Method for evaluating the degree of jaundice of a patient with liver disease based on image analysis and system therefor

Abstract

The application discloses a liver disease patient skin jaundice degree evaluation method and system based on image analysis, relates to the technical field of image data processing, and comprises the following steps: collecting a part image group containing a high-sensitivity part image and a low-sensitivity part image, and the two images are in the same frame, under the same illumination and the same camera response; performing brightness normalization on the two part images to obtain a high-sensitivity part chroma component and a low-sensitivity part chroma component; determining a cross-part consistent color cast as a common mode color cast; taking the low-sensitivity part chroma component as an individual self-referencing common mode reference; canceling the common mode color cast from the high-sensitivity part chroma component to obtain a difference mode jaundice component; and calculating a jaundice degree quantitative value according to the intensity of the difference mode jaundice component on a yellow-blue chroma axis. The application can separate the jaundice signal contaminated by the illumination and white balance common mode color cast without external standard color cards and invasive blood sampling baselines, and realizes objective and time-comparable jaundice degree quantification.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, specifically to a method and system for assessing the degree of jaundice in liver disease patients based on image analysis. Background Technology

[0002] Jaundice, the yellowing of the skin and sclera in patients with liver disease due to bilirubin deposition, is a crucial aspect of continuous observation in liver disease care and monitoring. In clinical practice, the degree of jaundice has long relied on visual assessment by nurses under natural light. However, this visual assessment is influenced by ambient lighting, differences in imaging equipment, and the individual patient's skin color, making the judgment subjective and difficult to formulate as an objective, continuously recordable quantitative indicator, as well as for comparable observations across different time points. To transform jaundice assessment from subjective visual observation to objective quantification, this field has proposed several schemes for quantifying the degree of jaundice using image acquisition and color analysis.

[0003] International patent application WO2018223069A1 discloses a method for estimating bilirubin using scleral color and its accessories. This method uses a smart device's camera to capture images of the subject's eyes, segments the scleral region from the images, extracts scleral color features, and then outputs an estimated bilirubin value using a regression model. It also uses a head-mounted hood or a calibration frame printed with known color blocks to calibrate the acquired colors, reducing the influence of external lighting on color measurement. However, this method requires the introduction of an external calibration frame or hood with known colors as an absolute reference for color calibration in each acquisition. Once this external reference is removed, it becomes impossible to separate the yellowing signal mixed with the white balance color cast under natural lighting in a nursing setting.

[0004] US Patent No. 10799150B2 discloses an image-based method for determining bilirubin. This method uses a camera to capture skin images, processes the captured colors using a skin optical diffusion model, and places a standard color chart with multiple color and grayscale blocks printed on it in the captured image. The standard color chart is used to detect changes in illumination and map the captured colors to a device-independent color space, thereby determining bilirubin levels for different skin tones. However, this approach also treats color calibration and yellowing quantization as two independent steps. The color calibration step heavily relies on the physical standard color chart in each frame, and when the standard color chart is missing, it cannot eliminate the interference of yellowing caused by light source color temperature drift on yellowing quantization.

[0005] Chinese patent application CN111434309A discloses a sensor and system for monitoring and managing neonatal jaundice. This system employs a contact optical sensor, which measures bilirubin levels in the blood after removing bilirubin deposits in tissues by decomposing a light source. An individual baseline is established upon initial contact with the sensor, combined with invasive blood sampling. Subsequent relative changes are used to calculate absolute levels and monitored over time. However, this approach relies on contact optical sensing and light decomposition, requiring specialized sensors and invasive blood sampling to establish an individual baseline. It does not employ non-contact image color quantization and cannot correct for individual differences without invasive procedures.

[0006] In summary, all the above-mentioned solutions, regardless of whether they employ external standard color charts, head-mounted hoods, or invasive blood sampling baselines, treat color calibration and yellowing quantification as two independent processes. They require the introduction of external references or invasive blood sampling in each frame to anchor the collected colors to a device-independent absolute color space. The fundamental bottleneck lies in the fact that under natural lighting in nursing settings, the yellowing caused by light source color temperature drift and the yellowing caused by bilirubin deposition fall in the same yellow-blue direction and become confused. Once removed from external references or invasive baselines, existing technologies cannot separate the common-mode color cast of illumination and white balance that contaminates the yellowing signal. This results in yellowing quantification being affected by both lighting conditions and individual skin color, and the quantification results collected at different times are not comparable. How to separate the yellowing signal contaminated by common-mode color cast of illumination and white balance from natural lighting images in nursing settings without relying on physical standard color charts and invasive blood sampling baselines for each frame, thereby obtaining objective, time-comparable quantitative values ​​of skin yellowing, is a technical problem urgently needing to be solved in this field. Summary of the Invention

[0007] Addressing the core bottleneck of existing technologies for assessing the degree of jaundice in liver disease care, which relies on physical standard color charts or invasive blood sampling baselines for each frame and cannot separate the common-mode color cast of illumination and white balance mixed with jaundice signals under natural lighting, this invention provides a method and system for assessing the degree of jaundice in liver disease patients based on image analysis. By using low-sensitivity jaundice areas as individual self-reference common-mode benchmarks within the same frame image and canceling the common-mode color cast of illumination and white balance in the colorimetric domain to obtain the differential-mode jaundice component, this invention achieves objective and time-comparable quantification of the degree of jaundice without relying on external standard color charts or invasive blood sampling baselines, based on the colorimetric decomposition principle of illumination common mode and jaundice differential mode.

[0008] The technical solution of this invention is: a method for assessing the degree of jaundice in liver disease patients based on image analysis, comprising the following steps: acquiring a group of images of the patient's body parts through an image acquisition unit, wherein the group of images includes images of highly sensitive jaundice areas and images of low-sensitive jaundice areas, wherein the highly sensitive jaundice areas are body parts with high jaundice load, and the low-sensitive jaundice areas are body parts with low jaundice load, wherein the images of the highly sensitive jaundice areas and the low-sensitive jaundice areas are images of the highly sensitive jaundice areas and the low-sensitive jaundice areas within the same frame, under the same lighting conditions and the same camera response; and processing the highly sensitive jaundice areas... The image and the image of the low-sensitivity yellow-stained region are respectively normalized to obtain the chromaticity components of the high-sensitivity region and the low-sensitivity region. The color cast consistent across the images of the high-sensitivity yellow-stained region and the low-sensitivity yellow-stained region is determined as the common-mode color cast. Using the chromaticity component of the low-sensitivity region as the individual self-reference common-mode reference, the common-mode color cast is canceled from the chromaticity component of the high-sensitivity region to obtain the differential-mode yellow-stained component, which is the difference in yellow-stained load that varies with region. Based on the intensity of the differential-mode yellow-stained component on the yellow-blue hue axis, the quantized value of the yellow-stained degree of the high-sensitivity yellow-stained region image is calculated and output.

