Personalized medical aesthetic scheme recommendation system based on medical aesthetic patient portrait construction

Abstract

The present application relates to the technical field of medical and cosmetic image analysis, and in particular to a personalized medical and cosmetic scheme recommendation system based on medical and cosmetic patient profiling. The present application first acquires a monocular video stream, maps a preset anatomical grid template to three-dimensional facial data of a subject, synchronously extracts a traction vector sequence and a vertical traction sequence of a malaris muscle, a slip vector sequence and a vertical following sequence of a fat pad, and a wrinkle depth curvature sequence; further extracts a ligament viscoelastic hysteresis area, a tissue relative slip angle ratio, and an apparent wrinkle depth eigenvalue; and finally performs hierarchical output of a facial dynamic deformation state label. The present application extracts and decouples three quantitative indicators of ligament hysteresis, fat slip, and dermal wrinkle from a dynamic video, and performs hierarchical determination based on the three indicators, and finally outputs a facial dynamic deformation state label, thereby automatically and accurately distinguishing the causes of facial dynamic mechanical characteristics.

Description

Technical Field

[0001] This invention relates to the field of medical aesthetic image analysis technology, specifically to a personalized medical aesthetic solution recommendation system based on the patient profile. Background Technology

[0002] The mechanisms underlying facial surface texture and deformation are complex, influenced by various physical factors such as the sluggishness of internal soft tissue movement, relative slippage of local structures, and substantial geometric depressions. Existing facial detection technologies primarily rely on static image analysis, determining surface texture depth by detecting grayscale gradients in shadowed areas. However, soft tissues in different physical states exhibit significant mechanical differences during movement. Tight tissues show strong deformation response, while aging tissues exhibit significant mechanical sluggishness. Simple static images cannot capture the dynamic biomechanical properties of tissues during reciprocating motion, making it difficult to distinguish between pseudo-creases formed by gravitational compression and substantial geometric depressions.

[0003] Furthermore, since static visual features cannot decouple the different physical movements of tissues in three-dimensional space, relying solely on image grayscale to make a single morphological judgment on facial regions with severe motion lag or structural slippage will ignore their inherent differences in mechanical deformation, leading to serious biases in the feature parameters output by the data calculation model. Currently, there is a lack of technical means to effectively extract the dynamic mechanical behavior of facial soft tissues based on video streams and accurately quantify the physical aging state of facial soft tissues accordingly. Summary of the Invention

[0004] To address the problem that existing technologies, relying solely on static image grayscale features, cannot effectively distinguish differences in facial visual deformation caused by different mechanical mechanisms, leading to inaccurate classification and identification of the physical aging state of facial soft tissues, the present invention aims to provide a personalized medical aesthetic solution recommendation system based on the patient profile. The specific technical solution adopted is as follows: A personalized medical aesthetic treatment plan recommendation system based on the patient profile of a medical aesthetic patient, the system comprising: The preparation module is used to acquire a monocular video stream of the subject's facial movements, map a preset anatomical mesh template onto the subject's three-dimensional facial data, and establish a head coordinate system based on the first frame image. In the head coordinate system, each frame image is compared with the first frame image, and the traction vector sequence and vertical traction sequence of the zygomaticus major muscle, the sliding vector sequence and vertical following sequence of the fat pad, and the crease depth curvature sequence are extracted simultaneously. The analysis module is used to obtain the area of ​​the closed trajectory formed by the vertical traction sequence of the zygomaticus major muscle and the vertical following sequence of the fat pad in a two-dimensional phase plane, which is used as the viscoelastic hysteresis area of ​​the ligament; at the peak time of the vertical traction sequence of the zygomaticus major muscle, the relative sliding angle ratio of the tissue is obtained based on the similarity characteristics between the traction vector of the zygomaticus major muscle and the sliding vector of the fat pad; the maximum increment of the crease depth curvature sequence is extracted, and combined with the viscoelastic hysteresis area of ​​the ligament, the apparent crease depth feature value is obtained. The output module is used to output facial dynamic deformation status labels in a graded manner based on the ligament viscoelastic hysteresis area, the ratio of the relative sliding angle of the tissue, and the apparent crease depth feature value.

[0005] Furthermore, the anatomical mesh template includes a pre-labeled set of key anatomical indexes, which includes: an index of the zygomaticus major muscle traction anchor point corresponding to the mid-section of the zygomaticus major muscle belly; an index of the center of gravity of the fat pad corresponding to the geometric center of the superficial fat pad on the outer side of the nasolabial fold; and a set of nasolabial fold crease mesh area indexes covering the bottom of the nasolabial fold and the skin on both sides.

[0006] Furthermore, the method for obtaining the crease depth curvature sequence includes: Iterate through all the grid vertices corresponding to the index set of the nasolabial fold grid region; For each frame of the image, the arithmetic mean of the average principal curvature of all vertices in the nasolabial fold region corresponding to the mesh vertex of the nasolabial fold is calculated as the value of the fold depth curvature sequence in the corresponding frame.

