Drug distribution and metabolism dynamic analysis system based on multispectral microscopic image

Abstract

The invention relates to the field of medical image analysis, in particular to a drug distribution and metabolism dynamic analysis system of a multispectral microscopic image, which comprises a light source module, a drug distribution module, a drug distribution module, a drug metabolism dynamic analysis module and a drug metabolism dynamic analysis module, and is characterized in that the light source module emits a multiband light source comprising broadband white light and time sequence switching narrow-band monochromatic exciting light; the data acquisition module is used for acquiring a multi-channel spectral image signal and generating a multi-spectral image containing a spatial dimension and a spectral dimension; the pharmacokinetic analysis module constructs a geometric characterization framework of drug metabolism based on a differential geometry theory, and comprises a Riemannian manifold characterization unit for mapping multispectral data to a Riemannian manifold space, a curvature tensor analysis unit for calculating manifold curvature characteristics and a parallel transmission optimization unit for evaluating drug transmission efficiency; the application analysis module carries out drug target co-localization analysis, space-time metabolic map analysis and delivery system optimization based on the analysis result, the spectrum channel limitation of traditional fluorescence imaging is broken through, and the precision and dimension of drug metabolism analysis are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical image analysis, in particular to a multi-spectral microscopic image drug distribution and metabolism dynamic analysis system, which is suitable for dynamic analysis of drug distribution and metabolism in the fields of drug research and development, pharmacodynamic research, and personalized medicine. BACKGROUND

[0002] In modern drug research and development and clinical application, accurate understanding of the distribution and metabolism dynamic process of drugs in the body is of great significance for evaluating drug efficacy, optimizing drug administration schemes, and guiding personalized treatment. Traditional drug metabolism kinetic analysis mainly relies on sampling and determination of biological samples, which can only obtain average concentration change information of drugs in the body, and it is difficult to intuitively present the distribution characteristics and metabolism process of drugs at the microscopic level.

[0003] With the development of fluorescence microscopic imaging technology, researchers can track the distribution of fluorescently labeled drug molecules in cells and tissues. However, traditional fluorescence imaging usually only uses 2-4 wavelength channels, which has limited information and is difficult to distinguish between drug substances and their metabolites, let alone accurate quantitative analysis. In addition, existing analysis methods are mostly based on simple signal intensity changes, lacking in-depth understanding and accurate characterization of drug metabolism processes.

[0004] In recent years, the emergence of multi-spectral imaging technology has provided a new idea for drug analysis. By obtaining the response information of the sample in multiple spectral channels, the information density and discrimination ability of the analysis can be greatly improved. However, how to effectively extract drug distribution and metabolism information from complex multi-spectral data, and how to quantitatively evaluate the drug transport process, are still important technical problems faced by the field. SUMMARY

[0005] The purpose of the present application is to provide a multi-spectral microscopic image drug distribution and metabolism dynamic analysis system, which innovatively introduces differential geometry theory to construct a high-dimensional Riemannian manifold drug metabolism characterization framework, realizing high-precision, multi-dimensional analysis of drug distribution and metabolism dynamic process in cells and tissues.

[0006] The present application provides a multi-spectral microscopic image drug distribution and metabolism dynamic analysis system, which comprises:

[0007] A light source module for emitting a multi-band light source, the multi-band light source comprising a wide-band white light for basic illumination and a plurality of narrow-band monochromatic excitation lights working in a time-sequential manner;

[0008] A data acquisition module connected to the light source module for acquiring multi-channel spectral image signals generated after the multi-band light source irradiates the sample, generating a multi-spectral image containing spatial and spectral dimensions;

[0009] a pharmacokinetics analysis module connected with the data acquisition module, configured to receive the multi-spectral images and construct a geometric characterization framework of drug metabolism based on differential geometry theory, the pharmacokinetics analysis module comprising:

[0010] a Riemannian manifold characterization unit configured to map the multi-spectral image data to a high-dimensional Riemannian manifold space and construct a geometric characterization of the drug metabolism process;

[0011] a curvature tensor analysis unit connected with the Riemannian manifold characterization unit, configured to calculate the curvature characteristics of the Riemannian manifold and generate a geometric fingerprint of drug metabolism;

[0012] a parallel transport optimization unit connected with the curvature tensor analysis unit, configured to construct an ideal drug transport geodesic line, evaluate the deviation between the actual drug transport path and the ideal drug transport geodesic line, and calculate a delivery system efficiency index; and

[0013] an application analysis module connected with the pharmacokinetics analysis module, configured to perform drug target co-localization analysis, spatiotemporal metabolic map analysis, and delivery system optimization based on the analysis results of the pharmacokinetics analysis module.

[0014] Preferably, the Riemannian manifold characterization unit comprises:

[0015] a data preprocessing component configured to perform noise removal, background correction, and spectral normalization processing on the multi-spectral images;

[0016] a manifold mapping component connected with the data preprocessing component, configured to vectorize and map the processed multi-spectral data to a high-dimensional Riemannian manifold space;

[0017] a local coordinate system construction component connected with the manifold mapping component, configured to establish a local tangent space and an orthogonal basis vector field on the Riemannian manifold; and

[0018] a drug metabolism trajectory extraction component connected with the local coordinate system construction component, configured to track the evolution trajectory of drug metabolism on the Riemannian manifold based on multi-spectral images at multiple time points.

[0019] Preferably, the manifold mapping component adopts a manifold learning method based on adjacency relationships to establish a manifold embedding representation that preserves local geometric structure, wherein the number of connected neighbors for each data point is 8 to 12.

[0020] Preferably, the curvature tensor analysis unit comprises:

[0021] a multi-scale curvature calculation component configured to calculate the curvature characteristics of the drug metabolism manifold at three spatial scales: micro, meso, and macro.

[0022] a principal curvature extraction component, connected with the multi-scale curvature calculation component, for extracting principal curvature values and principal directions of the drug metabolism manifold;

[0023] a curvature feature mapping component, connected with the principal curvature extraction component, for generating a spatial distribution curvature atlas of the drug metabolism activity; and

[0024] a metabolism fingerprinting component, connected with the curvature feature mapping component, for extracting time-varying features from the multi-time point curvature data, and constructing a metabolism fingerprint specific to the drug.

[0025] As preferred, the multi-scale curvature calculation component adopts a three-level calculation window size: a micro-scale window of 5x5 pixels for capturing fine structures within cells; a meso-scale window of 15x15 pixels for analyzing intercellular interactions; and a macro-scale window of 30x30 pixels for describing drug distribution at the tissue level.

