A real-time optimization method for tumor interventional radiotherapy planning

Abstract

This invention relates to the field of interventional tumor therapy technology and discloses a real-time optimization method for interventional radiotherapy planning. The method includes: dimensionality reduction processing of preoperative three-dimensional angiography images to generate an initial three-dimensional spatial radiation dose distribution matrix; acquisition of in-situ physical and electrophysiological physical quantities corresponding to the injection phase, outputting multimodal synchronous temporal features; analysis of the multimodal synchronous temporal features based on feature decoupling and cross-tissue dielectric cross-validation logic to exclude vasospasm and arteriovenous shunt states, and calculation of dynamic flow resistance increments; reconstruction of blood flow distribution ratios based on dynamic flow resistance increments, accumulating to generate a real-time cumulative radiation dose distribution matrix; monitoring the real-time cumulative radiation dose distribution matrix and catheter reflux pressure gradient, sending a low-level command when a threshold is reached to stop the operation of the radiation-proof high-pressure injection pump. This invention solves the problem of deviation between actual radiation dose distribution and static plan, preventing excessive radiation in non-target areas and microsphere reflux.

Description

Technical Field

[0001] This invention relates to the field of interventional tumor therapy technology, specifically a method for real-time optimization of interventional radiotherapy plans for tumors. Background Technology

[0002] Interventional radiotherapy for tumors relies on preoperative three-dimensional angiography to assess and pre-allocate radiation doses. During the actual intraoperative injection phase, the hemodynamic state of the tumor microvascular network changes with the intervention of fluids. Current treatment protocols lack synchronous feedback of dynamic physiological parameters such as in-situ physical and electrophysiological quantities during the procedure. Dose distribution calculations during the injection process remain at the static assessment level, leading to deviations between the actual radiation dose distribution and the preoperative static plan.

[0003] During the injection of radioactive microspheres, changes occur in the microcirculatory state of the target vessel. When the target vessel spasms due to physical stimulation, it manifests as a transient increase in local fluid resistance. Current intraoperative monitoring methods cannot separate the increased resistance caused by vascular spasm from the peripheral embolism saturation state, nor can they identify fluid loss caused by small arteriovenous shunts that were not visualized on preoperative angiography. This lack of state recognition prevents the system from determining the microcirculatory status, causing radioactive microspheres to enter the systemic circulation through unidentified shunt channels.

[0004] When localized radiation saturation occurs in the injection area or a backflow pressure gradient develops at the distal end of the catheter, the radiation-protective high-pressure injection pump must be stopped. Existing injection control systems rely on host computer software to issue stop commands, and signal transmission requires processing through the operating system's software scheduling cycle and communication protocol layer. This software-based communication response has a lag time; when facing the transient backflow risk under high pressure, the delay in control execution leads to delayed action blocking, which in turn causes excessive radiation exposure to non-target tissues and microsphere backflow. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a real-time optimization method for interventional radiotherapy planning for tumors. This method solves the problems caused by the lack of intraoperative dynamic physiological parameter feedback during interventional treatment, which makes it impossible to identify target vessel spasm or arteriovenous shunt status, leading to deviations in actual radiation dose distribution from the preoperative static plan and causing excessive radiation to non-target areas and microsphere reflux.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for real-time optimization of interventional radiotherapy planning for tumors, comprising:

[0008] The computational center performs dimensionality reduction processing on the three-dimensional angiography images before tumor interventional surgery, calculates and pre-allocates the initial flow resistance coefficient, and generates the initial three-dimensional spatial radiation dose distribution matrix.

[0009] The multi-channel synchronous data acquisition module acquires in-situ physical and electrophysiological physical quantities corresponding to the tumor intervention injection stage, and outputs aligned multimodal synchronous temporal characteristics by combining the initial three-dimensional spatial radiation dose distribution matrix.

[0010] The real-time solution module analyzes the multimodal synchronous timing features based on feature decoupling and cross-tissue dielectric cross-verification logic, excludes vasospasm and arteriovenous shunt states, and calculates the dynamic flow resistance increment corresponding to the diastolic phase of the target vessel terminal.

[0011] The blood flow distribution ratio is reconstructed by inputting the dynamic flow resistance increment and the particle distribution transfer function formula, and then accumulated in the time domain to generate a real-time cumulative radiation dose distribution matrix.

[0012] The system monitors the real-time cumulative radiation dose distribution matrix and the catheter reflux pressure gradient. When the radiation saturation threshold or reflux threshold is reached, a low-level command is sent to the treatment execution module to stop the operation of the radiation protection high-pressure infusion pump.

[0013] Preferably, the step of generating the initial three-dimensional spatial radiation dose distribution matrix specifically includes:

[0014] The preoperative three-dimensional angiography image of the tumor interventional procedure was processed into a voxel skeleton as a dimension reduction process by using a vascular centerline extraction algorithm to identify bifurcation points and terminal points to construct a one-dimensional lumped parameter fluid network topology.

