Exosome scar inhibiting effect prediction system based on deep learning
By using a deep learning-based exosome scar inhibition effect prediction system, which analyzes collagen fiber orientation and high-glucose edema factors through video stream analysis, calculates the drug diffusion blockage index, and generates targeted intervention instructions, the system solves the problem of waste of biological agents caused by blind drug administration and improves treatment efficacy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUBEI PICOLI MEDICAL TECH CO LTD
- Filing Date
- 2026-03-19
- Publication Date
- 2026-06-16
AI Technical Summary
In the current technology, blind administration of exosome biological agents leads to waste and poor therapeutic effects. It is impossible to assess the degree of physical obstruction of drug delivery by the wound in real time before administration, and clinical intervention lacks specificity.
A deep learning-based exosome scar inhibition effect prediction system was adopted. The system acquires the micro-motion video stream of the wound through the multi-dimensional data perception module, generates a tissue density stiffness map, analyzes the collagen fiber directional gradient field and the high sugar edema compression coefficient, and calculates the diffusion impedance between the drug supply point and the receptor node by combining the flow resistance simulation module. The prediction output module maps the drug diffusion retardation index to the effective uptake rate of exosomes and generates targeted intervention instructions.
It enables precise identification of areas where drug delivery is obstructed, avoids waste of biologics, improves the bioavailability of exosome drugs, and provides guidance for targeted intervention.
Smart Images
Figure CN122224501A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image analysis technology, and more specifically to a deep learning-based exosome scar inhibition effect prediction system. Background Technology
[0002] In the repair of chronic, difficult-to-heal wounds such as diabetic foot, exosomes, as nanobiochemical agents that regulate intercellular communication, have been widely used to inhibit pathological scarring. However, their clinical efficacy is often limited by the heterogeneity of the wound microenvironment. The increased osmotic pressure of tissue fluid caused by hyperglycemia leads to hydration and swelling of the extracellular matrix, compressing the tiny gaps between collagen fibers. The directional distribution of collagen fibers further exacerbates the difficulty of drug diffusion in specific directions. The combination of these two factors can easily form microscopic deadlock areas invisible to the naked eye.
[0003] Current clinical drug delivery methods rely heavily on physicians' visual or tactile experience, making it difficult to identify these microscopic physical barriers invisible to the naked eye. If drugs are blindly administered to areas with deadlocks, exosome particles not only fail to penetrate the matrix to reach deep receptor cells, but also become degraded by high concentrations of inflammatory enzymes due to prolonged retention, leading to ineffective treatment. Furthermore, while commonly used in vitro pathological examination methods are accurate, they are destructive and delayed, failing to assess the degree of physical obstruction to drug delivery in the wound in real time before administration. This results in a lack of targeted clinical intervention, leading to waste of exosome biologics and poor therapeutic effects. Summary of the Invention
[0004] To address the technical problems of indiscriminate drug administration in existing technologies, which leads to a lack of targeted clinical intervention, waste of exosome biological agents, and poor therapeutic effects, the present invention aims to provide a deep learning-based exosome scar inhibition effect prediction system. The specific technical solution adopted is as follows: This invention provides a deep learning-based exosome scar inhibition effect prediction system, the system comprising: The multidimensional data perception module acquires a time-series video stream of wound micro-motion, generates a tissue density stiffness map using time-frequency inversion, obtains the collagen fiber directional gradient field through the spatial gradient in the tissue density stiffness map, and determines the high-glucose edema compression coefficient based on the current blood glucose data. The flow resistance simulation analysis module, based on the tissue density stiffness map, determines the discretized drug supply point set and receptor node set within a local region. For each drug supply point and receptor node, it analyzes the coupling impedance gain influenced by the collagen fiber directional gradient field and the high-glucose edema compression coefficient along the connection path. Combined with the stiffness in the tissue density stiffness map, it obtains the interstitial diffusion local impedance. Based on the interstitial diffusion local impedance, it uses a bipartite graph matching algorithm to solve for the minimum total transmission cost of the drug supply point set and receptor node set, obtaining the regional drug diffusion hindrance index. The prediction output module maps the regional drug diffusion retardation index to the exosome effective uptake rate by combining pharmacokinetic parameters; it distinguishes the low exosome effective uptake rate based on the tissue density stiffness map and the high glucose edema compression coefficient and generates instructions; and it outputs the wound drug delivery efficiency map by superimposing the wound micro-motion video stream on the exosome effective uptake rate mapping map.
[0005] Furthermore, the method for obtaining the tissue density stiffness map includes: Euler video amplification processing was performed on the video stream of micro-motion of the wound, and non-cardiac noise was filtered out by a time bandpass filter to extract the micro-motion signal caused by cardiovascular pulsation; The instantaneous amplitude envelope of the micro-motion signal is obtained by Hilbert transform, and the time-averaged amplitude field is calculated. Based on the negative correlation between amplitude and stiffness, the stiffness is obtained by the reciprocal of the time-averaged amplitude field, thus obtaining the tissue density stiffness map.
