A method, apparatus and system for assessing the area of skin lesions on human extremities

Abstract

The application belongs to the technical field of medical image processing, and particularly relates to a human limb skin disease area evaluation method, device and system. The method comprises the following steps: collecting multi-view images of a target limb, adjacent view angles of the multi-view images containing overlapping areas; generating a three-dimensional point cloud of the target limb based on the multi-view images; constructing a cylindrical model with a radius changing along a limb axis based on the three-dimensional point cloud; projecting the multi-view images to the cylindrical model, and selecting pixels projected to the same position of the model based on an evaluation function, so that the model surface is attached with image texture; and unfolding the cylindrical model with the texture into a two-dimensional plane image for evaluating the skin disease area. The application realizes high-precision and high-fidelity lesion area evaluation without expensive three-dimensional scanning equipment.

Description

Technical Field

[0001] This application belongs to the field of medical image processing technology, specifically relating to a method, device, and system for assessing the area of ​​skin diseases on the human limbs. Background Technology

[0002] In clinical assessment of dermatological diseases, accurately calculating the area of ​​skin lesions on the limbs (such as the upper and lower extremities) is crucial for quantifying the severity of diseases like psoriasis and vitiligo (e.g., PASI scores) and tracking treatment effectiveness. Traditionally, doctors have relied on visual examination or manual measurement, methods that are highly subjective, have poor repeatability, and are difficult to use for objective quantitative analysis.

[0003] To address these issues, existing technologies are primarily developing in two directions: one is based on automated assessment systems using two-dimensional images, such as estimating limb area by stitching together front and back images; the other is using 3D scanners (such as structured light devices) to directly acquire high-precision 3D models. However, both of these approaches have significant drawbacks. For two-dimensional image systems, because human limbs are roughly cylindrical, when photographed and stitched from limited perspectives (such as front and back), severe geometric distortions (such as stretching or compression) inevitably occur on the sides of the image, leading to area calculation errors of over 20%. Furthermore, even when introducing cylindrical models for correction, it often assumes that the limbs are cylinders of uniform diameter, while actual limbs (such as forearms and lower legs) are actually truncated cone-like structures with gradually varying thicknesses. Such fixed-radius models will cause excessive compression at the thinner ends and excessive stretching at the thicker ends, again failing to guarantee the fidelity of lesion morphology and the accuracy of area calculation. For 3D scanning technology, although it can avoid geometric distortions and achieve high precision, its equipment is expensive, its acquisition speed is slow, and it is sensitive to ambient light, making it difficult to popularize in clinical settings.

[0004] Therefore, existing technologies cannot meet the urgent clinical needs for cost control and ease of operation while ensuring high measurement accuracy, which seriously restricts the objectivity and scalability of skin disease area assessment. Summary of the Invention

[0005] This application proposes a scheme for assessing the area of ​​skin diseases on the human limbs, aiming to provide a new technical approach to overcome the fundamental errors such as geometric distortion and model mismatch in assessment methods based on two-dimensional images, while avoiding the obstacles to the clinical popularization of three-dimensional scanning technology due to its high cost and complex process.

[0006] The first aspect of this application provides a method for assessing the area of ​​skin diseases on the human limbs, including: Acquire multi-view images of the target limb, wherein adjacent views of the multi-view images contain overlapping regions; Based on the multi-view images, a three-dimensional point cloud of the target limb is generated; Based on the aforementioned 3D point cloud, a cylindrical model with a radius varying along the limb axis is constructed. The multi-view image is projected onto the cylindrical model, and the pixels projected onto the same position of the model are selected based on an evaluation function, so that the surface of the model is covered with image texture. The textured cylindrical model is unfolded into a two-dimensional planar image for assessing the area of ​​the skin lesion.

[0007] In some embodiments of this application, the acquisition of multi-view images of the target limb includes: Before acquiring images, pre-coded markers are placed on the target limb; The feature points used to generate the 3D point cloud include the center point of the coded marker.

[0008] In some embodiments of this application, the acquisition of multi-view images of the target limb includes: According to the preset surround sequence, images are acquired at 45-degree intervals centered on the target limb, for a total of eight viewpoints; the eight viewpoints are as follows: front view at 0°, outer front view at 45°, outer side view at 90°, outer rear view at 135°, rear view at 180°, inner rear view at 225°, inner side view at 270°, and inner front view at 315°.

[0009] In some embodiments of this application, constructing a cylindrical model with a radius varying along the limb axis based on the three-dimensional point cloud includes: Based on the three-dimensional point cloud, the functional relationship between the radius of the cylindrical model and the axis position z is determined by linear fitting: R(z) = a * z + b, where a and b are constants obtained by fitting.