[0015] Extract the vascular lumen geometric parameters of the one-dimensional lumen-parameter fluid network topology, call the modified Casson fluid rheology empirical model to obtain the blood dynamic viscosity, perform calculation and pre-allocate the initial flow resistance coefficient, and combine the node constraint condition of inlet total flow conservation to normalize the allocation weight to obtain the initial blood flow allocation ratio.

[0016] The expected local radioactivity is obtained by multiplying the expected total activity of injected radioactive microspheres by the initial blood flow distribution ratio. The expected local radioactivity is then mapped to the corresponding three-dimensional voxel space coordinate system using a macroscopic medical internal radiation dose model, and the initial three-dimensional spatial radiation dose distribution matrix is ​​generated.

[0017] This step reduces the computational cost of three-dimensional full-scale fluid calculations by establishing a one-dimensional topology diagram combined with a rheological empirical model, providing a static reference base for subsequent high-frequency dynamic reconstruction.

[0018] Preferably, the steps for defining the output-aligned multimodal synchronization timing features specifically include:

[0019] The in-situ blood pressure signal, injection flow signal and transtissue bioelectrical impedance raw signal are obtained through the interventional sensing front end and the treatment execution module.

[0020] Low-frequency periodic baseline drift waveforms were extracted using surface-grounded electrode patches and combined with electrocardiogram waveforms to generate macroscopic physiological synchronization signals.

[0021] The highest channel sampling rate is used as a benchmark to construct a resampling time grid. Hardware timestamps are used to map the in situ blood pressure signal, the injection flow rate signal, the original cross-tissue bioelectrical impedance signal and the macroscopic physiological synchronization signal to the resampling time grid. After time-domain interpolation calculation using a cubic spline interpolation algorithm, the data is spliced ​​to generate aligned data.

[0022] The aligned data is then subjected to dimensional expansion and correlation mapping with the initial three-dimensional spatial radiation dose distribution matrix to output the aligned multimodal synchronous temporal features. The synchronous gridding processing mechanism for multi-source heterogeneous signals eliminates temporal misalignment caused by differences in sampling frequencies of the acquisition devices.

[0023] Preferably, the step of calculating the dynamic flow resistance increment corresponding to the diastolic phase of the target vessel terminal specifically includes:

[0024] Fast Fourier transform is performed on the in-situ blood pressure signal and the injection flow rate signal included in the multimodal synchronous timing features to extract the pressure complex spectrum and flow rate complex spectrum, and the complex fluid impedance at the harmonic frequency point is calculated using the complex fluid impedance extraction formula.

[0025] The phase angle of the complex fluid impedance corresponding to the fundamental frequency is extracted. When the flow rate advance corresponding to the phase angle exceeds the normal compliance phase threshold and the low-pass mean baseline of the in situ blood pressure signal shows a step upward, it is confirmed as a vasospasm state. The effective fluid resistance is calculated using the vasospasm transient compensation formula in feature decoupling.

[0026] A dual physiological gated time window is constructed using the macroscopic physiological synchronization signal contained in the multimodal synchronization time series feature, and combined with the original cross-tissue bioelectrical impedance signal to generate a net cross-tissue bioelectrical impedance sequence that eliminates the interference of macroscopic tissue volume changes.

[0027] The effective fluid resistance and the net transtissue bioelectrical impedance sequence are input into an arteriovenous shunt cross-validation observation model based on a Kalman filter framework for state determination. After excluding the arteriovenous shunt state, the mean transient fluid resistance within the dual physiological gating time window is extracted.

[0028] The difference between the average transient fluid resistance and the initial flow resistance coefficient of the pre-assigned corresponding node is calculated, and the dynamic flow resistance increment is output.

[0029] This step avoids misjudging increased resistance caused by spasm as peripheral embolism saturation by extracting phase angle features. At the same time, dielectric sequences are introduced for cross-validation to identify small arteriovenous shunts that are not visualized on angiography.

[0030] Preferably, in the state determination step:

[0031] In the filter recursive update logic, when the flow resistance feature maintains a low baseline and does not increase with the increase of injection volume, and the slope of the net trans-tissue bioelectrical impedance sequence is lower than the lower dielectric limit threshold, the estimated value of the true microsphere deposition density state approaches zero.

[0032] When the estimated value of the actual microsphere deposition density is lower than the preset safe deposition threshold, it is determined that there is arteriovenous shunt in the target blood vessel branch, and the topological weight of the corresponding pathway is forcibly reset to zero.

[0033] Preferably, the step of generating the real-time cumulative radiation dose distribution matrix includes:

[0034] Based on the initial flow resistance coefficient and the dynamic flow resistance increment of the corresponding node, the dynamic blood flow distribution ratio is calculated using the drag and particle distribution transfer function formula;

[0035] The time integral of the injection flow rate signal within the calculation period is extracted to obtain the injection physical volume, which is then multiplied by the specific activity constant of the radioactive microspheres to obtain the total radiation source intensity.