[0009] Furthermore, the low-sensitivity yellowing areas are selected from at least one of the palm skin and the sole skin, and the high-sensitivity yellowing areas are selected from at least one of the sclera and facial skin. The light color cast of the low-sensitivity yellowing areas is used as the individual self-reference common-mode reference to replace the external standard color chart. During acquisition, the low-sensitivity yellowing areas are guided to be placed in the same illumination area adjacent to the high-sensitivity yellowing areas for simultaneous imaging. A brightness gradient map is calculated for the high-sensitivity yellowing area image, and grazing pixels are excluded from the calculation. When the image group of the area is illuminated by multiple light sources, multiple low-sensitivity yellowing areas are extracted from the low-sensitivity yellowing area image. The spatial illumination chromaticity field is fitted to sensitive anchor points and extrapolated to the image of the high-sensitivity yellow staining region; the pixel-by-pixel chromaticity residuals are used to construct a spatial yellow staining distribution map, and the yellow staining progression grade is determined based on the expansion range of the connected regions exceeding the distribution threshold along the head-to-tail progression direction and incorporated into the quantification value of the yellow staining degree; the differential mode yellow staining component at the time of patient admission is used as the individual yellow staining baseline to obtain the yellow staining degree trend, and the spatial monotonicity of yellow staining in the head-to-tail progression direction is used as a constraint to fit the yellow staining degree trend, and when the head-to-tail progression front crosses a certain low-sensitivity yellow staining anchor point, the low-sensitivity yellow staining anchor point is removed from the fitting constraint or its weight is reduced.

[0010] This invention also provides an image analysis-based system for assessing the degree of jaundice in liver disease patients, comprising an image acquisition unit, a common-mode correction unit, and a jaundice quantification unit. The image acquisition unit includes a lens and an image sensor, used to acquire and output a group of images of the patient's affected areas. The common-mode correction unit performs luminance normalization on the images of the highly sensitive and low-sensitive jaundice areas, obtaining chromaticity components for the highly sensitive and low-sensitive areas. Consistent color casts across different areas are identified as common-mode color casts. The chromaticity component of the low-sensitive area is used as an individual self-reference common-mode baseline to cancel the common-mode color cast, resulting in a differential-mode jaundice component. The jaundice quantification unit calculates and outputs a quantified value of the jaundice degree in the highly sensitive jaundice area images based on the intensity of the differential-mode jaundice component on the yellow-blue hue axis. The group of images of the affected areas is output from the image acquisition unit to the common-mode correction unit, and the differential-mode jaundice component is output from the common-mode correction unit to the jaundice quantification unit.

[0011] The beneficial effects of this invention are as follows:

[0012] First, this invention uses low-sensitivity yellow areas as individual self-reference common-mode benchmarks within the same image frame. In the chromaticity domain, it cancels out common-mode color casts from the chromaticity components of high-sensitivity yellow areas to obtain differential-mode yellow components, thus separating the yellow signal without relying on external standard color charts. The mechanism is as follows: Light source color temperature drift, under the same light source and camera response, consistently affects all parts of the entire frame, constituting common-mode color casts. Bilirubin deposition loads differ significantly across different anatomical sites, forming differential-mode components that vary by site. Although these two are confused in the yellow-blue hue direction, they are separable across site dimensions. By taking the difference from low-sensitivity yellow areas, common-mode color casts can be canceled out, preserving the differential yellow component. This invention uses the patient's own body parts instead of external references, fundamentally eliminating dependence on external standard color charts.

[0013] Second, this invention calculates the quantification value of yellow staining degree by measuring the intensity of the differential yellow staining component on the yellow-blue hue axis. The mechanism is that bilirubin's strong absorption of short-wavelength light causes it to primarily enhance the yellow-blue component of the image. Projecting the differential yellow staining component onto the yellow-blue hue axis can stably characterize the degree of yellow staining, transforming the assessment of yellow staining from subjective visual observation to an objective color quantification index.

[0014] Third, this invention, in conjunction with an individual yellowing baseline, achieves cross-time comparability. The mechanism is as follows: common-mode color cast cancellation eliminates the influence of differences in illumination and white balance between successive acquisitions on the quantification results; the individual yellowing baseline further eliminates inherent individual skin color shifts; the superposition of these two ensures that the quantification values ​​of yellowing severity acquired at different times fall on the same comparison benchmark. This invention uses the differential yellowing component at admission as the individual yellowing baseline, achieving non-contact, non-invasive individual correction.

[0015] Fourth, this invention employs a layer-by-layer growth mechanism involving common-mode and differential-mode separation to generate a spatial illumination chromaticity field estimation, a spatial yellowing distribution map classification, and a self-verifying closed loop of anchor points at the beginning and end of the progression, forming a nonlinear synergy. The mechanism lies in the inconsistency of incident light geometry between low-sensitivity and high-sensitivity yellowing regions on the three-dimensional surface. This causes the single common-mode reference to fail under mixed illumination from multiple light sources, thereby forcing the development of a spatial illumination chromaticity field estimation constrained by low-sensitivity yellowing anchor points. The pixel-by-pixel residuals of the spatial illumination chromaticity field further emerge a spatial yellowing distribution map, allowing the degree of yellowing to be expanded from the intensity dimension to the spatial distribution dimension and classified according to the beginning and end expansion range. The spatial progression of yellowing along the beginning and end directions provides a physical monotonicity constraint on the trend of yellowing degree, and when the beginning and end of the progression crosses a low-sensitivity yellowing anchor point, the anchor point is eliminated to prevent it from being contaminated by yellowing and destroying the common-mode reference. The overall effect achieved by the combination of the above layers—no color card required, no blood sampling required, resistance to multiple light sources, classification capability, and trend tracking—exceeds the sum of the individual effects of each layer. Attached Figure Description

[0016] Figure 1 This is a flowchart of the image analysis-based method for assessing the degree of jaundice in liver disease patients according to the present invention.

[0017] Figure 2 This is an architecture diagram of the image analysis-based skin jaundice assessment system for liver disease patients according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are merely illustrative and are not intended to limit the invention.

[0019] The overall concept of this invention is as follows: Within the same image frame, both high-sensitivity yellow-stained regions with high yellow-stained loads and low-sensitivity yellow-stained regions with low yellow-stained loads are simultaneously acquired. Utilizing the fact that both regions are under the same illumination conditions and have the same camera response, the color cast consistent across regions caused by light source color temperature drift and white balance deviation is identified as the common-mode color cast. The color cast varying across regions due to bilirubin deposition is identified as the differential-mode component. Then, using the color cast carried by the low-sensitivity yellow-stained regions as an individual self-reference common-mode benchmark, the common-mode color cast is canceled out from the high-sensitivity yellow-stained regions in the chromaticity domain to obtain the differential-mode yellow-stained component. An objective quantification value of the degree of yellow-stainedness is obtained from the intensity of the differential-mode yellow-stained component on the yellow-blue hue axis. Based on this, spatial illumination chromaticity field estimation is used to address mixed illumination from multiple light sources, a spatial yellow-stained distribution map is used to address the spatial distribution of yellow-stainedness, and individual yellow-stained baselines and head-to-tail progression constraints are used to address trans-temporal trends, forming a coupled closed-loop evaluation process.

[0020] Before describing the specific implementation methods, the terminology used herein will be clarified. The term "highly sensitive yellowing area" refers to a body surface area with a high yellowing load, i.e., areas where bilirubin deposition is significant and yellowing appears early; the term "lowly sensitive yellowing area" refers to a body surface area with a low yellowing load, i.e., areas where bilirubin deposition is not significant and yellowing appears later. The term "area image group" refers to image data acquired and output by the image acquisition unit within the same image frame, simultaneously containing images of both highly sensitive and lowly sensitive yellowing areas. The term "yellow-blue chromaticity axis" refers to the chromaticity component axis characterizing the direction of color change from blue to yellow. The term "yellowing degree quantification value" refers to a quantitative value obtained from image analysis characterizing the degree of skin yellowing; it is a color quantification index, not an absolute measurement of bilirubin concentration in the blood, and is not used for disease diagnosis.

