An automatic lymphedema detection method

Abstract

The application provides an automatic lymphedema detection method, including: collecting optical image data of a target area; preprocessing the optical image data to obtain a preprocessed image; processing the preprocessed image using an image segmentation model to extract a target area contour; calculating a geometric feature parameter based on the target area contour and comparing it with a healthy benchmark model to obtain a deviation value; determining the degree of edema according to the deviation value and outputting a detection report.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an automatic method for detecting lymphedema. Background Technology

[0002] Lymphedema is a common and serious complication of cancer treatment, especially after breast cancer surgery. It directly affects patients' limb function, quality of life and even psychological state. Early and accurate identification and continuous monitoring of lymphedema are of great clinical significance.

[0003] Currently, the clinical assessment of lymphedema mainly relies on doctors or nurses measuring the circumference of multiple fixed points on the affected and healthy sides of the limbs with a soft measuring tape, and assessing the degree of edema by the difference in these measurements. However, this manual measurement method is easily affected by factors such as different measuring techniques, slight changes in patient posture, differences in soft tissue hardness, and inconsistent pressure applied during measurement. This leads to significant fluctuations in measurements taken by different medical personnel at different times for the same patient, making it difficult to obtain reliable continuous data. Furthermore, because only a few discrete points can be obtained for a rough circumference, it is difficult to detect and record subtle but important morphological abnormalities such as local bulges, changes in skin texture, or asymmetry. This is especially true in the subclinical stage when edema has just appeared, tissue fluid accumulation is not yet obvious, and the overall circumference change has not yet reached the diagnostic threshold. Manual measurement can hardly provide objective clues, often resulting in missed diagnoses of early lesions.

[0004] More importantly, the pathological process of lymphedema exhibits significant spatial heterogeneity. Fluid tends to preferentially accumulate in specific anatomical regions, such as the distal forearm, the back of the hand, above the elbow, or the inner side of the upper arm, while other areas may remain near normal for extended periods. This local morphological change and overall volume increase do not always occur synchronously, nor is there a fixed proportional relationship. Traditional methods, which can only collect perimeter data from a few fixed cross-sections, naturally lose a great deal of information about local shape, changes in surface curvature, and differences in symmetry between corresponding areas of the bilateral limbs. They cannot accurately reflect the true distribution characteristics and progression patterns of edema in three-dimensional space, thus making it difficult to make precise distinctions about the true severity of the disease and to provide consistent and comparable quantitative evidence at different stages of edema development.

[0005] Therefore, how to capture the true three-dimensional morphological information of the affected limb in a non-contact manner, especially how to accurately quantify the subtle geometric changes that are unevenly distributed locally, and establish an objective correspondence with the clinical stage, has become a key issue for the early identification, accurate grading and long-term dynamic monitoring of lymphedema. Summary of the Invention

[0006] This invention provides an automatic detection method for lymphedema, mainly comprising: Acquire optical image data of the target area; preprocess the optical image data to obtain a preprocessed image; process the preprocessed image using an image segmentation model to extract the contour of the target area; calculate geometric feature parameters based on the contour of the target area and compare them with a health benchmark model to obtain a deviation value; determine the degree of edema based on the deviation value and output a detection report.

[0007] Furthermore, the acquisition of optical image data of the target area includes: acquiring anteroposterior and lateral image data acquired by an image acquisition device under standard lighting conditions, wherein the image data represents the suspected edema area of ​​the patient; determining that the target area in the image data is fully presented without obstruction or distortion; and performing preliminary correction on the image data to unify the acquisition parameters.

[0008] Furthermore, the preprocessing of the optical image data to obtain a preprocessed image includes: converting the optical image data into grayscale image data; applying noise reduction filtering to the grayscale image data to remove noise interference; and performing size normalization on the filtered image data to obtain the preprocessed image, wherein the preprocessed image is adapted for subsequent model processing.

[0009] Furthermore, the step of processing the preprocessed image using an image segmentation model to extract the target region contour includes: inputting the preprocessed image into the image segmentation model, wherein the image segmentation model automatically identifies skin boundaries based on training samples; generating two-dimensional contour data of the target region in the preprocessed image; and applying morphological optimization to the two-dimensional contour data to correct edges and obtain the target region contour.

[0010] Furthermore, the step of calculating geometric feature parameters based on the target region contour and comparing them with the health benchmark model to obtain a deviation value includes: calculating girth parameters, area parameters, aspect ratio parameters, curvature distribution parameters, and symmetry parameters for the target region contour as the geometric feature parameters; obtaining the mean data of the corresponding geometric feature parameters in the health benchmark model; and comparing the geometric feature parameters with the mean data one by one to calculate the deviation value.

[0011] Furthermore, the step of determining the degree of edema and outputting a detection report based on the deviation value includes: applying weights to each deviation value to calculate a comprehensive deviation; determining the corresponding edema level if the comprehensive deviation is within a preset threshold range; and generating the detection report, wherein the detection report includes target area image data, geometric feature parameters, deviation value, and edema degree determination result.

