects:
[0038] (1) By jointly learning spatio-temporal features based on an improved three-dimensional convolutional neural network, the present invention can more accurately capture the spatial information and temporal evolution law of the monkeypox infection area, significantly improving the accuracy of monkeypox infection detection;
[0039] (2) Based on deep learning and an improved three-dimensional convolutional neural network, the present invention automatically completes the analysis of monkeypox-infected skin videos, reducing manual intervention, greatly improving work efficiency, and also supporting the analysis of real-time video streams, capable of quickly generating real-time monitoring results of the infection area;
[0040] (3) Through the training of a large number of monkeypox-infected video sequences, the present invention can extract stable features, which are not only applicable to common infection symptoms, but also can better identify and diagnose some different types or uncommon monkeypox cases, reducing the risk of missed diagnosis and misdiagnosis;
[0041] (4) Relying on deep learning and automation technologies, the system does not require professional image analysis experience to use. Just input the video sequence, and the system can automatically complete the video analysis and provide results. This feature makes the system highly user-friendly and scalable in actual clinical applications. Description of the Drawings
[0042] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:
[0043] Figure 1 is the overall architecture diagram of the intelligent analysis system for monkeypox-infected skin videos based on deep learning according to the embodiment of the present invention;
[0044] Figure 2Flowchart of the intelligent analysis method for monkeypox-infected skin videos based on deep learning according to an embodiment of the present invention;
[0045] Figure 3 Schematic diagram of the convolution operation of the three-dimensional convolutional neural network according to an embodiment of the present invention. Detailed implementation manners
[0046] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
[0047] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.
[0048] The present invention proposes an intelligent analysis system for monkeypox-infected skin videos based on deep learning, as Figure 1 , including:
[0049] Input end: used to obtain video sequence data to be detected;
[0050] Processing unit: used to input the video sequence data to be detected into an improved three-dimensional convolutional neural network for processing to obtain a video sequence analysis result;
[0051] Output end: used to display the dynamic evolution process, infection location and range of the monkeypox-infected area, and generate clinical analysis suggestions.
[0052] Further, the input end is a video input module, and the video input module is used to receive video sequence data to be detected; wherein, the video sequence data to be detected is a grayscale or color image of a fixed size.
[0053] Specifically, the video input module is responsible for receiving video sequence data from video acquisition devices (such as skin shooting devices, medical video monitoring systems, etc.). Each video sequence consists of multiple consecutive frames and contains dynamic information of the monkeypox-infected area. It is usually input into the system in the form of an RGB or grayscale image sequence. Each frame in the video sequence is passed into the processing unit as an independent image.
[0054] Further, the processing unit includes:
[0055] Preprocessing module: used to preprocess the video sequence data, including video image denoising, frame alignment, illumination equalization, and video image enhancement operations;
[0056] 3D Convolutional Neural Network Module: Used to perform joint spatial and temporal convolution on the preprocessed video images through convolutional kernels, extract spatio-temporal features, and generate spatio-temporal expressions of the monkeypox infection area;
[0057] Temporal Analysis Module: Used to further extract the features of the monkeypox infection area evolving over time, establish a time dynamic model of monkeypox infection, and identify the evolution law of the monkeypox infection area;
[0058] Post-processing Module: Used to receive the output results of the temporal analysis module and perform detail restoration, noise removal, and image smoothing operations.
[0059] Specifically, in this embodiment, the traditional 3D CNN is optimized, and deformable convolution is used to enhance the network's adaptability to irregular lesion areas. At the same time, residual connections are used to avoid the problem of gradient disappearance and improve the training efficiency.
[0060] Furthermore, the temporal analysis module includes:
[0061] Temporal Modeling Sub-unit: Used to model the features of consecutive time steps through a temporal model; in this embodiment, a Long Short-Term Memory network (LSTM) is used to model the features of consecutive time steps. The LSTM can capture the non-linear changes of the lesion area over time and retain long-term dependencies.
