An image enhancement-based general surgery postoperative rehabilitation effect monitoring system

Abstract

This invention relates to the field of medical image processing and intelligent monitoring technology, specifically to a postoperative rehabilitation effect monitoring system for general surgery based on image enhancement. The system includes: a spatiotemporal feature decoupling step: constructing a dynamic deformation field model, decomposing time-series image data containing non-rigid deformation into a mechanical deformation feature layer and a physiological texture feature layer, and calculating the texture deformation decoupling coefficient; an illumination consistency normalization step: calculating the time-series illumination consistency variance, and performing illumination component correction on the time-series image data based on the variance; an adaptive texture enhancement step: performing weighted enhancement on the physiological texture feature layer according to the decoupling coefficient, generating an enhanced rehabilitation texture image; and an evolutionary trend analysis step: extracting the orderliness index of texture evolution based on the enhanced image, generating monitoring results. This invention effectively solves the interference of mechanical deformation on the analysis of microscopic healing textures, significantly improving the image signal-to-noise ratio and feature extraction accuracy.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and intelligent monitoring technology, specifically to a postoperative rehabilitation effect monitoring system for general surgery based on image enhancement. Background Technology

[0002] In the application scenario of postoperative rehabilitation monitoring in general surgery, the intelligent monitoring system relies on high-fidelity wound texture analysis to ensure the accuracy of healing status assessment and clinical reference value. The monitoring terminal usually needs to combine continuously acquired wound temporal images with microscopic feature extraction algorithms to perceive the dynamic evolution of soft tissue healing in real time.

[0003] For texture enhancement and feature analysis of wound images, existing solutions generally adopt a global image enhancement architecture. This involves acquiring images of the wound area using a high-resolution camera, enhancing the local contrast of the image using contrast-limited adaptive histogram equalization or traditional sharpening operators, and directly using the enhanced image pixel values ​​as ground truth to extract pathological features. Although this approach is feasible to some extent when the patient is completely still or under ideal lighting conditions, it relies too heavily on the grayscale statistical characteristics of the image and lacks logical constraints on the physical deformation mechanisms of biological tissues. When encountering dynamic environments with non-rigid deformations such as patient breathing or turning over, traditional visual algorithms are prone to misidentifying mechanical geometric displacements caused by physical stretching or wrinkling of the skin surface as pathological texture details that should be enhanced, leading to erroneous signal amplification.

[0004] Furthermore, traditional data-driven enhancement algorithms struggle to distinguish between noise signals originating from body movement and valid signals from tissue growth. They also fail to eliminate motion artifacts that do not conform to the spatiotemporal patterns of wound healing. This makes the monitoring process highly susceptible to false positives caused by respiratory movements or the loss of crucial granulation tissue growth characteristics due to smoothing and denoising. Consequently, it is difficult to support the system in high-frequency, high-precision tracking of rehabilitation effects in uncontrolled home environments. Therefore, establishing an analysis mechanism that decouples physical movement from physiological texture features, effectively suppressing mechanical deformation interference while improving the robustness and signal-to-noise ratio of wound healing status capture, has become an urgent technical problem to be solved. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an image enhancement-based postoperative rehabilitation effect monitoring system for general surgery. Specifically, the technical solution of this invention includes:

[0006] A postoperative rehabilitation effect monitoring system for general surgery based on image enhancement includes a cloud processing center, which is communicatively connected to an image sequence acquisition module, a spatiotemporal feature decoupling module, an illumination consistency normalization module, an adaptive texture enhancement module, and an evolution trend analysis module.

[0007] The image sequence acquisition module is used to acquire time-series image data of the target wound area over a continuous period of time. The time-series image data contains geometric displacement information caused by non-rigid deformation.

[0008] The spatiotemporal feature decoupling module is used to construct a dynamic deformation field model containing a non-rigid deformation vector field. Based on the dynamic deformation field model, the time-series image data is decomposed into a mechanical deformation feature layer and a physiological texture feature layer, and the texture deformation decoupling coefficient, which characterizes the degree of independence between the two, is calculated.

[0009] The illumination consistency normalization module is used to calculate the temporal illumination consistency variance of the temporal image data and perform illumination component correction on the temporal image data based on the temporal illumination consistency variance.

[0010] The adaptive texture enhancement module is used to perform weighted enhancement processing on the physiological texture feature layer according to the texture deformation decoupling coefficient, and maintain the structural similarity of pathological features during the enhancement process to generate an enhanced rehabilitation texture image;

[0011] The evolution trend analysis module is used to extract the orderliness index of texture evolution based on the enhanced rehabilitation texture image and generate rehabilitation effect monitoring results.

[0012] Preferably, the process of calculating the texture deformation decoupling coefficient by the spatiotemporal feature decoupling module includes: Step 1: Estimating the optical flow field of adjacent frames in the temporal image data to obtain a non-rigid deformation vector field reflecting the stretching state of the skin surface; Step 2: Extracting high-frequency components in the temporal image data using the Laplacian operator to obtain a local texture gradient map characterizing the microstructure of the wound; Step 3: Calculating the cross-correlation between the non-rigid deformation vector field and the local texture gradient map in the temporal domain, and defining the difference between 1 and the absolute value of the cross-correlation as the texture deformation decoupling coefficient.

