Method for evaluating skin healing rate of burn model
By using a multimodal image acquisition and fusion evaluation model, the problem of insufficient information capture in burn models has been solved, enabling comprehensive characterization and dynamic quantitative evaluation of burn wounds, thus improving the scientific rigor and practicality of the evaluation.
Patent Information
- Application Number
- CN202511550040.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-28
- Publication Date
- 2026-02-10
AI Technical Summary
Existing technologies cannot simultaneously capture wound morphology, physiological and deep structural information in burn models. Furthermore, during data processing, image segmentation often suffers from low contrast between full-thickness defects and epithelialized areas, leading to the inability of existing technologies to effectively synchronize information. These technical problems cannot be accurately assessed.
By acquiring multimodal image data in a burn model, including visible light color images, thermal infrared images, and OCT images, image preprocessing and registration are performed, multidimensional feature parameters are extracted, and a dynamic healing index is output through a fusion evaluation model, which is then evaluated using a time-series attention mechanism.
It enables comprehensive characterization of burn wounds, enhances the scientific rigor and practicality of assessment, supports individual differences and adaptability, and provides dynamic quantification and forward-looking prediction.
Smart Images
Figure CN121506437A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of skin healing rate evaluation, in particular to a burn model skin healing rate evaluation method. BACKGROUND
[0002] Burn model is an experimental or simulation system used to simulate the occurrence, development process and related pathophysiological changes of human burns, which is a key tool connecting basic research and clinical application, mainly serving the fields of burn mechanism research, treatment method development, drug efficacy evaluation and clinical teaching, etc. Its core goal is to reproduce the key characteristics of burns (such as thermal injury, tissue necrosis, inflammatory response, etc.) under controllable conditions, providing scientific basis and technical support for burn prevention and treatment, and providing a standardized platform for scientific research.
[0003] Burn model skin healing rate evaluation is to measure the dynamic process of damaged skin from injury to repair and regeneration in a burn experimental model through quantitative or qualitative methods, aiming to objectively reflect the influence of treatment effect, pathological mechanism or intervention measures on wound repair. Its core is to quantify the rate and quality of the complex biological process of "skin healing" through standardized indicators and detection means, providing key data support for burn treatment research, drug / device efficacy evaluation and mechanism exploration.
[0004] The existing burn model skin healing rate evaluation method mainly relies on a single modality such as visible light, which is difficult to synchronously capture wound morphology, physiology and deep structure information, and the image segmentation in the data processing process often lacks sufficient precision due to low contrast between full-layer defects and epithelialization area. In addition, model training often ignores the weight of time sequence features and lacks multi-modal guidance, which is easily disturbed by environmental temperature and humidity, and the evaluation accuracy and generalization are limited. Therefore, the present application proposes a burn model skin healing rate evaluation method to solve the problems existing in the prior art. SUMMARY
[0005] To solve the above problems, the present application proposes a burn model skin healing rate evaluation method, which solves the problems that the existing burn model skin healing rate evaluation method is difficult to synchronously capture wound morphology, physiology and deep structure information, and the image segmentation in the data processing process often lacks sufficient precision due to low contrast between full-layer defects and epithelialization area, and model training often ignores the weight of time sequence features.
[0006] In order to achieve the purpose of the present application, the present application realizes the following technical scheme: a burn model skin healing rate evaluation method, comprising the following steps:
[0007] Step one: after modeling the burn model, synchronous multi-modal image data acquisition is performed on the same burn wound area at different preset time points, including visible light color image, thermal infrared image and OCT image;
[0008] Step two: Preprocessing the collected multi-modal image data, including color correction, noise reduction and scale normalization, and spatially registering the multi-modal image data based on the wound contour feature points;
[0009] Step three: Extracting multi-dimensional feature parameters related to wound healing from the spatially registered multi-modal image data, including morphological parameters extracted from visible light color images through a segmentation model based on U-Net architecture, physiological activity parameters extracted from thermal infrared images, and structural depth parameters extracted from OCT images;
[0010] Step four: Inputting the extracted multi-dimensional feature parameters into a pre-trained healing index fusion evaluation model, and outputting a comprehensive quantitative index, i.e., a dynamic healing index, through a weighted fusion algorithm;
[0011] Step five: Generating a wound healing process curve based on the dynamic healing index at multiple time points, and automatically generating an evaluation report containing the trends of each dimension parameter and the final healing evaluation.
