Tumor microenvironment multi-parameter monitoring method, device, medium and equipment

Abstract

The present application relates to the technical field of biomedical sensing, and discloses a tumor microenvironment multi-parameter monitoring method, device, medium and equipment, the method comprising applying an excitation magnetic field to a magnetic nano-probe in a target region, using a photomagnetic probe array to collect a nonlinear magnetization response signal; performing harmonic decomposition and feature extraction on the nonlinear magnetization response signal to obtain a harmonic feature vector for characterizing the nonlinear magnetization response; inputting the harmonic feature vector into a physical constraint inversion engine to solve, simultaneously decouple and output quantitative estimation values of at least two tumor microenvironment parameters of the target region, the at least two tumor microenvironment parameters being any two or more of a pH value, an oxygen partial pressure, a local temperature and a magnetic nano-probe concentration; the physical constraint inversion engine is constructed based on a Bayesian inference framework fused with a physical constraint regularization term. The above method realizes simultaneous, high-precision, in-situ decoupling and monitoring of multiple parameters in a tumor microenvironment.

Description

Technical Field

[0001] This invention relates to the field of biomedical sensing and imaging technology, and in particular to a method, device, medium and equipment for multi-parameter monitoring of the tumor microenvironment. Background Technology

[0002] Dynamic monitoring of the tumor microenvironment is crucial for understanding tumor biological behavior, assessing treatment efficacy, and achieving personalized treatment. Existing monitoring technologies have made some progress in terms of technological development. Specifically, the nonlinear magnetization response theory of magnetic nanoparticles has been used to describe their hysteresis behavior, harmonic analysis technology has been used for concentration imaging of nanoparticles, and the sensitivity of quantum sensing technology has been developed to the femtotes level, making it possible to detect weak magnetic signals.

[0003] However, existing technologies still face many challenges in achieving in-situ real-time monitoring of multiple parameters in the tumor microenvironment, including: First, existing models are mostly based on linear magnetization approximations, which cannot fully explore and utilize the rich physicochemical information carried by nonlinear harmonics under high-frequency excitation, resulting in low information utilization; Second, multiple key parameters such as pH, oxygen partial pressure (pO2), temperature, and nanoprobe concentration are coupled and interfere with each other in the measurement signals, making it difficult to achieve high-precision synchronous decoupling; Third, conventional solution algorithms have difficulty converging, and the solution results have poor robustness.

[0004] Therefore, a method that can overcome the above-mentioned defects and achieve synchronous, high-precision, in-situ decoupling and monitoring of multiple parameters such as pH value, oxygen partial pressure, temperature and probe concentration is urgently needed. Summary of the Invention

[0005] In view of this, this application provides a method, device, medium and equipment for multi-parameter monitoring of the tumor microenvironment. The main purpose is to solve the technical problems of low information utilization, difficulty in achieving high-precision synchronous decoupling and poor robustness of calculation results when monitoring multiple parameters of the tumor microenvironment in the prior art.

[0006] According to a first aspect of the present invention, a method for multi-parameter monitoring of the tumor microenvironment is provided, comprising:

[0007] An excitation magnetic field is applied to a magnetic nanoprobe within the target area, and the nonlinear magnetization response signal generated by the magnetic nanoprobe in response to the excitation magnetic field is acquired using an optically pumped magnetometer probe array.

[0008] Harmonic decomposition and feature extraction are performed on the nonlinear magnetization response signal to obtain a harmonic feature vector characterizing the nonlinear magnetization response;

[0009] The harmonic feature vector is input into the physical constraint inversion engine for solution, and quantitative estimates of at least two tumor microenvironment parameters of the target region are simultaneously decoupled and output. The at least two tumor microenvironment parameters are any two or more of pH value, oxygen partial pressure, local temperature and magnetic nanoprobe concentration.

[0010] The physical constraint inversion engine is built on a Bayesian inference framework that incorporates physical constraint regularization terms.

[0011] Optionally, the magnetization response of the magnetic nanoprobe is predicted based on a modified physical model, which introduces a first additional term and a second additional term on the Landau-Lifshitz-Gilbert equation. The first additional term characterizes the modulation effect of pH changes on the magnetic dipole interaction energy between the magnetic nanoprobes, and is manifested as an equivalent additional torque or equivalent magnetic field acting on the probe's magnetic moment. The second additional term characterizes the local thermodynamic state perturbation caused by the local oxygen partial pressure through photothermal effects or during energy transfer, and the local thermodynamic state perturbation and temperature work together to make the second additional term manifest as an equivalent additional torque or equivalent magnetic field term related to the local temperature gradient or energy state.

[0012] Optionally, the excitation magnetic field is a composite time-varying magnetic field, which includes a DC bias component, a first AC component, and a second AC component; wherein, the frequency value of the first AC component is set in the range of 1 Hz to 100 Hz to excite Brownian relaxation; the frequency of the second AC component is set in the range of 100 Hz to 10 kHz, and the frequency value of the second AC component is at least 10 times the frequency value of the first AC component to excite Niehr relaxation; there is an adjustable phase difference between the first AC component and the second AC component.

[0013] Optionally, the step of performing harmonic decomposition and feature extraction on the nonlinear magnetization response signal to obtain a harmonic feature vector characterizing the nonlinear magnetization response includes: converting the nonlinear magnetization response signal into a digital signal, and performing preprocessing operations such as bandpass filtering, synchronous averaging, and baseline correction on the digital signal in sequence; performing harmonic decomposition on the preprocessed digital signal and extracting all harmonic features of the nonlinear magnetization response signal in parallel, wherein all harmonic features include the amplitude and phase information of the fundamental and multiple harmonic components; selecting a target feature subset from all harmonic features based on a preset screening criterion, and constructing the harmonic feature vector based on the target feature subset, wherein the screening criterion is used to improve the ability of the harmonic feature vector to distinguish the tumor microenvironment parameters.

[0014] Optionally, the physical constraint inversion engine decouples the tumor microenvironment parameters by constructing and solving a target optimization problem. The objective function corresponding to the target optimization problem is constructed based on a data fitting term and a regularization constraint term. The data fitting term is used to measure the difference between the harmonic feature vector and the theoretical feature vector calculated based on the forward physics model. The regularization constraint term includes a spatial smoothing constraint sub-term and a physiological coupling constraint sub-term. The spatial smoothing constraint sub-term is used to constrain the continuity of the tumor microenvironment parameters in spatial distribution, and the physiological coupling constraint sub-term is used to constrain the cooperative change pattern between different tumor microenvironment parameters based on a preset physiological correlation.

[0015] Optionally, the execution method of the physical constraint inversion engine under the Bayesian inference framework includes: constructing a likelihood function based on the forward physical model and measurement noise characteristics, wherein the likelihood function is used to describe the probability of occurrence of the harmonic feature vector; establishing a hierarchical prior probability distribution of the tumor microenvironment parameters, wherein the hierarchical prior probability distribution limits the pH value to a preset physiological range and establishes a statistical dependency between oxygen partial pressure and pH value; sampling the posterior probability distribution of the tumor microenvironment parameters using a Markov chain Monte Carlo sampling method based on the likelihood function and the hierarchical prior probability distribution to obtain parameter samples; and calculating the statistical estimates and corresponding uncertainty intervals of the tumor microenvironment parameters based on the parameter samples.

[0016] Optionally, the physics constraint inversion engine is executed through a GPU parallel computing architecture, which is configured to create a first CUDA execution stream, a second CUDA execution stream, and a third CUDA execution stream. The first CUDA execution stream is used to perform batch calculations of the forward physics model, the second CUDA execution stream is dedicated to performing candidate point evaluation in the Markov chain Monte Carlo sampling process, and the third CUDA execution stream is dedicated to performing matrix operations in the physics constraint inversion process. The first CUDA execution stream, the second CUDA execution stream, and the third CUDA execution stream exchange data through the shared memory of the GPU, and the GPU and CPU use an asynchronous execution mechanism for data transmission.

[0017] According to a second aspect of the present invention, a multi-parameter monitoring device for the tumor microenvironment is provided, the device comprising:

[0018] A magnetic field application module is used to apply an excitation magnetic field to a magnetic nanoprobe within a target area, and to collect the nonlinear magnetization response signal of the magnetic nanoprobe in response to the excitation magnetic field using an optically pumped magnetometer probe array.

[0019] The feature extraction module is used to perform harmonic decomposition and feature extraction on the nonlinear magnetization response signal to obtain a harmonic feature vector characterizing the nonlinear magnetization response.

[0020] The parameter output module is used to input the harmonic feature vector into the physical constraint inversion engine for solving, and simultaneously decouple and output quantitative estimates of at least two tumor microenvironment parameters of the target region. The at least two tumor microenvironment parameters are any two or more of pH value, oxygen partial pressure, local temperature and magnetic nanoprobe concentration.

[0021] The physical constraint inversion engine is built on a Bayesian inference framework that incorporates physical constraint regularization terms.