[0013] Preferably, the adaptive texture enhancement module performs weighted enhancement processing based on the texture deformation decoupling coefficient, including: setting a preset decoupling threshold, comparing the texture deformation decoupling coefficient with the decoupling threshold; if the texture deformation decoupling coefficient is greater than the decoupling threshold, then the physiological texture feature layer is determined to be an independently evolving signal, and a sharpening enhancement operation is performed on the physiological texture feature layer; if the texture deformation decoupling coefficient is less than or equal to the decoupling threshold, then the physiological texture feature layer is determined to have deformation coupling interference, and a non-rigid deformation vector field is used to perform inverse motion compensation on the physiological texture feature layer, and a smoothing and denoising operation is performed after compensation.

[0014] Preferably, the illumination consistency normalization module performs illumination component correction by: converting the temporal image data to a logarithmic chromaticity space and extracting the temporal signal of the luminance channel; calculating the variance of the temporal signal of the luminance channel within a preset sliding window as the temporal illumination consistency variance; setting a preset illumination variance threshold; if the temporal illumination consistency variance is greater than the illumination variance threshold, then using a histogram specification algorithm to map the luminance distribution of the current frame to the luminance distribution of the reference frame to generate corrected image data; if the temporal illumination consistency variance is less than or equal to the illumination variance threshold, then keeping the luminance distribution of the current frame unchanged and directly outputting it as corrected image data.

[0015] Preferably, the adaptive texture enhancement module maintains the structural similarity of pathological features during enhancement by: acquiring the edges of key anatomical structures in the target wound area and constructing a structural tensor field; when enhancing the physiological texture feature layer, calculating the structural similarity between the enhanced image and the original image in the principal direction determined by the structural tensor field, which is defined as the pathological feature structural similarity index; the pathological feature structural similarity index is obtained by constructing a direction-adaptive Gaussian window in the principal direction of the structural tensor field and calculating the mean value of the structural similarity between the enhanced image and the original image across the entire image; constructing a loss function, which includes an enhancement gain term and a structural similarity constraint term, and solving for the optimal enhancement parameters by minimizing the loss function to ensure that the pathological feature structural similarity index is not lower than a preset fidelity threshold.

[0016] Preferably, the process of extracting the orderliness index of texture evolution by the evolution trend analysis module includes: performing gray-level co-occurrence matrix analysis on the enhanced rehabilitation texture image to extract entropy and energy features; constructing a time series model and inputting entropy and energy features at multiple consecutive time points; calculating the rate of decrease of entropy features over time and the rate of increase of energy features over time, and calculating the weighted sum of the rate of decrease and the rate of increase based on preset weight coefficients, which is defined as the orderliness index.

[0017] Preferably, the process of generating rehabilitation effect monitoring results by the evolution trend analysis module includes: setting an evolution rate threshold; if the orderliness index is greater than the evolution rate threshold, generating a monitoring signal representing benign wound healing; if the orderliness index is less than or equal to the evolution rate threshold, generating an early warning signal representing stagnant wound healing or infection risk, and marking the sub-region with the highest texture disorder in the target wound area.

[0018] Preferably, it also includes a feedback adjustment mechanism: the image sequence acquisition module dynamically adjusts the acquisition frequency according to the magnitude of the non-rigid deformation vector field output by the spatiotemporal feature decoupling module; a preset deformation warning line is set, and when the average magnitude of the non-rigid deformation vector field is greater than the preset deformation warning line, the acquisition frequency is increased to increase the computational density of the spatiotemporal feature decoupling module; when the average magnitude of the non-rigid deformation vector field is less than or equal to the preset deformation warning line, the current acquisition frequency is maintained.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. This invention constructs a spatiotemporal feature decoupling and dynamic deformation field model, effectively solving the problem that traditional image enhancement algorithms cannot distinguish between physiological texture changes and mechanical deformation interference. Unlike existing technologies that simply rely on grayscale statistical characteristics, this scheme generates texture deformation decoupling coefficients by calculating the cross-correlation between the non-rigid deformation vector field and the local texture gradient map in the time domain, thereby quantitatively distinguishing geometric displacement noise originating from respiration or body movement from real pathological signals originating from tissue growth. Based on this, the divide-and-conquer enhancement strategy can not only sharpen and enhance independently evolving granulation textures, but also perform inverse motion compensation and smoothing denoising on areas with coupling interference, significantly reducing the false positive rate caused by patient respiratory movements and improving the accuracy of texture feature extraction in non-rigid deformation areas.

[0021] 2. This invention introduces a pathological feature structural similarity constraint mechanism based on structural tensor fields, overcoming the defect of traditional sharpening algorithms that easily destroy key anatomical topologies. By identifying key edges such as surgical sutures to construct structural tensors, and calculating the structural similarity between the enhanced image and the original image in the principal direction during the optimization process, this method establishes a loss function that includes a sharpening gain term and a fidelity constraint term. This mechanism ensures that while significantly enhancing the granulation texture of granulation tissue, the system can strictly maintain the integrity of microstructures such as incision edges, avoiding artifacts caused by over-enhancement and guaranteeing high fidelity of the output image in clinical diagnosis.