[0012] Further improvement lies in that in the step one, the OCT is a swept source OCT, and the OCT is built-in with a Doppler blood flow imaging function to synchronously collect microvascular blood flow signals of the wound and the surrounding area, and to calculate the blood perfusion rate per unit area.
[0013] Further improvement lies in that in the step three, when calculating the collagen density score of the dermis layer of the OCT, first, polarization-sensitive analysis is performed on the original interference spectrum of the OCT to distinguish collagen fibers with different arrangement directions, then the scattering anisotropy coefficient is calculated based on the depolarization signal intensity, and then the scattering anisotropy coefficient is mapped to a 0-10 scale to obtain the collagen density score.
[0014] Further improvement lies in that in the step four, the training process of the healing index fusion evaluation model includes:
[0015] S1, collecting burn model samples with known final healing grades, obtaining multi-time point multi-modal images and standard healing quality scores of end-point histopathology thereof;
[0016] S2, extracting multi-dimensional feature parameters from the historical database, and screening the top N high-correlation feature subsets by using the recursive feature elimination method;
[0017] S3, pre-training the XGBoost algorithm with public skin pathology data sets, then fine-tuning with burn samples, optimizing hyperparameters through five-fold cross-validation, and finally obtaining a regression model that can accurately predict the dynamic healing index.
[0018] Further improvement lies in that in the step S3, the feature parameters at different time points of the same wound are regarded as a sequence, so that the model learns and weighs the contribution of the features at different healing stages to the final index.
[0019] Further improvement lies in that in the step three, the segmentation model based on the U-Net is trained by adopting a multi-modal guidance strategy and a hybrid loss function, wherein:
[0020] The training pair is a registered visible light + OCT pixel-level annotation image.
[0021] Feature fusion: the visible light extracts surface morphological features, the OCT extracts deep structure features, and the fusion outputs a layered segmentation result.
[0022] Hybrid loss: Dice loss + Focal Loss, so that the model can accurately distinguish full-layer defects, partial epithelialization and complete epithelialization zones only by inputting the visible light.
[0023] Further improvement lies in that in the step one, the multi-modal data acquisition further comprises high-frequency ultrasonic imaging, so as to obtain three-dimensional ultrasonic data and echo signals of the wound surface.
[0024] Further improvement lies in that in the step four, when obtaining the standard healing score of a new burn model sample, the healing index fusion evaluation model automatically inputs the feature parameters and the prediction error of the new sample into the model, fine-tunes the model weight by the gradient descent method, and improves the evaluation adaptability to new burn features.
[0025] The present application has the following advantages: the present application captures the surface healing process through visible light images, and the OCT and ultrasonic imaging respectively provide deep information such as dermal collagen remodeling, granulation filling and subcutaneous edema, and the thermal infrared monitoring dynamically monitors local blood flow and temperature, so that the multi-source data form a unified analysis benchmark after accurate registration, and the synchronous acquisition and cross-modal registration are completed, thereby realizing the all-round characterization of the macroscopic morphology, microscopic structure and physiological activity of the burn wound.
[0026] The evaluation model introduces a time sequence attention mechanism, weights the contribution of features at different stages to the final healing index, breaks through the limitation of static evaluation, and supports the model to be continuously optimized with new sample data, adapts to individual differences of different burn models, in addition, the prognosis prediction module combines the current index and the feature trend to provide a probabilistic reference for the healing time range, enhances the explainability and clinical guidance value of the result, and improves the scientificity and practicality of the burn healing evaluation, realizes the dynamic quantification and prospective prediction of the healing process. BRIEF DESCRIPTION OF DRAWINGS
[0027] Figure 1 is a flowchart of a burn model skin healing rate evaluation method of the present application. DETAILED DESCRIPTION
[0028] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.
[0029] At present, the evaluation of burn model skin healing rate has formed a comprehensive technical system from macro to micro and from structure to function. The most basic and core is macro morphometry, which accurately calculates the area change of the wound surface by regular digital photography and image analysis software, obtains the healing rate percentage and complete epithelialization time, and intuitively reflects the healing speed. However, macroscopic closure does not mean high-quality repair, so histopathology is needed for micro-precision judgment. Through hematoxylin-eosin staining, the degree of inflammation, granulation tissue growth and re-epithelialization is evaluated, and the deposition and arrangement of collagen fibers are observed by Masson trichrome staining, so as to scientifically score the healing quality. In order to further explore the internal mechanism, immunohistochemical technology is widely used, which reveals the cytological basis of tissue regeneration by labeling specific targets such as proliferating cells and new blood vessels. At the molecular level, techniques such as quantitative PCR and Western blotting can quantitatively analyze the expression dynamics of key genes such as growth factors and collagen. Finally, biomechanical testing evaluates the tensile strength and elasticity of the healed skin, and verifies whether it restores the mechanical properties of normal skin from the functional level. This multi-dimensional evaluation system complements each other and provides solid and comprehensive scientific basis for the research and development of new strategies for burn treatment.