[0026] As preferred, the parallel transmission optimization unit comprises:

[0027] a geodesic line calculation component for constructing an energy functional on the Riemannian manifold based on the administration location and the target site, and solving an optimal transmission path to generate an ideal drug transmission geodesic line;

[0028] a path extraction component for tracking a moving trajectory of an actual drug signal from time-series data of the multi-spectral images, and generating an actual drug transmission path;

[0029] a deviation analysis component, connected with the geodesic line calculation component and the path extraction component, for calculating a spatial deviation between the actual drug transmission path and the ideal drug transmission geodesic line; and

[0030] an efficiency evaluation component, connected with the deviation analysis component, for constructing a delivery system efficiency index based on the spatial deviation.

[0031] As preferred, the light source module comprises:

[0032] a wide-band white light source for providing basic illumination;

[0033] a narrow-band filter set, connected with the wide-band white light source, for filtering a plurality of monochromatic lights with a wavelength range of 400-700 nm from the wide-band white light;

[0034] a light source controller, connected with the narrow-band filter set, for controlling switching work of the plurality of monochromatic excitation lights in a time-series manner; and

[0035] a dichroic mirror connected with the wide-band white light source and the narrow-band filter set, for guiding the wide-band white light and the monochromatic excitation light to the sample.

[0036] As preferred, the data acquisition module comprises:

[0037] a charge-coupled device for capturing fluorescent signals emitted by the sample;

[0038] a spectral separation device connected with the charge-coupled device for separating the fluorescent signals into multiple wavelength channels;

[0039] an image acquisition controller connected with the charge-coupled device and the spectral separation device for controlling image acquisition timing, so that each monochromatic excitation wavelength corresponds to a detection channel; and

[0040] a data integrator connected with the image acquisition controller for integrating signals of the multiple detection channels into a multispectral image with a spectral dimension of 16-33.

[0041] As preferred, the application analysis module comprises:

[0042] a drug-target co-localization analysis component for quantitatively evaluating the binding of a drug to a specific target based on the metabolic fingerprint generated by the curvature tensor analysis unit;

[0043] a spatiotemporal metabolic map analysis component for generating a spatiotemporal dynamic change map of a drug at the tissue, cell, and subcellular levels based on the metabolic trajectory extracted by the Riemannian manifold representation unit;

[0044] a drug metabolic pathway analysis component for quantifying the dose-time pathway of drug metabolism in vivo based on changes in the multispectral image at different time points; and

[0045] a delivery system optimization component for parameter optimizing a drug delivery system based on the efficiency index calculated by the parallel transport optimization unit.

[0046] As preferred, the delivery system optimization component comprises:

[0047] a bottleneck identification unit for analyzing the spatial distribution of the efficiency index and identifying key links of delivery efficiency loss;

[0048] a parameter tuning unit connected with the bottleneck identification unit for generating delivery system parameter optimization suggestions based on the key links;

[0049] a performance prediction unit connected with the parameter tuning unit for simulating the transmission efficiency of the optimized delivery system under different conditions; and

[0050] An iterative optimization unit, connected with the performance prediction unit, is configured to further adjust the delivery system parameters based on the simulation result of the transmission efficiency, so as to gradually improve the delivery efficiency.

[0051] The beneficial effects of the present application include:

[0052] 1. Breaking through the spectral channel limit of traditional fluorescence imaging, realizing high-dimensional data acquisition of 16-33 spectral channels, greatly improving the information density and discrimination ability, and accurately distinguishing the drug entity and its metabolites.

[0053] 2. Innovatively applying differential geometry theory to drug metabolism kinetics analysis, establishing a drug metabolism characterization framework based on Riemannian manifold, and fundamentally improving the characterization accuracy and depth of drug metabolism process.

[0054] 3. Constructing drug metabolism fingerprint through curvature tensor analysis, which can capture the interaction characteristics of drugs and different cell structures, identify unknown metabolic pathways, and provide a new perspective for drug mechanism research.

[0055] 4. Based on the parallel transmission theory, a delivery system efficiency evaluation framework is established, which realizes the objective quantification of the performance of the drug delivery system and provides a theoretical basis for the optimization of the delivery system.

[0056] 5. Significantly improving the accuracy of drug-target co-localization (from 74% of traditional methods to 92%), increasing the sensitivity of metabolic pathway identification by 3.5 times, and improving the time resolution from minutes to seconds, providing strong technical support for drug research and precision medicine. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 It is the overall structure block diagram of the multi-spectral microscopic image drug distribution and metabolism dynamic analysis system of the present application;

[0058] Figure 2 It is the structure block diagram of the drug metabolism kinetics analysis module of the present application;

[0059] Figure 3 It is the structure block diagram of the Riemannian manifold characterization unit of the present application;

[0060] Figure 4 It is the structure block diagram of the curvature tensor analysis unit of the present application;

[0061] Figure 5 It is the structure block diagram of the parallel transmission optimization unit of the present application. DETAILED DESCRIPTION

[0062] Please refer to Figures 1-5The application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are only used to more clearly illustrate the technical solutions of the application, and cannot be used to limit the protection scope of the application.

[0063] Please refer to Figure 1 The multi-spectral microscopic image drug distribution and metabolism dynamic analysis system provided by the application comprises a light source module 1, a data acquisition module 2, a drug metabolism kinetics analysis module 3 and an application analysis module 4. The light source module 1 is connected with the data acquisition module 2, the data acquisition module 2 is connected with the drug metabolism kinetics analysis module 3, and the drug metabolism kinetics analysis module 3 is connected with the application analysis module 4.

[0064] The light source module 1 is used for emitting a multi-waveband light source, wherein the multi-waveband light source comprises a wide-waveband white light for basic illumination and a plurality of narrow-waveband monochromatic excitation lights which are switched in a time sequence. The data acquisition module 2 is used for acquiring multi-channel spectral image signals generated after the multi-waveband light source irradiates a sample, and generating a multi-spectral image containing spatial dimensions and spectral dimensions. The drug metabolism kinetics analysis module 3 is used for receiving the multi-spectral image, and constructing a geometric characterization framework of drug metabolism based on differential geometry theory. The application analysis module 4 is used for performing drug target co-localization analysis, spatiotemporal metabolism map analysis and delivery system optimization based on the analysis result of the drug metabolism kinetics analysis module 3.