[0022] 3. This invention establishes a texture evolution orderliness evaluation system and a visual early warning mechanism, achieving a leap from static image observation to dynamic trend analysis. The system abandons single-moment threshold judgment, instead extracting time series of entropy and energy characteristics through a gray-level co-occurrence matrix, calculating their rate of change to generate an orderliness index, and can keenly capture microscopic signs of stagnant wound healing or early deterioration. Simultaneously, by automatically marking the sub-regions with the highest texture disorder, the system can, like a thermodynamic... Figure 1 This allows for intuitive localization of potential infection foci or necrosis centers, significantly reducing doctors' reading time and improving their ability to warn of early complications.

[0023] 4. This invention designs an environmental perception and dynamic feedback adjustment mechanism, which enhances the robustness of the system in uncontrolled home monitoring environments. By monitoring the variance of time-series illumination consistency, the system achieves on-demand illumination correction, which not only eliminates interference from ambient light fluctuations but also preserves the original color information when the illumination is stable. Combined with the dynamic adjustment function of the acquisition frequency based on the non-rigid deformation vector field modulus, the system can automatically improve the temporal resolution under strenuous movement conditions such as coughing or turning over, exchanging the increase in computational density for the accuracy of spatial decoupling, and ensuring the consistency and reliability of monitoring results in complex dynamic scenarios. Attached Figure Description

[0024] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0025] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0027] Example 1:

[0028] Please see Figure 1 A postoperative rehabilitation effect monitoring system for general surgery based on image enhancement includes a cloud processing center, which is communicatively connected to an image sequence acquisition module, a spatiotemporal feature decoupling module, an illumination consistency normalization module, an adaptive texture enhancement module, and an evolution trend analysis module.

[0029] The image sequence acquisition module is used to acquire time-series image data of the target wound area over a continuous period of time. The time-series image data contains geometric displacement information caused by non-rigid deformation.

[0030] The spatiotemporal feature decoupling module is used to construct a dynamic deformation field model containing a non-rigid deformation vector field. Based on the dynamic deformation field model, the time-series image data is decomposed into a mechanical deformation feature layer and a physiological texture feature layer, and the texture deformation decoupling coefficient, which characterizes the degree of independence between the two, is calculated.

[0031] The illumination consistency normalization module is used to calculate the temporal illumination consistency variance of the temporal image data and perform illumination component correction on the temporal image data based on the temporal illumination consistency variance.

[0032] The adaptive texture enhancement module is used to perform weighted enhancement processing on the physiological texture feature layer according to the texture deformation decoupling coefficient, and maintain the structural similarity of pathological features during the enhancement process to generate an enhanced rehabilitation texture image;

[0033] The evolution trend analysis module is used to extract the orderliness index of texture evolution based on the enhanced rehabilitation texture image and generate rehabilitation effect monitoring results.

[0034] This embodiment details the overall architecture of the system and the collaborative working mechanism of each module; the image sequence acquisition module acquires temporal image data of the wound area using a high frame rate macro industrial camera. The data is in RGB color space format, with pixel values ​​ranging from 0 to 255. It originates from soft tissue areas such as the patient's abdomen and inevitably contains non-rigid deformation geometric displacement information caused by respiration or body movement. The spatiotemporal feature decoupling module constructs a dynamic deformation field model based on the physical assumption that observed image changes are the superposition of mechanical deformation and physiological texture changes. This model employs an additive decomposition strategy. Among them, the mechanical deformation feature layer Defined as the structural components obtained after motion compensation of a reference image using a non-rigid deformation vector field, physiological texture feature layer Defined as the residual component after subtracting the structural component from the original image. ;

[0035] Based on this mathematical definition, the model decomposes the original data into mechanical deformation feature layers that characterize the physical displacement of the skin. and physiological texture features characterizing granulation tissue growth And calculate the texture deformation decoupling coefficient. This coefficient aims to quantify the contribution of physical stretching to the current texture changes; the illumination consistency normalization module calculates the temporal illumination consistency variance to eliminate the interference of ambient light fluctuations on reflectivity analysis; based on this, the adaptive texture enhancement module uses the decoupling coefficient... Physiological texture feature layer Weighted enhancement is performed while constraining the structural similarity of pathological features to prevent the destruction of key topological structures such as surgical sutures, generating an enhanced rehabilitation texture image; the evolutionary trend analysis module extracts the orderliness index of texture evolution based on the enhanced image and outputs the final monitoring results;

[0036] In this embodiment, under the scenario of home monitoring after general surgery, a dynamic deformation field model is constructed and a decoupling coefficient is introduced to effectively solve the interference of mechanical deformation caused by patient respiration on the analysis of microscopic healing texture. The system can intelligently distinguish and suppress noise signals originating from body movement, while accurately enhancing effective signals originating from tissue growth, thereby avoiding false alarms caused by the mis-amplification of skin folds in traditional enhancement algorithms, ensuring accurate capture of wound healing status in dynamic and uncontrolled environments. To verify the above, this embodiment is verified through comparative experiments: 50 sets of postoperative wound time-series images containing respiratory motion interference are selected and processed by this system and the traditional CLAHE algorithm, respectively. The experimental results show that the signal-to-noise ratio (SNR) of the images processed by this system is improved by an average of 12.4 dB, the accuracy of texture feature extraction in non-rigid deformation areas is improved from 76.2% of the traditional algorithm to 94.5%, and the false positive rate caused by respiratory motion is significantly reduced from 18.5% to 2.1%, fully demonstrating the effectiveness of the system in solving mechanical deformation interference.