[0036] Using the dynamic blood flow distribution ratio as a weighting coefficient, combined with the equivalent physical density and the characteristic energy constant of isotope decay, a physical conversion is performed to generate an incremental dose distribution matrix.

[0037] Using the hardware sampling grid associated with the time step as the update frequency, the incremental dose distribution matrix is ​​three-dimensionally aligned with the spatial coordinate voxels of the initial three-dimensional spatial radiation dose distribution matrix, and then superimposed frame by frame onto the historical cumulative basis matrix using a weighted linear accumulation algorithm, thereby continuously refreshing the real-time cumulative radiation dose distribution matrix.

[0038] Preferably, in the step of 3D alignment of spatial coordinate voxels:

[0039] By combining the respiratory phase reference in the extracted macroscopic physiological synchronization signal, the spatial coordinate axes of the incremental dose distribution matrix and the initial three-dimensional spatial radiation dose distribution matrix are aligned and registered using a affine transformation. This step introduces a respiratory gating signal to counteract organ displacement caused by respiratory motion, ensuring the spatial accuracy of dose accumulation.

[0040] Preferably, the step of stopping the operation of the radiation-proof high-pressure injection pump specifically includes:

[0041] Extract the set of non-target voxels from the real-time cumulative radiation dose distribution matrix to calculate the average cumulative radiation dose in a local small neighborhood. When the radiation saturation threshold is exceeded, set the dose saturation flag in the internal register.

[0042] After performing time-domain alignment of the in-situ blood pressure signal, the injection flow rate signal and the dynamic flow resistance increment based on the hardware timestamp, the reflux risk assessment index is calculated using the reflux risk status assessment formula.

[0043] When the backflow risk assessment index exceeds the backflow threshold within a continuous judgment time window, a backflow danger sign is set.

[0044] A logical OR operation is performed on the dose saturation flag and the backflow hazard flag. When the abnormal flag is detected to be activated, a high-priority interrupt sequence is triggered to send the level-low instruction to stop the operation of the radiation protection high-pressure injection pump. The radiation saturation threshold is set to a value range of 30 Gy to 40 Gy, and the backflow threshold is set to a value range of 0.8 to 1.2.

[0045] Preferably, in the step of sending the low-level command to stop the operation of the radiation-proof high-pressure injection pump: the low-level command is sent via the control bus to the external hardware interrupt interface integrated on the main control circuit board; after receiving the command, the external hardware interrupt interface directly cuts off the power supply to the motor driver using the underlying hardware logic gate circuit and simultaneously activates the mechanical brake mechanism of the drive shaft to forcibly stop the injection. By combining the underlying hardware interrupt interface with the hardware logic gate circuit and the mechanical brake, the software scheduling cycle of the operating system is avoided, resulting in lower latency in the execution of the blocking action.

[0046] Preferably, in the step of extracting the vascular lumen geometric parameters of the one-dimensional lumened parameter fluid network topology: when the extracted vascular branch radius is less than the lower limit threshold of imaging resolution, the vascular branch radius is forcibly corrected to the lower limit threshold of imaging resolution; wherein, the lower limit threshold of imaging resolution is conventionally taken as 0.1 mm to 0.3 mm.

[0047] This invention provides a method for real-time optimization of interventional radiotherapy plans for tumors. It has the following beneficial effects:

[0048] 1. This invention acquires in-situ physical and electrophysiological physical quantities during the injection phase through a multi-channel synchronous data acquisition module, outputs multimodal synchronous temporal characteristics, and reconstructs the blood flow distribution ratio based on the calculated dynamic flow resistance increment, thereby accumulating and generating a real-time cumulative radiation dose distribution matrix. This method introduces real-time physiological parameter feedback during the operation, transforming the static assessment based on preoperative angiography into a dynamic calculation process, improving the accuracy of three-dimensional dose distribution estimation during the injection process, and solving the problem of deviation between the actual radiation dose distribution and the preoperative static plan.

[0049] 2. This invention extracts the complex fluid impedance phase angles of in-situ blood pressure and injection flow signals, combines them with a net transtissue bioelectrical impedance sequence that eliminates interference from macroscopic tissue volume changes, and inputs this sequence into a cross-validation observation model based on a Kalman filter framework for state determination. This processing logic eliminates interference from increased resistance caused by vasospasm, separates the actual fluid loss state caused by arteriovenous shunts, and avoids the problem of radioactive microspheres entering the system circulation due to unidentified abnormal shunts.

[0050] 3. This invention continuously monitors the real-time cumulative radiation dose distribution matrix and the catheter backflow pressure gradient. When the radiation saturation threshold or backflow threshold is reached, a command is sent to the external hardware interrupt interface. The underlying hardware logic gate circuit directly cuts off the power supply to the motor driver and simultaneously activates the mechanical brake mechanism. This control mechanism avoids the software scheduling cycle of the operating system, eliminates the delay caused by communication response lag, improves the execution speed of stopping injection in abnormal states, and prevents the risk of excessive radiation to non-target areas and microsphere backflow. Attached Figure Description

[0051] Figure 1 This is the architecture topology diagram of the real-time optimization system for interventional radiotherapy planning of the present invention;

[0052] Figure 2 This is a flowchart illustrating the real-time optimization method for interventional radiotherapy planning of tumors according to the present invention.