[0012] Furthermore, the image segmentation model processes the preprocessed image to extract the contour of the target region, including: determining the boundary features of each pixel in the preprocessed image; segmenting the set of pixels inside the target region based on the boundary features; and fitting a two-dimensional contour line based on the set of pixels to obtain the contour of the target region, wherein the two-dimensional contour line is used for subsequent geometric feature parameter calculation.

[0013] Furthermore, the step of calculating geometric feature parameters based on the target region contour includes: traversing the target region contour to calculate integral values ​​to obtain girth parameters and area parameters; comparing the corresponding positions of the two target region contours to determine symmetry parameters; and analyzing the rate of change of the target region contour line segments to obtain curvature distribution parameters, wherein the geometric feature parameters are jointly used in the comparison with the health benchmark model.

[0014] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an automated method for detecting lymphedema, addressing the clinical pain points of traditional manual measurement of circumference, which is susceptible to subjective interference, cannot accurately capture local morphological abnormalities, and is difficult to achieve objective quantitative assessment of early subclinical edema. This method achieves non-contact automated analysis of suspected edema areas using optical image data. First, the invention acquires and preprocesses anteroposterior and lateral images of the affected limb under standardized conditions. Then, a deep learning image segmentation model is used to accurately extract the contour of the target region. Based on the contour, multidimensional geometric feature parameters are calculated, including key layer circumference, area, aspect ratio, curvature distribution, and bilateral symmetry. These parameters are then compared item by item with a health benchmark model of the same population to obtain standardized deviation. Finally, a weighted comprehensive deviation is calculated and mapped to international staging standards, achieving objective grading of edema severity and outputting a structured report. This method significantly improves the consistency, repeatability, and early screening sensitivity of the detection, providing an efficient and quantitative new technical means for the auxiliary diagnosis, long-term follow-up, and efficacy evaluation of lymphedema. Attached Figure Description

[0015] Figure 1 This is a flowchart of an automatic detection method for lymphedema according to the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0017] like Figure 1 This invention provides an automated method for detecting lymphedema, aiming to accurately analyze and determine suspected edema areas in patients through optical image data, thereby assisting in the medical diagnostic process. The implementation process and details of this method will be described step-by-step in the logical order of its technical implementation.

[0018] This invention first introduces the overall process of its method. The method mainly includes acquiring optical image data of the target area, preprocessing the image data, extracting the contour of the target area using an image segmentation model, calculating geometric feature parameters based on the contour and comparing them with a health benchmark model, and determining the degree of edema based on the deviation value and outputting a detection report. These steps are interconnected, progressively extracting key information from the original image data, ultimately achieving automated assessment of the degree of edema. Each step will be described in detail below to ensure the feasibility of the technical solution.

[0019] Step S1: Acquire optical image data of the target area.

[0020] This step is the starting point of the entire testing process, aiming to acquire image data of the patient's suspected edema areas as the basis for subsequent analysis. The target area is typically the patient's arm, leg, or other areas where lymphedema may occur. The acquisition process needs to be conducted in a standardized environment to ensure the quality and consistency of the image data. Specifically, image acquisition can be performed using a high-resolution digital camera or dedicated medical imaging equipment. During acquisition, the patient must maintain a specific posture, such as standing or lying down, so that the target area can be fully displayed. Ambient lighting should be controlled to be uniform and without strong reflections to avoid shadows or overexposure in the image. Furthermore, the distance and angle between the acquisition device and the target area must be fixed; for example, the camera should be approximately 50 cm away from the patient's arm and shot at a perpendicular angle to reduce the possibility of image distortion.

[0021] Step S11: Acquire anteroposterior and lateral image data under standard lighting conditions using an image acquisition device, wherein the image data characterizes the suspected edema area of ​​the patient. In one embodiment, the image acquisition device can be a medical-grade camera equipped with a high-resolution sensor. During image capture, it is necessary to ensure that the light intensity is within a preset range, such as 500 to 800 lux, to avoid the impact of excessively strong or weak light on image quality. Anteroposterior and lateral image data refer to images captured from both the front and side angles to comprehensively capture the morphological features of the target area. For example, for the detection of a patient's arm, a frontal image of the arm can be captured first to record the morphology of the area from the elbow to the wrist; then a side image can be captured to record the lateral features of changes in arm thickness. This multi-angle acquisition method helps to more comprehensively assess the edema situation in subsequent analysis.

[0022] Step S12: Ensure the target area in the image data is fully presented, unobstructed, and free from distortion. In one implementation, after acquisition, a preliminary check of the image data is required to ensure the target area is fully visible in the image and unobstructed by clothing, accessories, or other objects. For example, if the target area is the patient's lower leg, it needs to be confirmed that the portion from the knee down to the ankle is clearly visible in the image, without any trouser legs or other obstructions. Simultaneously, check for image distortion, such as stretching or compression of the target area due to improper shooting angle. If the image does not meet the requirements, it needs to be retaken until the image quality meets the requirements for subsequent processing. This check can be completed manually or with the assistance of simple image processing software, such as checking whether the pixel ratio of the target area in the image reaches a preset proportion.