[0007] Furthermore, the method for obtaining the viscoelastic hysteresis area of ​​the ligament includes: A two-dimensional phase plane is constructed with the vertical traction sequence of the zygomaticus major muscle as the abscissa and the vertical following sequence of the fat pad as the ordinate. Connect the data points throughout the entire motion cycle in time sequence to form a closed trajectory loop; The geometric area enclosed by the trajectory loop is calculated using a discrete form of the shoelace formula, and is taken as the viscoelastic hysteresis area of ​​the ligament.

[0008] Furthermore, the method for obtaining the apparent crease depth feature value includes: Extract the maximum increment of each element in the crease depth curvature sequence relative to the first element; use the ligament viscoelastic hysteresis area as an inhibition factor to perform a logarithmic inverse weighted correction on the maximum increment, and calculate the apparent crease depth feature value.

[0009] Furthermore, before calculating the relative slip angle ratio of the tissue, the displacement validity is verified, and the verification method includes: At the peak moment of the vertical traction sequence of the zygomaticus major muscle, the magnitude of the fat pad slip vector is obtained; If the modulus is less than the preset micro-motion noise threshold, it is determined that there is no effective displacement, and the relative sliding angle ratio of the tissue is directly set to zero.

[0010] Furthermore, the method for obtaining the ratio of the tissue relative to the slip angle includes: At the peak moment of the vertical traction sequence of the zygomaticus major muscle, the cosine similarity between the traction vector of the zygomaticus major muscle and the slip vector of the fat pad is calculated; the cosine similarity is then negatively correlated and normalized as the ratio of the relative slip angle of the tissue.

[0011] Furthermore, the method for obtaining the facial dynamic deformation state tags includes: When the viscoelastic hysteresis area of ​​the ligament is greater than the preset relaxation determination threshold, the first-level status label Type-Region-A is output. When the viscoelastic hysteresis area of ​​the ligament is less than or equal to the relaxation determination threshold, if the relative sliding angle ratio of the tissue is greater than the preset displacement determination threshold, then the secondary status label Type-Region-B is output. When the viscoelastic hysteresis area of ​​the ligament is less than or equal to the relaxation determination threshold, and the ratio of the relative sliding angle of the tissue is less than or equal to the displacement determination threshold, if the apparent crease depth feature value is greater than the preset fracture determination threshold, then a three-level status label Type-Region-C is output.

[0012] Furthermore, the method for obtaining the head coordinate system includes: The Euclidean distance between the coordinates of the outer corners of the subject's eyes was used as the anatomical scale normalization factor. Based on the anatomical scale normalization factor, a three-dimensional Cartesian coordinate system is established with the direction of the line connecting the eyes as the X-axis, the vector perpendicular to the line connecting the eyes and located in the plane of facial symmetry as the Y-axis, and the facial normal vector as the Z-axis.

[0013] Furthermore, using a non-rigid registration algorithm based on 3DMM, the preset anatomical mesh template is mapped onto the three-dimensional facial data of the subject reconstructed from the monocular video stream, in order to lock the zygomaticus major muscle traction anchor point, fat pad centroid, and nasolabial fold mesh area that actually correspond to the subject's face.

[0014] The present invention has the following beneficial effects: This invention first acquires a monocular video stream of the subject's facial movements, maps a pre-set anatomical mesh template onto the subject's three-dimensional facial data, and simultaneously extracts the traction vector sequence and vertical traction sequence of the zygomaticus major muscle, the slip vector sequence and vertical following sequence of the fat pad, and the crease depth curvature sequence. The traction and slip vectors represent the anatomical direction of the muscle, the vertical component represents the tissue's ability to resist sagging, and the crease depth curvature represents the degree of physical folding of the skin surface, providing a foundation for subsequent analysis. Further, the viscoelastic hysteresis area of ​​ligaments is extracted to characterize the displacement hysteresis of soft tissue in the loading and unloading path, the relative sliding angle ratio of the tissue is obtained, the spatial offset of passive tissue relative to the traction direction of the active muscle is quantified, and the apparent crease depth feature value is obtained. Independent decoupling and dynamic feature quantification analysis of the supporting ligament relaxation mechanical properties, the spatial displacement of the fat pad structure, and the physical depression depth of the dermis are performed. Finally, a graded output of facial dynamic deformation state labels is performed. This invention extracts and decouples three quantitative indicators—ligament hysteresis, fat slippage, and dermal folds—from dynamic videos, and performs grading based on these three indicators, ultimately outputting a label for the dynamic deformation state of the face, thus enabling automated and precise differentiation of the causes of facial dynamic mechanical features. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system block diagram of a personalized medical aesthetic solution recommendation system based on a medical aesthetic patient profile, provided in one embodiment of the present invention. Figure 2 This is a flowchart illustrating a method for obtaining the relative slip angle ratio of a tissue, as provided in an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a personalized medical aesthetic solution recommendation system based on a patient profile constructed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific implementation of a personalized medical aesthetics solution recommendation system based on a patient profile.

[0020] Please see Figure 1 The diagram illustrates a system block diagram of a personalized medical aesthetic solution recommendation system based on a patient profile provided by an embodiment of the present invention. The system includes: a preparation module 101, an analysis module 102, and an output module 103.