[0065] Preferably, the light source module 1 comprises a wide-waveband white light source, a narrow-band filter set, a light source controller and a dichroic mirror. The wide-waveband white light source is used for providing basic illumination. The narrow-band filter set is connected with the wide-waveband white light source, and is used for filtering a plurality of monochromatic lights with a wavelength range of 400nm-700nm from the wide-waveband white light. The light source controller is connected with the narrow-band filter set, and is used for controlling the switching work of the plurality of monochromatic excitation lights in a time sequence. The dichroic mirror is connected with the wide-waveband white light source and the narrow-band filter set, and is used for guiding the wide-waveband white light and the monochromatic excitation lights to the sample.

[0066] In an embodiment of the application, the light source controller adopts a programmable controller, which can control the switching time sequence of the narrow-band filter set according to a preset program. Preferably, the narrow-band filter set comprises 16-33 narrow-band filters with different central wavelengths, which correspond to 16-33 different excitation wavelengths respectively. The light source controller controls the switching time interval of each filter, which can be set to 0.1 seconds per channel in a fast imaging mode, and can be set to 0.5 seconds per channel in a high signal-to-noise ratio mode. The dichroic mirror preferably adopts a special coating with a reflectivity varying with wavelength, which can efficiently reflect the excitation light to the sample, and at the same time let the fluorescence emitted by the sample pass through into the data acquisition module 2.

[0067] The data acquisition module 2 comprises a charge-coupled device, a spectral separation device, an image acquisition controller and a data integrator. The charge-coupled device is used to capture the fluorescent signal emitted by the sample. The spectral separation device is connected to the charge-coupled device and is used to separate the fluorescent signal into multiple wavelength channels. The image acquisition controller is connected to the charge-coupled device and the spectral separation device and is used to control the image acquisition timing, so that each monochromatic excitation wavelength corresponds to the formation of a detection channel. The data integrator is connected to the image acquisition controller and is used to integrate the signals of the multiple detection channels into a multispectral image with a spectral dimension of 16-33.

[0068] In an embodiment of the present application, the charge-coupled device uses a high-sensitivity scientific-grade CMOS camera with a resolution of 2048x2048 pixels, a dynamic range of 16 bits, and a maximum frame rate of 30 frames / second. The spectral separation device preferably uses an acousto-optic tunable filter (AOTF) or a liquid crystal tunable filter (LCTF), which can quickly switch between different wavelength channels without mechanical moving parts. The image acquisition controller works synchronously with the light source controller of the light source module 1 to ensure that the image acquisition of each excitation channel is perfectly matched with the corresponding excitation light. The data integrator organizes the images of each channel in spectral order to construct a four-dimensional data cube (x, y, λ, t), where x and y represent spatial coordinates, λ represents wavelength, and t represents time.

[0069] As shown in Figure 2 The pharmacokinetic analysis module 3 is the core innovative module of the present application, comprising a Riemannian manifold characterization unit 31, a curvature tensor analysis unit 32 and a parallel transport optimization unit 33. The Riemannian manifold characterization unit 31 is used to map the multispectral image data to a high-dimensional Riemannian manifold space to construct a geometric characterization of the drug metabolism process. The curvature tensor analysis unit 32 is connected to the Riemannian manifold characterization unit 31 and is used to calculate the curvature characteristics of the Riemannian manifold to generate a geometric fingerprint of drug metabolism. The parallel transport optimization unit 33 is connected to the curvature tensor analysis unit 32 and is used to construct an ideal drug transport geodesic line, evaluate the deviation between the actual drug transport path and the ideal drug transport geodesic line, and calculate the delivery system efficiency index.

[0070] Please refer to Figure 3 The Riemannian manifold characterization unit 31 comprises a data preprocessing component 311, a manifold mapping component 312, a local coordinate system construction component 313 and a drug metabolism trajectory extraction component 314.

[0071] The data preprocessing component 311 is configured to perform noise removal, background correction and spectral normalization on the multispectral images. In one embodiment of the present application, the data preprocessing component 311 first removes noise by combining adaptive median filtering and wavelet transform, then performs pixel-level difference correction by establishing a background model using drug-free control samples, and finally normalizes the intensity of each spectral channel to eliminate the influence of device sensitivity differences. Preferably, the wavelet transform uses Daubechies wavelet basis, the number of decomposition layers is 3-5 layers, the threshold selection uses soft threshold method, and the threshold is set to 3 times the standard deviation of the noise. The background correction uses a dynamic background modeling method to adapt to the autofluorescence characteristics of different samples.

[0072] The manifold mapping component 312 is connected to the data preprocessing component 311 and is configured to vectorize the processed multispectral data and map it to a high-dimensional Riemannian manifold space. Specifically, the manifold mapping component 312 first organizes the n spectral channel data at each spatial position (x, y) into an n-dimensional vector to construct a data matrix D, where the rows represent spatial positions and the columns represent spectral dimensions. Then, principal component analysis is applied for preliminary dimension reduction, retaining principal components that explain 95% of the variance.

[0073] Preferably, the manifold mapping component 312 uses a manifold learning method based on adjacency relationships to establish a manifold embedding representation that preserves local geometric structure, with each data point connected to 8-12 nearest neighbors. In one embodiment of the present application, the local linear embedding (LLE) algorithm is used to implement manifold learning, which is implemented by the following steps:

[0074] 1. Calculate the k nearest neighbors of each data point , preferably k = 10;

[0075] 2. Calculate the weight matrix W representing the linear combination of the nearest neighbors of each data point:

[0076] ,

[0077] where is the spectral vector of the i-th data point, representing the n-dimensional spectral data at the i-th spatial position; is an element of the weight matrix, representing the contribution weight of the j-th nearest neighbor to the i-th data point, and satisfies ; represents the Euclidean norm, used to measure the distance between vectors; represents the minimization objective function.

[0078] 3. Solve the corresponding embedding coordinates in the low-dimensional space:

[0079] ​ ,

[0080] wherein, is the coordinate of the i-th data point in the manifold space, representing the d-dimensional coordinate after dimension reduction of the original n-dimensional spectral data (usually d < n); the weight matrix is the same as in step 2, maintaining the local geometric relationship between data points.

[0081] The local coordinate system construction component 313 is connected with the manifold mapping component 312, and is used to establish a local tangent space and an orthogonal basis vector field on the Riemannian manifold. For each point p on the manifold, the local coordinate system construction component 313 constructs a tangent space as a local linear approximation space, and establishes an orthogonal basis vector field in the space. Specifically, for points p and q on the manifold, the geodesic from p to q is defined as Then, the local coordinate system transformation is realized by parallel transmission.