[0037] Example 2:

[0038] The process of calculating the texture deformation decoupling coefficients by the spatiotemporal feature decoupling module includes:

[0039] Step 1: Perform optical flow field estimation on adjacent frames in the time-series image data to obtain a non-rigid deformation vector field that reflects the stretching state of the skin surface;

[0040] Step 2: Use the Laplacian operator to extract high-frequency components from time-series image data to obtain a local texture gradient map characterizing the microstructure of the wound.

[0041] Step 3: Calculate the cross-correlation between the non-rigid deformation vector field and the local texture gradient map in the time domain, and define the difference between 1 and the absolute value of the cross-correlation as the texture deformation decoupling coefficient.

[0042] This embodiment is a further specification of the step in calculating the texture deformation decoupling coefficient in Embodiment 1; the system processes adjacent frames in the temporal image data. and Perform dense optical flow field estimation to obtain non-rigid deformation vector field Each vector in this vector field represents the displacement direction and magnitude of the corresponding pixel, aiming to directly reflect the physical stretching state of the skin surface; the Laplacian operator is applied to the current frame image to extract high-frequency components to obtain the local texture gradient map. The gradient map characterizes the microstructural edges and granular information of the wound area; the system performs feature extraction steps from multidimensional field data to a one-dimensional scalar sequence, and calculates the non-rigid deformation vector field. With local texture gradient map The Pearson correlation coefficient in the time domain is used, and 1 minus the absolute value of the correlation coefficient is defined as the texture deformation decoupling coefficient. Specifically, the system constructs a feature sequence within a sliding window and defines... For the first in the window The spatial mean of the non-rigid deformation vector field magnitude of a frame image. For the first The average intensity of the local texture gradient map of the frame image. Calculate both over the window length. The Pearson correlation coefficient within the model is used, and 1 minus the absolute value of the correlation coefficient is defined as the texture deformation decoupling coefficient. The calculation formula is as follows:

[0043]

[0044] In this embodiment, and The specific calculation formula is defined as follows:

[0045]

[0046] in, and Sequences and The arithmetic mean within the sliding window; These are the width and height in pixels, respectively. For the first Frame in coordinates The vector magnitude at that location;

[0047] Derived from optical flow calculation The spatial mean of the field magnitude of the non-rigid deformation vector at any given time, which physically represents the instantaneous deformation intensity;

[0048] Derived from Laplace extract Local texture gradient map at time step The average intensity, in physical terms, is the instantaneous texture complexity;

[0049] The statistical calculations originate from within the sliding window; in this embodiment, the length of the sliding window is explicitly set. A frame corresponds to approximately one second of data, and its physical meaning is the arithmetic mean of the corresponding variable within a window period.

[0050] : A preset small constant, physically meaning a numerical stability term to prevent the denominator from being zero, is set in this embodiment. ;

[0051] In processing images of abdominal wounds accompanied by respiratory movements, this embodiment utilizes the high positive correlation between physical deformation and pseudo-texture changes to construct a decoupling index based on correlation complementarity. This index can sensitively identify true tissue growth signals independent of physical stretching; that is, the lower the correlation, the more accurate the decoupling index. The closer the value is to 1, the more likely the current texture change is to originate from pathological evolution rather than mechanical movement, providing a quantitative confidence basis for subsequent differential enhancement.

[0052] Example 3:

[0053] The adaptive texture enhancement module performs weighted enhancement processing based on texture deformation decoupling coefficients, including the following steps:

[0054] A preset decoupling threshold is used to compare the texture deformation decoupling coefficient with the decoupling threshold.

[0055] If the texture deformation decoupling coefficient is greater than the decoupling threshold, the physiological texture feature layer is determined to be an independent evolution signal, and a sharpening enhancement operation is performed on the physiological texture feature layer.

[0056] If the texture deformation decoupling coefficient is less than or equal to the decoupling threshold, it is determined that there is deformation coupling interference in the physiological texture feature layer. The physiological texture feature layer is then compensated by inverse motion using a non-rigid deformation vector field, and a smoothing and denoising operation is performed after compensation.

[0057] This embodiment is a further specification of the adaptive enhancement logic in Embodiment 2; the system sets a decoupling threshold. This threshold is derived from the statistical distribution of deformation correlation caused by normal breathing in historical clinical data, and is specifically set in this embodiment. This value corresponds to the 85th percentile statistical value of the cross-correlation between texture changes and deformation displacement under normal and stable breathing conditions; the system will calculate the texture deformation decoupling coefficient in real time. With threshold Compare;

[0058] In response to Greater than The system determines that the physiological texture feature layer is an independent evolution signal, i.e., a true healing texture. It then applies an unsharpened masking or high-frequency boosting filter to this layer for sharpening enhancement, aiming to significantly enhance the granulation texture of the granulation tissue and the clarity of the epithelial edges; in response to... Less than or equal to The system determines that there is deformation coupling interference in the physiological texture feature layer, mainly due to artifacts caused by skin stretching; at this point, the system utilizes a non-rigid deformation vector field. The current image is subjected to inverse geometric mapping. Specifically, the bilinear interpolation algorithm is used to calculate the floating-point coordinates of the target pixel in the source image to offset the physical displacement. After compensation, Gaussian smoothing or median filtering is performed on the interpolation noise and residual small deformation edges generated during the inverse transformation to eliminate the pseudo-texture interference caused by stretching.

