Skin burn degree identification method
Through the skin burn degree recognition method based on the ResNet18 model, combined with technologies such as data enhancement, attention mechanism and transfer learning, the problems of insufficient accuracy and low efficiency in burn assessment are solved, and high-precision burn grade recognition and diagnostic support are achieved.
Patent Information
- Application Number
- CN202411430015.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-14
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has problems of insufficient accuracy, low efficiency and low recognition accuracy in burn skin assessment, making it difficult to accurately distinguish burn grades, especially in complex situations.
The skin burn degree recognition method based on the ResNet18 model is adopted, including data collection, data preprocessing, classification algorithm, evaluation and optimization, and post-processing modules. The model is improved through data augmentation, attention mechanism, transfer learning and premature stopping methods and other technologies. Generalization ability and classification accuracy.
Significantly improves the accuracy of burn skin classification, reduces the rate of misdiagnosis, reduces the workload of doctors, improves identification efficiency, provides a more accurate and reliable assessment of burn grades, suitable for the preliminary diagnosis and personalized treatment of various burn cases Plan formulation.
Smart Images

Figure HDA0005083397220000011 
Figure HDA0005083397220000021 
Figure HDA0005083397220000031
Abstract
Description
Technical field:
[0001] The present invention belongs to the technical field of medical image processing, and specifically relates to a method for identifying the degree of skin burns based on a deep learning algorithm, which aims to classify and analyze burned skin for assisting clinical diagnosis and treatment. Background technology:
[0002] In the process of burn treatment, accurate assessment of burn depth and range is the key to formulating a treatment plan. At present, the assessment of burned skin mainly relies on the doctor's experience and manual interpretation, which is greatly affected by subjective factors and has problems such as insufficient accuracy and low efficiency. Traditional image processing technology also faces challenges in burn classification, such as low recognition accuracy and difficulty in dealing with complex situations. With the development of artificial intelligence technology, deep learning methods have shown excellent performance in the field of image recognition, providing a new solution to the problem of burn classification. Summary of the invention:
[0003] The purpose of the present invention is to provide a method for identifying the degree of skin burns based on the ResNet18 model to solve the above-mentioned deficiencies in the prior art.
[0004] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: a method for identifying the degree of skin burns based on the ResNet18 model, comprising a data acquisition module, a data preprocessing module, a classification algorithm module, an evaluation and optimization module and a post-processing module.
[0005] The data acquisition module is used to collect skin image data of different burn levels and build a diverse data set. The module can obtain skin surface and deep layer image data through professional medical equipment. The data acquisition module can collect data for first-degree, second-degree and third-degree burns, provide data support for subsequent model training and verification, and can perform data annotation to improve the accuracy and effectiveness of training data.
[0006] The data preprocessing module is used to process the skin images collected by the data acquisition module, including image denoising, normalization and data enhancement. This module reduces image noise through Gaussian blur method, enhances image contrast through histogram equalization, and normalizes the image to the range of [0,1]. Through data enhancement operations such as random rotation, flipping, scaling, etc., the diversity of data is increased to improve the generalization ability and robustness of the model.
[0007] The classification algorithm module is used to classify the preprocessed burn skin images. The module is built based on the ResNet18 model and combines the attention mechanism to enhance the feature extraction capability. After inputting the image, the model can output the corresponding first-degree, second-degree or third-degree burn classification results, and use the Softmax classifier to provide the probability value of each category. The module can accurately distinguish different burn grades and improve the precision and accuracy of classification.
[0008] The evaluation and optimization module is used to evaluate the performance of the classification algorithm module and optimize the model. Through performance evaluation on the training set, validation set and test set, including indicators such as accuracy, recall rate and F1-score, the evaluation and optimization module can adjust the model parameters and hyperparameters in a timely manner, and use early stopping and learning rate scheduling methods to prevent overfitting, thereby ensuring that the performance of the model reaches the optimal level.