[0021] See Figure 1 The present invention provides a method for assessing the degree of jaundice in the skin of liver disease patients based on image analysis, comprising steps S1 to S5, which are described in detail below.

[0022] Step S1 involves acquiring a group of images of the patient's body parts using an image acquisition unit. This unit includes a lens and an image sensor, and can be implemented using a handheld terminal's camera module, a fixed camera module on a nursing cart, or a dedicated acquisition device. During acquisition, both highly sensitive and low-sensitive yellow areas are captured simultaneously in the same frame, resulting in a group of body part images containing both images of the highly sensitive and low-sensitive yellow areas. Because the two areas are within the same frame, they share the same light source, the same main distribution of illumination geometry, and the same camera's imaging response. This shared relationship forms the physical basis for subsequently dividing color cast into common-mode and differential-mode components.

[0023] In one embodiment, the highly sensitive areas for yellowing are selected from at least one of the sclera and facial skin. The sclera is rich in elastin, which has a high affinity for bilirubin, resulting in early yellowing that is unaffected by melanin. Facial skin has high blood perfusion and a thinner stratum corneum, leading to early yellowing as well. The low-sensitive areas for yellowing are selected from at least one of the palm and sole skin. The palms and soles have thicker stratum corneum and relatively lower blood perfusion, resulting in slower bilirubin deposition. They are usually the last to show obvious yellowing in the initial and final spread pattern of yellowing, and therefore can be approximated as areas that do not bear yellowing at most assessment points, making them suitable as areas bearing the individual self-reference common model benchmark. Selecting the palm or sole skin as low-sensitive areas for yellowing is a counterintuitive approach: they are not selected because they are sensitive to yellowing, but precisely because they are insensitive to yellowing and can purely bear the effects of light and white balance color cast.

[0024] In one embodiment, the standard lighting conditions for image acquisition include an illuminance of 300 lx to 1500 lx and a correlated color temperature of 3000 K to 6500 K. Setting the lower limit of illuminance to 300 lx ensures sufficient signal-to-noise ratio for the image sensor, preventing dark noise from overwhelming the differential mode yellow chromaticity component. Setting the upper limit of illuminance to 1500 lx prevents highlight clipping that would cause yellow-blue hue information to be clipped. An illuminance of 600 lx allows the image sensor to obtain a stable colorimetric response under normal nursing conditions, making it a preferred value. A correlated color temperature range of 3000 K to 6500 K covers the warm to cool white light sources commonly found in wards. A correlated color temperature of 5000 K is close to neutral illumination, minimizing common-mode color cast, making it a preferred value. It should be noted that the standard lighting conditions are only used to ensure image quality; this invention does not rely on precisely known lighting parameters. Residual color casts from lighting and white balance will be canceled as common-mode color casts in subsequent steps.

[0025] When acquiring a set of images of the same area, the patient can be prompted to naturally spread their palm and place it near their face, so that the palm, face, or eyes are in the same frame and in adjacent, illuminated areas, thereby enhancing the consistency of illumination received by the two areas. The acquired set of images is output by the image acquisition unit and used as input for step S2.

[0026] Step S2: Perform brightness normalization on the images of the high-sensitivity yellow staining region and the low-sensitivity yellow staining region respectively to obtain the chromaticity components of the high-sensitivity region and the chromaticity components of the low-sensitivity region. Use the chromaticity components of the low-sensitivity region as the individual self-reference common mode reference to cancel the common mode color cast and obtain the differential mode yellow staining component.

[0027] First, brightness normalization is performed on images of high-sensitivity and low-sensitivity yellow-stained areas. In one embodiment, brightness normalization involves transforming the area image from the acquired color space to the CIELAB color space and normalizing the L* component. The CIELAB color space represents color as a brightness component L* and two chromaticity components a* and b*, where a* represents the chromaticity change from green to red, and b* is the yellow-blue chromaticity axis, representing the chromaticity change from blue to yellow. The transformation from RGB acquired color to CIELAB color space uses a standard colorimetric transformation, first a linear transformation from standard RGB to CIEXYZ, and then a nonlinear transformation from CIEXYZ to CIELAB. This is a well-known transformation in the art and will not be elaborated further. After normalizing the L* component, the color of each pixel is represented by the chromaticity components a* and b*, thereby decoupling the color representation from the influence of light intensity on the pixel, retaining only information related to color direction. After brightness normalization, the high-sensitivity region image of the yellow-stained region is obtained as the chromaticity component of the high-sensitivity region, and the low-sensitivity region image of the yellow-stained region is obtained as the chromaticity component of the low-sensitivity region.

[0028] Next, the color cast consistent across images of high-sensitivity yellow-stained areas and low-sensitivity yellow-stained areas is defined as the common-mode color cast. This common-mode color cast originates from the light source color temperature and camera white balance, having a consistent effect on all areas within the same frame; however, the differential-mode component caused by bilirubin deposition varies depending on the area. To offset the common-mode color cast from the chromaticity components of high-sensitivity areas, this invention uses the chromaticity components of low-sensitivity areas as an individual self-reference common-mode baseline. Since low-sensitivity yellow-stained areas are approximately unaffected by yellow staining at most evaluation points, their chromaticity components are mainly composed of the individual's inherent skin tone and the common-mode color cast, serving as a robust source for estimating the common-mode color cast.

[0029] Considering the possibility of abnormal pixels such as shadows, wrinkles, or highlights in low-sensitivity images, a weighted center approach is used to estimate the common-mode color cast to obtain a robust individual self-reference common-mode baseline. The estimation of the individual self-reference common-mode baseline is shown in Equation (1):

[0030] (1),

[0031] in: The individual self-referenced common-mode reference is a two-dimensional vector, whose two components correspond to the common-mode color cast on the a* and b* axes, respectively. The value range is the CIELAB chromaticity coordinate range, meaning each component takes values ​​from... The units are CIELAB chromaticity units, which are calculated by weighted average of the chromaticity components of the low-sensitivity parts, and represent the common mode color shift of light and white balance carried by the low-sensitivity parts of yellow dyeing. The first of the colorimetric components in the low-sensitivity region The chromaticity vector of each pixel is a two-dimensional vector, with two components being the a* and b* values ​​of that pixel, taking values ​​ranging from 0 to 1. The values ​​are in CIELAB chromaticity units, obtained by normalizing the brightness of the image of the low-sensitivity yellow-stained area, and represent the chromaticity of the pixel. The pixel number of the chromaticity component of the low-sensitivity region is a scalar integer with a value range of [value range missing]. to , dimensionless, determined by pixel traversal, used to index each pixel; The total number of pixels in the chromaticity component of the low-sensitivity region is a scalar integer, which is a positive integer with no dimension. It is determined by the effective pixel count of the yellow-stained low-sensitivity region image and represents the size of pixels participating in the common-mode baseline estimation. For the first The robust weight of each pixel is a scalar with a value range of [value range missing]. , dimensionless, is calculated by the Gaussian kernel of this formula, and its function is to suppress the influence of abnormal pixels that deviate from the median on the common mode reference; Let be the component-wise median chromaticity vector of the chromaticity components of the low-sensitivity region, which is a two-dimensional vector with values ​​ranging from and . The same, the unit is CIELAB colorimetric units, which are obtained by taking the median of the colorimetric values ​​of all low-sensitivity regions, and are used as a robust reference center; It is the set of all pixel chromaticity components of the low-sensitivity region, representing the overall chromaticity components of the low-sensitivity region; Represents the L2 norm of a vector, a dimensionless operator used to measure the magnitude of chromaticity deviation; Here, is the chroma bandwidth parameter, which is a scalar and has a value range of . to The unit is CIELAB chromaticity units, determined by the dispersion of chromaticity in low-sensitivity regions, and its value is [value missing]. For most images in nursing scenarios, it can balance robustness and effective sample size. Its function is to control the suppression strength of abnormal pixels. If the value is too small, the effective sample size will be too small and the estimation will be unstable. If the value is too large, the suppression of abnormal pixels will be weakened. This represents an exponential function with the natural constant as its base.