[0053] Figure 3 This is a comparison curve showing the change of the real-time radiation dose distribution mapping error of the present invention with the injection time;

[0054] Figure 4 This is a transient response comparison chart of the backflow risk assessment index under sudden interference conditions according to the present invention. Detailed Implementation

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] See attached document Figure 1 This invention provides a real-time optimization method for interventional radiotherapy planning of tumors. The execution of this method relies on an interventional acquisition and control system, which includes an interventional sensing front end, a body surface receiving end, a treatment execution module, and a computing center.

[0057] The interventional sensing front end comprises a radiographic microcatheter, a miniature pressure sensor, and a monopolar microelectrode. The miniature pressure sensor is embedded in the distal sidewall of the radiographic microcatheter to acquire in-situ blood pressure signals within the target vessel. The monopolar microelectrode is attached in a ring shape to the outer wall of the radiographic microcatheter distal to the miniature pressure sensor. The radiographic microcatheter retains a working channel for the flow of radioactive microspheres. Signal transmission cables from the miniature pressure sensor and the monopolar microelectrode extend proximally along the wall interlayer of the radiographic microcatheter and connect to the computing center.

[0058] The surface receiver comprises an array of grounded surface-mounted electrodes. The computing center applies a high-frequency excitation current to the monopolar microelectrodes via a signal transmission cable. The grounded surface-mounted electrodes receive the transmitted current and generate a raw transtissue bioelectrical impedance signal. Simultaneously, the grounded surface-mounted electrodes acquire electrocardiogram (ECG) signals and baseline respiratory impedance drift signals, merging them into a macroscopic physiological synchronization signal. The grounded surface-mounted electrodes transmit the raw transtissue bioelectrical impedance signal and the macroscopic physiological synchronization signal to the computing center.

[0059] The treatment execution module includes a radiation-proof high-pressure injection pump and an external hardware interrupt interface. The power output shaft of the radiation-proof high-pressure injection pump is connected to the proximal sample loading interface of the radiation-specific microcatheter. The radiation-proof high-pressure injection pump generates an injection flow rate signal, which is transmitted to the computing center via a data bus. The external hardware interrupt interface is integrated into the main control circuit board of the radiation-proof high-pressure injection pump. This interface establishes a one-way command reception link via the control bus to receive control commands issued by the computing center.

[0060] The computational center comprises a multi-channel synchronous data acquisition module and a real-time calculation module. The multi-channel synchronous data acquisition module receives in-situ blood pressure signals, transtissue bioelectrical impedance analysis (TIA) raw signals, macroscopic physiological synchronization signals, and injection flow rate signals. It performs clock stamp alignment and analog-to-digital conversion on the input signals and outputs aligned digital signals to the real-time calculation module. The real-time calculation module receives the aligned digital signals, performs optimization calculations, and outputs a low-level command to the external hardware interrupt interface. The external hardware interrupt interface, upon receiving the low-level command, triggers the radiation-proof high-pressure injection pump to stop the injection.

[0061] See attached document Figure 2 This invention provides a method for real-time optimization of interventional radiotherapy plans for tumors, comprising the following steps:

[0062] S1. The computational center performs dimensionality reduction processing on the three-dimensional angiography images before tumor interventional surgery, calculates and pre-allocates the initial flow resistance coefficient, and generates the initial three-dimensional spatial radiation dose distribution matrix.

[0063] S2. The multi-channel synchronous data acquisition module acquires the in-situ physical and electrophysiological physical quantities corresponding to the tumor intervention injection stage, and outputs aligned multimodal synchronous temporal characteristics by combining the initial three-dimensional spatial radiation dose distribution matrix.

[0064] S3. The real-time solution module analyzes multimodal synchronous timing features based on feature decoupling and cross-tissue dielectric cross-verification logic, excludes vasospasm and arteriovenous shunt states, and calculates the dynamic flow resistance increment corresponding to the diastolic phase of the target vessel terminal.

[0065] S4. The real-time calculation module reconstructs the blood flow distribution ratio by inputting dynamic flow resistance increment and particle distribution transfer function, and performs time-domain accumulation to generate a real-time cumulative radiation dose distribution matrix.

[0066] S5. The real-time calculation module monitors the real-time cumulative radiation dose distribution matrix and the catheter reflux pressure gradient. When the radiation saturation threshold or reflux threshold is reached, it sends a low-level command to the treatment execution module to stop the operation of the radiation protection high-pressure injection pump.

[0067] The following section will provide a detailed explanation of each step in the real-time optimization method for interventional radiotherapy planning, combining specific physical constraints with a multi-field coupling computational model.