[0030] It should be noted that the technical means not described in detail in the following embodiments are conventional means in the art, which are not the key points of the invention and will not be described.
[0031] Reference Figure 1 The present embodiment provides a burn model skin healing rate evaluation method, comprising the following steps:
[0032] Step one, multi-modal data acquisition
[0033] After the burn model is modeled, the same burn wound area is synchronously imaged at different time points, and the multi-modal image data collected includes visible light color images, thermal infrared images and OCT (optical coherence tomography) images;
[0034] Step two, image preprocessing and registration
[0035] The collected multi-modal image data is pre-processed (color correction, noise reduction, scale normalization), and the visible light color image, thermal infrared image and optical coherence tomography (OCT) image are spatially registered based on the wound contour feature points, to ensure that the wound area pixels in each image are aligned at the pixel level;
[0036] Step three, multi-dimensional feature extraction
[0037] From the spatially registered multi-modal image data, multi-dimensional feature parameters related to wound healing are extracted, including:
[0038] Morphological parameters: the total wound area, epithelialization edge advancement distance, wound perimeter to area ratio are extracted from the visible light color image by a segmentation model based on U-Net architecture, the U-Net based segmentation model is trained using a multi-modal guided strategy + hybrid loss function:
[0039] Training pair: registered visible light + OCT pixel-level labeled image;
[0040] Feature fusion: visible light extracts surface morphological features, OCT extracts deep structure features, and the fused output is a layered segmentation result;
[0041] Hybrid loss: Dice loss (optimizes overall accuracy) + Focal Loss (solves class imbalance between full-thickness defect and epithelialization area), so that the model can accurately distinguish full-thickness defect, partial epithelialization and complete epithelialization area by only inputting visible light;
[0042] Physiological activity parameters: temperature difference between wound center and surrounding normal skin, average temperature of wound area and temperature distribution uniformity are extracted from thermal infrared image;
[0043] Structural depth parameters: neovascular epithelial layer thickness, dermal collagen density score and granulation tissue filling depth are extracted from optical coherence tomography (OCT) image, the calculation method of dermal collagen density score is:
[0044] Polarization-sensitive analysis is performed on the OCT original interference spectrum to distinguish collagen fibers with different arrangement directions;
[0045] Based on the depolarization signal intensity, the scattering anisotropy coefficient is calculated;
[0046] The scattering anisotropy coefficient is mapped to a scale of 0-10 to obtain the collagen density score (the higher the score, the more ordered and dense the collagen);
[0047] Step four, fusion evaluation model calculation
[0048] The extracted multi-dimensional feature parameters are input into the pre-trained healing index fusion evaluation model, and a comprehensive quantitative index, dynamic healing index, is output through a weighted fusion algorithm. The training process includes:
[0049] S1, training set construction: collect burn model samples with known final healing grades, obtain their multi-time point multi-modal images and standard healing quality scores of endpoint histopathology;
[0050] S2, feature selection: extract multi-dimensional feature parameters from the historical database, and use the recursive feature elimination method to select the top N highly correlated feature subsets;
[0051] S3, model training and verification: first, pre-train the XGBoost algorithm with public skin pathology data sets, and then fine-tune it with burn samples; through five-fold cross-validation to optimize the hyperparameters, finally get a regression model that can accurately predict the dynamic healing index. In the model training, the time series attention mechanism is introduced: the feature parameters at different time points of the same wound are regarded as a sequence, so that the model can learn and weigh the contribution of different healing stage features to the final index, and improve the time series accuracy;
[0052] The comprehensive quantitative index of the embodiment is the functional target: by integrating multi-dimensional information, the healing quality is comprehensively and objectively quantified (such as distinguishing the healing difference between "only epidermis regeneration" and "deep layer collagen remodeling");
[0053] The dynamic healing index of the embodiment is the implementation: through time series sampling and weighted fusion, the "comprehensive" evaluation is converted into a traceable dynamic value, reflecting the continuity of the healing process (such as index stagnation indicating delayed healing, and index surge indicating infection or excessive proliferation);
[0054] The healing index fusion evaluation model of the embodiment supports online updating:
[0055] When the standard healing score of a new burn model sample is obtained, the feature parameters and prediction error of the new sample are automatically input into the model, and the model weight is fine-tuned through gradient descent method to improve the evaluation adaptability to new burn features;
[0056] Step five, healing process visualization and report generation
[0057] According to the dynamic healing index at multiple time points, a wound healing process curve is generated, and the specific steps are as follows:
[0058] Take time as the horizontal axis and dynamic healing index as the vertical axis to construct the original data point set;
[0059] Use the cubic spline interpolation algorithm to smooth the data point set and generate a continuous healing index trend curve;
[0060] Calculate the 95% confidence interval of the index at each time point, and mark the reliability of the curve by the shaded area;
[0061] By combining the pathological stages of burn healing, key biological nodes such as the completion of epithelialization and the initiation of collagen remodeling are marked to generate a visual healing process curve.
