A method for analyzing fatty liver lesion images based on color texture features

Abstract

The present application relates to the technical field of image data processing, and more particularly, to a fat liver lesion image analysis method based on color texture features, comprising: extracting a rat liver region through image segmentation, converting to a CIELAB color space to separate luminance, red-green chroma and yellow-blue chroma components; calculating local texture difference features based on a pixel space neighborhood, coupling luminance values to construct pixel-level effectiveness weights to suppress specular reflection; using pixel-level effectiveness weights to update K-Means clustering centers, distinguishing between steatosis regions and non-lesion regions, and counting pixel points to realize analysis of fat liver lesion images. The present application accurately suppresses optical noise interference, improves clustering accuracy, and improves fat liver lesion analysis accuracy, providing reliable technical support for drug efficacy evaluation and pathological mechanism research.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology. More specifically, this invention relates to a method for analyzing fatty liver lesion images based on color and texture features. Background Technology

[0002] In drug development and pathological mechanism research, rat models are commonly used model organisms for studying fatty liver disease. The gross morphology of rat liver, especially its color and texture characteristics, is an important and intuitive basis for assessing the severity of fatty liver disease. Normal rat livers are mostly uniformly reddish-brown with a smooth and delicate surface. As the amount of fat accumulation in hepatocytes increases, the color of the liver surface gradually tends to be pale, accompanied by characteristic texture changes, i.e., fatty degeneration occurs.

[0003] Currently, using machine vision technology to analyze gross liver images has become a trend. Among them, the K-Means clustering algorithm is often used for color segmentation of liver images due to its unsupervised learning characteristics. The traditional K-Means algorithm mainly classifies isolated pixels based on the color value of individual pixels, dividing pixels into different clusters to achieve a preliminary distinction between fatty degeneration areas and non-lesion areas.

[0004] However, in practice, the surface of the excised liver is usually quite moist and has a certain curvature. Under standard laboratory light sources, the liver surface is prone to high-brightness specular reflection. These reflective areas appear as bright white in the color space, and their numerical characteristics are similar to the whitening characteristics of severe fatty liver. Because the traditional K-Means clustering algorithm ignores the spatial neighborhood information between pixels and lacks judgment on local texture features, it is easy to misclassify reflective points that belong to optical noise as fatty degeneration areas. This causes the calculation results of the fatty liver degeneration area to deviate from the true value, thereby reducing the accuracy of fatty liver lesion image analysis and ultimately affecting the accurate assessment of drug efficacy or pathological mechanisms. Summary of the Invention

[0005] To address the problem that traditional K-Means clustering algorithms misidentify fatty liver lesion areas due to specular reflection on the liver surface, thereby reducing the accuracy of fatty liver lesion image analysis and ultimately affecting the accurate assessment of drug efficacy or pathological mechanisms, this invention proposes a fatty liver lesion image analysis method based on color and texture features. This method includes the following steps:

[0006] Gross images of rat livers were acquired, and liver regions were extracted using image segmentation algorithms. The liver region images were then converted from the RGB color space to the CIELAB color space to separate color components, including luminance, red-green hue, and yellow-blue hue. Local texture difference features were calculated based on the spatial neighborhood information of pixels. Luminance values ​​were coupled with these local texture difference features to construct pixel-level effectiveness weights for suppressing specular reflection interference. Based on the color components, the iteration process of the K-Means clustering algorithm was constrained using these pixel-level effectiveness weights. Pixels within the liver region images were clustered by weighted updating of cluster centers to distinguish between fatty liver regions and non-lesion regions. The number of pixels in fatty liver regions was counted based on the clustering results, enabling the analysis of fatty liver lesion images.

[0007] This invention achieves effective separation of color information by converting images to the CIELAB color space, ensuring independent processing of luminance and chrominance values ​​and providing a reliable color space foundation for subsequent color analysis. By calculating local texture difference features and coupling them with luminance values, a pixel-level effectiveness weight is constructed, accurately distinguishing between real color changes and specular reflection, and effectively suppressing the interference of optical noise on clustering analysis. By updating cluster centers with weights, an improvement over the traditional K-Means algorithm is achieved, ensuring that the clustering process considers spatial neighborhood information and texture features, thus improving the accuracy of fatty degeneration region identification. Based on the statistical analysis of fatty degeneration regions using the improved clustering results, the accuracy of fatty liver lesion image analysis is effectively improved, providing reliable technical support for drug efficacy evaluation and pathological mechanism research.