[0059] This embodiment implements a divide-and-conquer enhancement strategy when dealing with complex dynamic wounds. For areas determined to be genuine healing signals, the system performs positive enhancement to highlight granulation features. For areas determined to be deformation artifacts, the system performs inverse compensation to eliminate interference. This mechanism effectively solves the contradiction that traditional global enhancement algorithms amplify noise when the patient is moving and may smooth out subtle healing features when the patient is stationary, ensuring that the enhanced images seen by doctors have extremely high clinical fidelity.

[0060] Example 4:

[0061] The process of performing illumination component correction by the illumination uniformity normalization module includes:

[0062] The temporal image data is converted to the logarithmic color space, and the temporal signal of the luminance channel is extracted.

[0063] The variance of the timing signal of the brightness channel within a preset sliding window is calculated and used as the timing illumination consistency variance.

[0064] Preset illumination variance threshold;

[0065] If the temporal illumination consistency variance is greater than the illumination variance threshold, the histogram specification algorithm is used to map the brightness distribution of the current frame to the brightness distribution of the reference frame to generate corrected image data.

[0066] If the temporal illumination consistency variance is less than or equal to the illumination variance threshold, the brightness distribution of the current frame remains unchanged, and the corrected image data is directly output.

[0067] This embodiment is a further specification of the illumination correction steps in Embodiment 1; the system converts RGB format temporal image data to logarithmic response space, aiming to use logarithmic transformation to convert multiplicative illumination components into additive components, and extract the temporal signal of the luminance channel. The specific conversion and extraction formulas are as follows:

[0068]

[0069] Calculate the variance of the signal within a preset sliding window to obtain the temporal illumination consistency variance. The formula is as follows:

[0070]

[0071] in, The preset total number of frames contained in the sliding window is set in this embodiment. , used for normalization calculations; : Extracted from the luminance channel, its physical meaning is the first The brightness value of the frame;

[0072] : Derived from window statistics, its physical meaning is the average brightness within the window; the system will Compared with the preset illumination variance threshold Comparison; in this embodiment, based on the normalized luminance channel, the value range is 0-255, and the setting is... ; in response Greater than This indicates significant fluctuations in ambient light. The system selects the first frame within the window as the reference frame and uses a histogram specification algorithm to force the current frame's luminance probability density function to be mapped to the distribution of the reference frame; in response to Less than or equal to This indicates that the lighting is stable, and the system directly outputs the current frame;

[0073] This embodiment targets uncontrolled lighting environments such as home wards, and achieves on-demand correction by monitoring temporal variance. The system only intervenes when it detects drastic fluctuations in lighting caused by shadow occlusion or light flickering, which ensures the consistency of color analysis, such as the benchmark for judging redness and swelling, and maximizes the preservation of the authenticity of the original data when the lighting is stable, thereby improving the robustness and computational efficiency of the system in complex lighting environments.

[0074] Example 5:

[0075] The adaptive texture enhancement module maintains the structural similarity of pathological features during enhancement by including:

[0076] Obtain the edges of key anatomical structures in the target wound area and construct a structural tensor field;

[0077] When enhancing the physiological texture feature layer, the structural similarity between the enhanced image and the original image in the principal direction determined by the structural tensor field is calculated and defined as the pathological feature structural similarity index. The pathological feature structural similarity index is obtained by constructing a direction-adaptive Gaussian window in the principal direction of the structural tensor field and calculating the average value of the structural similarity between the enhanced image and the original image across the entire image.

[0078] A loss function is constructed, which includes an enhancement gain term and a structural similarity constraint term. The optimal enhancement parameters are solved by minimizing the loss function to ensure that the structural similarity index of pathological features is not lower than the preset fidelity threshold.

[0079] This embodiment is a further specification of the structure preservation logic in Embodiment 1; the system uses the Canny edge detection operator to automatically identify the edges of key anatomical structures in the target wound area, such as incision sutures or drainage openings, and calculates the structural tensor of the image based on the identification results. To ensure stable extraction of the dominant directions of local texture and prevent singularities in the determinant of the structure tensor, this embodiment explicitly sets the Gaussian differential scale for gradient calculation. and the standard deviation of the Gaussian integral scale used to smooth the gradient product term. The corresponding integration window size The calculation formula is:

[0080]

[0081] in, This represents the convolution operation. This represents the tensor product operation. Indicated in scale The image gradient vector is calculated below; the eigenvalues ​​and eigenvectors of this tensor describe the dominant orientation and coherence of the local structure.

[0082] To construct an optimization model with computable gradients, this specification explicitly defines the enhanced image. With the parameters to be optimized Functional relationship between Specifically, a parametric Laplacian sharpening model is used:

[0083]

[0084] in, The input is a physiological texture feature layer image. For the second-order Laplace differential operator, The physical meaning of sharpening intensity is a scalar parameter, with the symbol... This represents scalar multiplication; before optimization begins, set... The initial value is The range of values ​​is constrained as follows: This formula establishes The linear weighting mechanism for high-frequency components of the image makes the loss function... right partial derivatives An analytical solution exists, thus supporting iterative solutions based on gradient descent; in this embodiment, the learning rate of gradient descent is set. Maximum number of iterations The iteration termination condition is the change in the loss function. ;