[0010] In some embodiments of this application, the selection of pixels projected to the same location on the model based on an evaluation function includes: The Laplacian variance function is used as the evaluation function to calculate pixel sharpness, and the pixel with the highest sharpness is selected.

[0011] In some embodiments of this application, projecting the multi-view image onto the cylindrical model includes: The image is projected onto the model surface using the Mercator projection method for coordinate transformation.

[0012] In some embodiments of this application, unfolding the textured cylindrical model into a two-dimensional planar image includes: The cylindrical model is divided into multiple projection units along its axis and unfolded in segments. Linear interpolation is used to achieve smooth splicing between adjacent projection units.

[0013] In some embodiments of this application, the acquisition of multi-view images of the target limb is performed under illumination conditions of a D65 standard light source.

[0014] A second aspect of this application provides a device for assessing the area of ​​skin diseases on the limbs, comprising: An image acquisition module is used to acquire multi-view images of a target limb, wherein adjacent views of the multi-view images contain overlapping areas; A point cloud generation module is used to generate a three-dimensional point cloud of the target limb based on the multi-view images; The model building module is used to construct a cylindrical model with a radius that varies along the limb axis based on the three-dimensional point cloud. An image projection module is used to project the multi-view image onto the cylindrical model, and select pixels projected onto the same position of the model based on an evaluation function, so that the surface of the model is covered with image texture. A planar unfolding module is used to unfold the textured cylindrical model into a two-dimensional planar image for assessing the area of ​​the skin lesion.

[0015] A third aspect of this application provides a system for assessing the area of ​​skin diseases on the limbs, comprising: Image acquisition equipment used to acquire multi-view images of a target limb under illumination conditions of a D65 standard light source; A processing device, communicatively connected to the image acquisition device, the processing device including a memory and a processor; The memory stores a computer program, and the processor is configured to execute the computer program to perform the method as described in any one of claims 1-8; An output device, communicatively connected to the processing device, is used to output the two-dimensional planar image or the evaluation result obtained based on the image.

[0016] In summary, the methods, devices, and systems for assessing the area of ​​skin diseases on the limbs provided in the embodiments of this application, through multi-view image sequence acquisition and structure of motion recovery (SfM) based 3D reconstruction technology, acquire sparse 3D point cloud data of the limbs intrinsically, providing a realistic 3D geometric foundation for subsequent modeling, thereby overcoming the geometric distortion error caused by finite-view 2D image stitching at the source; and further, through dynamic cylindrical surface model segmented fitting technology based on 3D point cloud, the geometric model can adaptively fit the real limb structure with gradual changes in thickness, such as the forearm and lower leg, achieving high accuracy in assessing truncated cone-shaped limbs. The fidelity characterization fundamentally solves the model mismatch error caused by the mismatch between the constant-diameter cylindrical model and the real gradual limb structure. Furthermore, by projecting multi-view images onto this dynamic cylindrical surface and unfolding them into a two-dimensional plane using the Mercator projection method, a continuous, smooth, and minimally distorted mapping from the three-dimensional limb surface to the two-dimensional plane image is achieved through intrinsic means. Finally, through the synergistic effect of the above technical features, high-precision area measurement comparable to that of expensive professional three-dimensional scanning equipment is achieved without relying on such equipment. This provides a brand-new solution for the objective assessment of the severity of skin diseases that combines high precision and high clinical applicability. Attached Figure Description

[0017] The features and advantages of this application will become clearer with reference to the accompanying drawings, which are illustrative and should not be construed as limiting the application in any way. In the drawings: Figure 1 A flowchart illustrating a method for assessing the area of ​​skin diseases on the limbs provided in one embodiment of this application; Figure 2 This is a schematic diagram of the structure of a human limb skin disease area assessment system provided in one embodiment of this application; Figure 3 This is a flowchart illustrating a method for assessing the area of ​​skin diseases on the human limbs, based on some embodiments of this application. Figure 4 This is a schematic diagram of a device for assessing the area of ​​skin diseases on the limbs of a human body, according to some embodiments of this disclosure. Detailed Implementation

[0018] In the following detailed description, numerous specific details of this application are illustrated by example to provide a thorough understanding of the relevant disclosure. However, it will be apparent to those skilled in the art that this application can be practiced without these details. It should be understood that the terms “system,” “apparatus,” “unit,” and / or “module” used in this application are one way of distinguishing different parts, elements, sections, or components at different levels in a sequential arrangement. However, these terms may be replaced with other expressions if other expressions can achieve the same purpose.