[0023] Step S13 involves performing preliminary corrections on the image data to standardize acquisition parameters. Specifically, since different devices or acquisition environments may cause differences in image brightness, contrast, or resolution, preliminary corrections are necessary to ensure consistency in subsequent processing. For example, image brightness and contrast can be adjusted to meet preset standard values; simultaneously, if image resolutions are inconsistent, they need to be adjusted to a uniform size, such as adjusting all images to a resolution of 1920 x 1080 pixels. Furthermore, if the shooting angle is slightly off, geometric transformations can be used to correct the image, rotating slightly tilted images to ensure the target area presents a standard orientation. The corrected image data will serve as input for subsequent preprocessing, ensuring consistency and accuracy in the analysis process.

[0024] In one embodiment, the preliminary correction process for detecting arm edema may include the following details: First, the brightness distribution of the image is checked. If the overall image is found to be too dark, the brightness parameters are adjusted to make the skin details in the arm area of ​​the image clearer. Second, if the image resolution is lower than a preset standard, the resolution is increased using interpolation methods to ensure that the subsequent segmentation model can recognize subtle boundary features. Finally, if the arm is slightly tilted in the image, rotation correction is used to align the arm's principal axis with the vertical axis of the image. These correction measures can effectively improve the usability of the image data and lay the foundation for subsequent processing.

[0025] Step S2: Preprocess the optical image data to obtain a preprocessed image.

[0026] This step aims to transform the raw image data into a format suitable for subsequent model processing through a series of image processing operations, while removing noise interference and improving image quality. Preprocessing is a crucial step in ensuring the accuracy of subsequent segmentation and analysis, involving multiple sub-steps, which will be explained in detail below.

[0027] Step S21: Convert the optical image data into grayscale image data. In one implementation, the original image data is typically a color image containing information from the red, green, and blue color channels. Subsequent segmentation models are usually based on grayscale information, therefore the color image needs to be converted into a single-channel grayscale image. Specifically, the grayscale value of each pixel can be calculated using a weighted average method, for example, assigning different weights to the red, green, and blue channels according to the differences in human eye sensitivity to different colors. The converted grayscale image retains the brightness information and morphological features of the target area while reducing the amount of data and lowering the computational complexity of subsequent processing. For example, for an image of an arm region, the brightness difference between the arm skin area and the background area is still significant in the converted grayscale image, facilitating subsequent boundary recognition.

[0028] Step S22: Apply noise reduction filtering to the grayscale image data to remove noise interference. In one embodiment, the original image data may contain noise, such as pixel anomalies caused by device sensor sensitivity or uneven ambient lighting. This noise can interfere with the accuracy of subsequent segmentation models, so noise reduction processing is required through filtering methods. Specifically, smoothing filtering techniques can be used to average or weight the pixel values ​​of each pixel in the image and its neighborhood, thereby smoothing out noise points. For example, for possible random noise, a 3x3 filtering window can be selected, and the pixel values ​​within the window area can be weighted and averaged to obtain the smoothed grayscale value. In the denoised image, small noise points are effectively removed, while the boundary features of the target area are still preserved. For example, in arm edema detection, the grayscale value distribution of the arm skin area is more uniform after noise reduction, avoiding boundary misjudgment caused by noise.

[0029] Step S23 involves performing size normalization on the filtered image data to obtain a preprocessed image, which is adapted for subsequent model processing. In one implementation, due to differences in acquisition devices or shooting distances, the image data sizes may be inconsistent, while subsequent image segmentation models typically require input images to have a uniform size standard. Therefore, the filtered grayscale image needs to be size normalized to meet the model input requirements. For example, if the model requires an input image size of 512 by 512 pixels, while the actual filtered image size is 600 by 800 pixels, it needs to be adjusted to the target size through scaling or cropping. During scaling, interpolation methods can be used to ensure that image details are not lost; during cropping, it is necessary to ensure that the target region is completely preserved in the center of the image. The size-normalized preprocessed image will serve as the input to the subsequent segmentation model, ensuring the uniformity and accuracy of model processing.

[0030] In one embodiment, the size normalization process for detecting leg edema may include the following details: First, determine the actual size of the filtered image, for example, 720 by 960 pixels; second, scale it to 512 by 512 pixels according to model requirements, using bilinear interpolation during scaling to ensure that the boundary details of the leg region are not distorted; finally, check whether the leg region in the scaled image is centered, and if there is an offset, adjust it by translation to place it in the center of the image. This size normalization process can effectively adapt to the model input requirements and improve the accuracy of subsequent segmentation.

[0031] Step S3: Use an image segmentation model to process the preprocessed image to extract the contour of the target region.