[0022] According to a third aspect of the present invention, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described method for monitoring multiple parameters of the tumor microenvironment.

[0023] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for multi-parameter monitoring of the tumor microenvironment.

[0024] This invention provides a method, device, medium, and equipment for multi-parameter monitoring of the tumor microenvironment. By applying an excitation magnetic field and acquiring nonlinear magnetization response signals, followed by harmonic decomposition and feature extraction, this invention abandons the traditional linear magnetization approximation model and directly acquires and utilizes the nonlinear harmonic components carrying rich physicochemical information generated by magnetic nanoprobes under the excitation magnetic field, thus improving information utilization from both the signal source and information extraction method. Furthermore, it introduces a physically constrained inversion engine based on a Bayesian inference framework incorporating physical constraint regularization terms. Specifically, the physical constraint regularization terms incorporate prior knowledge about parameter spatial distribution and physiological correlations as hard constraints into the calculation. The solution process restricts the arbitrariness of parameter solutions, providing a physical and physiological basis for decoupling. Simultaneously, the Bayesian inference framework provides a mathematical foundation for uncertainty quantification and robust estimation in multi-parameter, high-noise environments. The fusion of these two approaches improves the accuracy of synchronously decoupling multiple parameters from highly coupled nonlinear signals. The Bayesian framework is suitable for handling ill-conditioned inverse problems and noisy data; describing the uncertainty of the solution through probability distribution enhances the algorithm's robustness. Combined with physical constraints, it further guides the solution process towards a solution space that conforms to physical laws, overcoming the convergence difficulties or instability problems that traditional least squares and other algorithms easily encounter when facing complex nonlinear inversion. The above method achieves synchronous, high-precision, in-situ decoupling and monitoring of multiple parameters in the tumor microenvironment, fundamentally solving the problems of incomplete information, insufficient accuracy, and poor reliability in existing technologies, and providing a multi-dimensional quantitative tool for tumor biology research and treatment evaluation.

[0025] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0027] Figure 1 A flowchart illustrating a multi-parameter monitoring method for the tumor microenvironment provided in an embodiment of the present invention is shown.

[0028] Figure 2 The flowchart of the multi-parameter synchronous inversion algorithm in a multi-parameter monitoring method for tumor microenvironment provided by an embodiment of the present invention is shown.

[0029] Figure 3 This diagram illustrates the specific flow of the multi-parameter synchronous inversion algorithm in a multi-parameter monitoring method for the tumor microenvironment provided by an embodiment of the present invention.

[0030] Figure 4 This diagram illustrates a GPU parallel architecture in a multi-parameter monitoring method for the tumor microenvironment provided by an embodiment of the present invention.

[0031] Figure 5 This diagram illustrates an animal experiment procedure corresponding to a multi-parameter monitoring method for the tumor microenvironment provided in an embodiment of the present invention.

[0032] Figure 6 This invention provides a method for monitoring multiple parameters of the tumor microenvironment, which is shown in the time series diagram of animal experimental monitoring parameters.

[0033] Figure 7 This diagram illustrates the structure of a multi-parameter monitoring device for the tumor microenvironment provided in an embodiment of the present invention.

[0034] Figure 8 A schematic diagram of the device structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0035] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0036] This application provides a method for multi-parameter monitoring of the tumor microenvironment, such as... Figure 1 As shown, the method includes the following steps:

[0037] 101. Apply an excitation magnetic field to the magnetic nanoprobe in the target area and use an optically pumped magnetometer probe array to collect the nonlinear magnetization response signal generated by the magnetic nanoprobe in response to the excitation magnetic field.

[0038] The target region refers to the spatial range of the living tumor tissue or its phantom model to be monitored; the magnetic nanoprobe refers to superparamagnetic nanoparticles that have undergone special functionalization modification, whose magnetic properties can be specifically changed with the physicochemical parameters of the microenvironment, serving as a sensor to sense changes in the microenvironment; the excitation magnetic field refers to the time-varying magnetic field generated by an external coil system and acting on the target region, used to excite the magnetic nanoprobe to produce a measurable magnetic response, and this excitation magnetic field is usually a composite magnetic field containing specific frequency components; the optically pumped magnetometer (OPM) probe array is an ultra-high sensitivity magnetic sensor array based on the atomic spin effect, specifically utilizing laser manipulation and detection of the quantum state of alkali metal atoms in the gas chamber, capable of detecting extremely weak magnetic signals with a noise equivalent magnetic field sensitivity at the femtotes level; the nonlinear magnetization response signal refers to the complex response generated by the magnetic nanoprobe under the action of the excitation magnetic field, where the magnetization intensity and magnetic field intensity do not satisfy a simple linear proportional relationship, this response contains rich harmonic components, and the amplitude and phase carry detailed information about the microenvironment in which the probe is located.

[0039] Specifically, this step is the signal excitation and acquisition process in the monitoring process of this application. First, the magnetic nanoprobe that has been enriched in the target area is actively stimulated by an excitation magnetic field, driving the magnetic moment of the probe to move. Since the magnetization behavior of the probe is inherently nonlinear and is regulated by the local microenvironment, the response it generates is a nonlinear magnetization response signal containing the fundamental wave and multiple harmonics. Subsequently, the weak response signal is captured synchronously and in situ in a non-contact manner using an ultra-high sensitivity quantum sensor, namely an optically pumped magnetometer probe array. The multi-channel design of the OPM probe array allows for synchronous acquisition of signals from different spatial locations, providing a data basis for subsequent spatial resolution analysis. It can be seen that this step can convert the chemical and physical states of the tumor microenvironment into raw magnetic signals that can be quantitatively analyzed.

[0040] In this embodiment, the high sensitivity of the OPM probe array allows it to penetrate biological tissue and non-destructively detect extremely weak magnetic signals generated by nanoprobes deep within the body, achieving in-situ in vivo monitoring. By actively exciting and acquiring nonlinear magnetization response signals, the simplification assumptions of traditional linear models are abandoned, ensuring the capture of rich harmonic characteristics containing multiple parameters such as pH, oxygen partial pressure, and temperature from the signal source, laying a data foundation for subsequent high-precision decoupling. The excitation magnetic field and the OPM probe array work together to form a closed-loop process of response signal excitation and acquisition, constituting a reliable physical data source for the entire multi-parameter quantitative monitoring system.

[0041] 102. Perform harmonic decomposition and feature extraction on the nonlinear magnetization response signal to obtain the harmonic feature vector used to characterize the nonlinear magnetization response.

[0042] Harmonic decomposition is a signal processing method that decomposes a complex periodic or quasi-periodic signal into a series of sinusoidal components whose frequencies are integer multiples of each other. The fundamental frequency is the same as the excitation magnetic field frequency, and the frequencies of the second harmonic, third harmonic, etc., are integer multiples of the fundamental frequency, respectively. Feature extraction is the process of calculating or selecting quantitative indicators from the original data or the data after preliminary processing that can effectively characterize the essential attributes of the signal and have the highest discriminative power for subsequent analysis tasks. Each element in the harmonic feature vector represents a specific feature extracted from the nonlinear magnetization response signal. These features typically include, but are not limited to, the amplitude and phase of each harmonic, as well as their ratios or combinations.

[0043] Specifically, this step realizes the transformation from the original analog signal to digital information features. First, the time-domain nonlinear magnetization response signal acquired by the OPM probe array is digitized and preprocessed. Then, harmonic decomposition technology is used to perform frequency domain analysis, accurately separating and calculating the amplitude and phase information of the fundamental wave and each harmonic in the signal. Subsequently, through feature extraction algorithm, the most sensitive and representative subset of tumor microenvironment parameters is selected or constructed from a large amount of data such as all calculated harmonic amplitudes and phases. Finally, the selected features are arranged in order to form a low-dimensional harmonic feature vector, which serves as the sole input to the subsequent inversion algorithm, thus fully characterizing the core information of the original nonlinear response in a highly compressed form.

[0044] In this embodiment, the original time-domain waveform data is compressed into harmonic feature vectors with small dimensions, reducing the data processing burden of subsequent inversion algorithms and providing a technical foundation for real-time computing. By focusing on the amplitude and phase of nonlinear harmonics, the signal components sensitive to changes in tumor microenvironment parameters are effectively amplified, while common background noise is suppressed, significantly improving the signal-to-noise ratio and feature representation ability. The output harmonic feature vectors are input into the subsequent physical constraint inversion engine as a standardized and structured data form, ensuring the universality and stability of the algorithm processing.

[0045] 103. Input the harmonic feature vector into the physical constraint inversion engine to solve, and simultaneously decouple and output the quantitative estimates of at least two tumor microenvironment parameters of the target region. The at least two tumor microenvironment parameters are any two or more of pH value, oxygen partial pressure, local temperature and magnetic nanoprobe concentration. The physical constraint inversion engine is built based on a Bayesian inference framework that incorporates physical constraint regularization terms.