[0019] It should be understood that when a device, unit, or module is referred to as being "on," "connected to," or "coupled to" another device, unit, or module, it may be directly connected to or coupled to, or communicate with, other devices, units, or modules, or there may be intermediate devices, units, or modules present, unless the context explicitly indicates otherwise. For example, the term "and / or" as used herein includes any one and all combinations of one or more of the relevant listed items.

[0020] The terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the scope of this application. As shown in the specification and claims of this application, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate that explicitly identified features, integrals, steps, operations, elements, and / or components are included, and such expressions do not constitute an exclusive list, and other features, integrals, steps, operations, elements, and / or components may also be included.

[0021] Referring to the following description and accompanying drawings, these and other features and characteristics, operating methods, functions of related structural elements, combinations of parts, and economics of manufacture of this application can be better understood, wherein the description and drawings form part of the specification. However, it is clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. It is understood that the drawings are not drawn to scale.

[0022] Various structural diagrams are used in this application to illustrate various variations of the embodiments according to this application. It should be understood that the preceding or following structures are not intended to limit this application. The scope of protection of this application is determined by the claims.

[0023] As described in the background section, existing technologies face a core contradiction when assessing the area of ​​skin lesions on the human limbs: two-dimensional image systems based on limited perspectives suffer from insufficient measurement accuracy due to geometric distortion and model mismatch, while high-precision three-dimensional scanning technology is difficult to popularize in clinical practice due to cost and ease of use barriers.

[0024] To address the aforementioned contradictions, this application provides a scheme for assessing the area of ​​skin diseases on the human limbs. It aims to achieve high-precision and high-fidelity assessment of skin lesion area through computational imaging methods based on ordinary imaging equipment, without the need for expensive three-dimensional scanning equipment.

[0025] The core process of the solution is as follows: Figure 1As shown: First, a 3D point cloud of the target limb is generated using a multi-view image sequence and structure of motion restoration (SfM) technology to obtain a realistic 3D geometric basis, overcoming the limitations of 2D images from the source; then, a dynamic cylindrical model with a radius varying along the limb axis is constructed based on this point cloud to accurately fit the gradual limb structure; finally, the image is mapped onto the model and unfolded into a 2D plane using Mercator projection, thereby obtaining a high-fidelity image with minimal geometric distortion for accurate area calculation.

[0026] The following section will, in conjunction with the accompanying drawings, first provide a detailed description of the specific hardware implementation of the scheme, and then elaborate on the control logic and complete process under each working mode.

[0027] Figure 2 This is a schematic diagram of the structure of a human limb skin disease area assessment system provided in one embodiment of this application. Figure 2 As shown, the system mainly includes an image acquisition device 100, a processing device 200, and an output device 300. The image acquisition device 100, the processing device 200, and the output device 300 are sequentially and communicatively connected.

[0028] Specifically, the image acquisition device 100 is used to acquire multi-view sequence images of the target limb under controlled lighting conditions (such as D65 standard light source) according to a preset acquisition protocol, and transmit the image data to the processing device 200.

[0029] The processing device 200 is used to receive the image data and, by executing its stored computer program, implement the following: Figure 1 The steps include: generating a three-dimensional point cloud based on the image, constructing a dynamic cylindrical surface model, projecting the image onto the model and performing pixel optimization, and unfolding the model into a two-dimensional planar image.

[0030] The output device 300 is used to receive and output the two-dimensional planar image or the evaluation result obtained based on the image from the processing device 200.

[0031] Figure 3 This is a flowchart illustrating the steps of a method for assessing the area of ​​skin diseases on the human limbs, according to some embodiments of this application. Figure 3 As shown, the method includes: S310, acquire multi-view images of the target limb, wherein adjacent views of the multi-view images contain overlapping areas.

[0032] S310 involves acquiring multi-view image sequences of the target limb. Its core is to ensure that the acquired images can completely cover the limb surface and meet the input requirements for subsequent three-dimensional reconstruction through a standardized acquisition scheme.

[0033] S310 aims to provide high-quality image sequences that fully cover the limb surface and meet the input requirements of motion reconstruction algorithms for subsequent 3D reconstruction, thereby avoiding geometric distortions caused by limited viewing angles. Its implementation first requires establishing a standardized acquisition environment to ensure color consistency and repeatability. Preferably, a photographic lightbox using a D65 standard light source is used to provide illumination, ensuring uniform lighting without shadows and avoiding ambient light interference. The target limb (upper limb excluding the hand, lower limb excluding the foot) is placed under this controlled lighting condition.