[0006] Furthermore, the method for obtaining the collagen fiber directional gradient field includes: Differential operators were used to calculate the first-order spatial partial derivatives of the tissue density stiffness diagram in the horizontal and vertical directions, and a two-dimensional vector field was synthesized as the directional gradient field of collagen fibers.
[0007] Furthermore, the method for obtaining the drug supply point set and the receptor node set includes: A sliding window is used to traverse the tissue density stiffness map, and pixels are uniformly sampled in the local area within each sliding window at a preset spatial interval to obtain the drug supply point set. Calculate the local statistical characteristics of tissue density stiffness values within each sliding window, and construct an effective receptor screening range based on the local statistical characteristics; the local statistical characteristics are the mean and standard deviation. Pixels whose stiffness in the tissue density stiffness graph within the sliding window falls within the effective receptor screening value range are identified as receptor nodes, forming a receptor node set.
[0008] Furthermore, the method for obtaining the interstitial diffusion local impedance includes: For any drug supply point and receptor node, the straight path from the drug supply point to the receptor node is used as the connection path. For any pixel on the connection path, the parallelism between the direction vector of the connection path where the pixel is located and the corresponding vector direction in the collagen fiber orientation gradient field is analyzed, which is used as the inverse texture of the pixel; the inverse texture of the pixel is coupled with the high sugar edema compression coefficient to obtain the coupling impedance gain. The gain of the basic impedance and coupling impedance of the pixel stiffness analysis is superimposed and integrated along the connection path to obtain the local impedance of interstitial diffusion.
[0009] Furthermore, the method for obtaining the regional drug diffusion retardation index includes: Within each sliding window, the drug supply point set and the active tissue receptor point set are respectively used as two vertex sets of a bipartite graph, and the interstitial diffusion local impedance is mapped as the connection weight between vertices. When the number of drug supply point sets and active tissue receptor point sets are not equal, virtual nodes are introduced to complete the bipartite graph into a complete graph, and the connection cost between virtual nodes and real nodes is set as a preset penalty threshold. The minimum weight matching algorithm is used to find the optimal matching scheme, and the average connection weight in the optimal matching scheme is used as the regional drug diffusion inhibition index for each sliding window.
[0010] Furthermore, the method for obtaining the effective uptake rate of exosomes includes: The effective half-life under the given conditions was calculated using the baseline half-life of exosomes, the enzyme hydrolysis sensitivity coefficient, and the high glucose edema compression coefficient. The degree of attenuation was analyzed pixel by pixel in conjunction with the regional drug diffusion retardation index and the effective half-life to obtain the effective uptake rate of exosomes.
[0011] Furthermore, the attribution and generation of instructions based on the tissue density stiffness map and the high glucose edema compression coefficient for the low exosome effective uptake rate includes: Connected regions with exosome effective uptake rates less than a preset effective threshold baseline are identified as inefficient blocking regions. If the compression coefficient of high-glucose edema in the inefficient blockade zone is greater than the preset edema threshold, an intervention instruction for edema is generated; if the compression coefficient of high-glucose edema in the inefficient blockade zone is less than or equal to the preset edema threshold and the average tissue density stiffness is greater than the preset stiffness threshold, an intervention instruction for physical barrier is generated. Otherwise, intervention instructions related to dose adjustment or assisted delivery are generated.
[0012] Furthermore, the step of outputting a wound drug delivery efficiency map by superimposing an exosome effective uptake rate mapping map onto a wound micro-motion video stream includes: The effective uptake rate of exosomes is mapped to a pseudo-color encoded heatmap; a semi-transparent overlay technique is used to overlay the heatmap onto the first frame of the wound micro-motion video stream, and corresponding intervention instructions are simultaneously marked in the inefficient blockage area to form a wound drug delivery efficiency map.
[0013] Furthermore, the method for obtaining the high-glucose edema compression coefficient includes: The hyperglycemic edema compression coefficient is obtained based on the degree of deviation between the patient's current blood glucose data and the preset benchmark blood glucose value.
[0014] The present invention has the following beneficial effects: This invention acquires video streams, analyzes the coupling of collagen fiber orientation and hyperglycemic edema factors to calculate interstitial diffusion local impedance, and uses a bipartite graph matching algorithm to solve for the minimum total transmission cost, quantifying the regional drug diffusion obstruction index. Through anisotropic flow resistance simulation, it reveals the superimposed obstruction effect of two independent factors—collagen fiber alignment and tissue fluid osmotic pressure—under specific combinations, identifying specific regions where reverse texture channels are blocked due to edema compression. The obstruction index is mapped to the effective exosome uptake rate. By retrospectively analyzing microscopic parameters, the main causes of obstruction in inefficient regions are identified, and targeted intervention instructions are generated. Using physical-biological mapping and retrospective logic, the mechanisms leading to drug delivery obstruction are automatically distinguished, making intervention decisions more reliable. This invention, through anisotropic flow resistance analysis based on image-coupled texture and osmotic pressure, achieves attribution differentiation and intervention through microscopic analysis combined with physical-biological mapping, avoiding the waste of biological agents caused by indiscriminate drug administration and improving the bioavailability of exosome drugs. 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 structural diagram of a deep learning-based exosome scar inhibition effect prediction system provided in one 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 deep learning-based exosome scar inhibition effect prediction system proposed 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 a specific scheme for a deep learning-based exosome scar inhibition effect prediction system provided by the present invention.