[0062] Dynamic Evolution Analysis Sub-unit: Used to analyze the dynamic evolution trend of the monkeypox infection area through the hidden state of the temporal model, and predict the development direction and speed of monkeypox virus infection; among them, the hidden state of the temporal model refers to the potential representation of the development process of the monkeypox infection area inside the temporal model. This representation can capture the complex patterns and change trends during the infection process. By analyzing the changes in the hidden state, the evolution trajectory and potential trends of the lesion can be revealed, so as to predict the future expansion path and speed of the infection.
[0063] Stage Judgment Sub-unit: Used to judge the change stage of the monkeypox infection area through a judgment mechanism;
[0064] Among them, the judgment mechanism specifically refers to an evaluation algorithm based on the output of the temporal model. It analyzes the dynamic features of the infection area, combines historical data and statistical methods to determine the specific change stage of the infection area. This mechanism can automatically identify the key turning points in the infection process, thus providing a basis for subsequent infection trend prediction and prevention and control measures. The detailed process and parameter settings of the judgment mechanism will be further publicly described in the text to ensure transparency and reproducibility.
[0065] The judgment process is mainly divided into the following steps:
[0066] 1. Data preprocessing: Denoise, enhance contrast, and perform edge-preserving filtering on the input video sequence of monkey pox-infected skin to reduce background interference.
[0067] 2. Feature extraction: Use a three-dimensional convolutional neural network (3D-CNN) to extract spatio-temporal features, and combine it with a long short-term memory network (LSTM) for temporal dimension modeling to capture the changing trends of the infected areas.
[0068] 3. Key frame determination: Identify the key time nodes in the development of monkey pox infection (such as initial spread, lesion enlargement, healing trend) through inter-frame difference analysis and temporal feature clustering.
[0069] 4. Stage determination:
[0070] Use dynamic time warping (DTW) to compare the current video frame with the standard infection process model in the database and calculate the similarity score;
[0071] Combine a Bayesian classifier or a probabilistic neural network (PNN) to perform a hierarchical determination of the infected area.
[0072] 5. Result output: According to the determination result, the system outputs the infection stage (such as early stage, middle stage, late stage) and provides predicted infection trend information. The specific parameter settings are shown in Table 1.
[0073] Table 1
[0074]
[0075] The parameter settings in Table 1 are obtained through optimization based on a large number of experiments to ensure that the system can maintain a high determination accuracy under different lighting conditions, skin types, and video qualities.
[0076] Furthermore, the post-processing module uses a non-local means denoising method to remove random noise in the image, uses bilateral filtering to smooth the image, removes noise in flat areas while retaining edge information, and applies a Laplacian operator to enhance the details and edge sharpness of the image.
[0077] Furthermore, the output end is a result output module, including:
[0078] Visualization result subunit: Used to display the detection and change of the monkey pox-infected area, presented in the form of images and videos;
[0079] Clinical analysis suggestion subunit: Used to generate possible clinical conclusions based on the analysis results to provide auxiliary decision-making support for doctors.
[0080] This embodiment also provides a deep learning-based intelligent analysis method for monkey pox-infected skin videos, such as Figure 2 , including:
[0081] Preprocess the input video sequence of monkeypox-infected skin;
[0082] Extract and analyze the features of the preprocessed video frame data through an improved three-dimensional convolutional neural network, and simultaneously learn the spatial features and time series features of the video sequence to generate a spatio-temporal expression of the monkeypox-infected area;
[0083] Based on the output result of the improved three-dimensional convolutional neural network, use the time series analysis module to analyze the dynamic evolution of the monkeypox-infected area and establish a time dynamic model of monkeypox infection;
[0084] Post-process the output result of the time series analysis module, including noise removal, image smoothing, and detail enhancement;
[0085] Output the dynamic evolution process of the monkeypox-infected area, the location and scope of the infection, and clinical analysis suggestions. Further, the processing process of the improved three-dimensional convolutional neural network module is as follows (as Figure 3 )
[0086] y = Conv3D(V, K);
[0087] In the formula, V is the input three-dimensional video sequence, K is the convolutional kernel, and y is the output feature map after the convolutional operation.