[0085] When enhancing the physiological texture feature layer, the system calculates the structural similarity (SSIM) between the enhanced image and the original image along the principal direction determined by the structure tensor, and defines it as the pathological feature structural similarity index. The calculation in the principal direction of the structure tensor field is achieved by constructing a direction-adaptive Gaussian window. The specific algorithm steps are as follows: For each pixel in the image domain... For structure tensor Perform eigenvalue decomposition to extract the principal eigenvectors and their orientation angles corresponding to the largest eigenvalues. Construct an anisotropic Gaussian weighted kernel Clearly define its reference major axis standard deviation. Pixels, minor axis standard deviation The pixels satisfy the setting of a 2:1 ratio of the standard deviations of the major and minor axes, and the specific window size of the convolution kernel is defined as follows. ,Right now To ensure that the Gaussian kernel covers approximately [a certain amount] before truncation. Energy distribution; setting its major axis along Direction, minor axis perpendicular to Direction; utilizing the rotated nucleus As a statistical window, calculate respectively and The mean, variance, and covariance of local brightness;

[0086] Substitute the above statistics into the standard SSIM formula to calculate local similarity; in this formula, a constant term must be defined. and This embodiment sets , , ,Right now To prevent computational crashes caused by a denominator of zero; finally, the average of the entire graph is taken. This algorithm ensures that similarity measurement is performed only along the texture streamline direction, thereby filtering out noise interference perpendicular to the texture direction.

[0087] Constructing the loss function The formula used to solve for the optimal enhancement parameters is as follows:

[0088]

[0089] The first item adopts The aim is to transform the objective of maximizing texture sharpness gain into a unified mathematical form that minimizes the loss function;

[0090] : Preset weighting coefficients, given After normalization, the value of the term is extremely small, on the order of magnitude. The magnitude, in order to balance its relationship with structural similarity constraints, is usually in The gradient contribution is on the order of magnitude. In this embodiment, the weights are compensated for on the order of magnitude. This ensures that the sharpening gain term generates an effective gradient during the optimization process, while also... Constraints prevent over-sharpening;

[0091] : Derived from the enhancement algorithm settings, the physical meaning is the enhancement parameters to be optimized, such as sharpening intensity;

[0092] Source: Function computation, in order to solve the problem of the original gradient accumulation value, order of magnitude With structural similarity index orders of magnitude There is a serious imbalance in magnitude between the two, and this embodiment has performed strict normalization on this item. Firstly, the enhanced image needs to be... Perform numerical truncation to Interval; average gradient energy The calculation formula is revised as follows:

[0093]

[0094] in, This is the normalization constant, and its value is [value missing]. The calculation of this value is based on the following: for an 8-bit image, convolution kernel The maximum unidirectional response value is ; denominator terms in the formula The theoretical upper limit used to cover the sum of squared gradients in the horizontal and vertical directions. Thus forcing Strictly implement The range, making it consistent with the SSIM indicator. They are on the same order of magnitude; among them, and Defined as the pixel width and pixel height of the image, respectively, the summation symbol represents the summation of all pixel coordinates within the image domain. Perform traversal and accumulation; The enhanced images are respectively In coordinates Horizontal and vertical directions Gradient value, specifically defined as the image and Convolution results of operator templates: , ,in The convolution kernel is as follows:

[0095]

[0096] Symbols are used here To distinguish it from the aforementioned structural tensor Eliminate the ambiguity of multiple meanings of a word; the physical meaning is the normalized average enhancement of texture sharpness gain.

[0097] : Derived from SSIM calculations, its physical meaning is the topological similarity of key structures;

[0098] : Preset value, physically meaning a fidelity threshold, is set in this embodiment. By minimizing this loss function, the system improves texture sharpness while forcibly penalizing any factors that cause it to deteriorate. Below The operation.

[0099] Example 6:

[0100] The process of extracting the orderliness index of texture evolution by the evolution trend analysis module includes:

[0101] Gray-level co-occurrence matrix analysis was performed on the enhanced rehabilitation texture image to extract entropy and energy features; a time series model was constructed by inputting entropy and energy features from multiple consecutive time points.

[0102] The rate of decrease of entropy over time and the rate of increase of energy over time are calculated. The weighted sum of the rate of decrease and the rate of increase is calculated based on a preset weighting coefficient and defined as the orderliness index.

[0103] This embodiment further specifies the evolution index extraction steps in Embodiment 1; the system performs gray-level co-occurrence matrix analysis on the enhanced rehabilitation texture image; in this process, to ensure that the calculated entropy and energy values ​​are comparable in the time dimension and to provide a unified benchmark for subsequent rate calculations, this embodiment explicitly defines three core parameters for constructing the gray-level co-occurrence matrix: step size and distance. Set to 1 pixel to precisely capture subtle changes in granulation tissue texture during the early stages of recovery; scan direction Set to omnidirectional mode, i.e., calculate separately. The gray-level co-occurrence matrices in four directions were averaged to eliminate rotational differences caused by the wound imaging angle; the gray-level quantization level was set to 64 levels to suppress the interference of image noise on statistical features while preserving texture details; based on the above parameters, entropy features characterizing the randomness and disorder of the texture were extracted. And energy features characterizing texture uniformity and order. ; Construct a time series model, taking samples from multiple consecutive time points, such as the past 24 hours. and As input, calculate the rate of change of features over time and synthesize an orderliness index. The formula is as follows:

[0104]

[0105] in, Derived from the time series differential, its physical meaning is the slope of the entropy value as a function of time; a benign healing should be a negative value.