[0008] Furthermore, the pixel-level validity weights satisfy:

[0009] In the formula, For pixels Pixel-level validity weights, For brightness constraint terms, For texture constraints, For pixels Brightness values ​​in the CIELAB color space This represents the average brightness value of all pixels within the liver region image in the CIELAB color space. For pixels Local texture difference features, This represents the standard deviation of the brightness values ​​of all pixels within the liver region image in the CIELAB color space. The preset brightness sensitivity coefficient, The preset texture normalization coefficients, It is a natural exponential function.

[0010] This invention achieves a scientific evaluation of pixel-level effectiveness weights by constructing a product model that includes brightness constraints and texture constraints. The brightness constraint term is implemented through... The function accurately distinguishes the bright reflection areas, and the texture constraint term reflects the influence of local texture differences through an exponential function, thereby effectively suppressing specular reflection interference and ensuring that only pixels that conform to the true color and texture characteristics participate in cluster analysis.

[0011] Furthermore, in the step of clustering pixels within the liver region image using weighted updating of cluster centers, the updating of cluster centers is influenced by pixel-level validity weights, which reduces the contribution of low-weight pixels to the cluster centers. Specifically, the th... The cluster in the th order of ... Cluster centers after the second iteration satisfy:

[0012] In the formula, This represents the total number of pixels within the liver region image. For pixels Color components, For the first The cluster in the th order of ... Cluster centers after the next iteration For pixels Pixel-level validity weights, This is an indicator function.

[0013] This invention achieves adaptive updating of cluster centers by constructing a weighted average model that includes pixel-level validity weights. This ensures that the contribution of low-weight pixels to cluster centers is reduced, while the contribution of high-weight pixels to cluster centers is enhanced. The indicator function accurately identifies the pixels belonging to the current cluster, and the weighting mechanism effectively suppresses the influence of optical noise such as specular reflection, thereby improving the accuracy and robustness of the clustering results.

[0014] Furthermore, the analysis of fatty liver lesion images includes: calculating the ratio of the number of pixels in the fatty degeneration region to the total number of pixels in the liver region image; classifying the degree of fatty liver lesion into corresponding levels among a preset number of levels based on the ratio; and completing the analysis of fatty liver lesion images based on color and texture features.

[0015] This invention achieves a quantitative assessment of the severity of fatty liver disease by calculating the proportion of fatty degeneration areas, providing a reliable assessment indicator for disease grading. The automatic grading based on preset levels improves the objectivity and standardization of fatty liver disease analysis, and provides a scientific evaluation basis for pathological research and drug evaluation.

[0016] Furthermore, before the step of analyzing fatty liver lesion images, the method also includes: setting an effectiveness threshold, identifying pixels with a pixel-level effectiveness weight lower than the effectiveness threshold as specular noise and removing them, and not including the removed pixels in the statistics of fatty degeneration areas.

[0017] This invention achieves effective filtering of specular reflection noise by setting an effectiveness threshold and removing low-weight pixels, ensuring that subsequent statistical analysis is based only on high-confidence pixels, improving the accuracy of fatty degeneration region identification, and further enhancing the reliability of fatty liver lesion image analysis.

[0018] Furthermore, in the step of extracting the liver region image using the image segmentation algorithm, the Otsu's method is used to binarize the gross image to generate a mask, thereby removing the background region and obtaining the liver region image.

[0019] Furthermore, the method for obtaining the local texture difference features is as follows: for each pixel in the liver region image, a [feature name] is defined. A neighborhood window is defined; the standard deviation of the brightness values ​​of all pixels within the neighborhood window in the CIELAB color space is calculated to obtain the local texture difference features of the corresponding pixels.

[0020] Furthermore, the brightness sensitivity coefficient is used to control the slope of the exponential function in the brightness constraint term, so as to adjust the suppression strength of the pixel-level effectiveness weight for pixels with brightness values ​​higher than the mean value.

[0021] Furthermore, the texture normalization coefficient is used to adjust the decay rate of the texture constraint term to achieve a balance between suppressing pixels that generate specular reflection interference and preserving pixels in fatty degeneration regions.