[0062] It automatically generates an assessment report containing trends of parameters across various dimensions and a final healing evaluation. The specific steps are as follows:
[0063] Multi-dimensional parameter integration and trend analysis: Morphological parameters (such as the distance of epithelialization edge advancement, wound perimeter to area ratio), physiological activity parameters (such as temperature difference, blood perfusion rate), and structural depth parameters (such as the thickness of the new epithelial layer, collagen density score) of the same wound at different time points are imported into the time series analysis unit to calculate the slope of change of each parameter (reflecting the healing rate), fluctuation amplitude (reflecting the healing stability), and correlation coefficient (such as the correlation between temperature difference and collagen density).
[0064] Mapping of dynamic healing index to preset standards: Based on the dynamic healing index, combined with preset healing level thresholds (e.g., index < 0.3 is "slow healing", 0.3 to 0.7 is "normal healing", and > 0.7 is "rapid healing"), the healing stage at the current time point is determined, and the final healing level is predicted by the index trend at continuous time points;
[0065] The report is structured and generated by calling up multimodal images, parameter trend graphs (such as epithelialization advancement distance-time curves and collagen density-time curves) and correlation heatmaps (such as the correlation between temperature and blood perfusion) from the data storage module to generate visual content that combines text and graphics.
[0066] Based on the results of parameter trend analysis (such as "the epithelialization rate reaches its peak on day 7, with a slope of 0.15 mm / day"), the results of healing stage determination and prediction, the system automatically generates structured text descriptions (including "current healing status", "key parameter change trends", and "potential risk warnings" such as "decreased temperature uniformity, be alert for infection").
[0067] Integrate visual content with textual descriptions to output a standardized assessment report that includes "Healing Progress Summary", "Final Healing Evaluation", and "Treatment Recommendations".
[0068] In this embodiment, the OCT is a swept-source OCT (center wavelength 1300-1400nm, scanning depth 1.5-2.5mm, axial resolution ≥8μm). The OCT has a built-in Doppler blood flow imaging function, which simultaneously acquires microvascular blood flow signals of the wound and surrounding area, calculates the blood perfusion rate per unit area, extracts it as a supplementary physiological activity parameter in step three, and inputs it into the fusion model in step four.
[0069] In this embodiment, the multimodal data collected in step one also includes high-frequency ultrasound imaging (center frequency ≥ 20MHz) to obtain three-dimensional ultrasound data and echo signals of the wound.
[0070] In step three, the maximum thickness of granulation tissue, the total volume of the wound, and the hypoechoic volume of subcutaneous edema are extracted from three-dimensional ultrasound, and the ultrasound attenuation coefficient is calculated from the echo signal.
[0071] The above parameters are used as structural depth / inflammatory response parameters and input into the healing index fusion assessment model.
[0072] This embodiment also provides an evaluation system for assessing the skin healing rate of a burn model. This system comprises an integrated imaging module, a central control module, a data processing and analysis module, and a data storage and visualization module, wherein:
[0073] Integrated imaging module: a high-resolution color camera, a thermal infrared camera, and an OCT probe fixed to a programmable mobile platform. The programmable mobile platform ensures that the optical axes of all devices converge on the same focal plane.
[0074] In this embodiment, the integrated imaging module is placed in a controllable environment chamber. The environment chamber provides a standard D65 light source and a constant temperature environment. It integrates temperature, humidity and air pressure sensors. Before the system acquires data, it automatically reads the environmental parameters and compensates for the absolute temperature of thermal infrared and the optical path of OCT in real time according to the calibration curve to eliminate environmental fluctuation errors.