[0068] See attached document Figure 2 The computational center performs preoperative vascular topology dimensionality reduction and baseline radiation dose matrix generation for tumor interventional procedures. In this embodiment, the specific process by which the computational center performs dimensionality reduction processing on preoperative three-dimensional angiography images for tumor interventional procedures, calculates and pre-allocates initial flow resistance coefficients, and generates an initial three-dimensional spatial radiation dose distribution matrix includes the following sub-steps:

[0069] S11. Based on the spatial distribution characteristics of blood perfusion in the tumor target area, the three-dimensional angiography image received by the central nervous system before tumor interventional surgery is calculated, and the three-dimensional angiography image before tumor interventional surgery is reduced in dimension to construct a one-dimensional lumped parameter fluid network topology.

[0070] To accurately reflect the cascade structure of blood vessels, the computational center employs a vascular centerline extraction algorithm to perform voxel skeletonization processing on pre-tumor interventional 3D angiography images, thereby identifying the bifurcation points and terminal points of the 3D arterial tree. Subsequently, the computational center maps the 3D vascular lumens between adjacent bifurcation points as one-dimensional line segments, while defining the terminal points as independent RC network nodes, thus generating a one-dimensional lumped-parameter fluid network topology diagram containing the hierarchical relationships of vascular branches. Regarding the specific implementation of the vascular centerline extraction algorithm, those skilled in the art can refer to existing tubular structure filtering algorithms based on Hessian matrices; the process of extracting 3D image skeleton features is a well-known technique in the field and will not be elaborated upon here.

[0071] S12. Based on the geometric features contained in the one-dimensional lumped parameter fluid network topology, the calculation center is the RC network node to calculate and pre-allocate the initial flow resistance coefficient.

[0072] Within the realm of macroscopic fluid mechanics, the laminar flow of Newtonian fluids within a rigid cylindrical tube follows Poiseuille's law. Based on this general physical principle, the computational center extracts the geometric parameters of the vascular lumen of the corresponding branch in the one-dimensional lumped-parameter fluid network topology diagram. As a preferred engineering protection method, considering that the terminal branches of microvessels are susceptible to partial volume effects in imaging, when the radius of the extracted vascular branch is less than a preset lower limit threshold for imaging resolution, the computational center forcibly corrects it to this lower limit threshold. The range of this lower limit threshold is usually determined based on the physical voxel size of the three-dimensional angiography equipment used, with a conventional value of 0.1 mm to 0.3 mm. This prevents system overflow anomalies triggered by the denominator approaching zero in subsequent division calculations from the underlying data.

[0073] Based on the corrected geometric parameters, the computational center derives the baseline fluid resistance values ​​for each node in the capacitive-resistive network according to the physical relationships of Poiseuille's law (translated from text). Specifically, the initial resistance coefficient is set to be directly proportional to the product of blood dynamic viscosity and the length of the corresponding vascular branch, and inversely proportional to pi and the fourth power of the radius of the corresponding vascular branch. To ensure that the fluid dynamics calculation results accurately reflect the patient's actual physiological state, the blood dynamic viscosity is not a conventional fixed constant. Instead, the computational center uses preoperative data on hematocrit and total plasma protein concentration, obtained through interpolation calibration using a pre-set modified Casson fluid rheology empirical model. After obtaining the initial resistance coefficient, since local flow rate is inversely proportional to resistance in a parallel fluid network, the computational center normalizes the allocation weights of all terminal branches using the reciprocal of the initial resistance coefficient of each node as the base weight, combined with the node constraint of conserving total flow rate at the parent vessel inlet. This yields the initial blood flow allocation ratio for each node in the capacitive-resistive network.

[0074] S13. After obtaining the weights of the microscopic local blood flow distribution, the calculation center constructs the initial three-dimensional spatial radiation dose distribution matrix based on the initial blood flow allocation ratio of each resistance-capacitance network node.

[0075] To establish a mapping between physical space and radiation energy, the computational center obtains the total activity of the expected injected radioactive microspheres. By multiplying the expected total activity of the injected radioactive microspheres by the initial blood flow distribution ratio of each RC network node, the computational center obtains the expected local radioactivity of each RC network node. Subsequently, the computational center uses a macroscopic medical internal radiation dose model to map the expected local radioactivity of each RC network node to a three-dimensional voxel space coordinate system corresponding to the three-dimensional angiography image before tumor interventional procedures. Based on this spatial mapping relationship, the computational center calculates the expected radiation absorbed dose of the tumor target area and surrounding normal tissues, thereby generating an initial three-dimensional spatial radiation dose distribution matrix. For the spatial radiation energy attenuation calculation method of the macroscopic medical internal radiation dose model, those skilled in the art can use existing standard nuclear medicine radiation dosimetry models, whose process for calculating the energy deposition distribution of radioactive isotopes in biological tissues is well-known in the art and will not be elaborated here.