[0020] Please see Figure 1The diagram shows a structural diagram of a deep learning-based exosome scar inhibition effect prediction system provided by an embodiment of the present invention. The system includes: a multi-dimensional data perception module 101, a flow resistance simulation analysis module 102, and a prediction output module 103.
[0021] The multidimensional data perception module 101 acquires the video stream of wound micro-motion in time series, generates a tissue density stiffness map using time-frequency inversion, obtains the collagen fiber orientation gradient field through the spatial gradient in the tissue density stiffness map, and determines the high glucose edema compression coefficient through the current blood glucose data.
[0022] In this embodiment of the invention, the basic pharmacokinetic parameters of the exosomal drug are first obtained as global constants for subsequent calculations and analyses. These parameters mainly include: a baseline half-life constant, used to characterize the natural degradation rate of exosomes under standard physiological conditions; an enzymatic sensitivity coefficient, used to characterize the sensitivity of exosomes to matrix metalloproteinases in an inflammatory environment; and a preset effective threshold baseline, used to define the minimum drug concentration required to produce an inhibitory effect on scar hyperplasia. It should be noted that the basic pharmacokinetic parameters are derived from a pre-conducted database of in vitro tissue permeation experiments or clinical experience values, and are not limited here.
[0023] Furthermore, in this embodiment of the invention, an image acquisition device, such as a high-definition camera, is used to acquire a video stream of micro-motion of the examined wound. The acquisition time is preset to 10 seconds, and the frame rate is 30 frames per second. At the same time, a two-dimensional pixel coordinate system covering the wound area is established, and each frame of the acquired image is spatially calibrated to ensure that all pixels correspond to the same physical position on the time axis. At this time, the video stream of micro-motion of the wound can be represented as a three-dimensional matrix, including x and y pixel coordinates and the t time dimension.
[0024] Because the necrotic eschar, fibrotic scar, and edematous granulation tissue within diabetic wounds exhibit significantly different viscoelasticities, their deformation response amplitudes differ when driven by weak pressure waves transmitted from the patient's own cardiovascular system. According to the rheological properties of biological soft tissues, under the premise of a constant driving force, the deformation amplitude of a local tissue is negatively correlated with its elastic modulus, i.e., its stiffness.
[0025] Utilizing this characteristic, micro-motion signals are extracted from the video stream and stiffness distribution is inverted. In this embodiment of the invention, the micro-motion video stream of the wound is subjected to Euler video amplification processing, and the Laplacian pyramid decomposition method is used to separate image features at different spatial scales. Non-cardiac noise is filtered through a time bandpass filter, with the frequency band selectable from 0.8Hz to 2.5Hz, covering the normal heart rate range of adults at rest. This effectively filters out interference signals such as respiratory movements and changes in ambient light, extracts micro-motion signals caused by cardiovascular pulsation, and retains the micro-flutter signal component induced only by vascular pulsation.
[0026] The instantaneous amplitude envelope of the micro-motion signal is further obtained through Hilbert transform to eliminate signal phase interference and accurately capture the amplitude change pattern. The time-averaged amplitude field is then calculated. Specifically, the instantaneous amplitude envelope is arithmetically averaged over a preset acquisition period to obtain a time-averaged amplitude field covering the entire wound surface. Based on the negative correlation between amplitude and stiffness, the stiffness is obtained by calculating the reciprocal of the time-averaged amplitude field, resulting in a tissue density stiffness map. Specifically, a small constant is introduced into the reciprocal denominator when calculating the original stiffness value to avoid division by zero anomalies. This constant can be 0.01, and the implementer can adjust it as needed; no restrictions are imposed here.
[0027] The original stiffness values are then processed using the min-max normalization method, mapping them to a numerical range of 0 to 1, ultimately generating a tissue density stiffness map. A stiffness value closer to 1 indicates higher tissue density and greater stiffness, while a value closer to 0 indicates softer tissue. It should be noted that the min-max normalization method is a well-known technique among those skilled in the art. In other embodiments of this invention, standard normalization methods, etc., may also be used, and no limitation is made here.
[0028] Collagen deposition during wound healing is directional, and the resulting fiber bundle texture leads to anisotropy in drug diffusion resistance. At the microscale, tissue stiffness exhibits the greatest rate of change perpendicular to fiber alignment and the smallest rate of change parallel to fiber alignment. Therefore, the spatial gradient direction of the tissue density stiffness map is geometrically orthogonal to the local collagen fiber texture orientation.