[0032] After obtaining the individual self-reference common mode baseline, the common mode color cast is canceled pixel by pixel from the chromaticity component of the high-sensitivity part to obtain the differential mode yellow tint component, as shown in Equation (2):

[0033] (2), where: The first of the differential yellow staining components The differential mode chromaticity vector of n pixels is a two-dimensional vector, and its two components are denoted as . and The value range is the range of values ​​for each component. The value is between 1 and 2, and the unit is CIELAB chromaticity units. It is calculated by this formula and represents the difference in yellowing load of the pixel depending on the location after the common-mode color shift is offset. The first of the colorimetric components in the highly sensitive region The chromaticity vector of each pixel is a two-dimensional vector, and its values ​​range from 1 to 2. The values ​​are in CIELAB chromaticity units, obtained by normalizing the brightness of the image of the yellow-stained high-sensitivity area, and represent the chromaticity of the pixel. Here, is the pixel index of the chromaticity component of the highly sensitive region, which is a scalar integer with a value range of . to , dimensionless, determined by pixel traversal, used to index each pixel; The total number of pixels in the chromaticity component of the high-sensitivity region is a scalar integer, a dimensionless integer with a value range greater than zero, and is determined by the effective pixel count of the yellow-stained high-sensitivity region image. The definition is the same as in equation (1).

[0034] At this point, the common-mode color cast carried by the low-sensitivity yellow dyeing region is canceled out from the chromaticity component of the high-sensitivity yellow dyeing region, and the resulting differential-mode yellow dyeing component is the difference in yellow dyeing load that varies from region to region. The differential-mode yellow dyeing component is output by the common-mode correction unit and serves as the input for step S3.

[0035] Step S3: Calculate and output the quantized value of the yellowing degree in the image of the highly sensitive yellowing region based on the intensity of the differential yellowing component along the yellow-blue hue axis. Bilirubin has strong absorption of short-wavelength light, causing yellowing to primarily manifest as an enhancement in the yellow-blue hue direction in the image. Therefore, the differential yellowing component is projected onto the yellow-blue hue axis, and the mean value of its b* component is taken as the yellowing intensity and mapped to a quantized value of the yellowing degree. In one embodiment, the mean value of the b* component of each pixel of the differential yellowing component is taken and linearly mapped to a preset quantization interval. to On the quantization scale, the quantized value of the degree of yellowing is obtained. This mapping is a linear normalized mapping, and its mapping interval is determined by the differential modulus b* component distribution of the calibrated samples; the upper limit of the quantization scale is set to... For ease of nursing record keeping and presentation, the monotonic correspondence between the quantified value of yellowing severity and the differential yellowing component remains unchanged. The quantified value of yellowing severity monotonically increases with the intensity of the differential yellowing component on the yellow-blue hue axis, thus transforming the degree of yellowing severity from subjective visual observation into an objective color quantification index. When the highly sensitive area of ​​yellowing is selected as facial skin, to avoid errors caused by inconsistent incident light geometry on the three-dimensional body surface, during acquisition, the low-sensitivity yellowing area is guided to be placed in the same lighting area adjacent to the high-sensitivity yellowing area for simultaneous imaging within the same frame, ensuring that the illumination received by the two areas is as consistent as possible. Simultaneously, a brightness gradient map is calculated for the image of the highly sensitive yellow-stained area. The brightness gradient map is obtained pixel-by-pixel by calculating the brightness components of the image of the highly sensitive yellow-stained area using a gradient operator, representing the rate of change of brightness along the spatial direction. Pixels with brightness gradients higher than the gradient threshold in the brightness gradient map are marked as grazing pixels. Grazing pixels typically correspond to areas with incident angles close to grazing, such as the edges of facial contours and the sides of the nose. Their chromaticity is greatly affected by local shadows and specular reflections, and they do not satisfy the assumption of sharing the same illumination with the low-sensitivity yellow-stained areas. Grazing pixels are excluded from the calculation of the yellow-stained degree quantification value, and only non-grazing pixels are used in the mean statistics. In one embodiment, the gradient threshold is taken as the 80th percentile of the gradient magnitude of all pixels in the brightness gradient map. If the gradient threshold is too low, too many valid pixels will be excluded; if the gradient threshold is too high, it will not be sufficient to remove grazing pixels. The gradient operator is a first-order difference operator known in the art, which will not be elaborated further.