[0076] See attached document Figure 2 During the radioactive microsphere injection phase, the multi-channel synchronous data acquisition module collects the corresponding in-situ physical and electrophysiological physical quantities during the tumor interventional injection phase, and outputs aligned multimodal synchronous temporal features by combining the initial three-dimensional spatial radiation dose distribution matrix. In this embodiment, the above signal acquisition and temporal feature synchronization alignment process specifically includes the following sub-steps:

[0077] S21. Based on the dramatic changes in the intraluminal flow field and the dynamic evolution of the dielectric properties of the target tissue during the tumor interventional injection process, the multi-channel synchronous data acquisition module performs closed-loop acquisition of in-situ hydrodynamic signals and cross-tissue bioelectrical impedance raw signals through the interventional sensing front end and the treatment execution module.

[0078] In the specific hardware sensing chain, a miniature pressure sensor embedded in the sidewall of the distal end of a radiation-dedicated microcatheter senses the absolute fluid pressure within the target blood vessel via an internal piezoelectric Wheatstone bridge and converts it into an analog electrical signal using a local signal conditioning chip, thereby generating a real-time in-situ blood pressure signal. Simultaneously, to acquire the volume of the injected fluid, the main control circuit board of the radiation-proof high-pressure infusion pump records the physical displacement of the driving plunger via an internal servo motor encoder. A multi-channel synchronous data acquisition module calculates the instantaneous injection volumetric flow rate by multiplying the derivative of the plunger displacement by the cross-sectional area of ​​the syringe lumen, thereby generating a real-time injection flow rate signal. To further obtain the local tissue dielectric properties reflecting the microsphere deposition state, the computing center applies a high-frequency, microampere-level safety excitation current of a specific frequency and amplitude to the unipolar microelectrode via an internally integrated signal generation circuit and a signal transmission cable. As a preferred electrical safety configuration, the frequency of this excitation current is set between 50kHz and 100kHz to avoid low-frequency neuromuscular stimulation effects, and the amplitude is strictly limited to the human body leakage current safety threshold (e.g., 100 microamperes). After the excitation current penetrates the target blood vessel wall and surrounding tumor tissue, it is received by an array of surface-grounded electrode patches distributed on the body surface as the loop ground. The signal conditioning circuit at the rear end of the surface-grounded electrode patches amplifies the transmitted current through transimpedance and performs low-pass filtering and envelope detection, thereby generating a closed-loop original transtissue bioelectrical impedance signal characterizing the dynamic changes in local tissue impedance.

[0079] S22. After completing the measurement of the local microscopic physical field, eliminating the interference of the patient's own macroscopic physiological movements on the microscopic measurements becomes crucial to ensuring data accuracy. During interventional procedures, the patient's heartbeat and spontaneous breathing cause periodic, non-rigid physical displacements of the thoracic and abdominal organs. These motion artifacts directly lead to baseline drift and impedance characteristic distortion in microfluidic measurements. To eliminate these physiological motion artifacts, the multi-channel synchronous data acquisition module needs to simultaneously extract macroscopic physiological features independent of the target area.

[0080] In its implementation, the surface-grounded patch electrode, while physically multiplexing as a high-frequency transmission dielectric current receiver, uses a frequency-divided bandpass filter network to parallel acquire low-frequency ECG potential difference changes on the patient's body surface. In addition to ECG signal acquisition, considering the regular changes in thoracic cavity geometry and internal gas dielectric constant caused by lung inflation and deflation—specifically, the significant decrease in local conductivity during inhalation leading to an overall increase in transtissue impedance—the front-end circuit of the surface-grounded patch electrode extracts low-frequency periodic baseline drift waveforms in the 0.1Hz to 0.5Hz range from the overall transtissue impedance signal. This frequency band is chosen because it strictly corresponds to the normal physiological respiratory rate of 6 to 30 breaths per minute in adults, directly mapping the patient's physical respiratory state. The surface-grounded patch electrode merges the extracted ECG waveform with the baseline respiratory impedance drift waveform to generate a macroscopic physiological synchronization signal for subsequent timing gating decisions, which is then transmitted to a multi-channel synchronous data acquisition module for analog-to-digital conversion and channel synchronization alignment.

[0081] S23. Given the differences in sensing transduction mechanisms and data transmission physical paths between the aforementioned multi-source heterogeneous signals, there is an inherent phase delay and a mismatch between the analog-to-digital conversion sampling rate and the signals in each dimension. To ensure the rigor of the physical causal relationships in subsequent multi-field coupling calculations and to avoid false embolism misjudgments due to time window misalignment, the multi-channel synchronous data acquisition module executes global hardware clock stamp alignment logic.

