Dermoscope image processing and diagnosis system based on artificial intelligence
Through the artificial intelligence-based dermatoscope image processing and diagnosis system, the image features are automatically extracted using deep learning algorithms, and the problems of differences in diagnosis results and inefficiency in the existing technology are solved, efficient and accurate diagnosis of skin lesions is achieved, and resources are The scarce area provides high-quality services to ensure data security.
Patent Information
- Application Number
- CN202410947864.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-12-20
AI Technical Summary
The existing diagnostic methods for dermatosis rely on doctors’ experience and intuition, resulting in large differences in diagnostic results, low efficiency, and difficult to popularize in resource-scarce areas, and the lack of full use of dermatoscope image data to improve diagnostic accuracy.
Design a dermatoscope image processing and diagnostic system based on artificial intelligence, including data acquisition, image preprocessing, feature extraction, model training and diagnostic decision-making modules, use convolutional neural network to automatically extract image features, and realize diagnosis through deep learning algorithms. Provides a user-friendly interface and deploys on cloud servers to provide remote services.
It improves the accuracy and efficiency of diagnosis, realizes data-driven intelligent diagnosis, ensures that patients in remote areas can obtain high-quality diagnostic services, and ensures data security through encryption measures and complies with medical regulations.
Smart Images
Figure CN119170245A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of skin medical technology, and in particular to a dermatoscope image processing and diagnosis system based on artificial intelligence. Background Art
[0002] Existing methods for diagnosing skin lesions mainly rely on the doctor's experience and intuition, which to some extent limits the accuracy and consistency of diagnosis. In addition, as the types and manifestations of skin lesions increase, the challenges faced by doctors are also increasing. The main defects of existing technologies include:
[0003] Highly subjective: The diagnosis results depend largely on the doctor’s personal experience and technical level, which may lead to large differences in the diagnosis results.
[0004] Inefficiency: Traditional diagnostic methods require doctors to review images one by one, which is time-consuming and inefficient.
[0005] Uneven distribution of resources: In some remote areas or resource-poor settings, there may be fewer professional dermatologists, making diagnostic services difficult to popularize.
[0006] Insufficient data utilization: Traditional diagnostic methods fail to fully utilize the large amount of dermoscopic image data, missing the opportunity to improve diagnostic accuracy through data-driven approaches. Summary of the invention
[0007] The purpose of the present invention is to provide a dermatoscope image processing and diagnosis system based on artificial intelligence.
[0008] To achieve the above object, the present invention is implemented according to the following technical solutions:
[0009] The present invention includes a data acquisition module, an image preprocessing module, a feature extraction module, a model training module, a diagnosis decision module and a user interface. The signal output end of the data acquisition module is connected to the data input end of the image preprocessing module, the data output end of the image preprocessing module is connected to the data input end of the feature extraction module, the data output end of the feature extraction module is connected to the data input end of the model training module, the data output end of the model training module is connected to the data input end of the diagnosis decision module, and the data output end of the diagnosis decision module is connected to the data input end of the user interface.
[0010] Furthermore, the data acquisition module collects dermoscopic image data, which includes lesion images of different races, ages, genders and skin types. Dermatologists annotate the images, including lesion type, size and shape; the image preprocessing module uses a filter to remove image noise, standardizes the image size and pixel range, and enhances image contrast; the feature extraction module extracts color, texture, and shape features, and uses a convolutional neural network to automatically learn image features; the model training module uses annotated data sets to train the model, and uses a cross-validation method to optimize model performance. The trained model of the diagnostic decision module classifies new images, gives possible lesion types, and provides a confidence score for each classification result. The user interface uploads images and views results.
[0011] Specifically, the feature extraction module includes traditional feature extraction and deep learning feature extraction. The traditional feature extraction includes color histogram, texture analysis, and shape analysis; color histogram: calculates the distribution of each color in the image; texture analysis: uses gray co-occurrence matrix (GCM) or Gabor filter method; shape analysis: uses edge detection algorithm or contour tracking algorithm; the deep learning feature extraction uses convolutional neural network to automatically learn features; convolution layer: used to extract local features of the image; pooling layer: used to reduce the spatial dimension of the feature and increase the receptive field; fully connected layer: used to map the learned features to the final classification result.