[0036] Step S4: When the image group of a region is illuminated by a mixture of multiple light sources, multiple first-type low-sensitivity anchor points are extracted from the low-sensitivity yellow-stained region image. Simultaneously, multiple second-type low-sensitivity yellow-stained anchor points are extracted from the low-staining-load auxiliary region within the high-sensitivity yellow-stained region image. Using the positions and chromaticities of the two types of anchor points as common constraints, a spatial illumination chromaticity field covering the high-sensitivity yellow-stained region image is fitted within the image coordinate plane. In a ward environment, multiple light sources, such as natural light from outside the window and indoor fluorescent lights, often coexist, causing common-mode color cast to no longer be consistent throughout the image but to vary with spatial location. In this case, using a single individual self-reference common-mode benchmark to cancel the common-mode color cast will produce residual errors in space. Furthermore, it should be noted that low-sensitivity yellow areas (such as the palm) and high-sensitivity yellow areas (such as the face) correspond to three-dimensional curved surfaces of different depths and orientations of the human body. Under mixed illumination from multiple light sources, the relative contributions of each light source received by the two may be different. If the spatial illumination chromaticity field is fitted to the anchor point inside the low-sensitivity yellow area within its own image sub-region and extrapolated across the curved surface to the image sub-region where the high-sensitivity yellow area is located, then since the physical curved surfaces corresponding to the two image sub-regions are not part of the same continuous planar field, this extrapolation lacks physical rationality and may introduce systematic residual common-mode color shift in the high-sensitivity yellow area.To this end, the present invention employs a spatial illumination chromaticity field estimation strategy combining two types of anchor points: First, multiple first-type low-sensitivity anchor points are extracted from the image of low-sensitivity yellow staining areas. Each first-type low-sensitivity anchor point is a representative pixel of a locally dispersed and chromaticly stable region in the image of low-sensitivity yellow staining areas. The first-type anchor points undertake the macroscopic locking of common-mode color deviation across regions. Second, low-staining-load auxiliary regions are identified within the image of high-staining yellow staining areas, and multiple second-type low-staining-load auxiliary regions are extracted from them. The low-staining-load auxiliary regions are anatomical sub-regions within the image of high-staining yellow staining areas that have relatively low bilirubin deposition load, relatively stable reflectivity, and can be stably located by facial key point detection. These include at least one of the auricle root region, the nasal ala shadow boundary region, and the forehead hairline region. The auricle root region is located in the depression at the junction of the auricle and the temporal skin, where the stratum corneum is relatively thick and the blood perfusion pattern causes bilirubin deposition to be relatively delayed. The nasal ala shadow boundary region utilizes the nasal ala... The dark side boundary formed by its own diffuse reflection has a gentle change in surface normal, less chance of direct exposure to strong light, and relatively stable color. The skin near the hairline is long-term blocked by the hairline and has less chance of being exposed to strong direct light. At most evaluation points, it can be approximated as an auxiliary anchor for low yellowing load. The second type of anchor point undertakes the local constraint of the fine changes in spatial illumination chromaticity within the image of the high-sensitivity yellowing area. Third, with the position and chromaticity of the first and second types of low-sensitivity yellowing anchor points as common constraints, a spatial illumination chromaticity field is fitted in the image coordinate plane where the image of the high-sensitivity yellowing area is located. The fitting and pixel-by-pixel application of the spatial illumination chromaticity field are both limited to the same image coordinate plane where the image of the high-sensitivity yellowing area is located, and no longer extrapolate from the surface where the low-sensitivity yellowing area is located to the surface where the high-sensitivity yellowing area is located. Then, the pixel-by-pixel chromaticity value of the spatial illumination chromaticity field replaces the single individual self-reference common mode reference, realizing physically implementable spatial common mode color cast cancellation without extrapolation across the surface. When it is difficult to stably extract a sufficient number of second-type low-sensitivity anchor points in the image of the high-sensitivity yellow staining region, the system reverts to a single individual self-reference common-mode baseline mode as a downgrade scheme for spatial illumination chromaticity field estimation.

[0037] The spatial illumination chromaticity field is fitted using regularized spatial multinomial regression with pixel position as the independent variable and common mode color bias as the dependent variable. The fitted spatial illumination chromaticity field is shown in equation (3):

[0038] (3),

[0039] in: Let be the coefficient matrix of the spatial illumination chromaticity field, and be... The matrix, whose two columns correspond to the spatial distribution coefficients of the a* axis and b* axis respectively, takes values ​​in the range of real numbers, and is in CIELAB chromaticity units. It is obtained by least squares solution of this formula and characterizes the law of common mode color deviation with spatial position. Let be the spatial basis matrix of the yellow-stained low-sensitivity anchor points, and be... Matrix, its first Behavior No. Spatial basis vectors at the locations of the yellow-stained low-sensitivity anchor points The value range is a real number matrix, dimensionless, and is calculated from the normalized pixel coordinates of each yellow dye low-sensitivity anchor point. It is used to span a quadratic polynomial space. For position The spatial basis vectors at point are six-dimensional vectors, where , These are the normalized x and y coordinates of this position, with values ​​ranging from [value range missing]. , dimensionless, is obtained by dividing pixel coordinates by image width and height, and is used to express the spatial quadratic change of common mode color cast; For the first The position coordinates of each yellow-stained low-sensitivity anchor point are two-dimensional vectors, dimensionless, and determined by the coordinates of the anchor point pixels in the image of the yellow-stained low-sensitivity region. Here is the sequence number of the low-sensitivity anchor point for yellow staining, which is a scalar integer with a value range of [value range missing]. to Dimensionless, used to index each yellow staining low-sensitivity anchor point; The total number of low-sensitivity anchor points for yellow staining is a scalar integer, with a value range of not less than 100%. An integer, dimensionless, determined by the anchor point extraction step, whose value must be no less than the dimension of the spatial basis vectors to ensure solvability; its value is... A balance can be achieved between fitting stability and computational cost; Let be the chromaticity matrix of the yellow-dyed low-sensitivity anchor point. Matrix, its first Behavior No. The chromaticity vector of each low-sensitivity yellow anchor point takes values ​​ranging from the values ​​of each element. The values ​​are in CIELAB chromaticity units and are composed of the chromaticity of each yellow-dyed low-sensitivity anchor point. is the regularization coefficient, and is a scalar with a range of values ​​of . to Dimensionless, determined by the number and spatial distribution of anchor points, with a value of It can achieve a balance between overfitting and underfitting. Its function is to suppress coefficient oscillations caused by anchor noise. If the value is too large, the spatial illumination chromaticity field will tend to a constant and lose spatial adaptability. If the value is too small, it will be sensitive to anchor noise. for Identity matrix, dimensionless, used for regularization terms; superscript Indicates matrix transpose; superscript This represents the inversion of the matrix. After fitting the spatial illumination chromaticity field, it is extrapolated to each pixel position of the image of the yellow-stained high-sensitivity region to obtain the pixel-by-pixel spatial illumination chromaticity, as shown in equation (4):

[0040] (4),

[0041] in: The first in the image of the highly sensitive yellow staining region The spatial illumination chromaticity at each pixel location is a two-dimensional vector, with values ​​ranging from 1 to 2. Between these values, the unit is CIELAB chromaticity units, calculated by this formula, and used as the pixel-wise common-mode color shift that replaces the single individual self-reference common-mode reference at that pixel. The first in the image of the highly sensitive yellow staining region The position coordinates of each pixel are two-dimensional vectors, dimensionless, and are obtained by normalizing the pixel coordinates. Let be the spatial basis vector of the pixel location, defined in the same way as in equation (3). ; The definition is the same as in equation (3). By replacing the single individual self-reference common mode reference in equation (2) with the pixel-by-pixel spatial illumination chromaticity obtained in equation (4), the common mode color shift can be canceled pixel-by-pixel under multi-source mixed illumination.

[0042] Step S5: Construct a spatial yellowing distribution map from the pixel-by-pixel chromaticity residuals and classify them according to the head and tail expansion range. Obtain the trend of yellowing degree based on the individual yellowing baseline and perform head and tail progression constraints and anchor point self-verification.