[0032] This step is the core of the entire detection process. It aims to accurately extract the contour information of the target region from the pre-processed image using a pre-trained image segmentation model, providing a foundation for subsequent geometric feature parameter calculations. The image segmentation model is trained on a large amount of labeled image data and can automatically identify the boundary between the target region and the background region in the image. The specific implementation process of this step will be described in detail below.

[0033] Step S31: Input the preprocessed image into the image segmentation model, where the image segmentation model automatically identifies skin boundaries based on training samples. In one implementation, the image segmentation model can be a deep learning-based network structure, trained on a large amount of labeled medical image data to learn the distinguishing features between skin and background regions. After inputting the preprocessed image, the model classifies each pixel in the image to determine whether it belongs to the target region. For example, for arm edema detection, the model analyzes the grayscale distribution and texture features of the arm skin region in the preprocessed image, marking pixels belonging to the arm skin as target regions, and marking pixels in the background or other non-target regions as non-target regions. The model output is a binary image, where the pixel value of the target region is 1, and the pixel value of the background region is 0. In this way, the model can accurately identify the skin boundary of the target region, laying the foundation for subsequent contour extraction.

[0034] Step S32: Generate two-dimensional contour data of the target region in the preprocessed image. In one embodiment, based on the binary image output by the model, the boundary lines of the target region, i.e., two-dimensional contour data, can be further extracted. Specifically, a boundary tracing method can be used to traverse the regions with a pixel value of 1 in the binary image, recording the coordinate positions of the boundary pixels to form a continuous contour line. For example, for leg edema detection, the contour line of the leg region is extracted from the binary image, and the boundary coordinate data from the knee to the ankle is recorded to form a complete two-dimensional contour line. Such contour data can clearly characterize the external shape of the target region, providing a direct basis for subsequent geometric feature calculations.

[0035] Step S33: Apply morphological optimization to the 2D contour data to correct edges and obtain the target region contour. In one implementation, the initially extracted 2D contour data may contain minor discontinuities or jagged edges, such as broken contour lines or small protrusions caused by residual image noise. To improve the smoothness and integrity of the contour data, morphological optimization is required. Specifically, small breaks in the contour lines can be filled using morphological closure operations; simultaneously, small protrusions or isolated points on the contour lines can be removed using morphological opening operations. For example, for the contour data of the arm region, if there are small gaps at the boundary caused by noise, the break points are connected using closure operations to make the contour lines continuous; if there are isolated small protrusions, they are removed using opening operations. The optimized target region contour is smoother and more complete, accurately reflecting the true morphological characteristics of the target region.

[0036] In one embodiment, the morphological optimization process for arm edema detection may include the following details: First, the initial contour data is examined, revealing a small break in the elbow region caused by noise, approximately 5 pixels in length. Second, a 3x3 rectangular window is set as the structuring element for the closing operation to fill the broken area, restoring continuity to the contour line in the elbow region. Finally, the contour line is examined, revealing an isolated protrusion in the wrist region, approximately 10 pixels in area. An equally sized window is set as the structuring element for the opening operation to remove this protrusion. The optimized arm contour line is smoother, and the boundary features more closely resemble the actual shape, providing a reliable data foundation for subsequent geometric feature parameter calculations.

[0037] In another embodiment, the morphological optimization process for detecting calf edema can be further refined: First, analyzing the initial contour data reveals discontinuities in the knee region due to image shadows, with a break length of approximately 8 pixels. Second, a 5x5 rectangular window is set as the structuring element for the closing operation to fill the broken areas, restoring the knee region's contour to its integrity. Finally, examining the contour reveals multiple isolated small protrusions in the ankle region, each with an area of ​​approximately 6 to 12 pixels. A 5x5 window is set as the structuring element for the opening operation to remove these protrusions one by one. The optimized calf contour has smoother boundaries and its morphological features are more realistic, laying a solid foundation for subsequent analysis.

[0038] It's important to note that the training process of the image segmentation model is crucial for achieving accurate boundary recognition. In one implementation, the training data can include a large number of medical images from different patients, such as images of the arms and legs. Each image is manually labeled with the boundary information between the target region and the background region. By learning from this labeled data, the model gradually masters the distinguishing characteristics of the skin region, such as grayscale distribution and texture features. For example, for an arm image, the model learns that the grayscale values ​​of the arm skin region are usually within a specific range, and the texture features are relatively smooth, while the background region may contain complex texture features such as clothing and sheets. Through repeated iterative training, the model can quickly determine whether each pixel belongs to the target region when a new image is input, outputting an accurate binary image result.