[0022] Furthermore, the indicator function satisfies when belong The value is 1 if it is true, and 0 otherwise.

[0023] The present invention has the following beneficial effects:

[0024] (1) To address the shortcomings of the traditional K-Means clustering algorithm, which clusters based solely on the color values ​​of isolated pixels and ignores spatial neighborhood information, misjudging the bright white areas of specular reflection on the liver surface as fatty degeneration areas, this invention calculates local texture difference features based on pixel spatial neighborhood information. The reflective points of specular reflection are optical noise, and their local textures exhibit isolated and abrupt characteristics, while the white areas of fatty liver are true pathological features with continuous and uniform texture distribution. The texture difference features of the two are clearly distinguishable. By coupling brightness values ​​with texture features to construct pixel-level effectiveness weights, low weights are assigned to reflective points to suppress their influence in clustering, and high weights are assigned to true pathological areas to preserve their features. The optical noise interference of specular reflection is removed, avoiding misjudging non-pathological reflective areas as fatty degeneration areas, thus achieving accurate differentiation between pathological features and optical noise.

[0025] (2) Breaking through the limitations of the traditional K-Means clustering algorithm, which uses equal weight iteration and cluster center updates that are detached from actual image features, this algorithm uses pixel-level validity weights to constrain the iteration process of the K-Means clustering algorithm and adopts a weighted update method to complete pixel clustering. In the algorithm iteration, high-weight pixels in the effective pathological area contribute more to the cluster center, while the contribution of low-weight pixels with reflective interference is weakened. This ensures that the update of the cluster center always fits the true color and texture features of the liver, enabling the clustering results to accurately match the pathological morphological features of fatty degeneration areas and non-lesion areas, improving the clustering discrimination and effectively solving the problem of deviation between the clustering results of the traditional algorithm and the actual pathological area.

[0026] (3) The number of pixels in the fatty degeneration region is counted based on the accurate clustering results of the weighted K-Means clustering algorithm, which avoids the overestimation of fatty degeneration area caused by the misjudgment of reflective points in the traditional algorithm. The statistically obtained indicators such as the area and proportion of the lesion region can truly reflect the actual degree of fatty degeneration in rat liver. Compared with the analysis results of the traditional K-Means clustering algorithm that deviate from the true value, the indicators output by this invention are more reliable, providing accurate morphological data for the subsequent grading of the severity of fatty liver disease and the evaluation of drug efficacy, making the experimental analysis results more consistent with the actual pathological state. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the steps of an image analysis method for fatty liver lesions based on color and texture features according to an embodiment of the present invention.

[0028] Figure 2 This is a pixel effectiveness weight distribution histogram of a fatty liver lesion image analysis method based on color texture features according to an embodiment of the present invention.

[0029] Figure 3This is a schematic diagram of the feature spatial distribution of brightness and texture in a fatty liver lesion image analysis method based on color texture features according to an embodiment of the present invention.

[0030] Figure 4 This is a schematic diagram comparing the brightness distribution before and after weighting in an image analysis method for fatty liver lesions based on color texture features according to an embodiment of the present invention.

[0031] Figure 5 This is a schematic diagram comparing the cluster centers of traditional and improved K-Means in an image analysis method for fatty liver lesions based on color texture features according to an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below. The described embodiments are only a part of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0034] Please see Figure 1 The diagram illustrates a flowchart of an image analysis method for fatty liver lesions based on color texture features, according to an embodiment of the present invention. The method includes the following steps:

[0035] S001: Obtain a gross image of the rat liver, extract the liver region image using an image segmentation algorithm, and convert the liver region image from the RGB color space to the CIELAB color space to separate the color components, which include brightness value, red-green value, and yellow-blue value.

[0036] It should be noted that the acquired gross images of rat livers often contain background areas, and the luminance and color information in the RGB color space are highly coupled, making independent analysis of the two types of information impossible. Furthermore, uneven brightness can easily occur during the acquisition of ex vivo liver images, interfering with the subsequent extraction of lesion-related color features. Therefore, this step requires first separating the liver region from the background, and then decoupling luminance and chromaticity through color space conversion to ensure accurate analysis of liver color features in subsequent steps.