[0082] As a preferred underlying alignment method, the multi-channel synchronous data acquisition module incorporates a high-precision temperature-controlled crystal oscillator to provide a global reference hardware clock. At the moment of acquisition of each modal signal, the bus controller within the multi-channel synchronous data acquisition module assigns absolute hardware timestamps to the parallel in-situ blood pressure signal, injection flow signal, cross-tissue bioelectrical impedance raw signal, and macroscopic physiological synchronization signal. Subsequently, the multi-channel synchronous data acquisition module compares the data sampling rates of each channel and constructs a resampling time grid based on the sampling rate of the channel with the highest frequency. In a real medical environment, considering the potential data packet loss induced by bus transmission jitter, when the interval between adjacent valid timestamps exceeds a preset jitter tolerance threshold, the multi-channel synchronous data acquisition module prioritizes performing linear interpolation time compensation based on the preceding and following valid data to repair time axis breakpoints. In this embodiment, the preset jitter tolerance threshold is set to a range of 5 milliseconds to 10 milliseconds, determined based on approximately one percent of the typical cardiac cycle value (approximately 800 milliseconds). This setting can tolerate normal bus physical layer transmission delay fluctuations while promptly capturing real connection drops and packet loss anomalies.

[0083] For relatively low-frequency effective signal sequences, the multi-channel synchronous data acquisition module maps them to the resampling time grid based on the absolute hardware timestamp and applies a cubic spline interpolation algorithm for time-domain interpolation. After the upsampling interpolation calculation is completed, the multi-channel synchronous data acquisition module calls a digital anti-aliasing filter to eliminate spectral leakage that may be induced by the resampling process, thereby ensuring the physical authenticity of the high-frequency components after interpolation. For the node coefficient calculation of the cubic spline interpolation algorithm and the specific order design mechanism of the digital anti-aliasing filter, those skilled in the art can refer to the existing conventional standard procedures of digital signal processing. The process of multi-channel data resampling and reconstruction is a well-known technology in this field and will not be elaborated here. After clock calibration and resampling processing, the multi-channel synchronous data acquisition module performs matrix stitching of the physical data arrays of each dimension, and performs dimensional expansion and correlation mapping between the aligned data and the initial three-dimensional spatial radiation dose distribution matrix generated in S1, and outputs a comprehensive dataset that strictly corresponds under a unified absolute time coordinate system and integrates the static spatial prior distribution, that is, generates multimodal synchronous time series features.

[0084] See attached document Figure 2 The real-time calculation module built into the computational center performs fluid dynamic analysis and resistance increment extraction based on multi-field coupled state observations. In this embodiment, the specific process by which the real-time calculation module analyzes multimodal synchronous temporal features based on feature decoupling and cross-tissue dielectric cross-validation logic, excludes vasospasm and arteriovenous shunt states, and calculates the dynamic flow resistance increment corresponding to the diastolic phase of the target vessel terminal includes the following sub-steps:

[0085] S31. Within the realm of fluid dynamics, the pulsating flow state within an elastic pipe follows the linearized frequency domain solution of the Navier-Stokes equations, resulting in an inherent physical phase delay between the dynamic pressure and flow waveforms. Based on this theoretical mechanism, the real-time solution module receives the in-situ blood pressure signal and the injection flow rate signal contained in the multimodal synchronous timing characteristics.

[0086] To quantify the frequency domain dynamic response of the local flow field to cardiac pulsation and the injection source, the real-time calculation module performs a Fast Fourier Transform on the two time-domain signal waveforms to extract the corresponding complex pressure and flow rates. In conventional fluid dynamics resistance calculations, directly dividing by the time-domain mean often ignores the flow rate phase lead caused by vessel wall compliance. Therefore, the real-time calculation module calculates the ratio of the complex pressure and flow rates in the frequency domain. As a preferred algorithm fault-tolerant design, considering that the amplitude of the injection flow rate may be extremely small in certain non-fundamental harmonic frequency bands, direct division may trigger system data overflow when the denominator approaches zero. Based on this, the real-time calculation module uses a complex fluid impedance extraction formula to calculate the complex fluid impedance at specific harmonic frequencies. The complex fluid impedance extraction formula is as follows: ;

[0087] in, Indicates the first Complex fluid impedance at harmonic frequencies; Indicates the first The pressure complex spectrum at each harmonic frequency point; Indicates the first The conjugate of the complex spectrum of the flow at each harmonic frequency point; Indicates the first The square of the amplitude of the complex spectrum of the flow at each harmonic frequency; This represents the flow rate noise floor regularization parameter to prevent division by zero. Indicates the first The pressure complex spectrum at the nth harmonic frequency and the nth harmonic frequency The product of the conjugates of the complex spectra of the flow at each harmonic frequency; Indicates the first The sum of the squared amplitude of the complex spectrum of the flow at each harmonic frequency point and the flow rate noise floor regularization parameter to prevent division by zero.

[0088] The physical purpose of the aforementioned complex fluid impedance extraction formula is to accurately extract the complex domain dynamic impedance, which simultaneously includes frictional resistance and fluid inertia, by calculating the regularization ratio of the cross-power spectrum. In this embodiment, the specific value range of the flow rate noise floor regularization parameter to prevent division by zero is set as follows: to The physical basis for this determination is the equivalent variance of the background thermal noise power calibrated by the manufacturer of the miniature pressure sensor. This is intended to avoid the abnormal amplification of high-frequency measurement noise, which has no actual physical meaning, during the division, thus causing impedance spectrum distortion.