[0009] The post-processing module is used to further optimize the classification results output by the classification algorithm module. This module uses a time series smoothing method to smooth the classification results of multiple frames of images to reduce the misjudgment of single-frame images. Through this module, the stability and reliability of the classification results are improved, providing doctors with more accurate and reliable burn grade information, helping them to make accurate diagnoses in a short time.
[0010] Furthermore, the data preprocessing module is also used to uniformly resize images of different resolutions and sizes to ensure that the images input to the model have a consistent size. The module can automatically adjust the brightness, contrast and color balance of the image to reduce the impact of equipment or environmental differences during image capture. In addition, the data preprocessing module also includes an image alignment operation based on feature points to ensure that skin areas captured at different angles can be more accurately aligned to improve classification accuracy. The data preprocessing module also supports the segmentation of burn skin images, dividing large areas of burn areas into multiple groups of small skin pieces for subsequent small target recognition and improving the ability to detect subtle burn features.
[0011] Furthermore, the classification algorithm module is also used to improve the feature extraction capability of burned skin images by using transfer learning technology through the ResNet18 model pre-trained on a large image dataset. By using the attention mechanism, the module can automatically focus on the burn area and reduce the interference of the background area, thereby further improving the accuracy and stability of classification. The classification algorithm module supports independent classification and identification of small skin areas after segmentation, and generates refined classification results for each small area to ensure accurate capture of small-scale burn features. The classification algorithm module also supports adjustment through specific parameters to meet the needs of datasets of different types and sizes.
[0012] Furthermore, the evaluation and optimization module is also used to automatically generate a model performance report and automatically adjust training parameters such as optimizer type, learning rate, batch size, etc. based on the evaluation results. This module can dynamically adjust the learning rate scheduling strategy during the model training process to achieve rapid convergence in the early stages of training and fine-tune in the later stages. In addition, the evaluation and optimization module also supports the use of K-fold cross-validation to ensure that the evaluation results of the model are more stable and comprehensive. A comprehensive analysis is performed on the recognition results of each group of small pieces of skin to improve the overall classification performance, and the ability to distinguish different burn grades is improved by accurately identifying small burn areas.
[0013] Furthermore, the post-processing module is also used to perform weighted averaging on the classification results of multiple frames of images to reduce the impact of single-frame abnormal data on the overall classification results. The module can dynamically adjust the weight of the classification results according to the time series information, further smooth the result output, and reduce the impact of environmental changes and slight patient movements. The post-processing module provides more refined classification results through aggregation analysis for the classification results of multiple groups of small skin patches after segmentation, so as to accurately distinguish the burn grade of each small area, thereby significantly improving the stability of the overall classification results and providing doctors with a more continuous and reliable diagnostic reference. Beneficial effects:
[0014] The present invention enhances the model's ability to focus on key features by introducing a method based on the attention mechanism, thereby significantly improving the accuracy of burn skin classification, reducing the misdiagnosis rate, and ensuring that doctors can assess the degree of burns more accurately.
[0015] The present invention adopts a fully automated recognition process, which significantly reduces the workload of doctors and improves recognition efficiency in a short period of time, making clinical applications more convenient and capable of handling more cases.
[0016] The present invention adopts an advanced training strategy that combines the early stopping method and the learning rate scheduling method to ensure that the model can converge in the optimal state and effectively prevent overfitting, thereby improving the generalization ability of the model and adapting to different scenarios.
[0017] The design of the present invention has wide applicability, making the method suitable for the preliminary diagnosis of various burn cases, and can provide a reliable scientific basis for doctors to formulate personalized treatment plans, thereby improving clinical treatment effects. Description of the drawings:
[0018] In order to more clearly illustrate the embodiments of the present invention 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0019] Figure 1 This is a schematic diagram of the algorithm system structure of a method for identifying the degree of skin burns according to the present invention;
[0020] Figure 2 A schematic diagram of an algorithm and steps of a method for identifying the degree of skin burns according to the present invention;
[0021] Figure 3 A schematic diagram of the structure of a Resnet18 neural network model algorithm for a method for identifying the degree of skin burns in the present invention;
[0023] Figure 4 A schematic diagram of the data preprocessing flow of an algorithm for identifying the degree of skin burns in the present invention; Specific implementation method:
[0024] The present invention provides a method for identifying the degree of skin burns, comprising a data acquisition module, a data preprocessing module, a classification algorithm module, an evaluation and optimization module and a post-processing module. In order to clearly and completely describe the technical solution in the embodiment of the present invention, the following is a detailed description in conjunction with the drawings in the embodiment.