[0021] The preparation module 101 is used to acquire a monocular video stream of the subject's facial movements, map a preset anatomical mesh template onto the subject's three-dimensional facial data, and establish a head coordinate system based on the first frame image. In the head coordinate system, each frame image is compared with the first frame image, and the traction vector sequence and vertical traction sequence of the zygomaticus major muscle, the sliding vector sequence and vertical following sequence of the fat pad, and the crease depth curvature sequence are extracted simultaneously.

[0022] In one embodiment of the present invention, a monocular video stream of the subject's facial movements is first acquired. Since there are significant differences in the facial geometry of the subjects, and head posture changes are inevitable during video capture, directly using pixel coordinates would lead to a lack of comparability in subsequent biomechanical calculations. Therefore, a pre-set anatomical mesh template is mapped onto the subject's three-dimensional facial data, and then a head coordinate system is established based on the first frame image, creating a unified measurement reference system independent of individual head posture and absolute size.

[0023] Since facial expressions originate from the contraction and traction of deep muscles, quantifying muscle contraction behavior is a prerequisite for analyzing the conduction characteristics of soft tissue. Therefore, in the head coordinate system, by comparing each frame with the first frame, the traction vector sequence and vertical traction sequence of the zygomaticus major muscle, the slip vector sequence and vertical following sequence of the fat pad, and the crease depth curvature sequence are extracted simultaneously. The traction vector and slip vector are used to represent the anatomical direction of the muscle, the vertical component is used to represent the tissue's ability to resist sagging, and the crease depth curvature is used to represent the degree of physical folding of the skin surface, providing a basis for subsequent analysis.

[0024] Preferably, in one embodiment of the present invention, since the mechanical response of facial soft tissue is closely related to specific anatomical structures (such as muscle attachment points, deep fat pads and skin folds), a unified and standardized anatomical reference benchmark must be established in order to achieve accurate quantitative observation of dynamic behavior.

[0025] As an example, the anatomical mesh template is a 3D mesh model containing facial anatomical semantics, and its topology is pre-labeled with a set of key anatomical indices, including: the index of the zygomaticus major muscle traction anchor point corresponding to the mid-section of the zygomaticus major muscle belly, and the observation point as the active force source; The centroid index of the fat pad corresponding to the geometric center of the superficial fat pad on the outer side of the nasolabial fold is used as the observation point for the passively responding tissue. The set of indexes for the nasolabial fold grid region covering the bottom of the nasolabial fold and the skin on both sides is used as the observation area for surface morphology.

[0026] In one embodiment of the present invention, a monocular video stream of the entire process of a subject's natural smiling motion is acquired. The smiling motion includes the complete process from a naturally relaxed state to a maximum amplitude state and back to a naturally relaxed state. The subject is guided to lean against a vertical reference plane to limit torso movement. The lens height is adjusted to be level with the subject's eyes, and the subject is required to maintain a level gaze to reduce asymmetrical projection errors caused by head tilting during filming. The acquisition frequency is no less than 30 frames per second, and the video stream is a high-definition video with a resolution of no less than 1080P. The starting moment of the motion is recorded as [the moment of the motion's initiation]. The end frame of the motion restoration is recorded as .

[0027] Then, existing 3D face reconstruction techniques, such as using a non-rigid registration algorithm based on 3DMM, are employed to map a pre-set anatomical mesh template onto the 3D facial data of the subject reconstructed from a monocular video stream, in order to lock the actual corresponding zygomaticus major muscle traction anchor point, fat pad centroid, and nasolabial fold mesh area of ​​the subject's face.

[0028] Furthermore, a head coordinate system is established because interpupillary distance is a relatively stable, easily measurable anatomical parameter of the human head that is roughly proportional to the soft tissue area of ​​the face, making it suitable as a scale benchmark. Therefore, the coordinates of the outer corners of the subject's eyes are identified, and their Euclidean distance is calculated as an anatomical scale normalization factor. ; Based on an anatomical scale normalization factor, a three-dimensional Cartesian coordinate system is established with the line connecting the eyes as the X-axis, the vector perpendicular to the line connecting the eyes and located in the plane of facial symmetry as the Y-axis, and the facial normal vector as the Z-axis. The system divides the original spatial coordinate values ​​of all vertices in the mesh by... The coordinate values ​​were obtained after scale normalization, eliminating the influence of shooting distance and individual head size on displacement measurement.

[0029] Among them, the key anatomical index set is a fixed set of vertices defined according to the preset facial soft tissue anatomical features. It is pre-set based on the existing anatomical knowledge base. The non-rigid registration algorithm based on 3DMM (3D Morphable Models) is an existing computer vision technology. By iteratively optimizing the shape and texture parameters of the standard face model, it can achieve accurate reconstruction and topological correspondence of the three-dimensional geometric structure of the face in the input image or video frame.

[0030] In other embodiments of the present invention, the implementer may adjust the video stream acquisition settings according to needs and scenario limitations.