[0046] Among them, the physical constraint inversion engine, as the core computational module or algorithm system, functions to inversely deduce the unknown source parameters that generated the data from the observed data based on known physical laws and mathematical models. In this application, it inverses the tumor microenvironment parameters from the harmonic eigenvector. The physical constraint regularization term is a mathematical term added to the objective function when solving the inversion problem. Its role is to use prior knowledge about the solution to constrain and guide the range of the solution, thereby overcoming the inherent ill-conditioning of the inverse problem, stabilizing the solution process, and finally obtaining the solution. The Bayesian inference framework is a probabilistic statistical inference paradigm that treats all unknowns as random variables. By combining the likelihood function of the observed data and the prior probability distribution of the parameters, the posterior probability distribution of the parameters is calculated. This not only provides the most likely estimate of the parameters but also quantifies the uncertainty of the estimate. Synchronous decoupling refers to the process of simultaneously and independently separating and calculating multiple target parameters from a set of mutually coupled and mixed measurement signals.

[0047] Specifically, this step constructs an advanced physically constrained inversion engine that integrates physical constraint regularization terms with a Bayesian inference framework. First, within the Bayesian framework, the inversion problem is formalized as a probabilistic inference problem. A likelihood function is constructed using a forward physics model that describes how parameters generate harmonic features, combined with a prior distribution based on physiological knowledge. Simultaneously, physical constraint regularization terms are integrated into the framework, such as transforming constraints like spatial smoothness and physiological coupling into part of the prior distribution or as guidance for the posterior distribution. Then, efficient numerical algorithms, such as Markov chain Monte Carlo methods, are used to solve the fused model, simultaneously decoupling the optimal estimates of each independent tumor microenvironment parameter from the high-dimensional, coupled information space defined by the harmonic feature vector.

[0048] In this embodiment, by integrating physical constraint regularization terms and a Bayesian inference framework, knowledge constraints are combined with probabilistic statistical inference, enhancing the inversion engine's ability to handle parameter coupling and ill-conditioned inverse problems. This allows for the simultaneous decoupling of multiple key parameters from highly nonlinear and coupled signals, reducing cross-sensitivity. The output of the Bayesian inference framework is not only the point estimate of the parameters but also the complete posterior probability distribution, enabling the system to simultaneously output the confidence interval of each parameter. For example, the pH value has a 95% probability of being between 6.75 and 6.85, thus providing an important indicator for measuring the reliability of the results and improving the scientific nature of clinical decision-making. The engine connects the front-end signal features with deep physicochemical models and physiological knowledge bases, performing inference based on prior knowledge, enhancing the interpretability of the method and its adaptability in different application scenarios.

[0049] This invention provides a method, device, medium, and equipment for multi-parameter monitoring of the tumor microenvironment. By applying an excitation magnetic field and acquiring nonlinear magnetization response signals, followed by harmonic decomposition and feature extraction, this invention abandons the traditional linear magnetization approximation model and directly acquires and utilizes the nonlinear harmonic components carrying rich physicochemical information generated by magnetic nanoprobes under the excitation magnetic field, thus improving information utilization from both the signal source and information extraction method. Furthermore, it introduces a physically constrained inversion engine based on a Bayesian inference framework incorporating physical constraint regularization terms. Specifically, the physical constraint regularization terms incorporate prior knowledge about parameter spatial distribution and physiological correlations as hard constraints into the calculation. The solution process restricts the arbitrariness of parameter solutions, providing a physical and physiological basis for decoupling. Simultaneously, the Bayesian inference framework provides a mathematical foundation for uncertainty quantification and robust estimation in multi-parameter, high-noise environments. The fusion of these two approaches improves the accuracy of synchronously decoupling multiple parameters from highly coupled nonlinear signals. The Bayesian framework is suitable for handling ill-conditioned inverse problems and noisy data; describing the uncertainty of the solution through probability distribution enhances the algorithm's robustness. Combined with physical constraints, it further guides the solution process towards a solution space that conforms to physical laws, overcoming the convergence difficulties or instability problems that traditional least squares and other algorithms easily encounter when facing complex nonlinear inversion. The above method achieves synchronous, high-precision, in-situ decoupling and monitoring of multiple parameters in the tumor microenvironment, fundamentally solving the problems of incomplete information, insufficient accuracy, and poor reliability in existing technologies, and providing a multi-dimensional quantitative tool for tumor biology research and treatment evaluation.

[0050] Specifically, in the above embodiments, the magnetization response of the magnetic nanoprobe is predicted based on a modified physical model. The modified physical model introduces a first additional action term and a second additional action term on the basis of the Landau-Lifshitz-Gilbert equation. The first additional action term is used to characterize the modulation effect of pH value change on the magnetic dipole interaction energy between magnetic nanoprobes, and the first additional action term is manifested as an equivalent additional torque or equivalent magnetic field acting on the probe magnetic moment. The second additional action term is used to characterize the local thermodynamic state perturbation generated by the local oxygen partial pressure through photothermal effect or during energy transfer. The local thermodynamic state perturbation and temperature work together to make the second additional action term manifest as an equivalent additional torque term or equivalent magnetic field term related to the local temperature gradient or energy state.

[0051] In this embodiment, the magnetic behavior of the magnetic nanoprobe can respond sensitively and quantitatively to key parameters of the tumor microenvironment. Specifically, it is described and predicted by a modified Landau-Lifshitz-Gilbert (LLG) equation, which is a classical physical model describing the dynamic behavior of magnetic moments in a magnetic field. This application has modified the LLG equation to incorporate environmental response mechanisms.

[0052] Specifically, the evolution of the probe's magnetic moment m with time t is governed by a modified physical model, and the corresponding equation for the modified physical model is:

[0053]

[0054] In the formula, The term is the precession term, which is a standard term in the LLG equations, where... It is the gyromagnetic ratio. The total effective magnetic field acting on the magnetic moment describes the direction of the magnetic moment m around the effective magnetic field. Larmor precession is the main driving force for the motion of the magnetic moment; This is the damping term, which is also a standard term in the LLG equations, where... This is the Gilbert damping coefficient. The saturation magnetization is represented by this damping term, which describes the damping effect as the magnetic moment gradually moves toward the effective magnetic field due to energy dissipation during precession. This term determines the speed at which the magnetic moment eventually relaxes to equilibrium.

[0055] This application introduces a first additional action term and a second additional action term into the basic LLG equation. The first additional action term is the pH-induced torque term. The specific expression is:

[0056]

[0057] In the formula, It is a coupling coefficient related to the physical properties of the pH-sensitive polymer layer on the probe surface. It is a time-varying function whose amplitude or phase is directly modulated by the local pH value. This term describes the conformational change of the polymer layer modified on the probe surface caused by pH changes, which in turn alters the magnetic dipole-dipole interaction between neighboring probes. This modulated dipole interaction is mathematically equivalent to an additional torque applied to the magnetic moment of the probe. This allows the dynamic behavior of the magnetic moment to encode pH information.

[0058] The second additional term is the related photothermal effect torque term T. PO2 The specific expression is:

[0059]

[0060] In the formula, It is a coefficient related to the photothermal conversion efficiency of the oxygen-sensitive fluorescent dye doped in the probe; It is a local temperature gradient, related to the photothermal effect torque term T. PO2 The physical mechanism is that the luminescence efficiency of the fluorescent dye in the probe under specific photoexcitation is regulated by the local oxygen partial pressure. At the same time, the process is accompanied by a local photothermal effect generated by nonradiative relaxation. The difference in local oxygen partial pressure will change the intensity of the photothermal effect, thereby affecting the microscopic temperature distribution around the probe and generating a local temperature gradient. This temperature gradient, through thermal diffusion or other thermodynamic coupling mechanisms, is equivalent to generating an additional torque T acting on the magnetic moment. PO2 At the same time, temperature itself also directly affects the damping coefficient. And the effective field, thus it can be seen that the second additional action term simultaneously couples the changes in local oxygen partial pressure and temperature into the magnetic moment dynamics.

[0061] Furthermore, the total effective magnetic field It also includes an environmental response component, the specific expression of which is:

[0062]

[0063] In the formula, An excitation magnetic field applied externally; It is a pH-dependent magnetic dipole field, directly derived from the modulation of the average distance or relative orientation between probes by pH changes, and is a pH-induced torque term. One of the forms of field description; It is a temperature-dependent effective field, reflecting the effect of temperature changes on the probe's own magnetism.