[0075] The programmable mobile platform in this embodiment adopts a magnetic navigation + visual servo control strategy. It automatically adjusts its position by pre-calibrating the wound coordinates to achieve precise positioning of multimodal devices and reduce human operation errors.
[0076] Central control module: coordinates and controls the mobile platform and various imaging devices, automatically focusing and acquiring data sequentially according to a preset path;
[0077] Data processing and analysis module: It has built-in image registration unit, feature extraction unit, fusion evaluation model, and time series analysis unit, which automatically completes the entire process from image to index calculation;
[0078] Data storage and visualization module: Stores data and dynamically displays multimodal images, feature parameters, healing curves, and evaluation reports in an interactive interface.
[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for evaluating the skin healing rate of a burn model, characterized in that, Includes the following steps: Step 1: After the burn model is created, at different preset time points, synchronous multimodal image data acquisition is performed on the same burn wound area, including visible light color images, thermal infrared images and OCT images; Step 2: Perform preprocessing on the acquired multimodal image data, including color correction, noise reduction and scale normalization, and spatially register the multimodal image data based on the wound contour feature points; Step 3: Extract multidimensional feature parameters related to wound healing from the spatially registered multimodal image data, including morphological parameters extracted from visible light color images, physiological activity parameters extracted from thermal infrared images, and structural depth parameters extracted from OCT images using a segmentation model based on the U-Net architecture. Step 4: Input the extracted multidimensional feature parameters into the pre-trained healing index fusion evaluation model, and output a comprehensive quantitative index, namely the dynamic healing index, through a weighted fusion algorithm; Step 5: Generate a wound healing process curve based on the dynamic healing index at multiple time points, and automatically generate an assessment report that includes the trends of parameters in each dimension and the final healing evaluation.
2. The method for evaluating skin healing rate in a burn model according to claim 1, characterized in that: In step one, the OCT is a swept-source OCT, and the OCT has a built-in Doppler blood flow imaging function to simultaneously acquire microvascular blood flow signals of the wound and surrounding area and calculate the blood perfusion rate per unit area.
3. The method for evaluating skin healing rate in a burn model according to claim 1, characterized in that: In step three, when calculating the OCT dermal collagen density score, polarization sensitivity analysis is first performed on the original OCT interference spectrum to distinguish collagen fibers with different orientations. Then, the scattering anisotropy coefficient is calculated based on the depolarized signal intensity. Finally, the scattering anisotropy coefficient is mapped to a scale of 0 to 10 to obtain the collagen density score.
4. The method for evaluating skin healing rate in a burn model according to claim 1, characterized in that: In step four, the training process of the healing index fusion assessment model includes: S1. Collect burn model samples with known final healing grades, and obtain their multi-time point multimodal images and standard healing quality scores of endpoint histopathology. S2. Extract multidimensional feature parameters from the historical database and use the recursive feature elimination method to filter the subset of features with the highest correlation to the top N features. S3. First, the XGBoost algorithm is pre-trained using a publicly available dermatopathology dataset. Then, it is fine-tuned using burn samples. The hyperparameters are optimized through five-fold cross-validation, and finally, a regression model that can accurately predict the dynamic healing index is obtained.
5. The method for evaluating skin healing rate in a burn model according to claim 4, characterized in that: In step S3, the feature parameters of the same wound at different time points are regarded as a sequence, so that the model can learn and weigh the contribution of each feature at different healing stages to the final index.
6. The method for evaluating skin healing rate in a burn model according to claim 1, characterized in that: In step three, the U-Net-based segmentation model is trained using a multimodal guidance strategy and a hybrid loss function, wherein: Training pairs: Registered visible light + OCT pixel-level labeled images; Feature fusion: visible light is used to extract surface morphological features, and OCT is used to extract deep structural features. The fused features are then used to output a layered segmentation result. Hybrid loss: Dice loss + Focal Loss, enabling the model to accurately distinguish between full-layer defects, partial epithelialization, and fully epithelialized areas using only visible light input.
7. The method for evaluating skin healing rate in a burn model according to claim 1, characterized in that: In step one, multimodal data acquisition also includes high-frequency ultrasound imaging to acquire three-dimensional ultrasound data and echo signals of the wound.
8. The method for evaluating skin healing rate in a burn model according to claim 1, characterized in that: In step four, when the healing index fusion assessment model obtains the standard healing score of a new burn model sample, it automatically inputs the feature parameters and prediction error of the new sample into the model and fine-tunes the model weights through the gradient descent method to improve the assessment adaptability to new burn features.