[0012] The convolutional neural network convolutional layer formula:
[0013] y=f(Wx+b)
[0014] Where: y: output feature map; W: convolution kernel weight; x: input feature map; b: bias term; f: activation function;
[0015] The maximum pooling formula of the pooling layer is:
[0016] y=max(x)
[0017] Where: y: output feature map; x: local area of the input feature map;
[0018] The fully connected layer formula:
[0019] y=Wx+b
[0020] Where: y: output vector; W: weight matrix; x: input vector; b: bias term.
[0021] The loss function of the model training module is as follows:
[0022]
[0023] Where: L: loss value; y i : One-hot encoding of the true label; The probability distribution of the model's predictions.
[0024] The beneficial effects of the present invention are:
[0025] The present invention is a dermatoscope image processing and diagnosis system based on artificial intelligence. Compared with the prior art, the present invention has the following technical effects:
[0026] Improve diagnostic accuracy: Use deep learning algorithms to automatically extract features from images, reducing interference from human factors and improving diagnostic consistency and accuracy.
[0027] Improve diagnostic efficiency: The system can quickly process a large number of images, greatly shortening the diagnosis time and improving work efficiency.
[0028] Assisting resource-scarce areas: The system can be deployed on a cloud server and provide services through the Internet, so that patients in remote areas can also enjoy high-quality diagnostic services.
[0029] Data-driven diagnosis: The system analyzes a large amount of labeled image data, learns the characteristics of lesions, and realizes data-driven intelligent diagnosis.
[0030] User-friendly interface: A simple and intuitive user interface is designed to enable doctors to easily upload images and interpret diagnostic results.
[0031] Data security and regulatory compliance: The system uses advanced data encryption and security measures to ensure the security of patient data and comply with relevant medical regulatory requirements.
[0032] By implementing this technical solution, the quality and efficiency of diagnosis of skin lesions can be significantly improved, while providing doctors and patients with a better service experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 It is a system structure principle block diagram of the present invention. DETAILED DESCRIPTION
[0034] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.
[0035] like Figure 1As shown: the present invention includes a data acquisition module, an image preprocessing module, a feature extraction module, a model training module, a diagnosis decision module and a user interface, the signal output end of the data acquisition module is connected to the data input end of the image preprocessing module, the data output end of the image preprocessing module is connected to the data input end of the feature extraction module, the data output end of the feature extraction module is connected to the data input end of the model training module, the data output end of the model training module is connected to the data input end of the diagnosis decision module, and the data output end of the diagnosis decision module is connected to the data input end of the user interface.
[0036] Furthermore, the data acquisition module collects dermoscopic image data, which includes lesion images of different races, ages, genders and skin types. Dermatologists annotate the images, including lesion type, size and shape; the image preprocessing module uses a filter to remove image noise, standardizes the image size and pixel range, and enhances image contrast; the feature extraction module extracts color, texture, and shape features, and uses a convolutional neural network to automatically learn image features; the model training module uses annotated data sets to train the model, and uses a cross-validation method to optimize model performance. The trained model of the diagnostic decision module classifies new images, gives possible lesion types, and provides a confidence score for each classification result. The user interface uploads images and views results.
[0037] Data collection and preprocessing
[0038] Data collection: The system requires a large amount of dermatoscopic image data, which usually comes from medical institutions and needs to be labeled by professional doctors, including information such as lesion type, location, size, etc.
[0039] Image preprocessing: includes operations such as denoising, contrast enhancement, and normalization to improve image quality and prepare for subsequent processing.
[0040] Feature extraction:
[0041] Traditional feature extraction: Computer vision techniques are used to extract image features such as color, texture, and shape. For example, a color histogram can provide information about the color distribution of an image, while an edge detection algorithm can identify contours and boundaries in an image.
[0042] Deep learning feature extraction: Use convolutional neural networks (CNNs) to automatically learn complex features from images. CNNs abstract and extract image features layer by layer through multiple layers of convolution and pooling operations.
[0043] Model training:
[0044] Model selection: Choose a deep learning model architecture suitable for image classification tasks, such as ResNet, Inception, or VGG.
[0045] Training process: The model is trained using the preprocessed and feature extracted image data. During the training process, the model learns how to identify different skin lesions based on the features of the input image.
[0046] Hyperparameter tuning: Adjust the model’s hyperparameters, such as learning rate, batch size, convolution kernel size, etc., to improve the model’s performance.
[0047] Diagnostic decision making:
[0048] Classifier: The trained model classifies new dermoscopic images and predicts the possible lesion type in the image.
[0049] Confidence score: Provides a confidence score for the classification results to help doctors judge the diagnostic reliability of the model.
[0050] Multi-model fusion: Combining the prediction results of multiple models to improve the accuracy of diagnosis through methods such as voting or averaging.