[0088] Specifically, the improved three-dimensional convolutional neural network model processes the input video sequence through three-dimensional convolutional operations. In three-dimensional convolution, the convolutional kernel not only operates on the spatial dimension but also processes the time dimension to capture spatio-temporal features. Let the input video sequence be V = {I1, I2,..., I T}, where I T is the t-th frame image in the video sequence, and T is the total number of frames of the video.
[0089] Further, the time series analysis module uses a time series model to model the features of consecutive time steps, capture the non-linear changes of the lesion area over time, and retain long-term dependencies.
[0090] This embodiment is based on an improved three-dimensional convolutional neural network, and realizes the joint learning of spatio-temporal features by simultaneously processing the spatial features and time dynamic information of the video sequence. Compared with the traditional two-dimensional convolutional neural network, the improved three-dimensional convolutional neural network can not only extract spatial features from each frame of image, but also effectively learn and capture the time dimension information in the image sequence, improving the accuracy of the analysis of monkeypox-infected skin and the tracking ability of dynamic changes.
[0091] Structure and method based on an improved three-dimensional convolutional neural network for analyzing skin video sequences of monkeypox infection, capable of automatically extracting spatio-temporal features in the video, applicable to various clinical environments, and having high generalization ability. The joint processing of spatio-temporal features is the key innovation point of the present invention, with unique technical advantages.
[0092] Through the temporal modeling of monkeypox-infected skin, this embodiment can identify the dynamic changes in the skin lesion area, including the spread of infection, the evolution of lesions, and the trend of disease progression. It can capture the temporal dependence relationship between different frames in the video sequence, and improve the accuracy of skin infection analysis through the identification and modeling of dynamic changes. This temporal modeling method has important clinical significance for the continuous monitoring of infection and the prediction of future infection trends. Compared with traditional static image analysis methods, this embodiment can achieve the dynamic tracking and prediction of the lesion area through deep learning algorithms, improving the accuracy of disease assessment.
[0093] Through the training of deep learning algorithms, this embodiment can analyze multiple monkeypox infection areas in skin videos simultaneously, and can automatically extract the features of multiple lesion areas from the video sequence. Compared with traditional manual analysis, the system of this embodiment can improve the efficiency and accuracy of multi-region detection, especially showing excellent performance in cases of large-area skin infection. Through the multi-region automatic detection technology, it can identify and analyze multiple monkeypox infection areas, and provide a detailed analysis report. This technology enables the system to handle relatively complex cases, especially having more advantages in the analysis of large-area infections and multiple lesion areas.
[0094] This embodiment supports the analysis of real-time skin video sequences, can quickly process and generate a diagnostic report. The system has a high degree of integration and real-time performance, can analyze immediately after the doctor uploads the video, and provide the analysis results of monkeypox infection in a timely manner to support clinical decision-making. This function ensures that doctors can obtain the analysis results in the shortest time, thus accelerating the speed of diagnosis and treatment, and is especially suitable for emergency medical scenarios that require quick response.
[0095] The technical framework of this embodiment has high adaptability, can run on different hardware and software platforms, and can be integrated with existing medical imaging devices. The system not only supports the analysis of monkeypox videos with different lighting, angles, and skin types, but also can communicate with other medical diagnostic devices, enhancing its application value in different clinical environments. By designing a flexible technical architecture, it can not only process skin video sequences in various clinical scenarios, but also be seamlessly docked with existing medical imaging platforms or devices, having excellent scalability and being applicable to the analysis of various skin diseases
[0096] Although this embodiment is mainly applied to the analysis of monkeypox-infected skin, its technical framework and deep learning model have strong scalability and can adapt to the intelligent analysis of other skin diseases (such as chickenpox, measles, etc.). By retraining the model, it can be widely applied to the field of intelligent analysis of skin diseases.