[0031] Preferably, in one embodiment of the present invention, the system on the discrete time axis Above, track the zygomaticus major muscle traction anchor point in the normalized coordinate system. The system calculates the spatial location of the point in each frame. Relative to the initial time The three-dimensional displacement vector is defined as the zygomaticus major muscle traction vector sequence. : With the help of Accurately characterize the true three-dimensional spatial contraction direction of the active muscle group, providing a spatial reference for subsequent calculation of fat pad slippage deviation; Simultaneously, the system extracts the projection component of this vector along the Y-axis, defining it as the zygomaticus major muscle vertical traction sequence. : With the help of By stripping away deformation interference in the non-vertical direction, the vertical stretching amplitude of soft tissue against gravity is quantified and used as the reference actuation input for the subsequent construction of the actuation-response two-dimensional phase plane.

[0032] in, It is the unit vector along the Y-axis.

[0033] Since the nasolabial fat pad is a passively moving part driven by the zygomaticus major muscle, the degree of lag and directional deviation of its movement trajectory relative to muscle traction directly reflects the anchoring ability of the supporting ligaments and the integrity of the tissue structure. Therefore, the system, on the same time axis... Above, track the center of gravity of the nasolabial fat pad in the normalized coordinate system. The spatial location of the point is calculated in each frame. Relative to the initial time The three-dimensional displacement vector is defined as the fat pad slip vector sequence. : With the help of Characterizes the actual movement trajectory of the fat pad in three-dimensional space; Simultaneously, the system extracts the projection component of this vector along the Y-axis, defining it as the fat pad vertical following sequence. : With the help of The displacement response of the passive tissue in the anti-gravity direction is quantified and used as the response input for constructing the actuation-response two-dimensional phase plane.

[0034] Since dermal rupture and tissue laxity can both cause visual creases in the nasolabial fold area, the physical depth of these creases is difficult to quantify with the naked eye or two-dimensional images alone. The geometric curvature of a mesh surface can objectively characterize the degree of physical folding on the skin surface.

[0035] Therefore, the system iterates through all the grid vertices corresponding to the nasolabial fold grid region index set; For each frame of the image, the system uses discrete differential geometry methods (such as the Cotangent weighting method) to calculate the mean principal curvature of each vertex within the region. Then, it calculates the arithmetic mean of the mean principal curvatures of all vertices within the corresponding region, which is used as the value of the crease depth curvature sequence in the corresponding frame, thus obtaining the crease depth curvature sequence. The larger the value of this sequence, the deeper the depressions and folds on the skin surface.

[0036] It should be noted that the Cotangent weighting method for calculating the average principal curvature of mesh vertices is a mature existing technology in the fields of computer graphics and 3D geometry processing. In other embodiments of the present invention, implementers may use existing methods such as discrete Gaussian curvature estimation and average curvature flow calculation.

[0037] Analysis module 102 is used to obtain the area of ​​the closed trajectory formed by the vertical traction sequence of the zygomaticus major muscle and the vertical following sequence of the fat pad in the two-dimensional phase plane, which is used as the viscoelastic hysteresis area of ​​the ligament; at the peak moment of the vertical traction sequence of the zygomaticus major muscle, the relative sliding angle ratio of the tissue is obtained based on the similarity characteristics between the traction vector of the zygomaticus major muscle and the sliding vector of the fat pad; the maximum increment of the crease depth curvature sequence is extracted, and combined with the viscoelastic hysteresis area of ​​the ligament, the apparent crease depth feature value is obtained.

[0038] The essence of facial aging is the degradation of the biomechanical properties of soft tissue. Simple static visual depth cannot distinguish between pseudo-gravitational compression caused by ligament laxity, structural accumulation caused by fat pad displacement, and true depression caused by dermal rupture. Therefore, this module utilizes the nonlinear viscoelastic hysteresis characteristics of soft tissue during dynamic loading and unloading processes to construct a (actuation-response) two-dimensional phase plane. It obtains the area of ​​the closed trajectory formed by the vertical traction sequence of the zygomaticus major muscle and the vertical following sequence of the fat pad within the two-dimensional phase plane. This analysis examines the energy dissipation characteristics of facial soft tissue during anti-gravity movement due to the laxity of supporting ligaments, using this area as the ligament viscoelastic hysteresis area to characterize the displacement hysteresis of soft tissue in the loading and unloading path, and quantifies the resilience of supporting ligaments against gravity. Considering the lateral displacement that the fat pad may experience due to structural loosening when under traction, at the peak moment of the vertical traction sequence of the zygomaticus major muscle, the ratio of the relative sliding angle of the tissue is obtained based on the similarity between the traction vector of the zygomaticus major muscle and the sliding vector of the fat pad, so as to quantify the degree of spatial displacement of the passive tissue relative to the traction direction of the active muscle. To distinguish between pseudo-creases caused by gravitational compression and substantial depressions caused by dermal layer rupture, the maximum increment of the crease depth curvature sequence was extracted. Combined with the viscoelastic hysteresis area of ​​the ligament, the apparent crease depth feature value was obtained. By introducing a relaxation correction factor, the contribution component of relaxation compression to the surface crease was removed, thereby achieving independent decoupling and dynamic feature quantification analysis of the mechanical properties of the supporting ligament relaxation, the spatial displacement of the fat pad structure, and the physical depression depth of the dermis.