[0082] In an embodiment of the present application, the Riemannian metric tensor is used to describe the distance metric on the manifold, and is calculated in the following way:

[0083] ,

[0084] wherein, is the Riemannian metric tensor at point p, which is a dxd symmetric positive definite matrix, describing the local distance metric on the manifold; and are the basis vectors in the tangent space along the i-th and j-th coordinate directions, respectively; represents the inner product operation at point p, defining the angle and length between vectors.

[0085] The drug metabolism trajectory extraction component 314 is connected with the local coordinate system construction component 313, and is used to track the evolution trajectory of drug metabolism on the Riemannian manifold based on multiple multispectral images at multiple time points. Specifically, the drug metabolism trajectory extraction component 314 first identifies the initial drug distribution area and marks the starting point set. Then, for each starting point, the evolution path thereof over time is tracked in the manifold space to generate a drug metabolism trajectory γ(t).

[0086] In a preferred embodiment of the present application, in order to reduce the influence of noise, the trajectory is smoothed by using a locally weighted regression smoothing (LOESS) method, the window size is set to 15% of the length of the time series, and the weight function adopts cubic weighting:

[0087] ,

[0088] wherein, is the weight function, representing the weight of data points with distance x; x is the normalized distance, ranging from [-1, 1], representing the distance from the current point to the center point divided by the window radius; represents the absolute value of x. This weight function has the maximum weight (equal to 1) at the center point, and the weight gradually decreases as the distance increases, and the weight is 0 at the window edge.

[0089] In addition, the drug metabolism trajectory extraction component 314 also calculates the speed and acceleration features of each point on the trajectory, through the covariant differential operator The drug concentration gradient is calculated to ensure the invariance of the measurement under different coordinate systems:

[0090] ,

[0091] where, represents the covariant derivative of the vector field Y along the vector field X; X and Y are vector fields on the manifold, representing the direction and rate of change of drug concentration, respectively; and are the components of the vector fields X and Y in the i-th and j-th coordinate directions, respectively; represents the directional derivative of the k-th component of the vector field X acting on Y; is the Christoffel symbol, representing the connection coefficient in the curvilinear coordinate system, describing the rate of change of the coordinate system; represents the basis vector in the tangent space along the k-th coordinate direction; represents the summation over all coordinate directions k; represents the summation over all coordinate directions i and j.

[0092] In order to identify the main metabolic patterns, the drug metabolism trajectory extraction component 314 performs clustering analysis on similar trajectories. Preferably, a hierarchical clustering method based on dynamic time warping (DTW) is used, and the distance metric is defined as:

[0093] ,

[0094] where, represents the DTW distance between trajectory A and trajectory B; and represent two different drug metabolism trajectories, respectively, where and are the points on the trajectory; is the correspondence function from trajectory A to trajectory B, mapping each point in A to a certain point in B; is the local distance between point and point , usually using the Euclidean distance; denotes the selection of the one among all possible correspondence functions f that minimizes the total distance; denotes the summation of all n points in trajectory A. The clustering threshold is set to 50% of the average trajectory distance.

[0095] Please refer to Figure 4 The curvature tensor analysis unit 32 includes a multi-scale curvature calculation component 321, a principal curvature extraction component 322, a curvature feature mapping component 323, and a metabolic fingerprint identification component 324.

[0096] The multi-scale curvature calculation component 321 is used to calculate the curvature features of the drug metabolism manifold at three spatial scales: micro, meso, and macro. In a preferred embodiment of the present application, the multi-scale curvature calculation component 321 adopts a three-level calculation window size: a micro-scale window of 5x5 pixels is used to capture the fine structure within cells; a meso-scale window of 15x15 pixels is used to analyze the intercellular interaction; and a macro-scale window of 30x30 pixels is used to describe the drug distribution at the tissue level.

[0097] For each local region on the manifold, the multi-scale curvature calculation component 321 calculates the Riemann curvature tensor , which characterizes the degree of local spatial bending:

[0098] ,

[0099] wherein, is the Riemann curvature tensor, which is a fourth-order tensor representing the bending degree of the manifold in various directions; denotes the partial derivative along the jth coordinate direction; is the Christoffel symbol, which represents the connection coefficient in the curvilinear coordinate system; denotes the product of two Christoffel symbols and sums over the index m; similarly also denotes the product of two Christoffel symbols and sums over the index m.

[0100] The Christoffel symbol is defined as:

[0101] ,

[0102] wherein, is the Christoffel symbol, which describes the rate of change of the coordinate system; is the Riemann metric tensor, which describes the local distance metric on the manifold; is the inverse matrix element of ; denotes the partial derivative of the metric tensor element with respect to the kth coordinate. is the coefficient; the three terms in parentheses represent the rate of change of the metric tensor in different coordinate directions, and their combination describes the degree of distortion of the coordinate system.

[0103] The principal curvature extraction component 322 is connected to the multi-scale curvature calculation component 321 to extract the principal curvature values ​​and principal directions of the drug metabolism manifold. The principal curvature extraction component 322 calculates the Ricci curvature tensor using a condensed Riemann curvature tensor. :

[0104] ,

[0105] in, Let Ricci curvature tensor be a second-order tensor that represents the average curvature of the manifold in all directions. Let be the components of the Riemann curvature tensor, where the superscript k indicates that the first index is raised, i.e. ,in The inverse matrix elements of the metric tensor; This represents the summation over all coordinate directions k, achieving the reduction from the Riemann curvature tensor to the Ricci curvature tensor.

[0106] Furthermore, the scalar curvature is calculated. :

[0107] ,

[0108] in, Scalar curvature is a scalar value that represents the average degree of curvature of the manifold in all directions; The inverse matrix elements of the metric tensor; For Ricci curvature tensor elements; This represents the summation over all coordinate directions i and j, achieving the reduction of the Ricci curvature tensor to scalar curvature.

[0109] For a two-dimensional manifold, the principal curvatures and Obtained by solving the characteristic equation:

[0110] ,

[0111] in, Represents the determinant of a matrix; The shape operator, determined by the second and first fundamental forms, is a representation of the local shape of a surface. matrix; These are the eigenvalues, i.e., the principal curvatures; for The identity matrix; the entire equation represents finding the matrix. The determinant of the matrix is ​​equal to 0, which means finding the matrix. The eigenvalues ​​are represented by the principal directions, which correspond to the eigenvectors and indicate the directions of maximum and minimum curvature of the surface.