[0025] It should be noted that the described embodiments are only a part of the embodiments of the present invention, not all of them. Without doing creative work, all other embodiments that can be implemented by ordinary technicians in this field based on the content of the present invention are within the protection scope of the present invention.
[0026] Example embodiments will be described more fully below with reference to the accompanying drawings, but the example embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. On the contrary, the purpose of providing these embodiments is to make the present disclosure thorough and complete and to enable those skilled in the art to fully understand the scope of the present disclosure.
[0027] In the absence of conflict, the various embodiments of the present disclosure and the various features therein may be combined with each other.
[0028] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.
[0029] The terms used herein are only used to describe specific embodiments and are not intended to limit the present disclosure. As used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms, unless the context clearly indicates otherwise. It will also be understood that when the terms "comprising" and / or "made of" are used in this specification, the presence of the features, wholes, steps, operations, elements and / or components is specified, but the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or groups thereof is not excluded.
[0030] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted as having an idealized or overly formal meaning unless explicitly defined as such herein.
[0031] See also Figure 1 The skin burn degree recognition method of the present invention is based on the deep learning model ResNet18, which automatically recognizes the skin burn degree through the neural network model. The specific process is as follows:
[0032] Step 1: Data acquisition: Use a high-resolution binocular depth camera to collect skin burn images, obtain multimodal data including visible light images and depth maps, and synchronously record the shooting time information of each image for subsequent processing.
[0033] Step 2: Data preprocessing The collected skin burn images were preprocessed, including image denoising, enhancement, and standardization. Specifically, the original image was resized to a standard size of 480x480, Gaussian blur was used to remove image noise, and histogram equalization was used to enhance image contrast. At the same time, the image was normalized to standardize the pixel values to the range of [0,1] so that it could be input into the neural network for training and reasoning.
[0034] Step 3: Classification algorithm The preprocessed image is input into the ResNet18 network model for feature extraction and classification. The model extracts deep features such as edges, textures, and morphology in the image through multi-layer convolution operations, and enhances the recognition ability of key areas through the attention mechanism. Subsequently, the extracted deep features are input into the fully connected layer, and the burn images are classified using the Softmax classifier. The classification model outputs the probability values of first-degree, second-degree, and third-degree burn categories (corresponding to 0, 1, and 2, respectively), and finally selects the category with the highest probability value as the prediction result.
[0035] Step 4: Evaluation and Optimization Perform performance evaluation on the model, including analysis of classification accuracy and recall on the training set, validation set, and test set. During the training process, the generalization ability of the model is improved by using the early stopping method and learning rate scheduling strategy, and the classification effect of each category is analyzed through the confusion matrix and ROC curve, and the parameters are adjusted in a targeted manner to optimize the classification accuracy.
[0036] Step 5: Post-processing The classification results are post-processed to improve the stability and robustness of the classification. The prediction results of consecutive frames are smoothed by the sliding window method, and the results of multiple frames are voted, and the burn grade with the highest frequency is output as the final prediction result of the time period. This method can effectively reduce occasional errors in single-frame predictions and improve the consistency of time series.
[0037] Step 6: The output result processing unit outputs the final prediction result and its corresponding probability value to the display module and stores it in the database for subsequent analysis and evaluation. In actual application scenarios, if the classification result shows second-degree or third-degree burns, the treatment function of the intelligent therapeutic device can be further triggered to achieve personalized treatment plans for patients.