[0034] In a preferred embodiment, a Canon EOS 80D SLR camera is used as the image acquisition device and mounted on a tripod. To ensure consistent image scale, the center of the camera lens is kept horizontal with the center of the target limb, and the shooting distance is fixed at 50 cm. The camera parameters are set as follows: autofocus, image resolution of 6000×4000 pixels, and image files are stored in PNG format to retain more details.

[0035] During image acquisition, the operator holds a tripod and takes images sequentially around the target limb according to a pre-defined surround sequence. This sequence covers eight specific viewpoints: front view, back view, inner view, outer view, outer front view, inner front view, outer rear view, and inner rear view. To ensure image quality, 2-3 images are continuously acquired from each viewpoint, resulting in a total of 16-24 raw image sequences. To meet the feature matching requirements of the structure-of-motion reconstructive surgery (SORMS) algorithm, the overlap between images acquired from adjacent viewpoints must be guaranteed, with an overlap of no less than 40%. After acquisition, image files can be named according to viewpoint, such as image_front.png, image_back.png, etc., and are simultaneously uploaded to a local or cloud-based processing system.

[0036] To further improve the accuracy and robustness of subsequent feature point matching, coded markers can be pre-installed on the target limb before data acquisition. Specifically, circular or square QR code stickers with a diameter of 1 cm are affixed to key anatomical landmarks such as the ankle, knee, and groin. In a preferred embodiment, a 3 cm wide ring-shaped coded marker can be wrapped around and affixed to the ankle and groin. The center point of these coded markers will serve as the high-precision feature point used to generate the 3D point cloud.

[0037] S320, Based on the multi-view images, generate a three-dimensional point cloud of the target limb.

[0038] S320 aims to generate a 3D point cloud of a target limb based on the multi-view image sequence obtained by S310 using Structure of Motion (SfM) technology. This converts 2D image information into accurate 3D geometric data, providing a foundation for subsequent construction of a dynamic cylindrical model. The core of this method lies in recovering 3D geometric information from 2D images, effectively avoiding the inherent geometric distortion problem of stitching together 2D images from limited viewpoints.

[0039] In practice, the acquired image sequence is first preprocessed. A deep neural network model based on DeepLabV3 is used to segment the target limb from the image sequence, removing background interference. Then, feature points are extracted from the segmented image sequence; these feature points may include the center points of coded markers pre-attached to the skin, and / or natural skin texture feature points extracted using SIFT (Scale Invariant Feature Transform), SURF (Speed ​​Robust Feature Transform), or ORB (Oriented FAST and Rotated BRIEF) algorithms. Typically, approximately 5000 feature points are extracted per image on average.

[0040] Subsequently, a FLANN (Fast Approximate Nearest Neighbor) matcher is used for feature point matching, and the RANSAC (Random Sample Consensus) algorithm is employed to eliminate mismatched point pairs, ensuring the accuracy of the matched point pairs. Finally, based on the Structure of Motion Recovery (SfM) technique, the intrinsic and extrinsic parameters of the camera are calculated using the matched feature points to generate a sparse 3D point cloud of the target limb. This point cloud typically contains approximately 10,000 spatial points, covering the entire limb surface, with a preferred point cloud density of approximately 15 points per square centimeter.

[0041] This step successfully calculated camera parameters from the image sequence and generated a sparse 3D point cloud, providing a realistic 3D geometric basis for subsequent processing. The generated point cloud file can be saved in PLY format, and a camera parameter file is also output.

[0042] S330, Based on the three-dimensional point cloud, construct a cylindrical model whose radius varies along the limb axis.

[0043] S330 aims to construct a dynamic cylindrical model with a radius varying along the limb axis based on the 3D point cloud generated by S320, in order to accurately fit the anatomical structure of limbs with gradually varying thicknesses, such as the forearm and lower leg, and fundamentally solve the geometric mismatch error caused by using a cylindrical model with equal diameter.

[0044] In practice, firstly, principal component analysis (PCA) or a skeleton extraction algorithm is performed on the sparse 3D point cloud to extract the continuous central axis of the target limb. This axis is a spatial curve that reflects the natural curvature of the limb. Next, vertical cross-sections are made along the central axis at fixed intervals, preferably 10 cm, and the equivalent radius of curvature of the part where the cross-section intersects with the limb point cloud is calculated as the virtual cylindrical radius R(z) at that cross-section, where z is the coordinate position along the central axis.

[0045] Based on these discrete cross-sectional radii, the functional relationship between the radius of the cylindrical model and the axis position z is determined by linear fitting as: R(z) = a *z + b, where a and b are constants obtained from the fitting. For example, in a specific embodiment, at axis position z = 0 cm (near the ankle joint), the radius R(0) = 4.4 cm; at z = 10 cm, the radius R(10) = 6.2 cm; at z = 20 cm, the radius R(20) = 9.4 cm; and at z = 30 cm (near the knee joint), the radius R(30) = 8.6 cm.