[0043] The pixel-by-pixel chromaticity residuals of the high-sensitivity yellow-stained region image relative to the spatial illumination chromaticity field are used to construct a spatial yellow-stained distribution map. The value of the spatial yellow-stained distribution map at each pixel position is the residual of that pixel on the yellow-blue chromaticity axis relative to the spatial illumination chromaticity, that is, the b* component of the chromaticity component of the high-sensitivity region minus the b* component of the spatial illumination chromaticity. The spatial yellow-stained distribution map reveals the spatial distribution of yellow staining within the high-sensitivity yellow-stained region and is a natural derivative result of the spatial illumination chromaticity field estimation. Based on the spatial yellow-stained distribution map, its head-to-tail progression direction is determined, that is, the direction in which yellow staining spreads along the body surface from the head side to the foot side; according to the expansion range of the connected regions in the spatial yellow-stained distribution map where the degree of yellow staining exceeds the distribution threshold along the head-to-tail progression direction, the yellow staining progression grade is determined. The determination of the yellow staining progression grade is shown in Equation (5):

[0044] (5),

[0045] in: Let be the extent of the connected region along the head-to-tail progression direction, and let be a scalar with a value range of . , dimensionless, is calculated by this formula, and characterizes the degree of spatial spread of yellow staining in the head-to-tail direction; The largest connected region in the spatial yellow staining distribution map that exceeds the distribution threshold is obtained by thresholding the spatial yellow staining distribution map and taking the largest connected region. Let be the normalized position coordinates of pixels within a connected region, and let be a two-dimensional vector whose values ​​range from for each component to . Between them, there is no dimension; The normalized position coordinates of the head-side reference point are two-dimensional vectors, dimensionless, determined by the head-side endpoints in the high-sensitivity yellow staining region image, serving as the starting point for the head-to-tail progression direction; Let be a unit vector representing the direction of progress from head to tail, and be a two-dimensional vector with a value range of and a magnitude of . The vector is dimensionless and is determined by the direction of the head-to-tail axis in the image of the highly sensitive yellow-stained region. This represents the vector dot product operator, used to calculate the projected coordinates of a pixel in the head-to-tail direction; For the classification of yellow staining progress, is a scalar integer with a value range of . to , dimensionless, is obtained by comparing and summing the extended range with each grade threshold, and represents the spatial progression level of yellow staining; Here, is the threshold number for the classification, and is a scalar integer with a value range of . to Dimensionless; For each level, is a scalar integer with a range of values. to Dimensionless, determined by the graded needs of the nursing assessment, with a value of [value missing]. The time corresponds to three levels: light, medium, and heavy, which is a preferred value; For the first Each threshold value is a scalar, and its range is [value range missing]. And follow Increasing, dimensionless, determined by the quantiles of the extended range in the calibrated sample, its function is to delineate the hierarchical boundaries of the spatial progression of yellow staining; This indicates an indicator function that takes the value when the condition within the parentheses is true. Otherwise take Dimensionless. The classification of yellowing progression is incorporated into the quantitative value of yellowing degree, expanding the assessment result of yellowing degree from a single intensity dimension to a combination of intensity and spatial distribution.

[0046] In the cross-time assessment, the differential yellowing component at the time of patient admission was used as the individual yellowing baseline. The quantitative values ​​of yellowing severity in each assessment were relatively normalized to obtain the trend of yellowing severity. The individual yellowing baseline eliminated the bias caused by the patient's inherent skin color, so that the trend of yellowing severity only reflected the increase and decrease of yellowing severity itself over time. Considering that yellowing severity has spatial monotonicity along the head-to-tail progression direction, that is, as yellowing severity increases, the front of the head-to-tail progression advances towards the foot, this invention uses the spatial monotonicity of yellowing severity in the head-to-tail progression direction as a constraint to fit the trend of yellowing severity. The physical constraint fitting of the trend of yellowing severity is shown in Equation (6):

[0047] (6),

[0048] in: The fitted trend function for the degree of yellowing is a real-valued function with units of yellowing degree quantification scale. It is obtained by solving this formula and represents the change law of yellowing degree over time. Let be the trend function of yellow staining degree to be fitted, and be a scalar function; The evaluation time point number is a scalar integer with a value range of [value range missing]. to , dimensionless, used to index each evaluation; The total number of evaluations is a scalar integer, taking values ​​that are greater than zero, dimensionless, and determined by the number of evaluations already performed. For the first The quantified value of yellowing degree after relative normalization of individual yellowing baseline in the second assessment is a scalar, and the unit is the quantified scale of yellowing degree. It is obtained by subtracting the corresponding value of individual yellowing baseline from the quantified value of yellowing degree in this assessment, and is used as the fitted observation value. For the trend function at the th The time derivative at a given point in time is a scalar, with units of daily quantification of yellowing degree, representing the rate of change in yellowing degree. To extend the range in the first The time derivative at a given point in time is a scalar with a range of real numbers and a unit of 1 / day (i.e., every day, obtained by differentiating the dimensionless extended range with respect to time in days). It is obtained by the difference of the extended range of adjacent evaluations and represents the advancing speed of the frontier of the head and tail progress.

[0049] is the coupling coefficient, is a scalar, and its value range is a real number greater than zero. The unit is the quantification scale of yellow staining degree. It is determined by the correspondence between the rate of change and the rate of advancement of the front in the calibration sample. Its function is to couple the spatial advancement of the front of the head and tail with the temporal change of yellow staining degree. Here, represents the physical constraint weight, and is a scalar with a value range of . to Dimensionless, taking values ​​of The time balance between the data fitting term and the physical constraint term is used to control the strength of the spatial monotonicity constraint. If the value is too large, the trend will become overly dependent on spatial progress and deviate from the observation. If the value is too small, the physical constraint will be insufficient to suppress single-frame noise. This represents the independent variable that minimizes the objective function.

[0050] Furthermore, when the spatial yellow staining distribution map indicates that the front edge of the head-to-tail progression crosses a certain low-sensitivity yellow staining anchor point, it means that the area where the low-sensitivity yellow staining anchor point is located has been affected by yellow staining and no longer meets the assumption of low sensitivity to yellow staining. If it is continued to be used for common-mode baseline estimation, it will introduce yellow staining contamination. Therefore, the low-sensitivity yellow staining anchor point is removed from the fitting constraints of the spatial illumination chromaticity field or its weight is reduced, that is, the contribution of the anchor point in the fitting of equation (3) is set to zero or reduced. This feedback enables the estimation of the spatial illumination chromaticity field to adaptively exclude anchor points that have been affected by yellow staining, forming a closed loop in which the spatial yellow staining distribution map in step S5 influences the fitting of the spatial illumination chromaticity field in step S4. The anchor point self-verification is determined as follows: if the projected coordinates of the low-sensitivity yellow staining anchor point in the head-to-tail progression direction are less than the head-to-tail progression front edge position corresponding to the current expansion range, it is determined that the anchor point has been crossed by the front edge and is removed; otherwise, the anchor point is retained. Through the aforementioned closed loop, the fitting of the yellowing degree trend and the estimation of the spatial illumination color field support each other, enabling the assessment to remain stable as the yellowing gradually intensifies from the beginning to the end.

[0051] It should be noted that steps S4 and S5 above are further processing of multi-source mixed lighting and cross-time trends based on steps S1 to S3. In embodiments with a single light source and only a single intensity assessment, only steps S1 to S3 can be performed; in embodiments that need to address multi-source or cross-time assessments, steps S4 and S5 are performed accordingly.