[0097] The above technical points not only solve the limitations of the existing technology, improve the accuracy, efficiency and usability of monkeypox-infected skin analysis, but also provide new breakthroughs and directions for the development of intelligent analysis technology for skin diseases, and have broad application potential and market prospects.
[0098] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims
1. Intelligent analysis system for monkey pox-infected skin videos based on deep learning, characterized in that, Including: Input end: Used to receive video sequence data to be detected; Processing unit: Used to input the video sequence data to be detected into an improved three-dimensional convolutional neural network for processing to obtain a video sequence analysis result; Output end: Used to display the dynamic evolution process, infection location and scope of the monkeypox infection area, and generate clinical analysis suggestions.
2. The intelligent analysis system for monkey pox-infected skin videos based on deep learning according to claim 1, characterized in that, The input end is a video input module, and the video sequence data to be detected is a grayscale or color image of a fixed size.
3. The intelligent analysis system for monkey pox-infected skin videos based on deep learning according to claim 1, characterized in that The processing unit includes: Preprocessing module: Used to preprocess the video sequence data, including video image denoising, frame alignment, illumination equalization, and video image enhancement operations; Three-dimensional convolutional neural network module: Used to perform joint convolution in space and time on the preprocessed video image through a convolutional kernel, extract spatio-temporal features, and generate a spatio-temporal expression of the monkeypox infection area; Temporal analysis module: Used to further extract the features of the monkeypox infection area evolving over time, establish a time dynamic model of the monkeypox infection, and identify the evolution law of the monkeypox infection area; Post-processing module: Used to receive the output result of the temporal analysis module and perform detail restoration, noise removal, and image smoothing operations.
4. The intelligent analysis system for monkey pox-infected skin videos based on deep learning according to claim 3, wherein The temporal analysis module includes: Temporal modeling sub-unit: Used to model the features of consecutive time steps through a temporal model; Dynamic evolution analysis sub-unit: Used to analyze the dynamic evolution trend of the monkeypox infection area through the hidden state of the temporal model, and predict the development direction and speed of the monkeypox virus infection; where the hidden state of the temporal model refers to the potential representation of the development process of the monkeypox infection area inside the temporal model, used to capture complex patterns and change trends during the infection process; Stage determination sub-unit: Used to judge the change stage of the monkeypox infection area through a determination mechanism; where the determination mechanism determines the specific change stage where the infection area is located by analyzing the dynamic features of the infection area, combining historical data and statistical methods.
5. The intelligent analysis system for monkey pox-infected skin videos based on deep learning according to claim 3, characterized in that, The post-processing module uses a non-local mean denoising method to remove random noise in the image, uses bilateral filtering to smooth the image, removes noise in flat areas while retaining edge information, and applies a Laplacian operator to enhance and improve the details and edge sharpness of the image.
6. An intelligent analysis method for monkey pox-infected skin videos based on deep learning, characterized in that, Including: Preprocess the input monkeypox-infected skin video sequence; Extract and analyze the features of the preprocessed video frame data through an improved three-dimensional convolutional neural network, and at the same time learn the spatial features and time series features of the video sequence to generate a spatio-temporal expression of the monkeypox infection area; Based on the output result of the improved three-dimensional convolutional neural network, use the temporal analysis module to analyze the dynamic evolution of the monkeypox infection area and establish a time dynamic model of the monkeypox infection; Post-process the output result of the temporal analysis module, including noise removal, image smoothing, and detail enhancement; Output the dynamic evolution process, infection location and scope of the monkeypox infection area, and clinical analysis suggestions.
7. The intelligent analysis method for monkey pox-infected skin videos based on deep learning according to claim 6, characterized in that, The processing process of the improved three-dimensional convolutional neural network module is: y = Conv3D(V, K); In the formula, V is the input three-dimensional video sequence, K is the convolutional kernel, and y is the output feature map after the convolution operation.
8. The intelligent analysis method for monkey pox-infected skin videos based on deep learning according to claim 6, wherein The timing analysis module uses a timing model to model the features of consecutive time steps, capture the non-linear changes of the lesion area over time, and retain long-term dependencies.