[0046] Through the above process, a virtual cylindrical surface model with a radius that dynamically changes with the axis position z is constructed. This dynamic model can adaptively fit the gradually changing limb structure, achieving a continuous and smooth transition from the thin end to the thick end, providing a high-fidelity geometric basis for subsequent image projection and unfolding.

[0047] S340, the multi-view image is projected onto the cylindrical model, and the pixels projected onto the same position of the model are selected based on an evaluation function, so that the surface of the model is covered with image texture.

[0048] The S340 aims to project multi-view images acquired by the S310 onto the surface of a dynamic cylindrical model constructed by the S330. It then selects and fuses multiple pixels projected to the same location on the model based on an evaluation function, resulting in the clearest image texture adhering to the model surface, laying the foundation for subsequent high-fidelity 2D unfolding. Its core function is to solve the problems of texture overlap and inconsistent clarity caused by multiple image projections.

[0049] In specific implementation, the central axis of the cylindrical model generated by S330 is first divided into multiple continuous projection units for segmentation. Preferably, the length L of each projection unit is fixed at 10 cm. For each projection unit, the Mercator projection method is used for coordinate transformation. Specifically, for any point P(x1,y1) within the unit, its azimuth angle θ and axial position z (relative to the unit's starting point) in three-dimensional space are calculated. Its coordinates (x2,y2) projected onto the two-dimensional plane are calculated as: x2 = R(z)*θ, y2 = z. Wherein, R(z) is the radius of the section where the point is located, which can be obtained by linear interpolation from the starting radius and ending radius of the projection unit. In a specific calculation example, for unit 1 (z value is 0-10 cm), R(z) can be calculated as: R(z) = 4.4 + 0.18z (z ranges from 0 to 10).

[0050] Subsequently, for multiple pixels projected onto the same location on the cylindrical surface from images at different viewpoints, a sharpness evaluation function is used for optimal selection. Preferably, the Laplacian variance function is used as the evaluation function to calculate the sharpness value (i.e., Laplacian variance) of each pixel. In specific calculations, for a pixel block, its Laplacian variance can be obtained by calculating the variance of the convolution result of the Laplacian operator in that region; a larger variance value indicates higher local image sharpness. The system will automatically compare the Laplacian variance values ​​of all candidate pixels at the same location. For example, in a certain overlapping region, the Laplacian variance values ​​of three candidate pixels may be calculated as 1500, 3200, and 850 respectively. The system will then select the pixel with a variance value of 3200, i.e., the sharpest pixel, as the final texture value for that model surface location.

[0051] By traversing the entire model surface through the above process, a 3D cylindrical model with optimal image texture is finally generated. This step ensures that the final texture image used for unfolding is formed by fusing the clearest parts from each viewpoint, fundamentally improving the quality of the 2D unfolded image.

[0052] S350, the textured cylindrical model is unfolded into a two-dimensional planar image for assessing the area of ​​the skin lesion.

[0053] The S350 is designed to unfold the 3D cylindrical model generated by the S340, which already has high-resolution textures, into a complete 2D planar image. This 2D image has minimal geometric distortion and can realistically reflect the morphology and area of ​​skin lesions on the limb surface, providing direct and accurate input for subsequent automatic area calculation and disease assessment.

[0054] In practice, the textured cylindrical model generated by S340 is unfolded along its axial direction. Preferably, a segmented unfolding strategy is adopted, that is, the cylindrical model is divided into multiple continuous projection units along its central axis. A generatrix parallel to the central axis of the cylindrical model (e.g., the centerline of the outer view as the cutting line) is defined as the unfolding reference. Through the conversion between cylindrical coordinates and Cartesian coordinates, the three-dimensional surface texture map of each projection unit is unfolded into a two-dimensional rectangular image. Specifically, the width of the two-dimensional rectangular image is equal to the perimeter of its cylindrical segment (calculated by the formula 2πR(z)), and the height is equal to the length of the projection unit (e.g., 10 cm).

[0055] After each projection unit is independently unfolded, the two-dimensional unfolded images of all units are stitched together along the axis (Z direction). Since the radius change of the dynamic cylindrical model constructed in step S330 is continuous (defined by the linear function R(z)), a linear interpolation algorithm can be used to achieve a smooth transition between adjacent units during stitching, thereby ensuring that the final generated overall two-dimensional image is seamless and smooth, avoiding obvious seams or distortions.