[0038] The attention mechanism in this method enhances the recognition ability of key image areas by combining channel attention and spatial attention; the Softmax classifier in the classification algorithm module can accurately output the probability of each burn level to achieve refined classification; the post-processing module uses time series smoothing to effectively reduce the misjudgment caused by single-frame image noise.
[0039] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A method for identifying the degree of skin burns, characterized in that: include: Data acquisition module, data preprocessing module, classification algorithm module, evaluation and optimization module and post-processing module; The data acquisition module is used to collect diverse burned skin image data to improve the diversity of the data set and the generalization ability of the model; The data preprocessing module is used to perform denoising, normalization, data enhancement, size adjustment and image alignment on the collected images to improve image quality; The classification algorithm module is based on the ResNet18 model and combines the attention mechanism to classify the preprocessed skin images and accurately distinguish first-degree, second-degree and third-degree burns; The evaluation and optimization module is used to generate performance reports and adjust and optimize model parameters through cross-validation and dynamic learning rate to ensure the accuracy and stability of classification results; The post-processing module improves the continuity and stability of the classification results by weighted averaging and smoothing the multi-frame classification results, and provides doctors with a refined burn grade diagnosis reference.
2. The method for identifying the degree of skin burns according to claim 1, characterized in that: The data acquisition module comprises: Lighting condition control unit and multi-angle shooting unit; The lighting condition control unit is used to adjust the lighting in the shooting environment to ensure that the acquired image is consistent under lighting changes; The multi-angle shooting unit is used to shoot the same burn area at different angles to capture richer feature information.
3. The method for identifying the degree of skin burns according to claim 1, characterized in that: The data preprocessing module includes: an image denoising submodule, an image enhancement submodule and an image scaling submodule; The image denoising submodule is used to remove noise in the image and improve the image quality; The image enhancement submodule is used to improve the contrast and clarity of the image through techniques such as histogram equalization and Gaussian blur; the image scaling submodule is used to adjust the image to a uniform size for easy input into the classification algorithm module.
4. The method for identifying the degree of skin burns according to claim 1, characterized in that: The classification algorithm module further includes: a feature extraction submodule and a classifier submodule; The feature extraction submodule is used to extract important features from the preprocessed image for subsequent classification; The classifier submodule is used to classify the burned skin according to the extracted features and the classification model obtained through training.
5. The method for identifying the degree of skin burns according to claim 1, characterized in that: The evaluation and optimization module includes: a performance evaluation submodule and a model optimization submodule; The performance evaluation submodule is used to evaluate the accuracy and stability of the classification algorithm; The model optimization submodule is used to optimize the performance of the classification model by adjusting hyperparameters and using different training strategies.
6. The method for identifying the degree of skin burns according to claim 1, characterized in that: The post-processing module includes: a result smoothing submodule and a visualization submodule; the result smoothing submodule is used to smooth the multi-frame classification results to improve the continuity of the classification results; the visualization submodule is used to visualize the classification results to facilitate doctors to make judgments and decisions.
7. The method for identifying the degree of skin burns according to any one of claims 1 to 6, characterized in that: The classification algorithm module uses transfer learning to perform model training to improve the model's adaptability to new data.
8. A method for identifying the degree of skin burns, comprising the following steps: S1. Collect diverse burned skin image data; S2, denoising, normalizing, data enhancement, resizing and aligning the collected images; S3, classify the preprocessed images based on the ResNet18 model to identify first-degree, second-degree, and third-degree burns; S4, generate performance reports through evaluation and optimization modules, and optimize classification models; S5. Post-process the classification results to improve the continuity and stability of the results.
Citation Information
Cited By
Pressure damage classification system based on image processing
CN120656001A
Pressure injury classification system based on image processing
CN120656001B
Burn grading and typing prediction method, system and equipment and storage medium
CN120932923A
A burn grading and typing prediction method, system, device and storage medium
CN120932923B