[0064] Furthermore, the magnetic nanoprobes provided in this application are prepared through a precise and controllable multi-step chemical synthesis process, aiming to construct composite nanoparticles with a core-shell structure that can simultaneously exhibit specific magnetic responses to pH, oxygen partial pressure, and temperature. The specific preparation process includes:

[0065] First, a magnetic nanoparticle precursor with a well-defined core-shell structure was synthesized using a high-temperature thermal decomposition method. Using iron acetylacetone as the iron source precursor, the precursor was thermally decomposed for 2 hours in a phenyl ether solvent at 280°C under inert argon protection, generating monodisperse, magnetically uniform iron(III) oxide nanocrystals. Immediately afterward, silica was coated onto the same system using the Stöber method. Tetraethyl orthosilicate was added, and under ammonia catalysis, it underwent hydrolysis and condensation reactions, resulting in the uniform growth of a dense silica shell on the surface of the magnetic core. By precisely controlling the reaction conditions, the thickness of the silica shell could be controlled to 5 ± 0.5 nm, forming a standard Fe3O4@SiO2 core-shell structure. The silica shell not only provides abundant silanol active sites for subsequent functionalization modifications but also effectively improves the chemical stability of the probe and reduces the biotoxicity of the magnetic core.

[0066] Then, the surface of the obtained Fe3O4@SiO2 particles was specifically modified for pH response. Utilizing the silanol groups on the silica surface, a covalent coupling reaction was carried out with pHis-PEG-NHS active ester in DMF solvent, allowing the silanol groups to react with the NHS ester to form stable amide bonds. This covalently grafted pHis-PEG molecules onto the particle surface. The pHis-PEG used is a uniquely designed block copolymer with polyethylene glycol segments having a molecular weight of 5 kDa, responsible for providing biocompatibility and steric stability. Its histidine segments account for 40%, and the pKa value of the imidazole group of histidine is close to the physiological pH. Its protonated / deprotonated state is extremely sensitive to changes in the surrounding pH value, leading to significant changes in the polymer chain conformation, such as swelling and shrinkage. By controlling the reaction conditions, a grafting density of 1.2 chains / square nanometer was achieved, ensuring that the particle surface coverage is greater than 85%. This high-density grafted pH-sensitive polymer layer is the material basis for the sensitive modulation of magnetic dipole interaction with pH changes.

[0067] Finally, oxygen- and temperature-sensitive units are further introduced into the interior or surface of the silica shell to achieve a two-parameter response. Oxygen-sensitive fluorescent probe molecules are then incorporated using a doping method. Ruthenium complexes, incorporated into silica networks or modified onto their surfaces at a concentration of 0.1 mM, are classic oxygen-quenching fluorescent dyes. Their quenching efficiency follows the Stern-Volmer equation, with the relevant constant K... SV = 0.025 mmHg -1This indicates that its fluorescence intensity or lifetime is extremely sensitive to changes in the partial pressure of oxygen in the environment. Furthermore, the fluorescence lifetime τ of the dye exhibits a clear dependence on temperature T, specifically determined by the formula... Description, in which As the reference lifetime, β is the temperature coefficient, and ΔT is the temperature change. Therefore, the dye unit serves as both an oxygen partial pressure sensor and a temperature sensor. Under external light excitation, the fluorescence yield is regulated by the oxygen partial pressure, and the intensity of the photothermal effect generated by nonradiative transitions also changes accordingly, thereby affecting the local microthermal environment around the particles. This photothermal temperature disturbance, dominated by oxygen partial pressure and co-changing with temperature, ultimately affects the magnetic relaxation behavior of the probe through a thermodynamic coupling mechanism.

[0068] Specifically, in the above embodiments, the excitation magnetic field is a composite time-varying magnetic field, which includes a DC bias component, a first AC component, and a second AC component; wherein, the frequency value of the first AC component is set in the range of 1 Hz to 100 Hz to excite Brownian relaxation; the frequency of the second AC component is set in the range of 100 Hz to 10 kHz, and the frequency value of the second AC component is at least 10 times the frequency value of the first AC component to excite Niehr relaxation; there is an adjustable phase difference between the first AC component and the second AC component.

[0069] In this embodiment, an excitation magnetic field is formed by superimposing a DC magnetic field and two sinusoidal alternating magnetic fields of different frequencies in space or time. Specifically, the excitation magnetic field is a composite time-varying magnetic field, and its mathematical expression is:

[0070]

[0071] In the formula, the DC bias component Used to provide a stable magnetic field background, to adjust the magnetization state of magnetic nanoparticles, or for magnetization orientation; first AC component The magnetic field amplitude is The frequency is The first AC component is used to excite Brownian relaxation, which is related to the physical rotation of nanoparticles in a liquid. Its response frequency is low, typically in the range of 1 Hz to 100 Hz, preferably 25 Hz in this application. The low-frequency signal can effectively drive the rotation of the particles, generating a detectable magnetic response; the second AC component... The magnetic field amplitude is The frequency is The second AC component is used to excite Niehr relaxation, which is related to the magnetic moment reversal within the particle. Its response frequency is relatively high, typically in the range of 100 Hz to 10 kHz, and preferably 2.5 kHz in this application. To ensure effective separation of the excitation signals of the two relaxation mechanisms in the frequency domain and avoid mutual interference, the frequency f2 is required to be at least 10 times the frequency f1, thereby achieving harmonic separation and independent detection. Preferably, it is 100 times the frequency f1, and the phase difference is adjustable. It allows the extraction of phase information in signal processing, which can be used to enhance the identification of relaxation mechanisms or for encoding and decoding in imaging.

[0072] This application employs an OPM probe array as the core sensing unit for high-sensitivity, high-spatial-resolution measurement of magnetic nanoparticle signals excited by the aforementioned composite excitation magnetic field. The OPM probe array adopts a 5×5 planar array layout, with a total of 25 independent OPM sensing units. The spacing between each probe on the plane is 5 mm. This structural design aims to achieve dense spatial sampling of the target area, thereby obtaining a high spatial resolution magnetic field distribution map, providing a data foundation for subsequent signal reconstruction and imaging. The intrinsic magnetic field sensitivity of the OPM probe is determined by its physical principle, specifically expressed as follows:

[0073]

[0074] In the formula, σB represents the intrinsic magnetic field sensitivity of the OPM probe array. The lower the value, the stronger the probe's ability to detect weak magnetic fields. The electron gyromagnetic ratio is a physical constant representing the ratio of the electron's magnetic moment to its angular momentum, which determines the intensity of the atom's response to the magnetic field; N is the atomic number density in the sensing chamber, that is, the number of atoms participating in quantum state polarization per unit volume. Increasing the atomic number density helps to improve the signal-to-noise ratio. The transverse relaxation time of the atomic spin reflects the time during which the atomic spin coherence is maintained. A longer transverse relaxation time is beneficial for obtaining higher sensitivity. The Larmor precession frequency is proportional to the strength of the external magnetic field being measured. Based on the above physical optimizations, the OPM probe array used in this application achieves strong noise performance within the actual operating bandwidth, with the noise equivalent magnetic field in the 10Hz to 1000Hz frequency range being lower than [the specified value]. This indicator ensures that the system can effectively detect the typically extremely weak secondary magnetic field signal generated by the relaxation of nanoparticles.

[0075] This application also provides a multi-parameter synchronous inversion algorithm, such as... Figure 2 and Figure 3 As shown, the method includes the following steps:

[0076] 201. Signal preprocessing.

[0077] Specifically, the nonlinear magnetization response signal is converted into a digital signal, and the digital signal is then subjected to preprocessing operations including bandpass filtering, synchronous averaging, and baseline correction.

[0078] In this embodiment, the nonlinear magnetization response analog signal output by each OPM probe, which varies continuously with time, is converted into a digital signal by a high-precision analog-to-digital converter. The converted digital signal contains the target signal, low-frequency environmental noise, and high-frequency electronic noise. To initially extract the target frequency band, bandpass filtering is first performed using a digital bandpass filter. The passband frequency range of the filter is set to 0.1 Hz to 10 kHz to retain the core signal, as the nonlinear response signal generated by the magnetic nanoparticles under dual-frequency excitation is mainly distributed within this frequency band. Furthermore, ultra-low frequency drift below 0.1 Hz and high-frequency drift above 10 kHz are filtered out. The signal-to-noise ratio is significantly improved by removing unwanted high-frequency noise of kHz. Then, synchronous averaging is performed, using the frequency of the excitation magnetic field, especially the low-frequency component or its reference signal, as the trigger clock. Multiple periodic waveform superpositions are applied to the bandpass-filtered signal. Since the target signal is periodic and synchronized with the trigger clock, it coherently enhances during accumulation. Random noise, being incoherent, increases in amplitude much slower than the signal. Multiple averagings effectively extract weak, excitation-related deterministic responses, a crucial step in improving the system's detection limit. Finally, after the above processing, the signal may still contain slow baseline shifts or trend terms introduced by sensor characteristics or minor environmental disturbances. Baseline correction aims to eliminate these non-stationary DC shifts or low-frequency trends, ensuring a stable zero baseline. This is typically achieved using digital high-pass filtering, polynomial fitting subtraction, or moving window averaging subtraction. The corrected signal more accurately reflects the AC response components generated by the magnetic nanoparticles, varying around zero, providing clean data input for subsequent harmonic analysis, image reconstruction, or quantitative inversion.