[0039] In another embodiment, the training of the image segmentation model for leg edema detection can be further optimized: First, leg image data under different lighting conditions, such as bright lighting and shadow lighting, are collected to ensure the model adapts to various acquisition environments. Second, each image is manually labeled to clearly define the boundary between the leg skin area and the background area; for example, the skin area from the knee to the ankle is labeled as the target area. Finally, the labeled data is input into the model, and through multiple rounds of training, the model learns the characteristics of the leg skin area, such as the grayscale value distribution being concentrated in a certain range and the texture changes being relatively smooth. The trained model can quickly output a binary image when a new leg image is input, accurately distinguishing the target area from the background area, providing reliable support for subsequent contour extraction.

[0040] Specifically, the application of image segmentation models can significantly improve the efficiency and accuracy of target region contour extraction. Traditional manual annotation methods are time-consuming and susceptible to subjective influence, while deep learning-based segmentation models can automatically process large amounts of image data, outputting highly consistent and accurate binary image results. For example, in arm edema detection, the model can complete the segmentation of an image within seconds, outputting a binary image that distinguishes the arm region from the background region, whereas manual annotation may take several minutes or even longer. Through automated processing by the model, not only are labor costs saved, but the objectivity and consistency of the results are also ensured, providing high-quality data support for subsequent geometric feature analysis.

[0041] In one implementation, the image segmentation model's effectiveness in detecting arm edema can be further validated: First, a newly acquired arm image, 800 x 600 pixels, is input. Second, the model completes segmentation processing within approximately 2 seconds, outputting a binary image where the arm region's pixel value is 1 and the background region's pixel value is 0. Finally, the output result is examined, revealing clear arm region boundaries that highly match the actual shape, with a segmentation accuracy exceeding 95%. This efficient processing capability and high-precision result fully demonstrate the model's application value in target region contour extraction, laying a solid foundation for subsequent analysis.

[0042] Step S311: Determine the boundary features of each pixel in the preprocessed image. In one embodiment, based on the output of the image segmentation model, further detailed analysis can be performed to determine the boundary features of each pixel in the preprocessed image. Specifically, boundary features refer to whether a pixel is located at the boundary between the target region and the background region. For example, for arm edema detection, each pixel in the preprocessed image is traversed, and it is checked whether there is a change in pixel value from 1 to 0 or from 0 to 1 in its neighborhood. If so, the pixel is determined to have boundary features. In this way, all pixels that may belong to the boundary can be initially screened out, providing a reference for subsequent segmentation.

[0043] Step S312: Segment the pixel set inside the target region based on boundary features. In one implementation, based on determined boundary feature pixels, the pixel set inside the target region can be further segmented. Specifically, a region filling method can be used to fill in all pixels belonging to the target region inward from the boundary feature pixels. For example, for leg edema detection, a boundary feature pixel in the knee region is selected as the starting point, and all pixels with a value of 1 are traversed inward and marked until a boundary feature pixel or image edge is encountered. In this way, all pixels inside the leg region can be completely segmented, forming the pixel set inside the target region, providing data support for subsequent contour fitting.

[0044] Step S313 involves fitting a two-dimensional contour line based on the pixel set to obtain the target region contour, where the two-dimensional contour line is used for subsequent geometric feature parameter calculation. In one embodiment, based on the pixel set within the segmented target region, a two-dimensional contour line can be further fitted as the final target region contour. Specifically, a boundary tracking algorithm can be used to sequentially record the coordinate positions of boundary pixels along the edges of the pixel set within the target region, forming a continuous contour line.

[0045] Based on the optimized target region contour obtained above, the subsequent analysis process is carried out to achieve a quantitative assessment of the degree of lymphedema.

[0046] Step S4: Calculate the geometric feature parameters based on the target region contour and compare them with the health benchmark model to obtain the deviation value.

[0047] The core task of this step is to transform the previously extracted contour information into several quantifiable geometric indicators, and by comparing them with statistical benchmarks of healthy individuals, determine the degree to which the shape of the current target area deviates from the normal state.

[0048] Step S41: Calculate the girth parameter, area parameter, aspect ratio parameter, curvature distribution parameter, and symmetry parameter for the target region contour as the geometric feature parameters. In one embodiment, the girth parameter mainly reflects the perimeter of the target limb at a specific location. Typically, several key cross-sectional locations are selected for measurement. For example, in an upper arm detection scenario, multiple landmark layers such as 5 cm above the elbow, 5 cm below the elbow, and 5 cm above the wrist can be selected. The perimeter values ​​corresponding to these layers are calculated along the contour to form a girth sequence. The area parameter counts the total number of pixels in the two-dimensional region enclosed by the contour or the converted actual physical area, which can be approximated by pixel counting within the contour or the polygon area formula. The aspect ratio parameter typically represents the flatness or roundness of the overall shape as the ratio of the major axis to the minor axis of the smallest bounding rectangle of the target region. The curvature distribution parameter focuses on the severity of local curvature changes in the contour line. It is often calculated by sampling points at fixed intervals along the contour line, calculating the curvature value in the neighborhood of each sampling point, and then statistically analyzing the distribution characteristics of the curvature values, such as the mean, standard deviation, and peak position. Symmetry parameters are mainly used in bilateral limb comparison scenarios. By mirroring the contours of the affected side and the healthy side, the average distance deviation or area overlap rate between corresponding points on both sides is calculated to quantify the degree of asymmetry.