[0029] In this embodiment of the invention, the first-order spatial partial derivatives of the tissue density stiffness map in the horizontal and vertical directions are calculated using a differential operator to obtain the horizontal gradient component and the vertical gradient component. The gradient components in the two directions are synthesized into a two-dimensional vector field corresponding to the pixel positions as the collagen fiber orientation gradient field. In this two-dimensional vector field, the vector direction of any pixel points to the direction of the fastest increase in stiffness, that is, it is orthogonal to the local collagen fiber texture direction. The magnitude of the vector characterizes the severity of the tissue structure change at that position.
[0030] It should be noted that differential operator computation is a method well known to those skilled in the art, and the Sobel operator can be used to obtain it, which will not be restricted or elaborated here.
[0031] The hyperglycemic environment in diabetic patients causes an increase in tissue fluid osmotic pressure, leading to hydration and swelling of the extracellular matrix. This microscopic edema compresses intercellular spaces, thereby altering the effective porosity of the medium. The compression effect of edema on drug diffusion channels can be quantified by collecting patients' blood glucose data.
[0032] In this embodiment of the invention, a hyperglycemic edema compression coefficient is obtained based on the deviation between the patient's current blood glucose data and a preset benchmark blood glucose value. Specifically, the preset benchmark blood glucose value is set as the upper limit of normal fasting blood glucose. The relative difference between the current blood glucose data and the benchmark blood glucose value is calculated, and the relative difference is scalarized and mapped using a hyperbolic tangent function, resulting in a hyperglycemic edema compression coefficient with a value ranging from 0 to 1. A larger value indicates a higher degree of closure of the interstitial spaces due to edema compression, and a narrower drug diffusion channel. In this embodiment of the invention, the preset benchmark blood glucose value can be 5.5 mmol / L, and is not limited thereto. As an example, the expression for the hyperglycemic edema compression coefficient is: In the formula, This is expressed as the high-glucose edema compression factor. This represents the preset baseline blood glucose value. This represents the current blood glucose data. This is expressed as the relative difference between the current blood glucose value and the baseline blood glucose value. It is represented as the hyperbolic tangent function.
[0033] The flow resistance simulation analysis module 102, based on the tissue density stiffness map, determines the discretized drug supply point set and receptor node set within the local region. For each drug supply point and receptor node, it analyzes the coupling impedance gain affected by the collagen fiber directional gradient field and the high glucose edema compression coefficient on the connection path. Combined with the stiffness in the tissue density stiffness map, it obtains the interstitial diffusion local impedance. Based on the interstitial diffusion local impedance, it uses a bipartite graph matching algorithm to solve for the minimum total transmission cost of the drug supply point set and receptor node set, obtaining the regional drug diffusion hindrance index.
[0034] To accurately identify drug targets in highly heterogeneous wound environments, the system first needs to define discrete computational nodes within a continuous spatial field. Since baseline stiffness varies significantly among different patients and even different regions of the same wound, using a globally fixed threshold to screen receptor nodes often leads to misclassification. Therefore, an adaptive discretization strategy based on the statistical properties of a sliding window can be employed to pinpoint relatively suitable bioactive regions within the current local environment.
[0035] In this embodiment of the invention, based on the tissue density stiffness map, a sliding window is used to traverse the tissue density stiffness map. The size and step size of the sliding window are adaptively set according to the resolution of the wound image. In a specific embodiment of the invention, the size of the sliding window is set to 32×32 pixels and the step size is 8 pixels, and the sliding window is performed pixel by pixel on the global pixel coordinate system.
[0036] Within each sliding window, pixels are uniformly sampled at preset spatial intervals to obtain a drug supply point set. Specifically, assuming exosome drugs are uniformly applied to the wound surface, pixel coordinates within the window are extracted as drug supply nodes according to the preset spatial sampling interval, thus forming a drug supply point set. In this embodiment of the invention, the preset spatial interval can be set to 2 pixels, and is not limited thereto.
[0037] Furthermore, the local statistical characteristics of tissue density stiffness values within each sliding window are calculated. These local statistical characteristics are the mean and standard deviation, reflecting the overall level and dispersion of tissue stiffness within the window. Based on these local statistical characteristics, an effective receptor screening range is constructed. Specifically, the range is formed by floating above and below the local mean by a preset multiple of the standard deviation. The preset multiple is determined based on the distribution pattern of clinically active tissue stiffness. Pixels with stiffness within a reasonable range are then screened out through this range, eliminating necrotic eschar areas with excessively high stiffness and liquefied pus areas with excessively low stiffness, focusing on active tissue areas with therapeutic value. As an example, the effective receptor screening range is [uk×s, u+k×s], where u represents the mean, s represents the standard deviation, and k is the preset multiple, which can be set to 1.2 and adjusted by the implementer.
[0038] Finally, pixels whose stiffness in the tissue density stiffness map within the sliding window falls within the effective receptor screening range are identified as receptor nodes, forming a receptor node set. Specifically, the stiffness of each pixel within the window is determined by range. If the stiffness of the pixel falls within the effective receptor screening range, it is marked as a biologically active receptor node. All marked receptor nodes within the window are collected to form the receptor node set for that local region. As the sliding window traverses the entire wound, the discretization construction of the drug supply point set and receptor node set for the entire wound is completed.