[0039] Preferably, in one embodiment of the present invention, the aging and degradation of collagen fibers in facial supporting ligaments (such as the zygomatic arch ligament) leads to a decrease in energy storage modulus, causing the passive fat pad to exhibit significant energy dissipation during the movement of active muscles. Specifically, the elastic recoil path during the muscle relaxation (unloading) phase does not coincide with the traction path during the contraction (loading) phase, forming a closed hysteresis loop. The area of ​​this loop directly quantifies the displacement hysteresis of the soft tissue system during the reciprocating motion cycle, i.e., the degree of relaxation of the supporting ligaments; Based on this, the system constructs a two-dimensional phase plane coordinate system, using the zygomaticus major muscle vertical traction sequence. The horizontal axis represents the sequence of fat pads following vertically. Construct a two-dimensional phase plane with the vertical axis as the ordinate; Connect the data points throughout the entire motion cycle in time sequence to form a closed (actuation-response) trajectory loop; To calculate the geometric area enclosed by this irregular closed trajectory, the system uses a discrete form of the shoelace formula to calculate the geometric area enclosed by the trajectory loop, which is then used as the viscoelastic hysteresis area of ​​the ligament. .

[0040] As an example, initialize the cumulative variable ; System traversal of discrete time axis from to : ; In order to close the loop, the system performs the final calculation (connection) and ): ; Finally, the system calculates... Half of the absolute value is taken as the viscoelastic hysteresis area of ​​the ligament. ; the result The higher the value, the greater the degree of relaxation of the supporting ligaments and the weaker the tissue's elasticity.

[0041] In a preferred embodiment of the invention, considering that due to gravity and the loss of tension in the lateral supporting tissues, loose or free fat pads, when pulled by muscles, cannot move along the predetermined direction of muscle contraction vector, but instead undergo uncontrolled lateral slippage or sagging. By comparing the spatial motion vector directions of the actuation source and the response source at their peak moments, the degree of anatomical misalignment can be quantified, thereby obtaining the relative slip angle ratio of the tissue. Please refer to [link to relevant documentation]. Figure 2 The flowchart illustrates a method for obtaining the relative slip angle ratio of a tissue according to an embodiment of the present invention, specifically including: Step S201: Before calculating the relative sliding angle ratio of the tissue, the displacement validity is verified.

[0042] Since the degree of anatomical misalignment is only meaningful when the fat pad produces a sufficiently significant displacement, the peak moment is the observation point where the muscle traction force is strongest and the soft tissue dynamic response is most significant, and it can best expose structural problems. Therefore, at the peak moment of the zygomaticus major muscle vertical traction sequence, the magnitude of the fat pad slip vector is obtained. If the length of the model is less than the preset micro-motion noise threshold, it indicates that the movement distance of the fat pad is too small. This movement may be caused by minor vibrations due to non-expression factors such as slight head shaking, breathing, or minor sensor noise, which are insufficient to constitute a displacement signal with analytical value. In this case, it is determined that there is no effective displacement, and the relative sliding angle ratio of the tissue is directly set to zero.

[0043] As an example, since all spatial coordinates have been converted to dimensionless values ​​by dividing by the anatomical scale normalization factor (based on interpupillary distance), the micromotion noise threshold is also in normalized units. The interpupillary distance is usually 60-65mm, and the displacement of the fat pad caused by a normal smile is usually between 2mm and 15mm, corresponding to about 0.033 to 0.25 normalized units. Therefore, the micro-motion noise threshold is set to 0.05 normalized units. When the magnitude of the fat pad slip vector is greater than or equal to the preset micro-motion noise threshold, it indicates that the fat pad has indeed produced an effective displacement, and then the calculation in step S202 is performed.

[0044] It should be noted that, in order to suppress high-frequency noise or inter-frame jumps that may be generated in a single frame image by the 3D face reconstruction algorithm, the system performs a moving average filtering process on the vertical traction sequence of the zygomaticus major muscle before extracting the peak time. Specifically, the filtering window is set to 2k+1, k is 4, and the frame containing the maximum value in the smoothed sequence is determined as the peak time; k can be adjusted as needed, and virtual frames are supplemented for the sequence boundary using the existing mirror extension method.

[0045] Step S202: At the peak moment of the vertical traction sequence of the zygomaticus major muscle, calculate the cosine similarity between the traction vector of the zygomaticus major muscle and the slip vector of the fat pad; normalize the cosine similarity by negative correlation and use it as the ratio of the relative slip angle of the tissue.

[0046] In the scenario of muscle traction and fat pad response, the ideal structural correspondence is that the fat pad slides along the direction of muscle traction, at which point the cosine similarity approaches 1. When the supporting structure is relaxed or misaligned, the fat pad will slip laterally or sag, resulting in a cosine similarity of less than 1. Therefore, the cosine similarity is negatively correlated and normalized as the ratio of the relative sliding angle of the tissue. As an example, the difference between the constant 1 and the cosine similarity is subtracted and negatively correlated for mapping. Due to physical limitations in the scenario, the directions of muscle traction and fat pad response are unlikely to be completely opposite. The difference value is usually between 0 and 1. Therefore, the difference value is truncated, that is, when it exceeds 1, it is set to 1 to complete the normalization process. The truncated difference value is used as the ratio of the relative sliding angle of the tissue.