[0079] 202. Harmonic decomposition and feature extraction.

[0080] Specifically, harmonic decomposition is performed on the preprocessed digital signal, and all harmonic features of the nonlinear magnetization response signal are extracted in parallel. The all harmonic features include the amplitude and phase information of the fundamental and multiple harmonic components.

[0081] In this embodiment, preliminary demodulation is first performed using digital lock-in amplifier technology. This process uses a specific frequency or its derived frequency in the composite excitation magnetic field as a reference signal, and performs orthogonal multiplication with the preprocessed signal. This downconverts the signal components within the target frequency and its surrounding narrow band to DC or a low frequency, while suppressing all noise outside the reference frequency. This achieves highly selective extraction of the target frequency response, providing a foundation for subsequent accurate spectrum analysis. Then, to comprehensively acquire all frequency components of the signal, the system performs a Fast Fourier Transform (FFT) on the preprocessed or lock-in amplified signal, converting the time-domain signal to the frequency domain. This clearly reveals all frequency components and their energy distribution within the signal. Because magnetic nanoparticles exhibit a strong nonlinear magnetization response under composite excitation, their spectrum not only contains the fundamental excitation frequency but also generates abundant harmonics, sum frequencies, and difference frequencies—nonlinear products. FFT analysis achieves this. The system systematically decomposes and identifies these frequency components. Then, based on the frequency domain decomposition, it accurately reads and records the amplitude values ​​of all frequency components of interest from the spectrum, including the fundamental wave and each harmonic. These amplitude values ​​directly reflect the intensity of the corresponding nonlinear process and are key physical quantities for quantifying nanoparticle concentration or environmental characteristics. At the same time, it extracts the phase values ​​of each frequency component relative to the excitation source or common reference clock from the spectrum. The phase information is extremely sensitive to the magnetic field environment, particle aggregation state, or local viscosity changes, providing an independent and complementary contrast dimension in addition to the amplitude. Then, the amplitude and phase information of all extracted fundamental and harmonic components are organized into a multidimensional feature vector in a predetermined order. The vector comprehensively and structurally characterizes the complete nonlinear magnetization response fingerprint of the magnetic nanoparticle system in a single measurement. The feature vector is the direct input data for subsequent pattern recognition, quantitative inversion, or multi-parameter imaging reconstruction.

[0082] Based on this, this step, through a processing flow of lock-in amplification and demodulation, FFT spectrum analysis, parallel extraction of harmonics and phase, and construction of eigenvectors, achieves efficient, accurate, and parallel extraction of all harmonic features in the nonlinear magnetization response signal. It makes full use of the signal's spectral information, improves information utilization, and enhances the system's ability to distinguish and detect various physical and chemical parameters.

[0083] 203. Feature selection decision.

[0084] Specifically, based on preset screening criteria, a subset of target features is selected from all harmonic features, and a harmonic feature vector is constructed based on the subset of target features. The screening criteria are used to improve the ability of the harmonic feature vector to distinguish tumor microenvironment parameters.

[0085] In this application embodiment, the specific screening criteria include several methods. The first is screening based on variance analysis, which calculates the variance of each harmonic feature in multiple repeated measurements or measurements at different spatial locations, and then retains features with variances higher than a preset threshold. Excessively low variance usually means that the feature is not sensitive to changes in measurement conditions and carries limited effective information, possibly mainly reflecting system noise or a fixed background; therefore, it can be removed as a noise feature, effectively filtering out stable noise components in the signal. The second is screening based on correlation, which evaluates the correlation between each original feature and known prior information about the reference signal or target parameters, such as Pearson correlation coefficient, mutual information, etc. The aim is to retain features that are highly correlated with the reference excitation signal or have a strong statistical association with the target parameters. These features are considered important features, directly responding to the excitation or closely related to the physiological or pathological changes being measured. Conversely, features with weak correlations to both the reference signal and the target parameters are more likely to originate from random interference or irrelevant physiological processes and should be suppressed. The third is dimensionality reduction and screening based on principal component analysis. Principal component analysis is performed on the original high-dimensional feature space to find the main direction of data variation. For example, the top few principal components with cumulative contribution rates reaching a preset value are selected as new feature sets, which can retain the overall variation information of the dataset to the greatest extent, while automatically achieving decorrelation and dimensionality reduction. On this basis, the loadings of each original feature on important principal components are further analyzed. Features with extremely low absolute loading values ​​can be regarded as redundant or noisy features. The fourth method is feature importance ranking based on model feedback. A fast preliminary prediction model is used to evaluate the importance score of each feature to the prediction target parameter, such as a linear model or decision tree trained on historical data. Features are ranked according to importance scores, and the top-ranked feature subset is retained, realizing screening guided by the final application goal. Finally, through the comprehensive application of one or more of the above combined criteria, the system selects a target feature subset with a significantly reduced number but higher information density from a large number of original harmonic features. The selected features are then standardized and combined according to predetermined rules to form a refined harmonic feature vector for subsequent quantitative inversion.

[0086] 204. Construct an optimization problem and design regularization terms.

[0087] Specifically, the physical constraint inversion engine decouples tumor microenvironment parameters by constructing and solving a target optimization problem. The objective function corresponding to the target optimization problem is constructed based on data fitting terms and regularization constraint terms. Among them, the data fitting term is used to measure the difference between the harmonic eigenvector and the theoretical eigenvector calculated based on the forward physics model. The regularization constraint term includes a spatial smoothing constraint sub-term and a physiological coupling constraint sub-term. The spatial smoothing constraint sub-term is used to constrain the continuity of the spatial distribution of tumor microenvironment parameters, and the physiological coupling constraint sub-term is used to constrain the synergistic change pattern between different tumor microenvironment parameters based on the preset physiological correlation.

[0088] In this embodiment, the inversion is based on establishing a forward physics model that accurately describes the nonlinear magnetization response of magnetic nanoparticles. Starting from microscopic interactions, the model predicts the magnetic harmonic characteristics to be observed under a given parameter θ. The core formula of the model is:

[0089]

[0090] In the formula: N is the total number of interacting magnetic nanoparticles in the imaging region; Let be the instantaneous magnetic moment of the i-th particle; L is the Langevin function, which describes the statistical equilibrium magnetization behavior of superparamagnetic particles in a magnetic field; The effective magnetic field acting on the i-th particle includes the external excitation field and possible local fields; k B is Boltzmann constant; T is absolute temperature, which is one of the microenvironment parameters to be inverted; This is a nonlinear correction term used to characterize the nonlinear magnetization response that exceeds the Langevin linear approximation under high driving fields, and is the physical source of higher harmonics.

[0091] By numerically solving the dynamic response of the above model under the action of a combined excitation magnetic field, and performing the same harmonic analysis procedure as at the measurement end, the theoretical harmonic eigenvectors can be calculated. In the objective optimization problem, the data fitting term is used to quantify the consistency between experimental observations and physical model predictions. Its standard form is the least squares error between the two, i.e., the data fitting term is... That is, driving the inversion solution Its predicted harmonic characteristics To get as close as possible to the measured feature F.

[0092] To overcome the pathological nature of the inversion problem and incorporate prior anatomical and physiological knowledge, the objective function introduces a structured regularization constraint term, which is λ1R1(θ)+λ2R2(θ), where λ1 and λ2 are regularization parameters used to balance the weights of data fitting and different constraints.

[0093] Specifically, the regularization constraints include spatial smoothing constraints and physiological coupling constraints. The spatial smoothing constraint R1(θ) aims to constrain the continuity of microenvironment parameters in the spatial distribution of biological tissues and suppress non-physical turbulence caused by noise. Its specific form is gradient-based Tikhonov regularization.

[0094]

[0095] In the formula, For spatial gradient operators, , , These are the adaptive weighting coefficients for each parameter, which can be adjusted according to different tissue types to allow for reasonable spatial variation rates in different regions.

[0096] The physiological coupling constraint sub-term R2(θ), based on known physiological and pathological associations, constrains the synergistic change patterns among different microenvironment parameters, ensuring that the inversion results conform to biological rationality. Its expression is as follows:

[0097]

[0098]

[0099] In the formula, This is an empirical model learned from a large amount of physiological and pathological data, which quantitatively describes the negative correlation between oxygen partial pressure and acidity / alkalinity. Parameters a and b are the coefficients of the empirical model. This is a physiological model describing the relationship between oxygen partial pressure and local temperature. Basal body temperature; and These are the weighting coefficients for each physiological coupling constraint.

[0100] Ultimately, the physics-constrained inversion engine achieves the decoupling and estimation of parameter θ by solving the following objective optimization problem:

[0101]

[0102] The solution process can employ optimization algorithms such as gradient descent, conjugate gradient, or Gauss-Newton, leading to the optimal solution. This refers to the estimated parameter vector of the tumor microenvironment under multiple criteria, including optimal fitting of observation data, satisfaction of spatial smoothness, and compliance with physiological coupling relationships. Based on this, by systematically integrating a precise physical model with prior anatomical and physiological knowledge into a unified optimization framework, not only is high-precision and stable decoupling of multiple key parameters achieved, but the inversion results also have clear physical meaning and physiological rationality, thus improving the reliability of in vivo multi-parameter imaging of the tumor microenvironment.