[0039] For the constructed drug supply point set and receptor node set, the aim is to calculate the resistance on each possible micropath. The nonlinear modulation effect of the high sugar edema compression coefficient on the collagen fiber orientation gradient field is introduced, and the physical phenomenon of edema compression leading to inverse texture channel closure is simulated and analyzed.
[0040] In this embodiment of the invention, for any drug supply point and receptor node, the Euclidean straight line path from the drug supply point to the receptor node is used as the connection path. This connection path is the shortest geometric path for drug particles to diffuse from the supply point to the receptor node, which conforms to the principle of the shortest path for drug diffusion at the microscale. At the same time, the connection path is discretized into a continuous sequence of pixels with pixel precision to ensure the refinement of subsequent impedance calculations.
[0041] For any pixel on the connection path, the parallelism between the direction vector of the connection path where the pixel is located and the corresponding vector direction in the collagen fiber orientation gradient field is analyzed. This is used as the inverse texture degree of the pixel. In a specific embodiment of the present invention, the parallelism is characterized by calculating the dot product of the two vectors. The closer the dot product is to 1, the more parallel the path direction is to the direction of the collagen fiber orientation gradient field. That is, the more perpendicular the drug diffusion direction is to the collagen fiber texture direction, the higher the inverse texture degree, and the greater the directional resistance to drug diffusion.
[0042] Furthermore, by coupling the inverse texture degree of this pixel with the high sugar edema compression coefficient, the coupling impedance gain is obtained. When the inverse texture degree is high and the high sugar edema compression coefficient is large, the coupling impedance gain shows a significant amplification trend, simulating the resistance enhancement effect of edema compression causing the inverse texture diffusion channel to be locked. As an example, the expression for the coupling impedance gain is: ; Represented as the first in the connection path The coupling impedance gain of each pixel This is expressed as the high-glucose edema compression factor. Represented as the first in the connection path Inverse texture of each pixel This is represented as a structural sensitivity coefficient, used to adjust the model's penalty for inverse texture diffusion, and can be set to 3. It is expressed as an exponential function with the natural constant as the base. When hyperglycemic edema is present and the drug diffuses against the grain, the lock-up effect of the diffusion channel increases exponentially, significantly amplifying the resistance value, simulating the lock-up effect of physical channels.
[0043] The base impedance and coupling impedance gain of the pixel stiffness analysis are superimposed and integrated along the connection path to obtain the interstitial diffusion local impedance. Specifically, the base impedance is directly determined by the tissue density stiffness of the pixel; the greater the stiffness, the higher the base impedance. The base impedance and coupling impedance gain of each pixel are superimposed to obtain the instantaneous impedance of that point. As an example, the expression for the superimposed instantaneous impedance is: In the formula, Represented as the first in the connection path The instantaneous impedance after the superposition of individual pixels, Represented as the first in the connection path Coupling impedance gain per pixel Represented as the first in the connection path Stiffness of each pixel Represented as the first in the connection path The base impedance of each pixel represents the base impedance considering the theoretical minimum cost.
[0044] Then, the instantaneous impedance is sequentially accumulated along the pixel sequence of the connection path to complete the path integration operation, and finally the total resistance of the connection path is obtained, that is, the local impedance of interstitial diffusion.
[0045] Considering the resource competition characteristics of exosome therapy, namely that a limited number of drug particles need to cover a limited number of recipient cells, the microscopic path problem is transformed into a macroscopic resource allocation optimization problem to evaluate the overall feasibility of drug delivery in a local area. After obtaining the interstitial diffusion local impedance of the connection paths between all drug supply points and recipient nodes, the minimum total transport cost is solved using a bipartite graph matching algorithm based on the interstitial diffusion local impedance, resulting in the regional drug diffusion obstruction index.
[0046] In this embodiment of the invention, the drug supply point set and the receptor node set in each sliding window are respectively regarded as two vertex sets of a bipartite graph, and the interstitial diffusion local impedance of each connection path is regarded as the connection weight between the corresponding vertices to construct a complete bipartite graph weight matrix. This matrix intuitively reflects the diffusion resistance cost of each drug supply point-receptor node pair, providing a quantitative basis for subsequent resource allocation optimization.
[0047] Understandably, when the number of nodes in two point sets is unequal, virtual nodes are introduced to complete the bipartite graph into a complete graph. The connection cost between virtual nodes and real nodes is set as a preset penalty threshold. This penalty threshold is significantly higher than the average interstitial diffusion local impedance of the normal path to penalize supply-demand imbalances, avoid matching distortion caused by excessive drug supply or insufficient receptor nodes, and ensure that the matching result can truly reflect the resistance level of effective drug delivery. In this embodiment of the invention, it can be set to 100 times the average cost of the normal path, but this is not limited here.