[0112] In one embodiment of the present invention, when a principal curvature mutation exceeds a preset threshold, the system marks it as a key point in drug metabolism, corresponding to a specific metabolic event within the cell. Preferably, the curvature threshold is set to... This value is based on a large amount of experimental data and can effectively identify more than 95% of key metabolic events while keeping the false alarm rate below 5%.

[0113] The curvature feature mapping component 323 is connected to the principal curvature extraction component 322 to generate a spatial distribution curvature map of drug metabolic activity. The curvature feature mapping component 323 visualizes the spatial distribution of scalar curvature $R$, intuitively showing the differences in drug metabolic activity in different cellular regions, while the direction of the principal curvature indicates the dominant pathway for drug transport.

[0114] In a preferred embodiment of the present invention, the curvature feature mapping is represented by pseudo-color, and the curvature value range [-1,1] is mapped to a rainbow color map. The negative curvature region (concave region) represents the drug accumulation region, the positive curvature region (convex region) represents the drug diffusion region, and the region with curvature close to zero represents the drug metabolism stable region.

[0115] The metabolic fingerprint recognition component 324 is connected to the curvature feature mapping component 323 to extract time-varying features from curvature data at multiple time points, thereby constructing a drug-specific metabolic fingerprint. A metabolic fingerprint is a set of geometric features representing the drug's metabolic process, uniquely characterizing its metabolic properties.

[0116] In one embodiment of the present invention, the metabolic fingerprint construction process includes: firstly, extracting time-varying features from curvature data at multiple time points; then constructing a feature vector sequence to characterize the spatiotemporal pattern of drug metabolism; and finally extracting key metabolic markers to form a drug-specific metabolic fingerprint.

[0117] Metabolic fingerprint vector Defined as:

[0118] ,

[0119] in, This is a metabolic fingerprint vector, containing multiple features characterizing drug metabolism. and These are the mean and variance of the scalar curvature, respectively, representing the overall level and degree of fluctuation of the curvature; , , , These are the two principal curvatures. and The mean and variance of the surface characterize the bending properties of the surface in the principal direction; and The dominant direction of the main curvature direction, usually represented by an angle, characterizes the dominant direction of drug transport; The time characteristics of the curvature key points represent the time points when key events occur during drug metabolism.

[0120] Preferably, the metabolic fingerprint recognition component 324 also establishes a metabolic fingerprint database of known drugs, and evaluates the similarity between unknown drugs and known patterns by designing a similarity metric function. The similarity function is defined as the weighted Euclidean distance:

[0121] ,

[0122] in, This represents the distance between two metabolic fingerprint vectors; a smaller value indicates a higher similarity. and There are two metabolic fingerprint vectors, each containing multiple feature values; and They are vectors and The One element; The weight represents the importance of that feature in the similarity calculation. This represents summing over all features; This represents the square root operation. Based on experience, the weight of the curvature feature is set to 0.4, and the weight of the time feature is set to 0.6.

[0123] Please refer to Figure 5 The parallel transmission optimization unit 33 includes a geodesic calculation component 331, a path extraction component 332, a deviation analysis component 333, and an efficiency evaluation component 334.

[0124] The geodesic calculation component 331 is used to construct an energy functional on the Riemannian manifold based on the drug delivery location and target point, and solve for the optimal transport path to generate an ideal drug transport geodesic. The ideal transport path represents the optimal path for the drug from the drug delivery location to the target point without additional obstruction.

[0125] In one embodiment of the present invention, a geodesic line Solve by minimizing the energy functional:

[0126] ,

[0127] in, Let be the energy functional of the geodesic, and let represent the total energy of the curve γ(t); Riemannian metric tensor describing the local distance metric on the manifold; γ(t) is a parametrized curve representing a path from parameter a to b; γi represents the i-th coordinate component of the curve γ; and denote the i-th and j-th coordinate components of the velocity of the curve, respectively; denote the i-th and j-th coordinate components of the velocity of the curve, respectively;

[0128] By the variational method, the Euler-Lagrange equation can be obtained:

[0129] ,

[0130] where, denotes the second derivative of the k-th coordinate component of the curve γ, i.e., the acceleration; is the Christoffel symbol, representing the connection coefficients in the coordinate system of the curve; and denote the i-th and j-th coordinate components of the velocity of the curve, respectively; denotes the sum over all coordinate directions i and j. This equation describes the evolution of the geodesic, representing the natural trajectory of an object on the manifold under the action of no external force.

[0131] Preferably, the equation is numerically solved using the fourth-order Runge-Kutta method with adaptive step size, with a step size range of 0.01-0.1 and a relative error controlled within 10-6.

[0132] In order to better simulate the actual physiological environment, the geodesic calculation component 331 also considers the influence of transport barriers. Identify the transport barriers in the tissue (such as cell membranes, blood-brain barrier), and assign a high resistance value to the barrier region. In the modified Riemannian metric, the metric tensor of the barrier region is multiplied by a resistance factor

[0133] ,

[0134] where, is the modified Riemannian metric tensor; is the resistance factor, which is a positive real number representing the resistance of the barrier region; is the original Riemannian metric tensor. For normal regions , for slightly obstructed regions , for moderately obstructed regions , and for severely obstructed regions . These thresholds are based on experimental data and can reasonably reflect the penetration difficulty of different biological barriers.

[0135] The path extraction component 332 is configured to track the moving trajectory of the actual drug signal from the time-series data of the multispectral images, and generate the actual drug transport path. Specifically, the path extraction component 332 tracks the actual moving path of the drug in the body by analyzing the changes of the drug signal in the multispectral data at multiple time points.

[0136] In one preferred embodiment of the present application, the path extraction component 332 employs a particle filter algorithm for drug trajectory tracking, which is particularly suitable for target tracking under low signal-to-noise ratio conditions. Particle filtering is based on the Monte Carlo sampling method, which represents the posterior probability distribution by a set of weighted particles. Preferably, the number of particles is set to 1000, and the resampling threshold is set to 50% of the number of effective particles. The state transition model employs a constant velocity model, and the observation model is based on the drug fluorescence intensity.