[0106] Derived from the time series differential, its physical meaning is the slope of the energy characteristic changing with time; benign healing should be a positive value.

[0107] : Preset weighting coefficient, which physically means an adjustment factor to balance the contributions of the two dimensions. In this embodiment, given that the importance of energy characteristics in the early stage of healing is slightly higher than that of entropy characteristics, it is set as follows: ;

[0108] This embodiment transforms static image features into a dynamic evolution rate index. This index does not focus on the degree of disorder of the wound at a single moment, but rather on the trend of whether the wound is becoming more orderly. This is more in line with the biological healing law than a single threshold judgment, and can keenly capture microscopic signs of healing stagnation or early deterioration before visible morphological changes occur.

[0109] Example 7:

[0110] The process by which the evolutionary trend analysis module generates rehabilitation effect monitoring results includes:

[0111] A preset evolution rate threshold is set; if the orderliness index is greater than the evolution rate threshold, a monitoring signal characterizing benign wound healing is generated.

[0112] If the orderliness index is less than or equal to the evolution rate threshold, an early warning signal is generated that characterizes wound healing stagnation or infection risk, and the sub-region with the highest texture disorder in the target wound area is marked.

[0113] This embodiment is a further specification of the steps for generating monitoring results in Embodiment 6; the system presets an evolution rate threshold. This threshold is set based on the statistical average wound healing rate of the normal population. In this embodiment, it is set as follows: The calculated orderliness index and Comparison; response Greater than This indicates that the rates of entropy decrease and energy increase are as expected, and the system generates monitoring signals characterizing benign wound healing; in response to Less than or equal to This indicates that the texture evolution has stagnated or reversed, such as an increase in entropy value. The system generates an early warning signal that represents the stagnation of wound healing or the risk of infection. At the same time, the system calculates the local entropy value of each sub-region within the target wound area, identifies and marks the sub-region with the highest texture disorder, that is, the largest local entropy value, and highlights it on the visualization interface.

[0114] This embodiment achieves a closed loop from data trend analysis to clinical decision support; in particular, it automatically marks the area with the highest texture disorder when an alert is triggered, similar to thermal mapping. Figure 1 This method directly guides doctors to focus on early infection centers or necrotic foci that are difficult to detect with the naked eye, greatly shortening doctors' reading time and improving diagnostic accuracy. Clinical retrospective validation data shows that in a test on 120 cases of complex wound recovery, this marking function successfully located 95.8% of early infection foci. Compared with the control group that relied solely on manual reading, its warning advance for the risk of healing stagnation reached an average of 28 hours, significantly improving the timeliness of clinical intervention and diagnostic accuracy.

[0115] Example 8:

[0116] This system also includes a feedback adjustment mechanism:

[0117] The image sequence acquisition module dynamically adjusts the acquisition frequency based on the magnitude of the non-rigid deformation vector field output by the spatiotemporal feature decoupling module;

[0118] A preset deformation warning line is set. When the average magnitude of the non-rigid deformation vector field is greater than the preset deformation warning line, the acquisition frequency is increased to increase the computational density of the spatiotemporal feature decoupling module.

[0119] When the average magnitude of the non-rigid deformation vector field is less than or equal to the preset deformation warning line, the current acquisition frequency is maintained.

[0120] This embodiment is a further specification of the system feedback mechanism in Embodiment 1; the image sequence acquisition module reads the magnitude of the non-rigid deformation vector field output by the spatiotemporal feature decoupling module in real time. The system will calculate the average modulus over a period of time, such as 5 seconds. With the preset deformation warning line Comparison; this embodiment sets pixels / frame, this value corresponds to the typical displacement magnitude during a violent cough or turning over; in response to Greater than The system determines that the patient is in an active state, such as turning over or coughing, and automatically increases the acquisition frequency, for example, from 30fps to 120fps. This aims to increase the computational density of the spatiotemporal feature decoupling module and prevent optical flow calculation failure due to excessive inter-frame displacement. Less than or equal to The system determines that the patient is in a resting state and performs a frequency reset operation, that is, restores the acquisition frequency to the reference acquisition frequency, such as 30fps. If the current frequency is already the reference frequency, it will remain unchanged. This logic corrects the high-frequency lock-in problem that may be caused by simply maintaining the current frequency, and ensures that the system can automatically fall back to the low-power monitoring mode after the patient finishes moving.

[0121] This embodiment endows the monitoring system with the ability to sense and adapt dynamically. By actively increasing the temporal resolution when severe non-rigid deformation occurs, the system gains spatial decoupling accuracy by briefly increasing computing power, thereby ensuring that the monitoring results remain highly consistent and reliable in various activity states such as the patient coughing and turning over, and avoiding analysis failure caused by motion fuzziness.