[0049] In typical applications for upper limb lymphedema, the calculation of circumference parameters can be further refined as follows: First, locate the elbow joint on the contour (usually the local maximum of the contour curvature or pre-marked through key point detection). Then, using the elbow joint as a reference, draw approximately perpendicular lines to the principal axis at certain physical distances (e.g., 2 cm) upwards and downwards. Calculate the sum of the two arc lengths between the intersection of each line and the contour as the circumference value of that layer. Usually, circumference data from 8 to 12 layers are collected to form a curve reflecting the circumference variation from the shoulder to the wrist. When the edema is severe, this curve often shows a significant thickening at the proximal end and a relatively normal distal end.

[0050] Step S411: Traverse the contour of the target region and calculate the integral value to obtain the girth and area parameters. Specifically, the area parameter is often obtained by using the shoelace formula or pixel filling statistics method, accumulating the coordinates of the contour points arranged in clockwise or counterclockwise order to obtain an approximate value of the enclosed area. For the girth parameter, local integration calculation is mainly performed on several selected key sections, or the Euclidean distance between adjacent contour points is directly accumulated to approximate the perimeter of the entire contour. However, in clinical practice, the combination of local perimeters of key sections is more commonly used because the overall perimeter has lower sensitivity to severe local edema.

[0051] Step S412: Compare the corresponding positions of the contours of the target regions on both sides to determine the symmetry parameters. In one implementation, bilateral comparison is a very important auxiliary indicator for the diagnosis of lymphedema. Implementation typically requires simultaneously acquiring images of the affected and healthy sides under the same posture and lighting conditions. After extracting the contours of both sides, they are mirrored and aligned with key points (e.g., elbow joint, wrist center, acromion, etc.). Then, various indicators such as the average Euclidean distance, Hausdorff distance, or area difference ratio between corresponding sampling points are calculated. Hausdorff distance is more sensitive to severe local protrusions or depressions and is often used to detect unilateral severe localized edema; while the area difference ratio better reflects the degree of overall volume change. In practice, multiple symmetry indicators are usually combined and assigned different weights to form a comprehensive symmetry deviation score.

[0052] Step S413: Analyze the rate of change of the contour segments of the target region to obtain curvature distribution parameters. Curvature distribution is a morphological indicator of edema that has gained increasing attention in recent years because simple circumference increase may be caused by muscle development or fat accumulation, while edema is often accompanied by changes in local tissue tension, resulting in a characteristic distribution pattern of contour curvature. For example, in severe lymphedema of the arm, a more obvious "bulge" often appears in the area above the elbow, corresponding to a large positive peak value in the local curvature, while the area near the wrist is relatively flat. The analysis of curvature distribution usually includes: ① Calculate the local curvature value at each point on the profile; ② Smooth the curvature sequence to suppress noise; ③ Statistical analysis of the absolute values ​​of curvature, including mean, standard deviation, interquartile range, kurtosis, and skewness; ④ Locate the location of the curvature maxima and their normalized coordinates relative to the overall length; ⑤ Sometimes, the proportions of positive and negative curvature are calculated to determine whether the contour is more like "convex edema" or accompanied by "depressed scar" characteristics.

[0053] In one embodiment, taking a patient with severe forearm lymphedema as an example, the curvature distribution parameters may exhibit the following characteristics: the maximum curvature value occurs approximately 8–12 cm below the elbow, about 2.8–4.1 times that of the normal control group; the standard deviation of curvature is approximately 2.3 times that of the normal group; the proportion of positive curvature (convexity) reaches over 78%, while normal individuals typically fall within the 52%–64% range; the peak curvature position is mostly concentrated in the 38%–52% range of the normalized contour length (calculated with the wrist as 0% and the elbow as 100%). These distribution characteristics combined have certain diagnostic value in differentiating true lymphedema from simple venous return obstruction or fat accumulation.

[0054] Step S42: Obtain the mean data of the corresponding geometric feature parameters in the health benchmark model. The health benchmark model is typically constructed based on data from a large number of healthy volunteers (excluding those with lymphatic system diseases, abnormal cardiac or renal function, significant obesity, etc.) of the same sex, age group, height, and measurement site. Different sexes, age groups, and height ranges often require the establishment of stratified reference databases. For example, the baseline value for the average circumference of the upper arm 5 cm above the elbow for women aged 20–40 might be 24.5 ± 2.1 cm, while for women aged 60–75 it might decrease to 22.8 ± 2.6 cm. Therefore, in actual comparison, it is necessary to first select the most matching reference group from the benchmark database based on the patient's basic information, and extract the mean μ and standard deviation σ for the corresponding location and level.