[0137] The deviation analysis component 333 is connected with the geodesic line calculation component 331 and the path extraction component 332, and is configured to calculate the spatial deviation between the actual drug transport path and the ideal drug transport geodesic line. Specifically, the deviation analysis component 333 calculates the point-to-point distance between the actual path and the ideal geodesic line, and constructs a cumulative deviation curve.

[0138] In one embodiment of the present application, given the actual path and the ideal geodesic line , the path deviation at each time point t is calculated as:

[0139] ,

[0140] wherein, denotes the path deviation at time point ; denotes the actual path point at time point ; denotes the ideal geodesic line point at time point ; and d(·,·) is the geodesic distance on the manifold, which is used to measure the distance between two points.

[0141] The cumulative deviation is defined as:

[0142] ,

[0143] wherein, is the cumulative deviation, which represents the total deviation between the actual path and the ideal path during the entire transport process; denotes the integral from time 0 to T; is the path deviation at time t; and T is the total time of the transport process.

[0144] Preferably, further key statistical features are extracted: the maximum deviation , the average deviation , the rate of change of deviation These indicators collectively represent the transmission accuracy of the delivery system. Among them, is the maximum value of the path deviation, representing the maximum deviation degree; is the average value of the path deviation, representing the overall deviation degree; is the rate of change of deviation, i.e. the derivative of deviation with respect to time, representing the deviation speed.

[0145] The efficiency evaluation component 334 is connected with the deviation analysis component 333, and is used to construct a delivery system efficiency index based on the spatial deviation. The efficiency index is a dimensionless parameter that comprehensively evaluates the performance of the delivery system, with a range of [0, 1], and the closer the value is to 1, the closer the delivery system is to the ideal transmission effect.

[0146] In an embodiment of the present application, the efficiency index is defined as:

[0147] ,

[0148] Among them, is the efficiency index, with a range of [0, 1], representing the comprehensive performance of the delivery system; is the accuracy index, representing the path accuracy; is the timeliness index, representing the transmission speed; is the targeting specificity index, representing the targeting accuracy; , , is the weight coefficient, representing the importance of each indicator, and satisfies .

[0149] The accuracy index is defined as:

[0150] ,

[0151] Among them, is the accuracy index, with a range of [0, 1], and the larger the value, the higher the accuracy; is the cumulative deviation; is the preset maximum allowed deviation, which is usually set as the maximum deviation value observed in the experiment.

[0152] The timeliness index is defined as:

[0153] ,

[0154] Among them, is the timeliness index, which is usually in the range of (0, ∞), but is usually limited to the range of [0, 1] in actual application; Reference transit time, usually set as the ideal transit time; Actual transit time, refers to the time required for the drug to reach the target from the administration site.

[0155] Targeting specificity index is defined as:

[0156] ,

[0157] wherein, is the targeting specificity index, ranging from [0, 1], the larger the value, the higher the targeting specificity; is the non-target region drug concentration, refers to the average concentration of the drug in the non-target region; is the target region drug concentration, refers to the average concentration of the drug in the target region.

[0158] Preferably, the weight coefficient is adjusted according to the specific application scenario, for applications that require precise targeting, , for applications that require rapid onset of action, .

[0159] The application analysis module 4 includes a drug-target co-localization analysis component, a spatiotemporal metabolic map analysis component, a drug metabolic pathway analysis component, and a delivery system optimization component.

[0160] The drug-target co-localization analysis component is used to quantitatively evaluate the binding of the drug to the specific target based on the metabolic fingerprint generated by the curvature tensor analysis unit 32. Specifically, the drug-target co-localization analysis component decomposes the mixed drug / target co-localization spectral components onto the image of each monochromatic channel through spectral unmixing technology, and then performs cell distribution and subcellular localization analysis of the drug fluorescence signal.

[0161] In an embodiment of the present application, the drug-target co-localization coefficient is defined as:

[0162] ,

[0163] wherein, is the drug-target co-localization coefficient, ranging from [-1, 1], the closer to 1, the higher the co-localization degree; is the drug fluorescence intensity at position i; is the target fluorescence intensity at position i; indicates the summation over all spatial positions i; the numerator part calculates the dot product of the drug signal and the target signal, indicating their covariance; the denominator part is the geometric mean of the sum of squares of the two signal intensities, for normalization.

[0164] Preferably, the drug-target co-localization analysis component further employs machine learning techniques to improve the localization accuracy. In one example, a support vector machine (SVM) classifier is employed, with input features including fluorescence intensity, curvature features, and time series features, and the output is the co-localization probability. The SVM employs a radial basis function (RBF) kernel with parameters C = 10, = 0.1, which are optimized by 10-fold cross-validation.

[0165] The spatiotemporal metabolic map analysis component is used to generate spatiotemporal dynamic change maps of the drug at the tissue, cell, and subcellular levels based on the metabolic trajectories extracted by the Riemannian manifold representation unit 31. Specifically, the spatiotemporal metabolic map analysis component employs image processing techniques to present the dynamic changes of the drug in the body, forming intuitive visual results.

[0166] In one preferred embodiment of the present application, the spatiotemporal metabolic map analysis component constructs three-level spatiotemporal maps: macroscopic tissue level (resolution 1-10 pm), mesoscopic cell level (resolution 0.2-1 pm), and microscopic subcellular level (resolution <0.2 pm). Through this multi-scale representation, the spatial distribution characteristics of drug metabolism can be comprehensively captured. The time resolution is adaptively adjusted according to the metabolic rate, with a second-level resolution for fast metabolic processes and a minute-level resolution for slow metabolic processes.

[0167] The drug metabolic pathway analysis component is used to quantify the dose-time pathway of drug metabolism in the body based on the changes of the multi-spectral images at different time points. The drug metabolic pathway analysis component uses spectral unmixing techniques to unmix the fluorescence product signals of drug metabolism and the fluorescence product signals of drug metabolite metabolism onto the images of each monochromatic channel, and then performs intensity analysis of the drug and drug metabolite.

[0168] In one embodiment of the present application, the drug metabolic pathway analysis component obtains the metabolism of the drug in the body at a plurality of consecutive time points, and uses machine learning and spectral unmixing techniques to analyze the drug metabolism rate for each time point of drug metabolism, to obtain a multi-spectral drug metabolism time-fluorescence intensity curve graph for evaluating the drug metabolism kinetics.