[0052] To illustrate the effectiveness of this invention, a set of comparative embodiments are provided. Face and palm images of the same patient were captured under the same warm white light source. If existing technology is used without distinguishing between common-mode and differential-mode images, and the b* component of the facial image is directly used as the yellowing index, the yellowing caused by the warm white light source is included in the yellowing, causing even individuals without yellowing to be assessed as having significant yellowing. However, after using step S2 of this invention, which uses the palm as an individual self-reference common-mode benchmark to cancel common-mode color cast, the differential-mode yellowing component of individuals without yellowing approaches zero, and the interference from the warm white light source is effectively canceled. Under mixed window light and fluorescent lamp illumination, if only a single common-mode benchmark is used, the residual color cast on the side of the face facing the window differs from that on the side facing away from the window, causing spatial inconsistencies in the quantification value of yellowing degree within the face. However, after using step S4 of this invention, which uses multiple low-sensitivity yellowing anchor points to fit the spatial illumination chromaticity field, the spatially varying common-mode color cast is canceled pixel by pixel, improving the spatial consistency of the quantification value of yellowing degree within the face. Furthermore, in consecutive assessments of the same patient on different days, if the light conditions switch between warm white and cool white, using existing technology to directly use the b* component as the jaundice index will result in fluctuations in the quantification results as the light color temperature changes, making the jaundice levels on adjacent days incomparable. However, by using the present invention to offset common mode color bias and using the differential mode jaundice component at admission as the individual jaundice baseline for relative normalization, the quantification values ​​of jaundice levels on adjacent days fall on the same comparison benchmark, and the difference only reflects the rise and fall of jaundice itself. This allows nursing staff to objectively judge the aggravation or reduction of jaundice based on the trend of jaundice level.

[0053] It should be further noted that the selection of low-sensitivity areas for yellowing is of great significance to the robustness of this invention. The skin of the palm and sole not only has a thick stratum corneum, low blood perfusion, and delayed yellowing, but its surface is also relatively flat and free of hair obstruction, making it easy to obtain stable low-sensitivity chromaticity components. When the patient's palm has local abnormalities such as calluses or scars, the robust weight in equation (1) will automatically suppress the influence of these abnormal pixels on the individual self-reference common-mode benchmark. In the embodiment where both palm and sole skin are selected as low-sensitivity areas for yellowing, the chromaticity components of both locations can be included in the estimation of the individual self-reference common-mode benchmark to further improve the robustness of the common-mode color deviation estimation. In the embodiment where only palm skin can be collected, the palm skin is used alone as the bearing site of the individual self-reference common-mode benchmark. The above selection method enables this invention to stably obtain reliable common-mode benchmarks under diverse collection conditions in the nursing field.

[0054] See Figure 2 This invention also provides an image analysis-based system for assessing the degree of jaundice in liver disease patients, including an image acquisition unit, a common-mode correction unit, and a jaundice quantification unit. The embodiments of this system correspond to the above-described method embodiments, and the relevant descriptions in the method embodiments are applicable to the embodiments of this system; the similarities will not be repeated.

[0055] The image acquisition unit includes a lens and an image sensor, used to acquire and output a group of site images of the patient. The group of site images includes images of highly sensitive yellowing areas and images of low-sensitive yellowing areas. The highly sensitive yellowing areas are body surface areas with high yellowing load, and the low-sensitive yellowing areas are body surface areas with low yellowing load. The images of the highly sensitive and low-sensitive yellowing areas are formed within the same frame, under the same lighting conditions and with the same camera response. The image acquisition unit can be implemented as a camera component of a handheld terminal or a fixed camera component on a nursing cart. Its lens is used for imaging, and the image sensor is used to convert the light signal into a group of site images and output it. The image acquisition unit performs the function of step S1 in the method embodiment, and the acquired and output group of site images serves as the input to the common-mode correction unit.

[0056] The common-mode correction unit is connected to the image acquisition unit and is used to perform brightness normalization on the images of the high-sensitivity yellow-stained regions and the low-sensitivity yellow-stained regions respectively, to obtain the chromaticity components of the high-sensitivity regions and the low-sensitivity regions. The color cast that is consistent across the images of the high-sensitivity yellow-stained regions and the low-sensitivity yellow-stained regions is determined as the common-mode color cast. The chromaticity component of the low-sensitivity region is used as the individual self-reference common-mode reference, and the common-mode color cast is canceled from the chromaticity component of the high-sensitivity region to obtain the differential yellow-stained component. The differential yellow-stained component is the difference in yellow-stained load that varies with the region. The common-mode correction unit performs the function of step S2 in the method embodiment. It may further include a brightness normalization submodule, a common-mode reference estimation submodule, and a common-mode cancellation submodule, which respectively realize the functions of transforming the region image to the CIELAB color space and normalizing the L* component, estimating the individual self-reference common-mode reference according to formula (1), and canceling the common-mode color cast according to formula (2) to obtain the differential yellow-stained component. In the case of mixed illumination from multiple light sources, the common mode correction unit further extracts multiple low-sensitivity anchor points from the low-sensitivity yellow staining area image, fits the spatial illumination chromaticity field according to equation (3), and extrapolates it to the high-sensitivity yellow staining area image according to equation (4). The individual self-reference common mode reference is replaced by the pixel-by-pixel spatial illumination chromaticity, thus realizing the function of step S4 in the method embodiment.

[0057] The yellowing quantization unit is connected to the common-mode correction unit and is used to calculate and output the yellowing degree quantization value of the image of the high-sensitivity yellowing region based on the intensity of the differential yellowing component on the yellow-blue hue axis. The yellowing quantization unit performs the function of step S3 in the method embodiment, which takes the mean of the b* component of each pixel of the differential yellowing component on the yellow-blue hue axis and maps it to the yellowing degree quantization value, and calculates the brightness gradient map of the image of the high-sensitivity yellowing region when necessary, and excludes grazing pixels from the calculation of the yellowing degree quantization value. Further, the yellowing quantization unit can construct a spatial yellowing distribution map from the pixel-by-pixel chromaticity residual of the image of the high-sensitivity yellowing region relative to the spatial illumination chromaticity field, determine the yellowing progression grade according to formula (5) and incorporate it into the yellowing degree quantization value, and realize the function related to spatial distribution in step S5 in the method embodiment.

[0058] In one embodiment, the system further includes a trend recording unit connected to the yellowing quantification unit. This unit uses the differential yellowing component at patient admission as the individual yellowing baseline, relatively normalizes the yellowing quantification values ​​from each assessment to obtain a yellowing quantification trend, fits the trend to the spatial monotonicity of yellowing in the head-to-tail progression direction using equation (6), and feeds back to the common-mode correction unit when the head-to-tail progression front crosses a low-sensitivity yellowing anchor point. This low-sensitivity anchor point is then removed from the fitting constraint of the spatial illumination chromaticity field or its weight is reduced. The trend recording unit also summarizes the yellowing quantification values ​​obtained from multiple assessments into a yellowing quantification value sequence according to the assessment time and stores it, generating a curve showing the change of the yellowing quantification value sequence over time. The trend recording unit implements the function related to the cross-time trend in step S5 of the method embodiment, and its feedback to the common-mode correction unit constitutes a closed loop at the system level.

[0059] In the data stream, the image group of the location is output from the image acquisition unit to the common mode correction unit, the differential mode yellow staining component is output from the common mode correction unit to the yellow staining quantization unit, the yellow staining degree quantization value is output from the yellow staining quantization unit to the trend recording unit, and the determination result of the trend recording unit regarding the front edge of the head and tail progression crossing the low-sensitivity yellow staining anchor point is fed back to the common mode correction unit, thereby coupling each unit into a closed-loop whole that combines forward processing and reverse feedback.

[0060] In the above system embodiments, each unit can be implemented by a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments. The division of each unit is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple units can be integrated into the same processor, or the function of one unit can be implemented collaboratively by multiple processors.