[0055] Step S43: Compare the geometric feature parameters with the mean data one by one to calculate the deviation value. In one implementation, the deviation calculation often uses a standardized score (Z-score) form, that is, for each geometric indicator P... i Calculate its deviation: z i =(P i -μ i ) / σ i Where μ i and σ i These represent the mean and standard deviation of the indicator in the corresponding health reference group, respectively.

[0056] Then, different weights can be assigned to the Z-scores of different indicators according to their clinical significance (for example, the Z-score of the difference between the affected and healthy sides usually has a higher weight, the Z-score of the curvature abnormality has a lower weight, and the Z-score of the unilateral absolute value has a lower weight). Finally, the comprehensive deviation value D can be obtained by weighted summation or taking the maximum value / Euclidean norm.

[0057] In another implementation, the percentile method can be used, which converts the patient's current measurement value into a percentile rank P (0-100) within the distribution of healthy individuals. A higher value indicates a greater deviation from normal. For example, if the perimeter of a certain area is at the 96th percentile of the healthy population distribution, the indicator is considered severely abnormal; if it is at the 62nd percentile, it is considered to be within the upper-middle range of normal. Multiple deviation expression methods can be selected or combined according to the different medical institution's practices and reporting needs.

[0058] Step S5: Determine the degree of edema based on the deviation value and output a test report. This step is the output stage of the entire process, responsible for converting the previously calculated quantitative indicators into edema grading conclusions that clinicians can directly understand and use, and integrating all intermediate results to form a structured report.

[0059] Step S51: Apply weights to each of the aforementioned deviation values ​​to calculate the overall deviation. In one embodiment, the formula for calculating the overall deviation can be roughly expressed as: D 综合 =w1·Z 围度 +w2·Z 面积 +w3·Z 不对称 +w4·Z 曲率异常 +w5·Z 长宽比 The weights w are determined based on a large amount of clinical validation data using statistical methods (such as logistic regression coefficients, random forest feature importance, etc.).

[0060] Asymmetry and curvature abnormality indicators usually have a higher weight in the early identification of lymphedema, while circumference and area indicators contribute more to the staging of intermediate to severe edema.

[0061] Step S52: If the overall deviation is within a preset threshold range, the corresponding edema level is determined. The internationally used clinical staging of lymphedema mainly refers to the ISSL (International Lymphatic Society) standards, which divide it into stage 0 (subclinical), stage I (early pitting), stage II (non-pitting, with fibrosis), and stage III (elephantiasis stage, with severe skin and subcutaneous tissue changes). Automated detection systems typically map the overall deviation to these four levels; a typical threshold division example is as follows: D<1.5 indicates essentially normal or very mild; 1.5 ≤ D < 3.2 indicates suspected stage or stage 0 to I; 3.2 ≤ D < 6.8 indicates a clear stage I to II; A D ≥ 6.8 indicates a tendency towards stage II to III.

[0062] However, the actual threshold needs to be clinically verified and adjusted according to different body parts (upper limbs / lower limbs), different populations, and different devices. There is no universally applicable fixed value.

[0063] In one implementation, a two-tier threshold strategy can be used to improve the sensitivity of early screening: setting a lower threshold D... alert To alert doctors to potential subclinical edema, a relatively high threshold is set (D). confirm This indicates a high probability of clinically diagnosed edema. The area between these two is a "gray zone," and it is recommended to make a comprehensive judgment based on the patient's chief complaint, medical history, physical examination, and other auxiliary examinations.

[0064] Step S53: Generate the detection report, which includes target area image data, geometric feature parameters, deviation values, and edema degree determination results. A complete detection report typically includes the following structured modules: ① Patient basic information and detection time; ② Acquired images (including both original images and segmented annotated contour images); ③ Key measurement data table (circumference, area, symmetry indicators, curvature features, etc. of each layer); ④ Comparison results with a healthy baseline (Z-values ​​or percentiles of each indicator); ⑤ Comprehensive deviation values ​​and preliminary staging suggestions; ⑥ Disclaimer (stating that this system is an auxiliary tool, and the final diagnosis is the responsibility of the physician); ⑦ A trend chart of the patient's historical detection data can be attached to facilitate observation of edema progression or treatment effects.

[0065] In one embodiment, for patients with long-term follow-up of postoperative lymphedema of the upper limb, the test report can be specially equipped with a trend analysis module, which compares the perimeter values ​​of each layer of the current test with the corresponding data of the previous tests, and displays the trend of change in the form of a line graph, while marking key time nodes (such as the end of radiotherapy, the start of compression therapy, surgical intervention, etc.), so that doctors can intuitively judge the effectiveness of the intervention measures.

[0066] In another implementation, for secondary lymphedema of the lower extremities (especially after tumor surgery or radiotherapy), the report can be supplemented with a stratified bar chart of the difference in circumference between the affected limb and the healthy limb, and the bar chart can be marked with commonly used international diagnostic criteria (such as the circumference of the affected side being ≥2 cm or ≥10% greater than that of the healthy side), so that doctors can quickly compare the consistency and differences between traditional diagnostic criteria and machine learning quantitative results.