[0169] Preferably, the drug metabolism kinetics parameters are obtained by fitting a two-compartment model:

[0170]

[0171]

[0172] wherein, and represent the change rates of the drug concentrations in the central and peripheral compartments, respectively; and ​​Drug concentration in the central and peripheral compartments, respectively, in mg / L or μmol / L; Ktr is the transfer rate constant of the drug from the central to the peripheral compartment in 1 / h; Ktr is the transfer rate constant of the drug from the central to the peripheral compartment in 1 / h; Ktr is the transfer rate constant of the drug from the central to the peripheral compartment in 1 / h;

[0173] The delivery system optimization component is used to optimize the parameters of the drug delivery system based on the efficiency index calculated by the parallel transport optimization unit 33. The delivery system optimization component quantifies the delivery system in terms of cell penetration ability by distinguishing different formulation delivery systems, and distinguishes the metabolic rate difference between the delivery system drug and the free drug according to the quantification result of the fluorescent label and the fluorescence spectrum.

[0174] Preferably, the delivery system optimization component includes a bottleneck identification unit, a parameter tuning unit, a performance prediction unit, and an iterative optimization unit. The bottleneck identification unit is used to analyze the spatial distribution of the efficiency index and identify the key link of the loss of delivery efficiency. The parameter tuning unit is connected with the bottleneck identification unit and is used to generate a delivery system parameter optimization suggestion based on the key link. The performance prediction unit is connected with the parameter tuning unit and is used to simulate the transport efficiency of the optimized delivery system under different conditions. The iterative optimization unit is connected with the performance prediction unit and is used to further adjust the parameters of the delivery system based on the simulation result of the transport efficiency, so as to gradually improve the delivery efficiency.

[0175] In an embodiment of the present application, the key parameters of the delivery system include particle size distribution, surface charge, hydrophilic / hydrophobic balance, and targeting ligand density. The parameter tuning adopts the Response Surface Methodology to construct the mapping relationship between the parameters and the efficiency index. Preferably, the response surface model adopts a second-order polynomial:

[0176] ,

[0177] wherein, Efficiency index, is the dependent variable of the model; The i-th parameter, such as particle size, surface charge, etc., is the independent variable of the model; Constant term, indicating the value of the dependent variable when all independent variables are 0; First-order term coefficient, indicating the linear effect of the i-th parameter; Second-order term coefficient, indicating the quadratic effect of the i-th parameter; Interaction term coefficient, indicating the interaction effect of the i-th parameter and the j-th parameter; Total number of parameters; is the random error, representing the variation that the model cannot explain; represents the summation of all parameters i from 1 to k; represents the sum of the quadratic terms of all parameters; represents the summation of all parameter pairs (i, j) where i < j, that is, the interaction term.

[0178] The optimization of the delivery system is an iterative process. In each iteration, the parameters are adjusted according to the bottleneck analysis, the performance prediction unit simulates the efficiency after adjustment, and the iterative optimization unit further optimizes according to the simulation results until the efficiency index reaches the preset target or stabilizes without further improvement. Preferably, the iterative termination condition is that the efficiency improvement in three consecutive iterations is less than 1% or the preset maximum number of iterations (usually 10 times) is reached.

[0179] The following uses specific application examples to illustrate the working process and effects of the system of the present invention.

[0180] Example 1: Evaluation of the cell targeting of an anti-cancer drug

[0181] In this example, the system of the present invention is used to evaluate the targeting ability of a fluorescently labeled anti-cancer drug to cancer cell mitochondria. First, cancer cells are treated with the fluorescently labeled anti-cancer drug at a drug concentration of 10 μm and an incubation time of 0 - 24 hours. At time points of 0, 0.5, 1, 2, 4, 8, 12, and 24 hours, multi-spectral images are collected. The number of spectral channels is 24, the wavelength range is 400 - 700 nm, and the interval is 12.5 nm.

[0182] The light source module 1 excites the sample in a time-sequential manner, and the exposure time for each channel is 0.2 seconds. The data acquisition module 2 records the fluorescence images of each channel and constructs a four-dimensional data set (x, y, λ, t). Subsequently, the pharmacokinetic analysis module 3 processes the data: the Riemannian manifold characterization unit 31 maps the multi-spectral data to a 24-dimensional manifold space to construct a geometric characterization of drug metabolism; the curvature tensor analysis unit 32 calculates the curvature characteristics of the manifold to generate a drug metabolism fingerprint; the parallel transport optimization unit 33 analyzes the transport path of the drug from the cell membrane to the mitochondria and calculates the delivery efficiency index.

[0183] The results show that the drug gradually accumulates in the mitochondria within 4 hours, and the scalar curvature value in the mitochondrial region gradually increases from 0.05 to 0.32, indicating a significant accumulation rate of the drug in this region. The drug-target co-localization coefficient increases from 0.35 (1 hour) to 0.87 (4 hours) and then remains stable, indicating good mitochondrial targeting. The drug transport efficiency index is 0.78, higher than 0.45 of conventional drugs, indicating that the drug has excellent mitochondrial targeting ability.

[0184] Example 2: Optimization of the liposome delivery system

[0185] In this example, the system is applied to optimize a liposome drug delivery system. Four different formulations of fluorescently labeled liposomes (A-D) are prepared, differing mainly in lipid composition, surface charge, and PEGylation level. Delivery experiments are performed in an in vitro cell model, and multispectral time-lapse images are collected from 0-12 hours.

[0186] System analysis shows that the delivery efficiency index of formulation A is 0.42, with the main bottleneck being cellular uptake; the delivery efficiency index of formulation B is 0.56, with the main bottleneck being endosomal escape; the delivery efficiency index of formulation C is 0.63, with the main bottleneck being intracellular degradation; and the delivery efficiency index of formulation D is 0.58, with the main bottleneck being non-specific distribution.

[0187] Based on the bottleneck analysis, the delivery system optimization component optimizes the four formulations: adjusting the surface charge of formulation A (from -15 mV to -5 mV), enhancing the pH responsiveness of formulation B (increasing the DOPE content), improving the stability of formulation C (increasing the cholesterol content), and increasing the targeting ligand density of formulation D (from 0.2 to 0.5 per square nanometer).

[0188] After optimization, the delivery efficiency indexes of the four formulations are increased to 0.61, 0.72, 0.75, and 0.69, with an average increase of 25%. Further iterative optimization is based on formulation C, adjusting the lipid ratio and PEG length, and the final efficiency index reaches 0.83, an increase of 32% compared to the initial formulation. Experimental verification shows that the optimized liposome improves the drug delivery efficiency in vivo by about 30%, consistent with the system prediction results.