[0037] Specifically, gross images of rat livers are acquired. Due to the experimental environment and the moist nature of the liver surface, the acquired images typically contain background, liver tissue, and unavoidable specular reflection areas. To remove background interference, this embodiment uses an image segmentation algorithm, such as the maximum inter-class variance (MOV) method, to binarize the gross images. The MOV method automatically determines the optimal threshold by maximizing the inter-class variance, generating a binary mask. The binary mask is then used to remove the background area, retaining only the liver parenchyma region, thus obtaining the liver region image. After acquiring the liver region image in RGB format, considering the high coupling between color and brightness information in the RGB color space, which is not conducive to separately analyzing brightness abrupt changes caused by strong light irradiation, i.e., specular reflection, this embodiment utilizes a standard transformation function, such as cv2.cvtCol. Alternatively, the liver region image can be converted from the RGB color space to the CIELAB color space. In the CIELAB color space, three independent color components are separated: luminance value, typically ranging from 0 to 100, representing the brightness of a pixel; in specular reflection areas, the luminance value is usually extremely high, close to 100; in fatty degeneration areas, the luminance value is relatively high, about 70-90; in normal areas, the luminance value is moderate, about 50; red-green luminance value, representing the red-green color dimension, with positive values ​​indicating a red bias and negative values ​​indicating a green bias; normal livers appear reddish-brown due to congestion, resulting in a high red-green luminance value; and yellow-blue luminance value, representing the yellow-blue color dimension, with positive values ​​indicating a yellow bias and negative values ​​indicating a blue bias; fatty degeneration areas appear whitish-yellowish due to lipid deposition, and the yellow-blue luminance value differs from that of normal areas.

[0038] S002: Calculate local texture difference features based on the spatial neighborhood information of pixels, couple the brightness value with the local texture difference features, and construct pixel-level effectiveness weights to suppress specular reflection interference.

[0039] It should be noted that the surface of the rat liver is moist and has a natural curvature after being removed from the body. Under laboratory light, it is prone to specular reflection. The brightness characteristics of the reflected area overlap with those of the fatty degeneration area. Furthermore, relying solely on the brightness or color characteristics of a single pixel cannot distinguish between brightness abnormalities caused by optical interference and brightness changes caused by pathological changes. Additionally, the spatial neighborhood texture information of the pixels has not been mined, lacking a feature index that can quantify the difference. Therefore, this step requires mining the spatial neighborhood texture features of pixels and constructing pixel-level effectiveness weights in conjunction with brightness features to provide an evaluation basis for subsequently distinguishing specular reflection interference from true pathological features.

[0040] Specifically, the method for obtaining the local texture difference features is as follows:

[0041] For each pixel in the liver region image, define a In this embodiment, the neighborhood window, The preferred value is 5. This value is based on the microscopic texture scale of the rat liver surface, with a liver lobule diameter of about 0.3 mm. It can capture texture fluctuations while avoiding excessive smoothing that would cause the lesion edges to be lost.

[0042] The standard deviation of the brightness values ​​of all pixels in the neighborhood window in the CIELAB color space is calculated to obtain the local texture difference features of the corresponding pixels. In the specular reflection area, although the brightness is high, the surface is extremely smooth, and the local texture difference features are close to 0. In the fatty degeneration area, the brightness is high and accompanied by uneven texture fluctuations caused by lipid deposition, and the local texture difference features will be significantly increased.

[0043] Specifically, the pixel-level validity weights satisfy:

[0044] ;

[0045] In the formula, For pixels Pixel-level validity weights, For brightness constraint terms, For texture constraints, For pixels Brightness values ​​in the CIELAB color space The average brightness value of all pixels in the liver region image in the CIELAB color space is used as the brightness reference. For pixels Local texture difference features, This represents the standard deviation of the brightness values ​​of all pixels within the liver region image in the CIELAB color space. The preset brightness sensitivity coefficient is set to 2.5 in this embodiment. Multiple sets, such as 20 sets, of rat liver images containing specular reflection are selected, and iterative tests are performed within the range of 2 to 3, using pathological section results as the standard. The value that maximizes the specular reflection suppression rate and minimizes the false rejection rate of actual lesions is selected as the preset value. The brightness sensitivity coefficient is used to control the slope of the exponential function in the brightness constraint term, thereby adjusting the sensitivity to brightness values ​​higher than... The suppression intensity of the pixels; The preset texture normalization coefficient is set to 0.8 in this embodiment. The texture normalization coefficient determines the algorithm's criterion for judging smooth surfaces; a smaller value indicates a smooth surface. A higher value makes the algorithm more sensitive to subtle texture variations, while a higher value... The value tends to preserve texture features. The texture normalization coefficient is used to adjust the decay rate of the texture constraint term in order to achieve a balance between suppressing pixels that produce specular reflection interference and preserving pixels in fatty degeneration regions. It is a natural exponential function.