[0056] To meet the accuracy requirements for skin lesion area assessment, the resolution of the unfolded image needs to be controlled. By controlling the acquisition resolution of the original image (e.g., 6000x4000 pixels) and the projection parameters, it can be ensured that the spatial resolution of the final unfolded two-dimensional planar image at any part of the limb is not lower than a preset minimum threshold. Preferably, this minimum threshold is set to 60 pixels / cm. This resolution is sufficient to support subsequent automatic lesion identification and pixel-level area calculation. In a specific embodiment, for a limb approximately 80 cm long and with an average circumference of approximately 30 cm, the final unfolded two-dimensional image resolution can reach 4800x1800 pixels.

[0057] Finally, the stitched and optimized two-dimensional planar image is output to a display device (such as a medical monitor) or a storage device (such as a hospital PACS system). The image format can be lossless compression formats such as PNG. This image can be directly used in subsequent computer vision-based automatic lesion segmentation and area calculation modules to ultimately achieve high-precision skin disease area assessment.

[0058] Furthermore, the system can also be configured to directly output assessment results. In this mode, the processing device can further integrate or invoke an automatic lesion identification and area calculation module. This module processes the two-dimensional planar image, automatically identifies and segments the lesion area, calculates assessment results such as lesion area value and severity grade, and outputs them to a display device or hospital information management system.

[0059] One embodiment of this application records the clinical testing process, specific parameters, and quantitative results of the proposed solution in the Department of Dermatology, Beijing Hospital of Traditional Chinese Medicine, Capital Medical University, as follows: I. Test Environment, System Configuration and Objects The tests were conducted in the clinical environment of the Department of Dermatology, Beijing Hospital of Traditional Chinese Medicine, affiliated with Capital Medical University. The testing system includes an image acquisition subsystem, an image processing subsystem (including an image processing unit, a projection and unfolding unit), an output unit, and a storage unit.

[0060] Image acquisition subsystem: Image acquisition module: A Canon EOS 80D SLR camera is used as the image acquisition device.

[0061] Lighting module: A ring-shaped LED photographic light box using a standard D65 light source, with a color temperature of 6500K and illuminance uniformity ≥90% to eliminate shadows and reflections. The inner wall of the light box is coated with a diffuse reflection coating to ensure uniform light distribution.

[0062] Markings: Circular coded markers, 3 cm wide, are wrapped and affixed to the patient's ankles and groin. The markers are pre-coded with gradient color blocks.

[0063] Image processing subsystem: Hardware platform: The processing device is a computer workstation equipped with an Intel Core i7 processor and an NVIDIA RTX 3060 graphics card.

[0064] Storage: DDR4-based RAM and a Samsung 1TB NVMe SSD are used as storage units.

[0065] Test subjects: 10 psoriasis patients from the Department of Dermatology, Beijing Hospital of Traditional Chinese Medicine, Capital Medical University.

[0066] II. Testing Methods and Specific Implementation Process Assess one side of the patient's limb (upper limb excluding the hand, lower limb excluding the foot). The following details the steps using a patient's left lower limb as an example: Image acquisition: The patient stands on one leg with the other leg raised and placed on a support or chair in front of them. The operator takes pictures of the target lower limb from eight preset viewpoints (i.e., front view (0°), lateral anterior view (45°), lateral view (90°), lateral posterolateral view (135°), posterior view (180°), medial posterolateral view (225°), medial view (270°), and medial anterior view (315°). Three images are continuously acquired from each viewpoint, each with a resolution of 6000×4000 pixels. The overlap area between adjacent images is ensured to be ≥40%, resulting in a total of 24 sequential images. The images are stored in PNG format and the image data is transferred to the storage unit and image processing unit via USB data cable.

[0067] Image processing and 3D reconstruction: Using a deep neural network model and algorithm based on DeepLabV3, the target limb is segmented from the sequence of images.

[0068] Subsequently, the ORB feature detection algorithm was implemented using the OpenCV 4.5 library; the center point of the pre-pasted marker color block was also detected; approximately 5000 feature points were extracted from each image; the FLANN matcher was used for feature matching, and the RANSAC algorithm was used to remove false matches.

[0069] Feature point matching is performed using the FLANN matcher, and mismatched point pairs are removed using the RANSAC algorithm.

[0070] The 3D reconstruction module is based on motion recovery structure technology to calculate the camera's intrinsic and extrinsic parameters; it generates a sparse 3D point cloud containing approximately 10,000 spatial points; the point cloud density is an average of 15 points per square centimeter.