[0067] It should be noted that although the automatic extraction and quantitative analysis of geometric feature parameters can significantly improve the consistency and objectivity of detection, the pathogenesis of lymphedema is complex and is affected by a variety of factors (venous function, fat distribution, degree of fibrosis, skin tension, measurement posture, etc.). Therefore, any automated system can only serve as auxiliary information for clinical decision-making at present, and cannot replace the physician's comprehensive assessment and judgment.

[0068] In actual clinical deployment, it is generally recommended to combine the results of this method with other examination methods such as traditional tape measure, bioelectrical impedance analysis (L-Dex), lymphoscintigraphy, MRI or high-frequency ultrasound to form a multimodal diagnostic strategy to improve the early detection rate and staging accuracy.

[0069] If the technical solution of this application involves the collection, processing, or application of personal information, the relevant products have, before implementing any personal information processing activities, fully and clearly informed individuals of the processing rules in accordance with the "Personal Information Protection Law of the People's Republic of China" and other current laws and regulations, and obtained their voluntary and explicit consent. If sensitive personal information is involved, the product has obtained the individual's separate consent before processing, and such consent is given in an explicit manner. For example, prominent signs are set up in the area where information collection devices such as cameras are located, clearly indicating "Entering is considered as consent to the collection of personal information"; or through pop-ups, checkboxes, user-initiated uploads, etc., under the premise of clearly listing the processor's identity, processing purpose, processing method, and information type, the user actively completes the authorization operation. The above mechanisms ensure that all personal information processing activities are based on legal authorization and fully comply with national compliance requirements regarding personal information protection.

[0070] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. The present invention has been described in detail with reference to preferred embodiments. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An automated method for detecting lymphedema, characterized in that, include: Acquire optical image data of the target area; The optical image data is preprocessed to obtain a preprocessed image; The preprocessed image is processed using an image segmentation model to extract the contour of the target region; Geometric feature parameters are calculated based on the contour of the target region and compared with the health benchmark model to obtain the deviation value; The degree of edema is determined based on the deviation value, and a test report is output.

2. The method as described in claim 1, characterized in that, The optical image data of the target area acquired includes: Acquire frontal and lateral image data acquired by an image acquisition device under standard lighting conditions, wherein the image data characterizes the suspected edema area of ​​the patient; The image data is confirmed to contain a complete target region that is unobstructed and free from distortion. Preliminary corrections are performed on the image data to unify the acquisition parameters.

3. The method as described in claim 1, characterized in that, The preprocessing of the optical image data to obtain a preprocessed image includes: Convert the optical image data into grayscale image data; A noise reduction filter is applied to the grayscale image data to remove noise interference; Size normalization is performed on the filtered image data to obtain the preprocessed image, wherein the preprocessed image is adapted for subsequent model processing.

4. The method as described in claim 1, characterized in that, The step of processing the preprocessed image using an image segmentation model to extract the contour of the target region includes: The preprocessed image is input into the image segmentation model, wherein the image segmentation model automatically identifies skin boundaries based on training samples; Generate two-dimensional contour data of the target region in the preprocessed image; Morphological optimization is applied to the two-dimensional contour data to correct the edges and obtain the contour of the target region.

5. The method as described in claim 1, characterized in that, The step of calculating geometric feature parameters based on the contour of the target region and comparing them with a health benchmark model to obtain a deviation value includes: The girth parameter, area parameter, aspect ratio parameter, curvature distribution parameter, and symmetry parameter of the target region are calculated as the geometric feature parameters. Obtain the mean data of the corresponding geometric feature parameters in the health benchmark model; The deviation value is calculated by comparing the geometric feature parameters with the mean data one by one.

6. The method as described in claim 1, characterized in that, The process of determining the degree of edema based on the deviation value and outputting a test report includes: A weighted average deviation is calculated by applying weights to each deviation value. If the overall deviation is within a preset threshold range, the corresponding edema level is determined. The detection report is generated, wherein the detection report includes target area image data, geometric feature parameters, deviation value and edema degree determination result.

7. The method as described in claim 4, characterized in that, The image segmentation model processes the preprocessed image to extract the contour of the target region, including: Determine the boundary features of each pixel in the preprocessed image; The set of pixels inside the target region is segmented based on the boundary features; The target region contour is obtained by fitting a two-dimensional contour line based on the pixel set, wherein the two-dimensional contour line is used for subsequent geometric feature parameter calculation.

8. The method as described in claim 5, characterized in that, The calculation of geometric feature parameters based on the contour of the target region includes: The integral value is calculated by traversing the contour of the target region to obtain the girth parameter and area parameter; Compare the corresponding positions of the target region contours on both sides to determine the symmetry parameters; The rate of change of the contour line segments of the target region is analyzed to obtain curvature distribution parameters, wherein the geometric feature parameters are used in comparison with the health benchmark model.

Images