[0046] The above relationship constructs weights through the product of brightness constraint terms and texture constraint terms, with the brightness constraint term... Much higher The time quickly approaches 0, achieving initial suppression of highlight areas, and the texture constraint term in When it approaches 0, it approaches 0, while... When it is large, it approaches 1. The combination of these two factors allows for the detection of high-brightness and low-variance specular reflection points during scanning. It decays exponentially to a minimum value, while in areas of high brightness but with texture fluctuations in fatty degeneration, Maintaining a high value directly targets the specular reflection characteristics caused by the moisture on the surface of the rat liver, ensuring that reflections can still be accurately distinguished from real lesions under the influence of light.

[0047] like Figure 2 As shown, the weight values ​​of most pixels are concentrated in the range of 0.6-1.0, accounting for more than 80% of the total number of pixels. These pixels correspond to the real area of ​​liver tissue, and their brightness and texture features are consistent with the inherent properties of pathological tissue. The high weight values ​​ensure that they play a dominant role in clustering. The number of pixels with weight values ​​below 0.2 is extremely small. These pixels are specular reflection noise. Their high brightness and low texture difference characteristics result in extremely small weights. They will be removed by the validity threshold. This verifies that the pixel-level validity weight constructed in this invention can accurately distinguish between valid pixels and noise pixels, providing a reliable basis for clustering constraints.

[0048] like Figure 3 As shown in the figure, two types of regions are clearly distinguished: the suppressed region (reflection) and the effective region. The pixels in the suppressed region are concentrated in the high brightness and low texture difference range, corresponding to the specular reflection noise on the liver surface. Although its brightness is close to that of the severely steatotic region, its texture is extremely smooth, which is consistent with the physical characteristics of optical reflection. The pixels in the effective region are distributed in the medium-high brightness and medium-high texture difference range, corresponding to the normal liver region and the steatotic region. Its texture difference comes from the microstructure of the liver tissue, such as the arrangement of liver lobules and uneven lipid deposition. This proves that by coupling the two-dimensional features of brightness and texture, specular reflection noise and the real pathological region can be separated more accurately.

[0049] S003: Based on color components, the iteration process of the K-Means clustering algorithm is constrained by pixel-level validity weights. The pixels in the liver region image are clustered by weighted updating of cluster centers to distinguish between fatty degeneration areas and non-lesion areas.

[0050] It should be noted that when performing unconstrained K-Means clustering directly on pixels in the liver region, specularly reflected pixels will participate in the cluster center update with their aberrant color components. This can easily cause the cluster centers to deviate from the true color characteristics of steatotic and non-lesion areas of the liver, making it impossible to accurately classify the pixel assignments between the two types of regions, thus affecting the accuracy of lesion area identification. Therefore, this step needs to constrain the K-Means clustering iteration process through pixel-level validity weights, so that the update of the cluster centers is affected by the authenticity of pixel features, achieving more accurate pixel clustering of steatotic and non-lesion areas.

[0051] Specifically, in the step of clustering pixels in the liver region image by weighted updating of cluster centers, the update of cluster centers is affected by pixel-level validity weights, which reduces the contribution of low-weight pixels to the cluster centers. Specifically, the th... The cluster in the th order of ... Cluster centers after the second iteration satisfy:

[0052] ;

[0053] In the formula, This represents the total number of pixels within the liver region image. For pixels Color components, For the first The cluster in the th order of ... Cluster centers after the next iteration For pixels Pixel-level validity weights, For indicator functions, when belong The value is 1 if it is true, and 0 otherwise.