[0051] user interface:
[0052] Interaction Design: Design an easy-to-use user interface that enables doctors to conveniently upload images and receive diagnostic results.
[0053] Result presentation: Clearly display the model’s diagnostic results and confidence scores to help doctors make decisions.
[0054] Specifically, the feature extraction module includes traditional feature extraction and deep learning feature extraction. The traditional feature extraction includes color histogram, texture analysis, and shape analysis; color histogram: calculates the distribution of each color in the image; texture analysis: uses gray co-occurrence matrix (GCM) or Gabor filter method; shape analysis: uses edge detection algorithm or contour tracking algorithm; the deep learning feature extraction uses convolutional neural network to automatically learn features; convolution layer: used to extract local features of the image; pooling layer: used to reduce the spatial dimension of the feature and increase the receptive field; fully connected layer: used to map the learned features to the final classification result.
[0055] The convolutional neural network convolutional layer formula:
[0056] y=f(Wx+b)
[0057] Where: y: output feature map; W: convolution kernel weight; x: input feature map; b: bias term; f: activation function;
[0058] The maximum pooling formula of the pooling layer is:
[0059] y=max(x)
[0060] Where: y: output feature map; x: local area of the input feature map;
[0061] The fully connected layer formula:
[0062] y=Wx+b
[0063] Where: y: output vector; W: weight matrix; x: input vector; b: bias term.
[0064] The loss function of the model training module is as follows:
[0065]
[0066] Where: L: loss value; y i : One-hot encoding of the true label; The probability distribution of the model's predictions.
[0067] Convolution layer: Through the convolution operation, the input image is element-wise multiplied by the convolution kernel and then summed to generate a new feature map that captures the local features of the image.
[0068] Activation function: Setting all negative values to 0 helps alleviate the vanishing gradient problem during training.
[0069] Pooling layer: Through pooling operation, the spatial dimension of the feature map is reduced, the number of parameters is reduced, and the invariance of the features is improved.
[0070] Fully connected layer: maps the high-level features extracted by the convolutional layer and the pooling layer to the final classification result.
[0071] Loss function: measures the difference between the model's predicted value and the true value, used to guide model training.
[0072] Optimization algorithm: By calculating the gradient of the loss function, the network parameters are updated to reduce the prediction error.
[0073] Backpropagation: An algorithm for training neural networks that uses gradient descent to minimize a loss function by iteratively adjusting the weights and biases in the network.
[0074] Example:
[0075] Development of a dermatoscope image diagnosis system based on artificial intelligence
[0076] 1. Project initiation and planning
[0077] Objective definition: Develop an artificial intelligence system that can assist doctors in diagnosing skin lesions.
[0078] Team Building: Assemble a multidisciplinary team including data scientists, software engineers, dermatologists, and UI / UX designers.
[0079] 2. Data Collection
[0080] Partners: Establish partnerships with multiple hospitals and skin clinics to collect dermoscopic image data.
[0081] Data annotation: Dermatologists annotate the collected images, including lesion type, boundaries, severity, etc.
[0082] 3. Data Preprocessing
[0083] Denoising: Remove image noise using a median filter.
[0084] Contrast enhancement: Apply histogram equalization technology to improve image contrast.
[0085] Normalize: Resize all images to a uniform size and resolution.
[0086] 4. Feature extraction and model design
[0087] Feature extraction: Use deep learning frameworks (such as TensorFlow or PyTorch) to design CNN models to automatically extract image features.
[0088] Model architecture: Choose a pre-trained CNN model (such as ResNet-50) as a basis and adjust it according to the characteristics of dermoscopic images.
[0089] 5. Model training and optimization
[0090] Training set: Use labeled image data to train the CNN model.
[0091] Validation set: Use an independent validation set to evaluate model performance and perform hyperparameter tuning.
[0092] Loss function: The cross entropy loss function is used to optimize the classification accuracy of the model.
[0093] 6. Diagnostic decision system
[0094] Classifier: The trained CNN model is used as a classifier to predict the lesion type of new images.
[0095] Confidence: Generate a confidence score for each classification result for the doctor’s reference.
[0096] 7. User Interface Development
[0097] UI Design: Design a simple and intuitive user interface to facilitate doctors to upload images and view diagnosis results.
[0098] Interaction: Realize the interactive process of image uploading, result display and user feedback.
[0099] 8. System Integration Testing
[0100] Module testing: Each module is tested individually to ensure correct functionality.