[0071] Dynamic modeling, projection, and unfolding: First, principal component analysis was performed on the point cloud to extract the central axis of the limb. Vertical sections were made along the axis at 10cm intervals (a total of 8 sections) to form multiple continuous projection units (a total of 8). Calculate the equivalent radius of each cross section: First cross-section (near the ankle joint): R = 4.4 cm Second cross-section (mid-calf): R = 6.2cm Third section (mid-calf): R = 9.4cm Fourth section (near the knee joint): R = 8.6cm Construct the radius function: First projection unit (z represents the axial position, z value is 0-10 cm): R(z) = 4.4 + 0.18*z (z ranges from 0 to 10); Subsequent projection units follow the same pattern; Generate a dynamic cylindrical mesh model with a mesh resolution of 1 mm.

[0072] Next, Mercator projection is performed: The central axis is divided into 8 projection units, each unit being 10cm in length.

[0073] Perform Mercator projection on each element: Coordinate transformation: x2 = R(z)×θ, y2 = z where R(z) is obtained by linear interpolation of the element's initial and final radii; The Laplacian variance sharpness evaluation function is used to select the best pixels in the overlapping area; Generate texture maps for each unit and store them in 32-bit floating-point format; Then unfold in two dimensions: Using the centerline of the outer view as the cutting line, perform a transformation from cylindrical coordinates to Cartesian coordinates; The projection unit splicing uses bilinear interpolation for smooth transition.

[0074] Final output: Display interface: HDMI 2.0, supports 4K resolution output; Storage interface: PCI-E 3.0; Software interface: Provides APIs for third-party lesion segmentation or disease identification algorithms; III. Test Results and Comparative Analysis To quantify the assessment, the results obtained by this method were compared with the results of manual, refined measurement of patient skin lesions (as a baseline reference), and also with traditional manual assessment methods based on two-dimensional images. The test results for 10 patients are summarized below: Measurement accuracy: The average error of the lesion area measured by the method of this application compared with the benchmark reference value is 5%. Under the same test conditions, the average error of the traditional manual assessment method is 24%.

[0075] Measurement efficiency: The average time to complete a single measurement (from image acquisition to result output) using the method of this application is approximately 3 minutes. The average time to complete a single measurement using traditional manual evaluation methods is approximately 8 minutes.

[0076] IV. Conclusion Test results show that, compared with traditional methods, this approach can significantly improve measurement accuracy (average error reduced from 24% to 5%) and greatly improve assessment efficiency (single measurement time reduced from 8 minutes to 3 minutes) in assessing the area of ​​skin lesions on the limbs, demonstrating its outstanding clinical application value and technical advantages.

[0077] Figure 4This is a schematic diagram of a device for assessing the area of ​​skin diseases on the human limbs, according to some embodiments of this disclosure. Figure 4 As shown, the human limb skin disease area assessment 400 includes an image acquisition module 410, a point cloud generation module 420, a model construction module 430, an image projection module 440, and a plane unfolding module 450. Wherein: Image acquisition module 410 is used to acquire multi-view images of the target limb, wherein adjacent views of the multi-view images contain overlapping areas; The point cloud generation module 420 is used to generate a three-dimensional point cloud of the target limb based on the multi-view images. The model building module 430 is used to build a cylindrical model with a radius that varies along the limb axis based on the three-dimensional point cloud. The image projection module 440 is used to project the multi-view image onto the cylindrical model, and select the pixels projected onto the same position of the model based on an evaluation function, so that the surface of the model is covered with image texture. Planar unfolding module 450 is used to unfold the textured cylindrical model into a two-dimensional planar image for assessing the area of ​​the skin lesion.

[0078] In summary, the methods, devices, and systems for assessing the area of ​​skin diseases on the limbs provided in the embodiments of this application, through multi-view image sequence acquisition and structure of motion recovery (SfM) based 3D reconstruction technology, acquire sparse 3D point cloud data of the limbs intrinsically, providing a realistic 3D geometric foundation for subsequent modeling, thereby overcoming the geometric distortion error caused by finite-view 2D image stitching at the source; and further, through dynamic cylindrical surface model segmented fitting technology based on 3D point cloud, the geometric model can adaptively fit the real limb structure with gradual changes in thickness, such as the forearm and lower leg, achieving high accuracy in assessing truncated cone-shaped limbs. The fidelity characterization fundamentally solves the model mismatch error caused by the mismatch between the constant-diameter cylindrical model and the real gradual limb structure. Furthermore, by projecting multi-view images onto this dynamic cylindrical surface and unfolding them into a two-dimensional plane using the Mercator projection method, a continuous, smooth, and minimally distorted mapping from the three-dimensional limb surface to the two-dimensional plane image is achieved through intrinsic means. Finally, through the synergistic effect of the above technical features, high-precision area measurement comparable to that of expensive professional three-dimensional scanning equipment is achieved without relying on such equipment. This provides a brand-new solution for the objective assessment of the severity of skin diseases that combines high precision and high clinical applicability.