[0189] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A drug distribution and metabolic dynamics analysis system based on multispectral microscopic images, characterized in that, include: A light source module is used to emit multi-band light sources, including wide-band white light for basic lighting and multiple narrow-band monochromatic excitation lights that switch in a time sequence. The data acquisition module is connected to the light source module and is used to acquire the multi-channel spectral image signal generated after the multi-band light source irradiates the sample, and generate a multispectral image containing spatial and spectral dimensions. A drug metabolism kinetic analysis module, connected to the data acquisition module, is used to receive the multispectral images and construct a geometric characterization framework for drug metabolism based on differential geometry theory. The drug metabolism kinetic analysis module includes: Riemannian manifold characterization unit is used to map the multispectral image data to a high-dimensional Riemannian manifold space to construct a geometric characterization of the drug metabolism process; The curvature tensor analysis unit, connected to the Riemannian manifold characterization unit, is used to calculate the curvature characteristics of the Riemannian manifold and generate a geometric fingerprint of drug metabolism. A parallel transport optimization unit, connected to the curvature tensor analysis unit, is used to construct an ideal drug transport geodesic, evaluate the deviation between the actual drug transport path and the ideal drug transport geodesic, and calculate the delivery system efficiency index; and An application analysis module, connected to the pharmacokinetic analysis module, is used to perform drug target colocalization analysis, spatiotemporal metabolic mapping analysis, and delivery system optimization based on the analysis results of the pharmacokinetic analysis module.

2. The system according to claim 1, characterized in that, The Riemannian manifold representation unit includes: A data preprocessing component is used to perform noise removal, background correction, and spectral normalization on the multispectral image; A manifold mapping component, connected to the data preprocessing component, is used to vectorize and map the processed multispectral data to a high-dimensional Riemannian manifold space. A local coordinate system construction component, connected to the manifold mapping component, is used to establish a local tangent space and orthogonal basis vector field on the Riemannian manifold; and A drug metabolism trajectory extraction component, connected to the local coordinate system construction component, is used to track the evolution trajectory of drug metabolism on the Riemann manifold based on multispectral images at multiple time points.

3. The system according to claim 2, characterized in that, The manifold mapping component employs a manifold learning method based on adjacency relationships to establish a manifold embedding representation that preserves the local geometric structure, wherein each data point is connected to 8 to 12 nearest neighbors.

4. The system according to claim 1, characterized in that, The curvature tensor analysis unit includes: A multi-scale curvature calculation component for calculating the curvature characteristics of drug metabolism manifolds at three spatial scales: micro, meso, and macro. The principal curvature extraction component, connected to the multi-scale curvature calculation component, is used to extract the principal curvature values ​​and principal directions of the drug metabolism manifold; A curvature feature mapping component, connected to the principal curvature extraction component, is used to generate a spatial distribution curvature map of drug metabolic activity; and A metabolic fingerprint recognition component, connected to the curvature feature mapping component, is used to extract time-varying features from curvature data at multiple time points to construct a drug-specific metabolic fingerprint.

5. The system according to claim 4, characterized in that, The multi-scale curvature calculation component employs a three-level calculation window size: a 5×5 pixel microscale window for capturing fine intracellular structures; a 15×15 pixel mesoscale window for analyzing intercellular interactions; and a 30×30 pixel macroscale window for describing tissue-level drug distribution.

6. The system according to claim 1, characterized in that, The parallel transmission optimization unit includes: A geodesic calculation component is used to construct an energy functional on the Riemannian manifold and solve for the optimal transport path based on the drug delivery location and target point, generating an ideal drug transport geodesic. A path extraction component is used to track the movement trajectory of the actual drug signal from the time-series data of the multispectral image and generate the actual drug transport path. A deviation analysis component, connected to the geodesic calculation component and the path extraction component, is used to calculate the spatial deviation between the actual drug delivery path and the ideal drug delivery geodesic; and An efficiency evaluation component, connected to the deviation analysis component, is used to construct a delivery system efficiency index based on the spatial deviation.

7. The system according to claim 1, characterized in that, The light source module includes: A wide-band white light source is used to provide basic lighting; A narrowband filter array, connected to the wideband white light source, is used to filter out multiple monochromatic lights with wavelengths ranging from 400nm to 700nm from the wideband white light; A light source controller, connected to the narrowband filter group, is used to control the switching of the multiple monochromatic excitation lights in a timing manner; and A dichroic mirror, connected to the broadband white light source and the narrowband filter group, is used to guide the broadband white light and the monochromatic excitation light to the sample.

8. The system according to claim 1, characterized in that, The data acquisition module includes: Charge-coupled devices are used to capture the fluorescence signal emitted by a sample. A spectral separation device, connected to the charge-coupled device, is used to separate the fluorescence signal into multiple wavelength channels; An image acquisition controller, connected to the charge-coupled device and the spectral separation device, is used to control the image acquisition timing, so that each monochromatic excitation wavelength corresponds to a detection channel; and A data integrator, connected to the image acquisition controller, is used to integrate the signals from the multiple detection channels into a multispectral image with a spectral dimension of 16-33.

9. The system according to claim 1, characterized in that, The application analysis module includes: A drug-target colocalization analysis component is used to quantitatively assess the binding of drugs to specific targets based on the metabolic fingerprint generated by the curvature tensor analysis unit. The spatiotemporal metabolic mapping analysis component is used to generate spatiotemporal dynamic change maps of drugs at the tissue, cellular and subcellular levels based on the metabolic trajectories extracted by the Riemann manifold characterization unit. A drug metabolism pathway analysis component is used to quantify the dose-time pathway of drug metabolism in vivo based on changes in the multispectral images at different time points; and A delivery system optimization component is used to optimize the parameters of the drug delivery system based on the efficiency index calculated by the parallel transport optimization unit.

10. The system according to claim 9, characterized in that, The delivery system optimization components include: The bottleneck identification unit is used to analyze the spatial distribution of the efficiency index and identify the key links that cause delivery efficiency loss. A parameter tuning unit, connected to the bottleneck identification unit, is used to generate delivery system parameter optimization suggestions based on the key links; A performance prediction unit, connected to the parameter tuning unit, is used to simulate the transmission efficiency of the optimized delivery system under different conditions; and An iterative optimization unit, connected to the performance prediction unit, is used to further adjust the delivery system parameters based on the simulation results of the transmission efficiency, so as to gradually improve the delivery efficiency.

Images