[0101] Integration test: After all modules are integrated, an overall test is performed to ensure system stability.
[0102] 9. Security and Compliance
[0103] Data security: Implement encryption measures to protect patient data security.
[0104] Compliance checks: Ensure that the system complies with regulatory requirements for relevant medical devices.
[0105] 10. Deployment and maintenance
[0106] Cloud deployment: Deploy the system on a cloud server to facilitate remote access and maintenance.
[0107] Maintenance plan: Develop a system maintenance and update plan to ensure long-term stable operation.
[0108] 11. User training and support
[0109] Training materials: Produce training videos and manuals to help doctors quickly master how to use the system.
[0110] Technical support: Provide technical support services to answer questions users encounter during use.
[0111] 12. Performance evaluation and clinical trials
[0112] Evaluation metrics: Define evaluation metrics such as diagnostic accuracy, user satisfaction, etc.
[0113] Clinical trials: Conduct clinical trials in partner medical institutions, collect feedback and optimize the system.
[0114] 13. Commercialization and Marketing
[0115] Pricing strategy: Develop a reasonable pricing strategy based on market research and cost analysis.
[0116] Promotional activities: Promote the system by participating in medical industry conferences, publishing research papers, etc.
[0117] 14. Continuous Improvement
[0118] User feedback: Collect user feedback and continuously improve system functions and user experience.
[0119] Technology update: Follow the latest artificial intelligence technology and continuously improve system performance.
[0120] Through the above embodiment, the whole process from project initiation to system development, testing, deployment and commercialization is demonstrated. This embodiment can be adjusted according to actual conditions to meet the needs of specific projects.
[0121] The technical solution of the present invention is not limited to the above-mentioned specific embodiments. All technical variations made according to the technical solution of the present invention fall within the protection scope of the present invention.
Claims
1. A dermatoscope image processing and diagnosis system based on artificial intelligence, characterized by: It includes a data acquisition module, an image preprocessing module, a feature extraction module, a model training module, a diagnosis decision module and a user interface, wherein the signal output end of the data acquisition module is connected to the data input end of the image preprocessing module, the data output end of the image preprocessing module is connected to the data input end of the feature extraction module, the data output end of the feature extraction module is connected to the data input end of the model training module, the data output end of the model training module is connected to the data input end of the diagnosis decision module, and the data output end of the diagnosis decision module is connected to the data input end of the user interface.
2. The artificial intelligence-based dermoscopic image processing and diagnosis system according to claim 1, characterized in that: The data acquisition module collects dermoscopic image data, which includes lesion images of different races, ages, genders and skin types. Dermatologists annotate the images, including lesion type, size and shape. The image preprocessing module uses a filter to remove image noise, standardizes image size and pixel range, and enhances image contrast. The feature extraction module extracts color, texture, and shape features, and uses a convolutional neural network to automatically learn image features. The model training module uses annotated data sets to train models and uses a cross-validation method to optimize model performance. The trained model of the diagnostic decision module classifies new images, gives possible lesion types, and provides a confidence score for each classification result. The user interface uploads images and views results.
3. The artificial intelligence-based dermoscopic image processing and diagnosis system according to claim 2, characterized in that: The feature extraction module includes traditional feature extraction and deep learning feature extraction. The traditional feature extraction includes color histogram, texture analysis, and shape analysis; color histogram: calculates the distribution of each color in the image; texture analysis: uses gray co-occurrence matrix (GCM) or Gabor filter method; shape analysis: uses edge detection algorithm or contour tracking algorithm; the deep learning feature extraction uses convolutional neural network to automatically learn features; Convolutional layer: used to extract local features of the image; Pooling layer: used to reduce the spatial dimension of the features and increase the receptive field; Fully connected layer: used to map the learned features to the final classification results.
4. The artificial intelligence-based dermoscopic image processing and diagnosis system according to claim 3, characterized in that: The convolutional neural network convolutional layer formula: y=f(Wx+b) Where: y: output feature map; W: convolution kernel weight; x: input feature map; b: bias term; f: activation function; The maximum pooling formula of the pooling layer is: y=max(x) Where: y: output feature map; x: local area of the input feature map; The fully connected layer formula: y=Wx+b Where: y: output vector; W: weight matrix; x: input vector; b: bias term.
5. The artificial intelligence-based dermoscopic image processing and diagnosis system according to claim 4, characterized in that: The loss function of the model training module is as follows: Where: L: loss value; y i : One-hot encoding of the true label; The probability distribution of the model's predictions.