[0122] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system for monitoring postoperative rehabilitation effects in general surgery based on image enhancement, characterized in that, It includes a cloud processing center, which is communicatively connected to an image sequence acquisition module, a spatiotemporal feature decoupling module, an illumination consistency normalization module, an adaptive texture enhancement module, and an evolution trend analysis module; The image sequence acquisition module is used to acquire time-series image data of the target wound area over a continuous time period. The time-series image data includes geometric displacement information caused by non-rigid deformation. The spatiotemporal feature decoupling module is used to construct a dynamic deformation field model containing a non-rigid deformation vector field. Based on the dynamic deformation field model, the time-series image data is decomposed into a mechanical deformation feature layer and a physiological texture feature layer, and a texture deformation decoupling coefficient representing the degree of independence between the two is calculated. The steps include: Step 1: Estimating the optical flow field of adjacent frames in the time-series image data to obtain a non-rigid deformation vector field reflecting the stretching state of the skin surface; Step 2: Extracting high-frequency components in the time-series image data using the Laplacian operator to obtain a local texture gradient map representing the microstructure of the wound; Step 3: Calculating the cross-correlation between the non-rigid deformation vector field and the local texture gradient map in the time domain, and defining the difference between 1 and the absolute value of the cross-correlation as the texture deformation decoupling coefficient. The illumination consistency normalization module is used to calculate the temporal illumination consistency variance of the temporal image data, and to perform illumination component correction on the temporal image data based on the temporal illumination consistency variance. The adaptive texture enhancement module is used to perform weighted enhancement processing on the physiological texture feature layer according to the texture deformation decoupling coefficient, and maintain the structural similarity of pathological features during the enhancement process to generate an enhanced rehabilitation texture image. The process of maintaining the structural similarity of pathological features during the enhancement process includes: acquiring the edges of key anatomical structures in the target wound area and constructing a structural tensor field; when enhancing the physiological texture feature layer, calculating the structural similarity between the enhanced image and the original image in the principal direction determined by the structural tensor field, which is defined as the pathological feature structural similarity index; the pathological feature structural similarity index is obtained by constructing a direction-adaptive Gaussian window in the principal direction of the structural tensor field and calculating the mean value of the structural similarity between the enhanced image and the original image across the entire image; constructing a loss function, which includes an enhancement gain term and a structural similarity constraint term, and solving for the optimal enhancement parameters by minimizing the loss function to ensure that the pathological feature structural similarity index is not lower than a preset fidelity threshold. The evolution trend analysis module is used to extract the orderliness index of texture evolution based on the enhanced rehabilitation texture image and generate rehabilitation effect monitoring results. The process of extracting the orderliness index of texture evolution by the evolution trend analysis module includes: performing gray-level co-occurrence matrix analysis on the enhanced rehabilitation texture image to extract entropy features and energy features; constructing a time series model, inputting the entropy features and energy features at multiple consecutive time points; calculating the decrease rate of the entropy features and the increase rate of the energy features over time, and calculating the weighted sum of the decrease rate and the increase rate based on preset weight coefficients, which is defined as the orderliness index.

2. The image enhancement-based postoperative rehabilitation effect monitoring system for general surgery according to claim 1, characterized in that, The process by which the adaptive texture enhancement module performs weighted enhancement processing based on the texture deformation decoupling coefficients includes: A preset decoupling threshold is used to compare the texture deformation decoupling coefficient with the decoupling threshold. If the texture deformation decoupling coefficient is greater than the decoupling threshold, the physiological texture feature layer is determined to be an independent evolution signal, and a sharpening enhancement operation is performed on the physiological texture feature layer. If the texture deformation decoupling coefficient is less than or equal to the decoupling threshold, it is determined that there is deformation coupling interference in the physiological texture feature layer. The non-rigid deformation vector field is used to perform inverse motion compensation on the physiological texture feature layer, and a smoothing and denoising operation is performed after compensation.

3. The image enhancement-based postoperative rehabilitation effect monitoring system for general surgery according to claim 1, characterized in that, The process of performing illumination component correction by the illumination uniformity normalization module includes: The time-series image data is converted to a logarithmic color space, and the time-series signal of the luminance channel is extracted. Calculate the variance of the timing signal of the brightness channel within a preset sliding window, and use it as the timing illumination consistency variance; A preset illumination variance threshold is set. If the temporal illumination consistency variance is greater than the illumination variance threshold, the brightness distribution of the current frame is mapped to the brightness distribution of the reference frame using a histogram specification algorithm to generate corrected image data. If the temporal illumination consistency variance is less than or equal to the illumination variance threshold, the brightness distribution of the current frame is kept unchanged, and the corrected image data is directly output.

4. The image enhancement-based postoperative rehabilitation effect monitoring system for general surgery according to claim 1, characterized in that, The process by which the evolution trend analysis module generates rehabilitation effect monitoring results includes: Preset evolution rate threshold; If the degree of order is greater than the evolution rate threshold, a monitoring signal characterizing benign wound healing is generated. If the degree of order is less than or equal to the evolution rate threshold, an early warning signal is generated to characterize the stagnation of wound healing or the risk of infection, and the sub-region with the highest texture disorder in the target wound area is marked.

5. The image enhancement-based postoperative rehabilitation effect monitoring system for general surgery according to claim 1, characterized in that, It also includes a feedback adjustment mechanism: The image sequence acquisition module dynamically adjusts the acquisition frequency based on the magnitude of the non-rigid deformation vector field output by the spatiotemporal feature decoupling module. A preset deformation warning line is set. When the average magnitude of the non-rigid deformation vector field is greater than the preset deformation warning line, the acquisition frequency is increased to increase the computational density of the spatiotemporal feature decoupling module. When the average magnitude of the non-rigid deformation vector field is less than or equal to the preset deformation warning line, the current acquisition frequency is maintained.

Images