[0047] Furthermore, to obtain the apparent crease depth feature value, since the curvature of the surface texture alone cannot distinguish between the physical depressions formed by the collagen breakage in the dermis and the pseudo-fold features formed by the gravitational compression of the loose tissue, the higher the degree of tissue laxity, the deeper the visual fold formed by the gravitational compression. In order to obtain the true defect degree contributed only by the physical breakage of the dermis, it is necessary to use ligament laxity as an inhibitory factor to correct the curvature factor. Based on this, the crease depth curvature sequence is extracted. Inner element relative to the first element The maximum increment is calculated by using the viscoelastic hysteresis area of ​​the ligament as an inhibition factor and performing a logarithmic inverse weighted correction on the maximum increment to calculate the apparent crease depth characteristic value.

[0048] As an example, within the crease depth curvature sequence, calculate the maximum value and the initial value over the entire motion cycle. The difference is used as the maximum increment of crease depth. ; because It is an area quantity, whose numerical magnitude is usually much larger than the curvature increment, and the effect of relaxation on the crease is non-linear. Therefore, the system uses logarithmic function compression. Based on the dynamic range, an inverse weighted model is constructed to calculate the apparent crease depth feature value. The calculation formula includes: ; in, This is a preset non-zero adjustment constant, which is set to 1 in this example to ensure that the denominator is not zero and to set the reference sensitivity. This represents the logarithmic function with the natural constant e as the base.

[0049] When the examinee has severe relaxation (i.e. When the denominator is larger, the calculation result... Significantly suppressed, corresponding to relaxed pseudocrease characteristics; only when the subject has low relaxation (i.e. When the curvature increment is small and the calculation result is large, the result is... It exhibits high values, specifically characterizing the physical fracture of the dermis.

[0050] It should be noted that the calculation process of cosine similarity is a well-known technique. In other embodiments of the present invention, the implementer may also subtract the difference of cosine similarity from the constant 1 and divide it by 2 to obtain the ratio of the relative slip angle of the tissue, adjust the negative correlation normalization method, and adjust the subsequent displacement judgment threshold accordingly.

[0051] Output module 103 is used to output facial dynamic deformation status labels in a graded manner based on the ligament viscoelastic hysteresis area, the ratio of tissue relative sliding angle, and the apparent crease depth feature value.

[0052] The ligament viscoelastic hysteresis area, the ratio of relative tissue sliding angle, and the apparent crease depth feature values ​​characterize the degree of decrease in the ligament's anti-gravity rebound ability, the degree of structural misalignment of the fat pad relative to the muscle traction direction, and the true physical depth of indentation after eliminating the interference of gravity compression, respectively, from the physical dimensions of soft tissue dynamic energy dissipation, three-dimensional spatial motion vector offset, and surface geometric morphology correction. Therefore, based on these parameters, a graded output of facial dynamic deformation state labels is finally made, and a mutually exclusive graded logic based on dynamic biomechanics and spatial geometric features is established to provide a reliable basis for personalized medical aesthetic plan recommendations.

[0053] Preferably, in one embodiment of the present invention, before system deployment, statistical analysis is performed on an empirical sample library containing a large number (e.g., no less than 200 cases) of manually labeled visual feature samples to determine that each sample is assigned a "pathological status" label based on independent and authoritative clinical assessment (e.g., a comprehensive judgment by two or more senior plastic surgeons through palpation, dynamic assessment and / or imaging examinations), such as "supporting ligament dysfunction / normal", "significant fat pad prolapse / normal", "substantial dermal defect / normal". The three indicators of ligament viscoelastic hysteresis area, tissue relative sliding angle ratio, and apparent crease depth were analyzed as target indicators; the values ​​of the target indicators for each confirmed sample were extracted. Using the clinical "pathology" label as the gold standard, ROC curves for the target indicators were plotted. By analyzing the ROC curves, the system selected the indicator value corresponding to the maximum value of the Youden index and established it as the threshold for determining the target indicator.

[0054] The relaxation threshold, displacement threshold, and fracture threshold are obtained and stored in non-volatile memory as preset values ​​for system calls.

[0055] For the current examinee, since ligament laxity means the loss of the soft tissue's ability to suspend against gravity, which is the most basic and priority factor affecting the mechanical stability of facial contours, laxity judgment is performed first. When the viscoelastic hysteresis area of ​​the ligament is greater than the preset laxity judgment threshold, the first-level status label Type-Region-A is output. Assuming ligament laxity is excluded, since the relative movement between soft tissue components becomes the key factor affecting morphology when the supporting structure is intact, if the fat pad slips laterally, it will cause structural accumulation on the outer side of the nasolabial fold. Therefore, displacement judgment is performed next. When the viscoelastic hysteresis area of ​​the ligament is less than or equal to the laxity judgment threshold, if the ratio of the relative sliding angle of the tissue is greater than the preset displacement judgment threshold, the secondary status label Type-Region-B is output. Only when neither of the first two conditions is met (i.e., the ratio of the ligament viscoelastic hysteresis area to the relative sliding angle of the tissue does not exceed its corresponding threshold) does it indicate that the mechanical support and spatial alignment of the soft tissue are within the normal range. The final fracture determination is performed at this point because, after excluding the two main dynamic mechanical causes mentioned above, the visual depression in the nasolabial fold region is more likely attributed to changes in the static physical properties of the local tissue. When the ligament viscoelastic hysteresis area is less than or equal to the relaxation determination threshold, and the ratio of the relative sliding angle of the tissue is less than or equal to the displacement determination threshold, if the apparent crease depth feature value is greater than the preset fracture determination threshold, a third-level status label, Type-Region-C, is output.