[0103] 205. Bayesian framework construction.

[0104] Specifically, a likelihood function is constructed based on the forward physical model and measurement noise characteristics, where the likelihood function is used to describe the probability of occurrence of harmonic eigenvectors; a hierarchical prior probability distribution of tumor microenvironment parameters is established, where the hierarchical prior probability distribution limits the pH value to a preset physiological range and establishes the statistical dependence between oxygen partial pressure and pH value.

[0105] In this embodiment, the inversion process is based on measurement data. Assuming the measurement noise is additive Gaussian noise, a likelihood function is constructed based on the forward physical model and the characteristics of the measurement noise. The likelihood function describes the probability density of observing a specific harmonic eigenvector under a given parameter θ. The likelihood function is defined as follows:

[0106]

[0107] In the formula, K is the dimension of the feature vector F, that is, the total number of harmonic features after screening; The k-th feature component in the observed feature vector F; Forward physics model Among the calculated theoretical predictions, and The corresponding k-th feature component; The variance of the noise measured for the k-th characteristic component can be estimated by the noise power spectrum of the experimental data, reflecting the difference in measurement accuracy for different characteristic components.

[0108] In order to incorporate prior knowledge of physiology and pathology into the inference process, a hierarchical prior probability distribution of tumor microenvironment parameters was systematically established. The hierarchical prior model has a hierarchical structure and can reflect the dependencies between parameters. The specific model is as follows:

[0109]

[0110] In the formula, For parameter vectors The joint prior probability distribution is a truncated Gaussian distribution with a mean of =7.0, standard deviation =0.5, and restrict its value range to a physiologically or pathologically reasonable range [5.5, 7.5], and a priori limit the pH value to a preset physiological range; For a given pH value, the conditional prior distribution of oxygen partial pressure is given by a Gaussian distribution. ,in, It is a function of pH. The formula establishes the statistical dependence between oxygen partial pressure and pH value. Let be the variance of the conditional distribution; The conditional prior distribution of temperature for a given partial pressure of oxygen reflects the possible relationship between metabolic heat production and oxygen supply. The prior distribution of magnetic nanoparticle concentration can typically be based on the injection dose or a broader distribution.

[0111] 206. MCMC sampling, convergence diagnosis and posterior analysis.

[0112] Specifically, based on the likelihood function and hierarchical prior probability distribution, the posterior probability distribution of tumor microenvironment parameters is sampled using the Markov chain Monte Carlo sampling method to obtain parameter samples; based on the parameter samples, the statistical estimates of tumor microenvironment parameters and the corresponding uncertainty intervals are calculated.

[0113] In this embodiment, the physical constraint inversion engine employs a Markov chain Monte Carlo (MCMC) sampling method based on the adaptive Metropolis-Hastings algorithm to efficiently and robustly sample and infer the posterior distribution of tumor microenvironment parameters. First, MCMC sampling initialization is performed, based on the forward physical model and hierarchical prior distribution. Multiple independent Markov chains are simultaneously launched, each starting from a different initial position in the parameter space. During sampling, the algorithm dynamically adjusts the covariance matrix of the proposed distribution to match the geometry of the parameter space already explored by the current chain. A hybrid proposal strategy is also adopted: high probability for local fine-grained search and low probability for global exploration, effectively preventing the sampling chain from getting trapped in local optima and ensuring a thorough exploration of the posterior distribution. Afterward, convergence diagnosis is performed, continuously monitoring the... The Gelman-Rubin statistic with parallel sampling chains is used. When the statistic values ​​of all parameters are less than a preset threshold, the sampling process is considered to have reached convergence. Otherwise, sampling continues until the convergence condition is met. Finally, posterior distribution analysis and parameter extraction are performed. Using the converged sampling chain samples, the mean of the posterior distribution is calculated as the optimal statistical estimate of the four parameters of the tumor microenvironment (pH, pO2, T, C). Based on the sampling chain samples, the posterior variance or standard deviation of each parameter is calculated to objectively evaluate the reliability of the inversion results. Then, based on the posterior distribution of the parameters, the 95% confidence interval of each parameter is calculated and output, providing complete parameter estimates and their uncertainty ranges for clinical decision-making. This step, through adaptive sampling and convergence diagnosis, ensures the accuracy and statistical reliability of parameter estimates, realizing a robust inversion from harmonic characteristics to multiple parameters of the tumor microenvironment.

[0114] 207. Calculation optimization.

[0115] Specifically, the GPU parallel computing architecture is configured to create a first CUDA execution stream, a second CUDA execution stream, and a third CUDA execution stream. The first CUDA execution stream is used to perform batch computations of the forward physics model, the second CUDA execution stream is dedicated to performing candidate point evaluation in the Markov chain Monte Carlo sampling process, and the third CUDA execution stream is dedicated to performing matrix operations in the physics constraint inversion process. The first, second, and third CUDA execution streams exchange data through the GPU's shared memory, and the GPU and CPU use an asynchronous execution mechanism for data transmission.

[0116] In the embodiments of this application, such as Figure 4 As shown, the GPU computing unit is configured to create three independent CUDA execution streams to achieve task-level pipelined parallelism. The first CUDA execution stream is used for batch execution of the forward physics model computation, packaging multiple candidate parameter vectors and computing the corresponding theoretical harmonic responses in parallel, accelerating likelihood function evaluation. The second CUDA execution stream is used for candidate point evaluation in the Markov chain Monte Carlo sampling process, processing transition proposals and acceptance probability calculations for multiple MCMC chains in parallel, improving the throughput of posterior sampling. The third CUDA execution stream is used for matrix operations in the physics constraint inversion process, including linear algebra operations such as gradient calculation and covariance matrix update involved in the regularization constraint terms. Based on the three CUDA execution streams, inter-stream data exchange based on shared memory can be achieved. For example, the forward model results completed by the first stream can be directly stored in shared memory for the second stream to read in real time when evaluating the acceptance probability. The covariance matrix of stream computing can also be directly used to adjust the proposal distribution of the second stream, thus avoiding the latency caused by repeated data transfer between streams through global memory and achieving efficient on-chip communication. In addition, the GPU and CPU use an asynchronous execution mechanism for data transfer. When the GPU is executing the computation tasks of three CUDA execution streams in parallel, the CPU can asynchronously transfer the parameter results of the completed inversion from the GPU memory to the host memory and execute subsequent post-processing tasks at the same time. This asynchronous transmission and execution mechanism effectively hides the data I / O latency, realizes the overlap of computation and data transmission, and maximizes the overall system throughput. Through the above multi-stream parallelism, shared memory communication and asynchronous execution mechanism, the GPU parallel architecture provided in this application achieves deep collaborative acceleration of forward model computation, MCMC sampling and matrix operations, enabling the complex Bayesian inversion process to be completed within a clinically acceptable time and meeting the needs of short-time imaging.

[0117] In one embodiment, such as Figure 5As shown, to verify the effectiveness of this application's in vivo monitoring, a complete experimental verification was conducted using a 4T1 breast cancer mouse model. The experimental procedure specifically included: injecting functionalized magnetic nanoprobes via the tail vein and waiting for them to specifically accumulate at the tumor site; performing initial (baseline) multi-parameter measurements at t=0 to record the initial pH, pO2, temperature T, and probe concentration C of the tumor; after injecting chemotherapy drugs, starting the system for continuous 24-hour monitoring to dynamically track the treatment response; euthanizing the animals and collecting tissue samples at key time points during the monitoring period; and finally, using the following simultaneous verification techniques for in vitro verification: microelectrodes to directly measure the pH and pO2 of the tumor tissue; infrared thermography to measure the temperature distribution on the tumor surface; and inductively coupled plasma mass spectrometry (ICP-MS) to quantify the iron content in the tumor tissue to verify the probe concentration C.

[0118] To achieve clinically usable real-time processing speed, the system implements the following key optimizations based on the GPU parallel architecture: Forward model batch processing, packaging 1024 different parameter combinations and performing forward computation in parallel on the GPU at once, greatly improving the efficiency of likelihood assessment; Matrix operation optimization, utilizing the NVIDIA cuBLAS library to highly optimize linear algebra operations such as the Jacobian matrix (sensitivity matrix); Memory access optimization, using GPU texture memory to accelerate interpolation operations in the forward model, reducing video memory access latency; Convergence acceleration, the initial value of MCMC sampling is not randomly set, but is quickly provided by a pre-trained neural network based on the current observed features, which is close to the posterior pattern, thereby significantly reducing the number of iterations required for convergence.