[0048] The optimal matching scheme is found using a minimum-weight matching algorithm. The Kuhn-Munkres algorithm or the Hungarian algorithm can be selected. This algorithm iterates through all possible node matching combinations to find a one-to-one mapping that minimizes the global total connection weight. This mapping corresponds to the optimal allocation strategy of a finite number of drug particles among receptor nodes, i.e., allowing the drug to cover as many active receptor nodes as possible with the lowest diffusion resistance. It should be noted that the matching algorithm is a well-known technique in the art and will not be elaborated upon here.
[0049] After obtaining the optimal matching scheme, the sum of the connection weights of all matching paths under this scheme is calculated. The total weight is divided by the actual total number of nodes in the recipient node set to eliminate virtual nodes used for completion, and the average connection weight is obtained. This average connection weight is used as the regional drug diffusion barrier index for the local region corresponding to each sliding window. This index is a dimensionless value that characterizes the average physical energy barrier for drug delivery from the supply point to the recipient node in the local region, reflecting the overall drug diffusion difficulty in the region. The larger the value, the more severe the drug delivery is blocked and the more difficult it is for exosomes to reach the recipient cells.
[0050] It should be noted that, in order to quantify the coverage of different local areas, there is overlap between sliding windows. Therefore, the drug diffusion retardation index of each region is ultimately the drug diffusion retardation index of the center point of the corresponding sliding window, which facilitates subsequent overall analysis.
[0051] Thus, the transformation from microscopic path impedance calculation to macroscopic regional blockage index generation was completed. The heterogeneity of the local wound environment, the coupling effect of collagen fiber orientation and high glucose edema, and the resource competition characteristics of drug-receptor were uniformly quantified into physical indicators that can be directly used for efficacy prediction, laying the core foundation for the calculation of exosome effective uptake rate in the subsequent prediction output module.
[0052] The prediction output module 103 maps the regional drug diffusion retardation index to the exosome effective uptake rate by combining pharmacokinetic parameters; it distinguishes the low exosome effective uptake rate based on the tissue density stiffness map and the high sugar edema compression coefficient and generates instructions; it outputs the wound drug delivery efficiency map by superimposing the wound micro-motion video stream on the exosome effective uptake rate mapping map.
[0053] During exosome therapy, the longer the drug remains in the delivery pathway, the higher the probability of it being degraded by matrix metalloproteinases (MMPs) in the inflammatory environment. Therefore, it is necessary to transform the physical blockage indicators obtained from flow resistance simulation into biological indicators reflecting the drug survival rate, that is, to transform physical diffusion blockage into effective dose attenuation at the biological level, providing intuitive predictive evidence for clinical practice.
[0054] In this embodiment of the invention, the first-order elimination kinetic equation from pharmacokinetics is introduced to quantify the drug decay over time. The regional drug diffusion retardation index can be equated to the "equivalent transport time" of the drug from the supply point to the receptor node. A higher regional drug diffusion retardation index indicates a longer drug transport time and a higher risk of degradation. The effective half-life under the given conditions is calculated using the exosome baseline half-life, the enzymatic sensitivity coefficient, and the high glucose edema compression coefficient. Specifically, high glucose edema is usually accompanied by increased inflammation and elevated matrix metalloproteinase concentrations, which accelerate exosome degradation. Therefore, the effect of correcting the enzymatic sensitivity coefficient using the high glucose edema compression coefficient is applied. The product of the high glucose edema compression coefficient and the enzymatic sensitivity coefficient is normalized to obtain the acceleration coefficient. Standard normalization can be selected to map the acceleration coefficient to [0,1). Furthermore, the product of the baseline half-life and the acceleration coefficient is used as the acceleration regulation degree, and the difference between the baseline half-life and the acceleration regulation degree is taken as the effective half-life.
[0055] As an example, the expression for the effective half-life is: In the formula, This is expressed as the effective half-life. This is expressed as the baseline half-life. This is expressed as the enzyme hydrolysis sensitivity coefficient. This is expressed as the high-glucose edema compression factor. Represented as the acceleration coefficient, This is expressed as the acceleration adjustment degree. This is represented as a normalization function.
[0056] The effective uptake rate of exosomes is obtained by analyzing the decay degree of regional drug diffusion retardation index and effective half-life pixel by pixel. Based on the exponential decay characteristics of first-order kinetics, the longer the equivalent transport time and the shorter the effective half-life, the more significant the drug decay and the lower the uptake rate, and vice versa. Therefore, the ratio of regional drug diffusion retardation index to effective half-life is used as the exponential decay function to obtain the effective uptake rate of exosomes. As an example, the expression for the effective uptake rate of exosomes is: In the formula, Represented as the first Effective uptake rate of exosomes per pixel Represented as the first The drug diffusion resistance index of a region per pixel This is expressed as the effective half-life. It is represented as an exponential function with the natural constant as the base.