[0079] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding descriptions in the foregoing device embodiments, and will not be repeated here.

[0080] Although the subject matter described herein is provided in the general context of execution on a computer system in conjunction with an operating system and applications, those skilled in the art will recognize that other implementations can also be executed in conjunction with other types of program modules. Generally, program modules include routines, programs, components, data structures, and other types of structures that perform specific tasks or implement specific abstract data types. Those skilled in the art will understand that the subject matter described herein can be practiced using other computer system configurations, including handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, minicomputers, mainframes, etc., and can also be used in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules may reside on both local and remote memory storage devices.

[0081] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0082] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of this application and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of this application should be included within the protection scope of this application. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. A method for assessing the area of ​​skin diseases on the human limbs, characterized in that, include: Acquire multi-view images of the target limb, wherein adjacent views of the multi-view images contain overlapping regions; Based on the multi-view images, a three-dimensional point cloud of the target limb is generated; Based on the aforementioned 3D point cloud, a cylindrical model with a radius varying along the limb axis is constructed. The multi-view image is projected onto the cylindrical model, and the pixels projected onto the same position of the model are selected based on an evaluation function, so that the surface of the model is covered with image texture. The textured cylindrical model is unfolded into a two-dimensional planar image for assessing the area of ​​the skin lesion.

2. The method according to claim 1, characterized in that, The multi-view images of the target limb acquired include: Before acquiring images, pre-coded markers are placed on the target limb; The feature points used to generate the 3D point cloud include the center point of the coded marker.

3. The method according to claim 1, characterized in that, The multi-view images of the target limb acquired include: According to the preset surround sequence, images are captured at 45-degree intervals centered on the target limb, for a total of eight viewpoints; the eight viewpoints are as follows: front view at 0°, outer front view at 45°, outer side view at 90°, outer rear view at 135°, rear view at 180°, inner rear view at 225°, inner side view at 270°, and inner front view at 315°.

4. The method according to claim 1, characterized in that, The construction of a cylindrical model with a radius varying along the limb axis based on the three-dimensional point cloud includes: Based on the three-dimensional point cloud, the functional relationship between the radius of the cylindrical model and the axis position z is determined by linear fitting: R(z) = a * z + b, where a and b are constants obtained by fitting.

5. The method according to claim 1, characterized in that, The selection of pixels projected to the same position on the model based on an evaluation function includes: The Laplacian variance function is used as the evaluation function to calculate pixel sharpness, and the pixel with the highest sharpness is selected.

6. The method according to claim 1, characterized in that, The step of projecting the multi-view image onto the cylindrical model includes: The image is projected onto the model surface using the Mercator projection method for coordinate transformation.

7. The method according to claim 1, characterized in that, Unfolding the textured cylindrical model into a two-dimensional planar image includes: The cylindrical model is divided into multiple projection units along its axis and unfolded in segments. Linear interpolation is used to achieve smooth splicing between adjacent projection units.

8. The method according to claim 1, characterized in that: The acquisition of multi-view images of the target limb was carried out under illumination conditions of a D65 standard light source.

9. A device for assessing the area of ​​skin diseases on the human limbs, characterized in that, include: An image acquisition module is used to acquire multi-view images of a target limb, wherein adjacent views of the multi-view images contain overlapping areas; A point cloud generation module is used to generate a three-dimensional point cloud of the target limb based on the multi-view images; The model building module is used to construct a cylindrical model with a radius that varies along the limb axis based on the three-dimensional point cloud. An image projection module is used to project the multi-view image onto the cylindrical model, and select pixels projected onto the same position of the model based on an evaluation function, so that the surface of the model is covered with image texture. A planar unfolding module is used to unfold the textured cylindrical model into a two-dimensional planar image for assessing the area of ​​the skin lesion.

10. A system for assessing the area of ​​skin diseases on the human limbs, characterized in that, include: Image acquisition equipment used to acquire multi-view images of a target limb under illumination conditions of a D65 standard light source; A processing device, communicatively connected to the image acquisition device, the processing device including a memory and a processor; The memory stores a computer program, and the processor is configured to execute the computer program to perform the method as described in any one of claims 1-8; An output device, communicatively connected to the processing device, is used to output the two-dimensional planar image or the evaluation result obtained based on the image.

Images