[0054] The above relationship is a vector form of a weighted average, where the numerator represents the first... The sum of the products of the color vectors of the pixels in each cluster and their weights, with the denominator representing the sum of the values ​​of the first cluster. The sum of the weights of pixels in each cluster, when When the size is small, such as in specular reflection regions, the contribution of a pixel to the cluster center is significantly weakened. Even with extreme color values, it will not significantly affect the location of the cluster center. In rat liver images, specular reflection points typically appear as isolated, bright pixels. The size is relatively small, so when calculating the cluster centers of fatty degeneration, the specular reflection points will not have an impact, ensuring that the cluster centers truly reflect the color characteristics of fatty degeneration tissue, rather than optical artifacts.

[0055] like Figure 4As shown, the original brightness distribution curve has a significant peak in the high brightness range, corresponding to the concentrated distribution of specular reflection noise. However, the normalization degree of the weighted brightness distribution curve in the high brightness range is significantly reduced, while the normalization degree in the medium-high brightness range is relatively stable. This verifies that the weighting mechanism of the present invention can effectively suppress the influence of high-brightness specular reflection pixels, reduce their weight ratio in the clustering process, and at the same time retain the medium-high brightness characteristics of the fatty degeneration region, avoiding the problem of misjudging reflection as lesions in traditional algorithms, and providing data support for improving the accuracy of subsequent clustering.

[0056] like Figure 5 As shown in the figure, where L represents brightness, a represents red-green hue, and b represents yellow-blue hue, it can be seen from the figure that: the traditional algorithm's fatty degeneration cluster centers have significantly higher L-channel feature values, and the feature values ​​of channels a and b deviate from the actual pathological range. This is because the abnormal color components of specular reflection pixels pull the cluster centers off-center. In contrast, the fatty degeneration cluster centers of the present invention have significantly lower L-channel feature values, and the feature values ​​of channels a and b are consistent with the color characteristics of fatty degeneration tissue verified by pathological sections. At the same time, the normal liver tissue cluster centers of the present invention are more consistent with the color attributes of reddish-brown normal liver in all three channels. This indicates that by constraining the clustering iteration with pixel-level effectiveness weights, the present invention can generate cluster centers that are more consistent with the actual pathological features, thus improving the accuracy of region division.

[0057] S004: Based on the clustering results, count the number of pixels in the fatty degeneration region to achieve the analysis of fatty liver lesion images.

[0058] It should be noted that a small number of specular noise pixels may still remain in the weighted clustering results. Directly counting all pixels clustered as fatty degeneration regions would lead to a discrepancy between the statistical results and the actual situation, failing to accurately reflect the actual degree of fatty degeneration in rat livers and making it difficult to accurately quantify and grade the lesion based on the statistical results. Therefore, this step requires first removing noisy pixels from the clustering results, and then performing statistical analysis based on high-confidence pixels to provide accurate and reliable analytical data for assessing the degree of fatty liver lesions.

[0059] Specifically, prior to the step of analyzing images of fatty liver lesions, the following steps are also included:

[0060] An effectiveness threshold is set, and pixels with a pixel-level effectiveness weight lower than the threshold are judged as specular reflection noise and removed. Removed pixels are not included in the statistics of fatty degeneration areas. In this embodiment, the effectiveness threshold is set to 0.05, which is determined based on pathological experience and statistical experiments. By selecting multiple groups, such as 20 groups of rat liver images and their corresponding pathological slide standards for comparative analysis, 0.05 is selected as the threshold. This ensures that most reflective noise points are removed while ensuring that real lesion tissue is not mistakenly removed, thereby achieving the best signal-to-noise ratio.

[0061] Specifically, the analysis of fatty liver lesion images includes:

[0062] The ratio of the number of pixels in the fatty degeneration region to the total number of pixels in the liver region image is calculated. Based on this ratio, the severity of fatty liver lesions is classified into corresponding levels among several preset grades. This completes the analysis of fatty liver lesion images based on color and texture features. The corresponding grades are based on relevant standards for rat fatty liver pathology, such as the NAS scoring standard: Normal: no obvious lipid deposition in liver tissue, ratio <5%; Mild fatty degeneration: ratio between 5% and 33%, consistent with the characteristics of mild lesions in a rat high-fat model; Moderate fatty degeneration: ratio between 34% and 66%, with mild disorder of liver lobule structure; Severe fatty degeneration: ratio >66%, with obvious compression of hepatic sinusoids, consistent with the pathological manifestations of severe fatty liver in rats.