[0057] To accurately locate ineffective treatment areas and identify the causes of blockage, guiding targeted clinical intervention, ineffective blockage areas need to be differentiated based on physical attribution logic after screening. In this embodiment of the invention, connected regions with exosome effective uptake rates lower than a preset effective threshold baseline are extracted as ineffective blockage areas. The preset effective threshold baseline is determined based on the uptake rate level corresponding to the effective concentration in clinical treatment, ensuring that the screened areas are those where scar inhibition is difficult to achieve in actual treatment. It can be set to 0.4, and the specific value can be adjusted by the implementer.
[0058] If the high-glucose edema compression coefficient in the inefficient blockade zone is greater than the preset edema threshold, it indicates that the main cause of drug blockade in this area may be the increase in tissue fluid osmotic pressure caused by hyperglycemia. The extracellular matrix hydration and swelling leads to the closure of drug diffusion channels, i.e., the hydraulic barrier, generating intervention instructions for edema, such as local dehydration treatment and negative pressure drainage, to relieve tissue edema and widen the drug diffusion channels.
[0059] If the high-glucose edema compression coefficient in the inefficient blockade zone is less than or equal to the preset edema threshold and the average tissue density stiffness is greater than the preset stiffness threshold, it indicates that there may be dense necrotic eschar or severe fibrosis in the area, i.e., a physical barrier. The high density of the tissue itself hinders drug penetration, generating intervention instructions for the physical barrier, such as mechanical debridement or enzymatic debridement, to remove dense tissue and reduce drug diffusion resistance.
[0060] Otherwise, it indicates that there is no obvious edema or physical barrier in the area, and the inefficient drug delivery may be due to excessive transmission distance or insufficient drug dosage. In this case, intervention instructions related to dosage adjustment or delivery assistance will be generated, such as increasing the exosome dosage or combining microneedle delivery technology to enhance drug penetration. In this embodiment of the invention, the preset edema threshold can be set to 0.6, and the preset stiffness threshold can be set to 0.7. Both the preset edema threshold and the preset stiffness threshold are determined based on clinical pathological data and a large number of experimental statistics, and can be adaptively adjusted according to different wound types to ensure the accuracy of attribution determination.
[0061] To visually present the prediction results, a wound drug delivery efficiency map is output by overlaying an exosome effective uptake rate mapping map onto a wound micro-motion video stream. In this embodiment, the exosome effective uptake rate values are mapped as a pseudo-color encoded heatmap, using a gradient color scheme to distinguish between high and low uptake rates. For example, high uptake rate areas are encoded in green, medium uptake rate in yellow, and low uptake rate in red, making the delivery effect in different areas simple and clear. A semi-transparent overlay technique is used to distribute and overlay the heatmap onto the first frame of the wound micro-motion video stream, preserving the original anatomical location information of the wound while clearly overlaying the efficacy prediction results. Simultaneously, corresponding intervention instructions, such as dehydration, debridement, and dose increase, are labeled in inefficient blockage areas, forming a wound drug delivery efficiency map. This efficiency map can be directly output from a clinical display terminal, assisting doctors in quickly developing more accurate and reliable pre-treatment plans for drug administration.
[0062] In summary, this invention calculates the local impedance of interstitial diffusion by coupling collagen fiber orientation with the factor of high glucose edema, and uses a bipartite graph matching algorithm to solve for the minimum total transport cost, quantifying the regional drug diffusion blockage index. Through anisotropic flow resistance simulation, it reveals the superimposed blockage effect of two independent factors—collagen fiber alignment and tissue fluid osmotic pressure—under specific combinations, identifying specific regions where reverse texture channels are blocked due to edema compression. The blockage index is mapped to the effective uptake rate of exosomes. By retrospectively analyzing microscopic parameters, the main causes of blockage in inefficient regions are identified, and targeted intervention instructions are generated. Using physical-biological mapping and retrospective logic, the mechanisms leading to drug delivery obstruction are automatically distinguished, making intervention decisions more reliable. This invention, through anisotropic flow resistance analysis coupled with texture and osmotic pressure, achieves attribution differentiation and intervention through microscopic analysis combined with physical-biological mapping, avoiding the waste of biological agents caused by indiscriminate drug administration and improving the bioavailability of exosome drugs.
[0063] 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.
[0064] 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 deep learning-based exosome scar inhibition effect prediction system, characterized in that, The system includes: The multidimensional data perception module acquires a time-series video stream of wound micro-motion, generates a tissue density stiffness map using time-frequency inversion, obtains the collagen fiber directional gradient field through the spatial gradient in the tissue density stiffness map, and determines the high-glucose edema compression coefficient based on the current blood glucose data. The flow resistance simulation analysis module, based on the tissue density stiffness map, determines the discretized drug supply point set and receptor node set within a local region. For each drug supply point and receptor node, it analyzes the coupling impedance gain influenced by the collagen fiber directional gradient field and the high-glucose edema compression coefficient along the connection path. Combined with the stiffness in the tissue density stiffness map, it obtains the interstitial diffusion local impedance. Based on the interstitial diffusion local impedance, it uses a bipartite graph matching algorithm to solve for the minimum total transmission cost of the drug supply point set and receptor node set, obtaining the regional drug diffusion hindrance index. The prediction output module maps the regional drug diffusion retardation index to the exosome effective uptake rate by combining pharmacokinetic parameters; it distinguishes the low exosome effective uptake rate based on the tissue density stiffness map and the high glucose edema compression coefficient and generates instructions; and it outputs the wound drug delivery efficiency map by superimposing the wound micro-motion video stream on the exosome effective uptake rate mapping map.
2. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The method for obtaining the tissue density stiffness map includes: Euler video amplification processing was performed on the video stream of micro-motion of the wound, and non-cardiac noise was filtered out by a time bandpass filter to extract the micro-motion signal caused by cardiovascular pulsation; The instantaneous amplitude envelope of the micro-motion signal is obtained by Hilbert transform, and the time-averaged amplitude field is calculated. Based on the negative correlation between amplitude and stiffness, the stiffness is obtained by the reciprocal of the time-averaged amplitude field, thus obtaining the tissue density stiffness map.
3. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The method for obtaining the collagen fiber directional gradient field includes: Differential operators were used to calculate the first-order spatial partial derivatives of the tissue density stiffness diagram in the horizontal and vertical directions, and a two-dimensional vector field was synthesized as the directional gradient field of collagen fibers.
4. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The method for obtaining the drug supply point set and the receptor node set includes: A sliding window is used to traverse the tissue density stiffness map, and pixels are uniformly sampled in the local area within each sliding window at a preset spatial interval to obtain the drug supply point set. Calculate the local statistical characteristics of tissue density stiffness values within each sliding window, and construct an effective receptor screening range based on the local statistical characteristics; the local statistical characteristics are the mean and standard deviation. Pixels whose stiffness in the tissue density stiffness graph within the sliding window falls within the effective receptor screening value range are identified as receptor nodes, forming a receptor node set.
5. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The method for obtaining the interstitial diffusion local impedance includes: For any drug supply point and receptor node, the straight path from the drug supply point to the receptor node is used as the connection path. For any pixel on the connection path, the parallelism between the direction vector of the connection path where the pixel is located and the corresponding vector direction in the collagen fiber orientation gradient field is analyzed, which is used as the inverse texture of the pixel; the inverse texture of the pixel is coupled with the high sugar edema compression coefficient to obtain the coupling impedance gain. The gain of the basic impedance and coupling impedance of the pixel stiffness analysis is superimposed and integrated along the connection path to obtain the local impedance of interstitial diffusion.
6. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The method for obtaining the regional drug diffusion resistance index includes: Within each sliding window, the drug supply point set and the active tissue receptor point set are respectively used as two vertex sets of a bipartite graph, and the interstitial diffusion local impedance is mapped as the connection weight between vertices. When the number of drug supply point sets and active tissue receptor point sets are not equal, virtual nodes are introduced to complete the bipartite graph into a complete graph, and the connection cost between virtual nodes and real nodes is set as a preset penalty threshold. The minimum weight matching algorithm is used to find the optimal matching scheme, and the average connection weight in the optimal matching scheme is used as the regional drug diffusion inhibition index for each sliding window.
7. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The method for obtaining the effective uptake rate of exosomes includes: The effective half-life under the given conditions was calculated using the baseline half-life of exosomes, the enzyme hydrolysis sensitivity coefficient, and the high glucose edema compression coefficient. The degree of attenuation was analyzed pixel by pixel in conjunction with the regional drug diffusion retardation index and the effective half-life to obtain the effective uptake rate of exosomes.
8. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The attribution and generation of instructions based on tissue density stiffness maps and high glucose edema compression coefficients for low exosome effective uptake rates includes: Connected regions with exosome effective uptake rates less than a preset effective threshold baseline are identified as inefficient blocking regions. If the compression coefficient of high-glucose edema in the inefficient blockade zone is greater than the preset edema threshold, an intervention instruction for edema is generated; if the compression coefficient of high-glucose edema in the inefficient blockade zone is less than or equal to the preset edema threshold and the average tissue density stiffness is greater than the preset stiffness threshold, an intervention instruction for physical barrier is generated. Otherwise, intervention instructions related to dose adjustment or assisted delivery are generated.
9. The deep learning-based exosome scar inhibition effect prediction system according to claim 8, characterized in that, The method of outputting a wound drug delivery efficiency map by superimposing an exosome effective uptake rate mapping map onto a wound micro-motion video stream includes: The effective uptake rate of exosomes is mapped to a pseudo-color encoded heatmap; a semi-transparent overlay technique is used to overlay the heatmap onto the first frame of the wound micro-motion video stream, and corresponding intervention instructions are simultaneously marked in the inefficient blockage area to form a wound drug delivery efficiency map.
10. The deep learning-based exosome scar inhibition effect prediction system according to claim 1, characterized in that, The method for obtaining the high-glucose edema compression coefficient includes: The hyperglycemic edema compression coefficient is obtained based on the degree of deviation between the patient's current blood glucose data and the preset benchmark blood glucose value.