[0119] Finally, system performance verification was conducted, comprising three aspects. The first aspect was numerical simulation verification, establishing a multi-scale simulation model covering the micro to macro scales. At the micro scale, the magnetic moment dynamics of a single magnetic nanoparticle were simulated based on the Landau-Lifshitz-Gilbert (LLG) equations. At the mesoscale, the Monte Carlo method was used to simulate the magnetic dipole interactions between a large number of particles. At the macroscale, the finite element method was used to solve for the distribution of the excitation magnetic field in biological tissue and the induced signal of the detection coil. Specifically, the number of simulated particles was 10. 4 Up to 10 6 One, with a time step of 10. -9 The total simulation time is 10 seconds. -3The second aspect is the phantom experiment verification. Specifically, a layered agarose gel phantom was designed, consisting of 5 layers, each with a preset different pH and pO2 values. Magnetic nanoprobes are distributed in the phantom with a concentration gradient (0.1–10 μM). Then, a temperature gradient from 0 to 2°C is generated and precisely controlled in the phantom using Peltier elements. The third aspect is the algorithm performance evaluation index. The system uses cross-sensitivity as the index, with a target value of less than 5%, ensuring that each parameter is independently and accurately inverted. Under GPU acceleration, the target is that the inversion calculation time per frame is less than 100 milliseconds and the number of iterations is less than 1000. The coverage rate of the 95% confidence interval is used as the index, with a target of greater than 90%, that is, the probability that the uncertainty interval reported by the algorithm can truly cover the true value of the parameter exceeds 90%.

[0120] In addition, the system integrates a real-time early warning module based on time-series data analysis. The module maintains a sliding time window, continuously analyzes the dynamic change trend of parameters, and automatically triggers alarms based on a preset physiological and pathological threshold library. The early warning rules include: triggering when the average pH value is continuously lower than the threshold for more than 30 minutes and shows a downward trend; triggering when pO2 shows a rapid decrease exceeding the set slope in a short period of time; and triggering when the temperature exceeds the basal body temperature by 1.5°C. Through the complete design and verification from animal models, software and hardware optimization, multi-dimensional verification to clinical auxiliary early warning, this application can provide a high-precision, high-efficiency, and high-reliability in vivo multi-parameter dynamic monitoring platform for the tumor microenvironment.

[0121] In another embodiment, tumor-bearing mice were continuously and dynamically monitored for up to 24 hours, and the monitoring procedure was as follows: Figure 6 As shown, the entire dynamic change from pre-injection baseline to post-chemotherapy intervention was fully recorded. Specifically, the increase in targeted nanoprobe concentration was monitored during the probe enrichment period (t=30-90 minutes). After chemotherapy started (t=2 hours), the system successfully captured the typical multi-parameter response pattern of the acute phase (t=2-8 hours). The pH value slowly increased and eventually stabilized in the range of 6.8-7.0, pO2 showed a recovery change of first decreasing and then increasing, the temperature showed a brief slight increase, and the probe concentration decreased rapidly due to metabolism. During the recovery period (t=8-24 hours), all parameters entered a stable phase. This sequential monitoring fully presented the multi-dimensional dynamic evolution of the tumor microenvironment under treatment intervention. Regarding quantitative validation, based on large-scale simulation tests, the performance of this application on 1000 sets of composite data is shown in the table below:

[0122]

[0123] The results show that the estimation errors of all parameters are at extremely low levels, the cross sensitivity is less than the design target of 5%, and the single-frame full parameter inversion time is only 42 milliseconds. This embodiment, from the two dimensions of actual dynamic monitoring and objective quantitative testing, jointly confirms the effectiveness and superiority of this application in achieving high-precision, strongly decoupled, and real-time multi-parameter dynamic monitoring of the tumor microenvironment.

[0124] Furthermore, as Figure 1 In terms of specific implementation, this application provides a multi-parameter monitoring device for the tumor microenvironment, such as... Figure 7 As shown, the device includes: a magnetic field application module 301, a feature extraction module 302, and a parameter output module 303.

[0125] The magnetic field application module 301 is used to apply an excitation magnetic field to the magnetic nanoprobe in the target area and to collect the nonlinear magnetization response signal of the magnetic nanoprobe in response to the excitation magnetic field using an optically pumped magnetometer probe array.

[0126] The feature extraction module 302 is used to perform harmonic decomposition and feature extraction on the nonlinear magnetization response signal to obtain a harmonic feature vector that characterizes the nonlinear magnetization response.

[0127] The parameter output module 303 is used to input the harmonic feature vector into the physical constraint inversion engine for solving, and simultaneously decouple and output quantitative estimates of at least two tumor microenvironment parameters of the target region. The at least two tumor microenvironment parameters are any two or more of pH value, oxygen partial pressure, local temperature and magnetic nanoprobe concentration.

[0128] Among them, the physical constraint inversion engine is built on a Bayesian inference framework that incorporates physical constraint regularization terms.

[0129] In specific application scenarios, the magnetic field application module 301 is specifically used to predict the magnetization response of the magnetic nanoprobe based on a modified physical model. The modified physical model introduces a first additional action term and a second additional action term on the basis of the Landau-Lifshitz-Gilbert equation. The first additional action term is used to characterize the modulation effect of pH value change on the magnetic dipole interaction energy between magnetic nanoprobes, and the first additional action term is manifested as an equivalent additional torque or equivalent magnetic field acting on the probe magnetic moment. The second additional action term is used to characterize the local thermodynamic state perturbation generated by the local oxygen partial pressure through photothermal effect or during energy transfer. The local thermodynamic state perturbation and temperature work together to make the second additional action term manifest as an equivalent additional torque term or equivalent magnetic field term related to the local temperature gradient or energy state.

[0130] In specific application scenarios, the magnetic field application module 301 is also used to excite a composite time-varying magnetic field, which includes a DC bias component, a first AC component, and a second AC component. The frequency of the first AC component is set in the range of 1 Hz to 100 Hz to excite Brownian relaxation. The frequency of the second AC component is set in the range of 100 Hz to 10 kHz, and the frequency of the second AC component is at least 10 times that of the first AC component to excite Niehn relaxation. There is an adjustable phase difference between the first AC component and the second AC component.

[0131] In specific application scenarios, the feature extraction module 302 can be used to convert the nonlinear magnetization response signal into a digital signal, and perform preprocessing operations such as bandpass filtering, synchronous averaging, and baseline correction on the digital signal in sequence; perform harmonic decomposition on the preprocessed digital signal, and extract all harmonic features of the nonlinear magnetization response signal in parallel, wherein all harmonic features include the amplitude and phase information of the fundamental wave and multiple harmonic components; based on the preset screening criteria, select a target feature subset from all harmonic features, and construct a harmonic feature vector based on the target feature subset, wherein the screening criteria are used to improve the ability of the harmonic feature vector to distinguish tumor microenvironment parameters.

[0132] In specific application scenarios, the parameter output module 303 can be used by the physical constraint inversion engine to decouple tumor microenvironment parameters by constructing and solving a target optimization problem. The objective function corresponding to the target optimization problem is constructed based on data fitting terms and regularization constraint terms. Among them, the data fitting term is used to measure the difference between the harmonic feature vector and the theoretical feature vector calculated based on the forward physical model. The regularization constraint terms include spatial smoothing constraint sub-terms and physiological coupling constraint sub-terms. The spatial smoothing constraint sub-terms are used to constrain the continuity of the spatial distribution of tumor microenvironment parameters, and the physiological coupling constraint sub-terms are used to constrain the synergistic change patterns between different tumor microenvironment parameters based on preset physiological correlations.

[0133] In specific application scenarios, the parameter output module 303 can also be used for the execution method of the physical constraint inversion engine under the Bayesian inference framework, including: constructing a likelihood function based on the forward physical model and measurement noise characteristics, wherein the likelihood function is used to describe the probability of occurrence of harmonic feature vectors; establishing a hierarchical prior probability distribution of tumor microenvironment parameters, wherein the hierarchical prior probability distribution limits the pH value to a preset physiological range and establishes a statistical dependence between oxygen partial pressure and pH value; based on the likelihood function and the hierarchical prior probability distribution, sampling the posterior probability distribution of tumor microenvironment parameters through the Markov chain Monte Carlo sampling method to obtain parameter samples; and calculating the statistical estimates of tumor microenvironment parameters and the corresponding uncertainty intervals based on the parameter samples.

[0134] In specific application scenarios, the parameter output module 303 can also be used for the physical constraint inversion engine to be executed through a GPU parallel computing architecture. The GPU parallel computing architecture is configured to create a first CUDA execution flow, a second CUDA execution flow, and a third CUDA execution flow. The first CUDA execution flow is used to perform batch calculations of the forward physical model, the second CUDA execution flow is dedicated to performing candidate point evaluation in the Markov chain Monte Carlo sampling process, and the third CUDA execution flow is dedicated to performing matrix operations in the physical constraint inversion process. The first CUDA execution flow, the second CUDA execution flow, and the third CUDA execution flow exchange data through the shared memory of the GPU, and the GPU and CPU use an asynchronous execution mechanism for data transmission.