[0063] 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 principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing fatty liver lesion images based on color and texture features, characterized in that, include: A gross image of a rat liver was obtained, and the liver region image was extracted using an image segmentation algorithm. The liver region image was then converted from the RGB color space to the CIELAB color space to separate the color components, which include brightness value, red-green value, and yellow-blue value. Local texture difference features are calculated based on the spatial neighborhood information of pixels. Brightness values ​​are coupled with these local texture difference features to construct pixel-level effectiveness weights for suppressing specular reflection interference, satisfying the following: , For pixels Pixel-level validity weights, For brightness constraint terms, For texture constraints, For pixels Brightness values ​​in the CIELAB color space This represents the average brightness value of all pixels within the liver region image in the CIELAB color space. For pixels Local texture difference features, This represents the standard deviation of the brightness values ​​of all pixels within the liver region image in the CIELAB color space. The preset brightness sensitivity coefficient, The preset texture normalization coefficients, It is a natural exponential function; Based on color components, pixel-level validity weights are used to constrain the iterative process of the K-Means clustering algorithm. The pixels in the liver region image are clustered by weighted updating of cluster centers to distinguish between fatty degeneration areas and non-lesion areas. The number of pixels in fatty degeneration regions is counted based on clustering results to achieve the analysis of fatty liver lesion images.

2. The method for analyzing fatty liver lesion images based on color and texture features according to claim 1, characterized in that, In the step of clustering pixels in the liver region image by weighted updating of cluster centers, the update of cluster centers is affected by pixel-level effectiveness weights, which reduces the contribution of low-weight pixels to the cluster centers. Specifically, the... The cluster in the th order of ... Cluster centers after the second iteration satisfy: ； In the formula, This represents the total number of pixels within the liver region image. For pixels Color components, For the first The cluster in the th order of ... Cluster centers after the next iteration For pixels Pixel-level validity weights, This is an indicator function.

3. The method for analyzing fatty liver lesion images based on color and texture features according to claim 1, characterized in that, The analysis of fatty liver lesion images includes: The ratio of the number of pixels in the fatty degeneration region to the total number of pixels in the liver region image is calculated. Based on the ratio, the degree of fatty liver lesions is classified into corresponding levels among multiple preset levels, thus completing the analysis of fatty liver lesion images based on color and texture features.

4. The method for analyzing fatty liver lesion images based on color and texture features according to claim 1, characterized in that, Before the step of analyzing images of fatty liver lesions, the following steps are also included: A validity threshold is set, and pixels with a pixel-level validity weight lower than the validity threshold are judged as specular reflection noise and removed. Pixels that are removed are not included in the statistics of fatty degeneration regions.

5. The method for analyzing fatty liver lesion images based on color and texture features according to claim 1, characterized in that, In the step of extracting liver region images using image segmentation algorithms, the Otsu's method is used to binarize the gross image to generate a mask, thereby removing background regions and obtaining liver region images.

6. The method for analyzing fatty liver lesion images based on color and texture features according to claim 1, characterized in that, The method for obtaining the local texture difference features is as follows: For each pixel in the liver region image, define a The neighborhood window; Calculate the standard deviation of the brightness values ​​of all pixels within the neighborhood window in the CIELAB color space to obtain the local texture difference features of the corresponding pixels.

7. The method for analyzing fatty liver lesion images based on color and texture features according to claim 1, characterized in that, The brightness sensitivity coefficient is used to control the slope of the exponential function in the brightness constraint term, so as to adjust the suppression strength of the pixel-level effectiveness weight for pixels with brightness values ​​higher than the mean value.

8. The method for analyzing fatty liver lesion images based on color and texture features according to claim 1, characterized in that, The texture normalization coefficient is used to adjust the decay rate of the texture constraint term to achieve a balance between suppressing pixels that generate specular reflection interference and preserving pixels in fatty degeneration regions.

9. The method for analyzing fatty liver lesion images based on color and texture features according to claim 2, characterized in that, The indicator function satisfies when belong The value is 1 if it is true, and 0 otherwise.

Images