[0061] 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 and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing the degree of jaundice in liver disease patients based on image analysis, characterized in that, include: The image acquisition unit acquires a group of images of the patient's body parts, which includes images of highly sensitive yellow staining areas and images of low-sensitive yellow staining areas. The highly sensitive yellow staining areas are the body surface areas with high yellow staining load, and the low-sensitive yellow staining areas are the body surface areas with low yellow staining load. The images of the highly sensitive yellow staining areas and the low-sensitive yellow staining areas are images of the highly sensitive yellow staining areas and the low-sensitive yellow staining areas within the same frame, under the same lighting conditions and the same camera response. Brightness normalization is performed on the images of the high-sensitivity yellow staining regions and the low-sensitivity yellow staining regions respectively to obtain the chromaticity components of the high-sensitivity regions and the chromaticity components of the low-sensitivity regions. The color cast that is consistent across the image of the high-sensitivity region of yellow staining and the image of the low-sensitivity region of yellow staining is defined as the common-mode color cast. Using the chromaticity component of the low-sensitivity region as the individual self-reference common mode standard, the common mode color deviation is canceled from the chromaticity component of the high-sensitivity region to obtain the differential mode yellow staining component, which is the difference in yellow staining load that varies with region. Based on the intensity of the differential yellow staining component on the yellow-blue hue axis, the quantization value of the yellow staining degree of the image of the high-sensitivity yellow staining region is calculated and output.

2. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 1, characterized in that, The low-sensitivity yellowing areas are selected from at least one of the palm skin and the sole skin, and the high-sensitivity yellowing areas are selected from at least one of the sclera and the facial skin; the light color cast carried by the low-sensitivity yellowing areas is used as the individual self-reference common mode benchmark to replace the external standard color card in the shooting image.

3. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 2, characterized in that, When acquiring the image group of the region, guide the low-sensitivity yellow staining region to be placed in the same illumination area adjacent to the high-sensitivity yellow staining region for same-frame imaging; calculate the brightness gradient map of the image of the high-sensitivity yellow staining region, mark the pixels with brightness gradients higher than the gradient threshold in the brightness gradient map as grazing pixels, and exclude the grazing pixels from the calculation of the quantification value of the yellow staining degree.

4. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 3, characterized in that, When the image group of the affected area is illuminated by a mixture of multiple light sources, multiple first-type low-sensitivity yellow staining anchor points are extracted from the low-sensitivity yellow staining area image. Furthermore, a low-yellowing-load auxiliary region is identified within the high-sensitivity yellow staining area image, and multiple second-type low-sensitivity yellow staining anchor points are extracted from the low-yellowing-load auxiliary region. The low-yellowing-load auxiliary region is an anatomical sub-region within the high-sensitivity yellow staining area image with relatively low bilirubin deposition load and relatively stable reflectivity, including at least one of the auricular root region, the nasal ala shadow boundary region, and the forehead hairline region. The first-type low-sensitivity yellow staining anchor points are used as the basis for further analysis. The location and chromaticity value of the second type of low-sensitivity anchor point are jointly constrained. A spatial illumination chromaticity field covering the high-sensitivity region image is fitted within the image coordinate plane where the high-sensitivity region image is located. The fitting and application of the spatial illumination chromaticity field are both limited to the same image coordinate plane where the high-sensitivity region image is located, so as to avoid the spatial illumination chromaticity field being extrapolated from the surface where the low-sensitivity region is located to the surface where the high-sensitivity region is located. The pixel-by-pixel chromaticity value of the spatial illumination chromaticity field is used to replace the individual self-reference common mode reference to cancel the common mode color shift.

5. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 4, characterized in that, A spatial yellowing distribution map is constructed by the pixel-by-pixel chromaticity residual of the image of the high-sensitivity yellowing region relative to the spatial illumination chromaticity field; the head and tail progression direction of the spatial yellowing distribution map is determined; based on the expansion range of the connected regions in the spatial yellowing distribution map whose yellowing degree exceeds the distribution threshold along the head and tail progression direction, the yellowing progression grade is determined, and the yellowing progression grade is incorporated into the yellowing degree quantification value.

6. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 5, characterized in that, Using the differential yellowing component at the time of patient admission as the individual yellowing baseline, the quantitative values ​​of yellowing degree in each assessment were relatively normalized to obtain the trend of yellowing degree. Using the spatial monotonicity of yellow staining in the head-to-tail progression direction as a constraint, the trend of yellow staining degree is fitted; when the spatial yellow staining distribution map indicates that the front edge of head-to-tail progression crosses a certain yellow staining low-sensitivity anchor point, the yellow staining low-sensitivity anchor point is removed from the fitting constraint of the spatial illumination chromaticity field or its weight is reduced.

7. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 1, characterized in that, The brightness normalization involves transforming the part image to the CIELAB color space and normalizing its L* component. The yellow-blue hue axis is the b* component axis of the CIELAB color space.

8. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 1, characterized in that, The standard lighting conditions for acquiring the image group of the aforementioned area include an illuminance of 300 lx to 1500 lx and a correlated color temperature of 3000 K to 6500 K.

9. The method for assessing the degree of jaundice in liver disease patients based on image analysis according to claim 1, characterized in that, The quantified values ​​of yellow staining degree obtained from multiple assessments are summarized into a sequence of quantified values ​​of yellow staining degree according to the assessment time and stored. The change curve of the quantified values ​​of yellow staining degree over time is then generated and presented.

10. An image analysis-based system for assessing the degree of jaundice in liver disease patients, used to implement the image analysis-based method for assessing the degree of jaundice in liver disease patients as described in any one of claims 1-9, characterized in that, It includes an image acquisition unit, a common-mode correction unit, and a yellow staining quantization unit; The image acquisition unit includes a lens and an image sensor, used to acquire and output a group of images of the patient's body parts. The group of images of the body parts includes images of highly sensitive yellow staining areas and images of low-sensitive yellow staining areas. The highly sensitive yellow staining areas are body surface areas with high yellow staining load, and the low-sensitive yellow staining areas are body surface areas with low yellow staining load. The images of the highly sensitive yellow staining areas and the low-sensitive yellow staining areas are images of the highly sensitive yellow staining areas and the low-sensitive yellow staining areas within the same frame, under the same lighting conditions and with the same camera response. The common-mode correction unit is used to perform brightness normalization on the images of the high-sensitivity yellow dyeing region and the low-sensitivity yellow dyeing region respectively, to obtain the chromaticity components of the high-sensitivity region and the low-sensitivity region. The color cast that is consistent across the images of the high-sensitivity yellow dyeing region and the low-sensitivity yellow dyeing region is determined as the common-mode color cast. The chromaticity component of the low-sensitivity region is used as the individual self-reference common-mode reference, and the common-mode color cast is canceled from the chromaticity component of the high-sensitivity region to obtain the differential-mode yellow dyeing component. The differential-mode yellow dyeing component is the difference in yellow dyeing load that varies with region. The yellow staining quantization unit is used to calculate and output the quantization value of the yellow staining degree of the image of the high-sensitivity yellow staining region based on the intensity of the differential yellow staining component on the yellow-blue hue axis. The image group of the affected area is output from the image acquisition unit to the common mode correction unit, and the differential mode yellow staining component is output from the common mode correction unit to the yellow staining quantization unit.

Images