[0056] When no status label is output, the backup output strategy is triggered, and the label Type-Normal is output to indicate that no significant anomaly was detected.

[0057] It should be noted that the calculation of the Youden index and the ROC curve are existing, standard evaluation and parameter calibration methods, and will not be elaborated here.

[0058] In another embodiment of the invention, after outputting facial dynamic deformation state tags, a personalized list of recommended medical aesthetic treatments can be further provided. This function is implemented based on a pre-built experience sample library with treatment plan tags, specifically including: Each sample in the experience sample library not only contains the calculated values ​​of three indicators (ligament viscoelastic hysteresis area, tissue relative sliding angle ratio, and apparent crease depth characteristic value) and clinical facial dynamic deformation status labels, but also is associated with one or more medical aesthetic solutions that have been clinically validated and are targeted at their specific pathological conditions (such as "focused ultrasound skin tightening", "linear suspension repositioning", "hyaluronic acid filling", etc.).

[0059] The system first uses a preset judgment threshold to classify all samples in the database into the same pathological categories as in real-time analysis, classifying them into Type-Region-A, Type-Region-B, Type-Region-C, or Type-Normal.

[0060] Once the system outputs a specific facial dynamic deformation state label (e.g., Type-Region-A) for the current subject, it only searches the experience sample library for samples with the same facial dynamic deformation state label. For each sample of the same type in the library, the system calculates the absolute value of the difference between the core indicator that leads to that classification and the corresponding indicator for the current subject.

[0061] For example, for Type-Region-A samples, the core indicator is the viscoelastic hysteresis area of ​​the ligament, so the absolute difference between the hysteresis area of ​​the samples in the library and the hysteresis area of ​​the current subject is calculated; for Type-Region-C samples, the absolute difference between the apparent crease depth feature values ​​is calculated.

[0062] The smaller the absolute value of the difference, the more similar the quantitative characteristics of the historical sample are to the current user. The system sorts all samples of the same type from smallest to largest based on the calculated absolute value of the difference. Then, it extracts the medical aesthetic treatments associated with the top-ranked samples (e.g., the top 10) and generates a personalized recommendation list for the user.

[0063] In summary, to address the technical problem that existing technologies, relying solely on static image grayscale features, cannot effectively distinguish differences in facial visual deformation caused by different mechanical mechanisms, leading to inaccurate classification and identification of the physical aging state of facial soft tissues, this invention aims to provide a personalized medical aesthetic solution recommendation system based on a patient profile. This invention first acquires a monocular video stream of the subject's facial movements, maps a pre-set anatomical mesh template onto the subject's three-dimensional facial data, and simultaneously extracts the traction vector sequence and vertical traction sequence of the zygomaticus major muscle, the slip vector sequence and vertical following sequence of the fat pad, and the crease depth curvature sequence; further, it extracts the ligament viscoelastic hysteresis area, the tissue relative slip angle ratio, and the apparent crease depth feature values; finally, it performs hierarchical output of facial dynamic deformation state labels. This invention extracts and decouples three quantitative indicators—ligament hysteresis, fat slip, and dermal creases—from the dynamic video, and performs hierarchical judgment based on these three indicators, ultimately outputting facial dynamic deformation state labels, thus automating and accurately distinguishing the causes of facial dynamic mechanical features.

[0064] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A personalized medical aesthetic treatment plan recommendation system based on the patient profile of a medical aesthetic patient, characterized in that, The system includes: The preparation module is used to acquire a monocular video stream of the subject's facial movements, map a preset anatomical mesh template onto the subject's three-dimensional facial data, and establish a head coordinate system based on the first frame image. In the head coordinate system, each frame image is compared with the first frame image, and the traction vector sequence and vertical traction sequence of the zygomaticus major muscle, the sliding vector sequence and vertical following sequence of the fat pad, and the crease depth curvature sequence are extracted simultaneously. The analysis module is used to obtain the area of ​​the closed trajectory formed by the vertical traction sequence of the zygomaticus major muscle and the vertical following sequence of the fat pad in a two-dimensional phase plane, which is used as the viscoelastic hysteresis area of ​​the ligament; at the peak time of the vertical traction sequence of the zygomaticus major muscle, the relative sliding angle ratio of the tissue is obtained based on the similarity characteristics between the traction vector of the zygomaticus major muscle and the sliding vector of the fat pad; the maximum increment of the crease depth curvature sequence is extracted, and combined with the viscoelastic hysteresis area of ​​the ligament, the apparent crease depth feature value is obtained. The output module is used to output facial dynamic deformation status labels in a graded manner based on the ligament viscoelastic hysteresis area, the ratio of the relative sliding angle of the tissue, and the apparent crease depth feature value.