[0135] It should be noted that other corresponding descriptions of the functional units involved in the multi-parameter monitoring device for the tumor microenvironment provided in this embodiment can be found in [reference]. Figure 1 and Figure 2 The corresponding description in [the document] will not be repeated here.

[0136] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method for monitoring multiple parameters of the tumor microenvironment.

[0137] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), including several instructions to enable a computer device (such as a personal computer, server, or network device) to execute the multi-parameter monitoring method for the tumor microenvironment in various implementation scenarios of this application.

[0138] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 7 The illustrated embodiment of the tumor microenvironment multi-parameter monitoring device, in order to achieve the above objectives, such as... Figure 8 As shown, this embodiment also provides a physical device for multi-parameter monitoring of the tumor microenvironment. This device includes a communication bus, a processor, a memory, and a communication interface. It may also include input / output interfaces and a display device. The various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the multi-parameter monitoring method for the tumor microenvironment described in the above embodiment.

[0139] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0140] Those skilled in the art will understand that the structure of the physical device for multi-parameter monitoring of the tumor microenvironment provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0141] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs to be identified. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0142] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. By applying the technical solution of this application, by applying an excitation magnetic field and acquiring nonlinear magnetization response signals, and performing harmonic decomposition and feature extraction, the traditional linear magnetization approximation model is abandoned. The nonlinear harmonic components carrying rich physicochemical information generated by the magnetic nanoprobe under the excitation magnetic field are directly obtained and utilized, improving the information utilization rate from the signal source and information extraction method. Furthermore, a physical constraint inversion engine based on a Bayesian inference framework that incorporates physical constraint regularization terms is introduced. Specifically, the physical constraint regularization terms incorporate prior knowledge about parameter spatial distribution and physiological correlation as hard constraints into the solution process, limiting the parameters. The arbitrariness of the solution provides a physical and physiological basis for decoupling. Simultaneously, the Bayesian inference framework provides a mathematical foundation for uncertainty quantification and robust estimation in multi-parameter, high-noise environments. The fusion of these two approaches improves the accuracy of synchronously decoupling multiple parameters from highly coupled nonlinear signals. The Bayesian framework is suitable for handling ill-conditioned inverse problems and noisy data. Describing the uncertainty of the solution through probability distributions enhances the robustness of the algorithm. Combined with physical constraints, it further guides the solution process towards a solution space that conforms to physical laws, overcoming the convergence difficulties or instability problems that traditional algorithms such as the least squares method easily encounter when facing complex nonlinear inversion. The above method achieves synchronous, high-precision, in-situ decoupling and monitoring of multiple parameters in the tumor microenvironment, fundamentally solving the problems of incomplete information, insufficient accuracy, and poor reliability in existing technologies, and providing a multi-dimensional quantitative tool for tumor biology research and treatment evaluation.

[0143] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0144] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for multi-parameter monitoring of the tumor microenvironment, characterized in that, include: An excitation magnetic field is applied to a magnetic nanoprobe within the target area, and the nonlinear magnetization response signal generated by the magnetic nanoprobe in response to the excitation magnetic field is acquired using an optically pumped magnetometer probe array. Harmonic decomposition and feature extraction are performed on the nonlinear magnetization response signal to obtain a harmonic feature vector characterizing the nonlinear magnetization response; The harmonic feature vector is input into the physical constraint inversion engine for solution, and quantitative estimates of at least two tumor microenvironment parameters of the target region are simultaneously decoupled and output. The at least two tumor microenvironment parameters are any two or more of pH value, oxygen partial pressure, local temperature and magnetic nanoprobe concentration. The physical constraint inversion engine is built on a Bayesian inference framework that incorporates physical constraint regularization terms. The magnetization response of the magnetic nanoprobe is predicted based on a modified physical model, which introduces a first additional action term and a second additional action term on the basis of the Landau-Lifshitz-Gilbert equation. The first additional action term is used to characterize the modulation effect of pH value change on the magnetic dipole interaction energy between magnetic nanoprobes, and the first additional action term is manifested as an equivalent additional torque or equivalent magnetic field acting on the magnetic moment of the probe. The second additional action term is used to characterize the local thermodynamic state disturbance generated by the local oxygen partial pressure through the photothermal effect or during energy transfer, and the local thermodynamic state disturbance works together with temperature, so that the second additional action term is an equivalent additional torque term or equivalent magnetic field term related to the local temperature gradient or energy state.

2. The method according to claim 1, characterized in that, The excitation magnetic field is a composite time-varying magnetic field, which includes a DC bias component, a first AC component, and a second AC component. The frequency value of the first AC component is set in the range of 1 Hz to 100 Hz to excite Brownian relaxation; The frequency of the second AC component is set in the range of 100 Hz to 10 kHz, and the frequency value of the second AC component is at least 10 times the frequency value of the first AC component, for exciting Niel relaxation; There is an adjustable phase difference between the first AC component and the second AC component.

3. The method according to claim 1, characterized in that, The harmonic decomposition and feature extraction of the nonlinear magnetization response signal to obtain a harmonic feature vector characterizing the nonlinear magnetization response includes: The nonlinear magnetization response signal is converted into a digital signal, and the digital signal is then subjected to preprocessing operations of bandpass filtering, synchronous averaging, and baseline correction in sequence. Harmonic decomposition is performed on the preprocessed digital signal, and all harmonic features of the nonlinear magnetization response signal are extracted in parallel. The all harmonic features include the amplitude and phase information of the fundamental and multiple harmonic components. Based on a preset screening criterion, a target feature subset is selected from all harmonic features, and the harmonic feature vector is constructed based on the target feature subset. The screening criterion is used to improve the ability of the harmonic feature vector to distinguish the tumor microenvironment parameters.

4. The method according to any one of claims 1 to 3, characterized in that, The physical constraint inversion engine achieves the decoupling of the tumor microenvironment parameters by constructing and solving a target optimization problem. The objective function corresponding to the target optimization problem is constructed based on data fitting terms and regularization constraint terms. The data fitting term is used to measure the difference between the harmonic feature vector and the theoretical feature vector calculated based on the forward physics model. The regularization constraint term includes a spatial smoothing constraint sub-term and a physiological coupling constraint sub-term. The spatial smoothing constraint sub-term is used to constrain the continuity of the tumor microenvironment parameters in spatial distribution. The physiological coupling constraint sub-term is used to constrain the synergistic change pattern between different tumor microenvironment parameters based on a preset physiological correlation.

5. The method according to claim 4, characterized in that, The execution method of the physical constraint inversion engine under the Bayesian inference framework includes: A likelihood function is constructed based on the forward physical model and the measurement noise characteristics, wherein the likelihood function is used to describe the probability of occurrence of the harmonic feature vector; A stratified prior probability distribution of the tumor microenvironment parameters is established, wherein the stratified prior probability distribution limits the pH value to a preset physiological range and establishes a statistical dependence between oxygen partial pressure and pH value. Based on the likelihood function and the hierarchical prior probability distribution, the posterior probability distribution of the tumor microenvironment parameters is sampled using the Markov chain Monte Carlo sampling method to obtain parameter samples; Based on the parameter sample, the statistical estimates of the tumor microenvironment parameters and the corresponding uncertainty intervals are calculated.

6. The method according to claim 5, characterized in that, The physical constraint inversion engine is executed through a GPU parallel computing architecture, which is configured as follows: Create a first CUDA execution flow, a second CUDA execution flow, and a third CUDA execution flow, wherein the first CUDA execution flow is used to perform the calculation of the forward physics model in batches, the second CUDA execution flow is dedicated to performing candidate point evaluation in the Markov chain Monte Carlo sampling process, and the third CUDA execution flow is dedicated to performing matrix operations in the physics constraint inversion process; The first CUDA execution stream, the second CUDA execution stream, and the third CUDA execution stream exchange data through the GPU's shared memory, and the GPU and CPU use an asynchronous execution mechanism for data transmission.

7. A multi-parameter monitoring device for the tumor microenvironment, characterized in that, The device includes: A magnetic field application module is used to apply an excitation magnetic field to a magnetic nanoprobe within a target area, and to collect the nonlinear magnetization response signal of the magnetic nanoprobe in response to the excitation magnetic field using an optically pumped magnetometer probe array. The feature extraction module is used to perform harmonic decomposition and feature extraction on the nonlinear magnetization response signal to obtain a harmonic feature vector characterizing the nonlinear magnetization response. The parameter output module is used to input the harmonic feature vector into the physical constraint inversion engine for solving, and simultaneously decouple and output quantitative estimates of at least two tumor microenvironment parameters of the target region. The at least two tumor microenvironment parameters are any two or more of pH value, oxygen partial pressure, local temperature and magnetic nanoprobe concentration. The physical constraint inversion engine is built on a Bayesian inference framework that incorporates physical constraint regularization terms.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images