2. The personalized medical aesthetic treatment plan recommendation system based on a patient profile as described in claim 1, characterized in that, The anatomical mesh template includes a pre-labeled set of key anatomical indexes, which includes: an index of the zygomaticus major muscle traction anchor point corresponding to the middle section of the zygomaticus major muscle belly; an index of the centroid of the fat pad corresponding to the geometric center of the superficial fat pad on the outer side of the nasolabial fold; and an index set of nasolabial fold crease mesh areas covering the bottom of the nasolabial fold and the skin on both sides.

3. The personalized medical aesthetic solution recommendation system based on a patient profile as described in claim 2, characterized in that, The method for obtaining the crease depth curvature sequence includes: Iterate through all the grid vertices corresponding to the index set of the nasolabial fold grid region; For each frame of the image, the arithmetic mean of the average principal curvature of all vertices in the nasolabial fold region corresponding to the mesh vertex of the nasolabial fold is calculated as the value of the fold depth curvature sequence in the corresponding frame.

4. The personalized medical aesthetic solution recommendation system based on the patient profile of a medical aesthetic patient as described in claim 1, characterized in that, The method for obtaining the viscoelastic hysteresis area of ​​the ligament includes: A two-dimensional phase plane is constructed with the vertical traction sequence of the zygomaticus major muscle as the abscissa and the vertical following sequence of the fat pad as the ordinate. Connect the data points throughout the entire motion cycle in time sequence to form a closed trajectory loop; The geometric area enclosed by the trajectory loop is calculated using a discrete form of the shoelace formula, and is taken as the viscoelastic hysteresis area of ​​the ligament.

5. The personalized medical aesthetic solution recommendation system based on the patient profile of a medical aesthetic patient as described in claim 1, characterized in that, The method for obtaining the apparent crease depth feature value includes: Extract the maximum increment of each element in the crease depth curvature sequence relative to the first element; use the ligament viscoelastic hysteresis area as an inhibition factor to perform a logarithmic inverse weighted correction on the maximum increment, and calculate the apparent crease depth feature value.

6. The personalized medical aesthetic treatment plan recommendation system based on a patient profile as described in claim 1, characterized in that, Before calculating the relative slip angle ratio of the tissue, the displacement validity is verified. The verification methods include: At the peak moment of the vertical traction sequence of the zygomaticus major muscle, the magnitude of the fat pad slip vector is obtained; If the modulus is less than the preset micro-motion noise threshold, it is determined that there is no effective displacement, and the relative sliding angle ratio of the tissue is directly set to zero.

7. The personalized medical aesthetic solution recommendation system based on the patient profile of a medical aesthetic patient as described in claim 1, characterized in that, The method for obtaining the relative slip angle ratio of the tissue includes: At the peak moment of the vertical traction sequence of the zygomaticus major muscle, the cosine similarity between the traction vector of the zygomaticus major muscle and the slip vector of the fat pad is calculated; the cosine similarity is then negatively correlated and normalized as the ratio of the relative slip angle of the tissue.

8. The personalized medical aesthetic treatment plan recommendation system based on a patient profile as described in claim 1, characterized in that, The method for obtaining the facial dynamic deformation state tags includes: When the viscoelastic hysteresis area of ​​the ligament is greater than the preset relaxation determination threshold, the first-level status label Type-Region-A is output. When the viscoelastic hysteresis area of ​​the ligament is less than or equal to the relaxation determination threshold, if the relative sliding angle ratio of the tissue is greater than the preset displacement determination threshold, then the secondary status label Type-Region-B is output. When the viscoelastic hysteresis area of ​​the ligament is less than or equal to the relaxation determination threshold, and the ratio of the relative sliding angle of the tissue is less than or equal to the displacement determination threshold, if the apparent crease depth feature value is greater than the preset fracture determination threshold, then a three-level status label Type-Region-C is output.

9. The personalized medical aesthetic treatment plan recommendation system based on a patient profile as described in claim 1, characterized in that, The method for obtaining the head coordinate system includes: The Euclidean distance between the coordinates of the outer corners of the subject's eyes was used as the anatomical scale normalization factor. Based on the anatomical scale normalization factor, a three-dimensional Cartesian coordinate system is established with the direction of the line connecting the eyes as the X-axis, the vector perpendicular to the line connecting the eyes and located in the plane of facial symmetry as the Y-axis, and the facial normal vector as the Z-axis.

10. A personalized medical aesthetic treatment plan recommendation system based on a patient profile as described in claim 2, characterized in that, Using a non-rigid registration algorithm based on 3DMM, the preset anatomical mesh template is mapped onto the three-dimensional facial data of the subject reconstructed from the monocular video stream, so as to lock the zygomaticus major muscle traction anchor point, fat pad centroid and nasolabial fold mesh area that actually correspond to the subject's face.

Images