[0089] S32. During interventional procedures, the physical and mechanical stimulation of the tip of the radiotherapy-specific microcatheter often triggers sudden contraction of the smooth muscle in the local target blood vessel. This vasospasm leads to a significant decrease in vessel wall compliance and an abnormal increase in local fluid resistance. To eliminate this non-realistic microsphere embolism characteristic and avoid the system relying on a single extreme value for one-sided judgment, the real-time calculation module performs multi-dimensional vasospasm identification and transient attenuation compensation based on a joint determination of phase drift boundary and pressure baseline.

[0090] Specifically, the real-time calculation module extracts the fundamental frequency (i.e., The phase angle corresponding to the complex fluid impedance is defined as follows: When the flow rate lead corresponding to this phase angle exceeds the normal compliance phase threshold that a healthy blood vessel can accommodate, and the real-time calculation module simultaneously detects a step increase of more than 10% in the low-pass mean baseline of the in-situ blood pressure signal, the system state is comprehensively confirmed as triggering vasospasm. The normal compliance phase threshold typically ranges from -15 degrees to -5 degrees, and its determination is based on the statistical distribution characteristics of the prior elastic modulus of the target area vascular network in healthy individuals. After confirming the occurrence of vasospasm, the real-time calculation module... The real part of the complex fluid impedance at each harmonic frequency point is reconstructed using an inverse discrete Fourier transform to generate the measured fluid resistance in the time domain. To counteract the abnormal unsteady resistance changes caused by smooth muscle contraction, the real-time calculation module uses a transient compensation formula for vasospasm to calculate the compensated effective fluid resistance. The transient compensation formula for vasospasm is as follows: ;

[0091] in, Indicates time The effective fluid resistance after compensation; Indicates time Measuring fluid resistance; The base of the natural logarithm; This represents the spasm decay time constant; Indicates the current time variable; Indicates the moment when vasospasm is triggered; Indicates the duration of vasospasm; This represents the product of a negative spasm decay time constant and the duration of vasospasm. This represents the negative exponential compensation factor based on spasticity decay. The range of values ​​for the aforementioned spasticity decay time constant is... to Its value is determined based on the physiological half-life characteristics of the relaxation of calcium ion channels in vascular smooth muscle, so that the exponential compensation curve can fit the physical process of the natural dissipation of tissue stress.

[0092] S33. After obtaining the effective features of the fluid dimension, in order to obtain the pure dielectric features, the real-time calculation module parses the macroscopic physiological synchronization signal from the multimodal synchronization time sequence features.

[0093] Changes in blood volume caused by cardiac pumping and geometric deformations caused by thoracic respiration both introduce high-amplitude physiological low-frequency artifacts into the original transtissue bioelectrical impedance signal. To eliminate these macroscopic volume artifacts, the real-time calculation module delineates the end-diastolic period from the end of the T wave to the beginning of the next P wave in the electrocardiogram waveform, and simultaneously delineates the end-expiratory resting period using the trough and flattening region of the baseline respiratory impedance drift waveform. By intersecting these two periods, the real-time calculation module constructs a dual physiological gating time window and deploys data sampling points only within this window. For the extracted discrete dielectric data, the system uses median filtering to remove outliers, thereby generating a net transtissue bioelectrical impedance sequence that eliminates interference from macroscopic tissue volume changes.

[0094] S34. Abnormally proliferating microvascular networks within tumors are often accompanied by arteriovenous shunting. Once shunting occurs, radioactive microspheres will directly escape into the venous return system instead of depositing in the target area. To cross-validate this abnormal shunting state, a real-time solution module was constructed and run based on the Kalman filter framework to build and run an arteriovenous shunting cross-validation observation model.

[0095] This filtering model defines the estimated state of the actual microsphere deposition density as an internal one-dimensional state variable of the system, and synchronously inputs the time step. The compensated effective fluid resistance and net trans-tissue bioelectrical impedance sequence are used as external two-dimensional observation variables. To avoid singularity issues that may occur in matrix operations during the observation update phase, the real-time solution module injects a small regularization scalar on the main diagonal of the matrix when calculating the Kalman gain to invert the observation noise covariance matrix, ensuring full-rank invertibility. In the filter-recursive update logic, if the flow resistance characteristic remains at a low baseline and does not increase with the increase of the injection volume, and the ramp rate of the net trans-tissue bioelectrical impedance sequence is lower than the lower dielectric threshold characterizing the insulation properties of the microsphere material, the state update equation will, through dynamic adjustment of the Kalman gain, cause the estimated value of the true microsphere deposition density to converge sharply and approach zero.

[0096] In this embodiment, the dielectric lower limit threshold is set to a range of 0.1 to 0.5 Ω / mL, determined based on the effective polarization hysteresis coefficient of the insulating polymer microspheres used in a blood plasma environment. When this estimated value is lower than the preset safe deposition threshold, the real-time calculation module determines that there is a severe arteriovenous shunt in the current target vessel branch. As a reliable parameter setting, the safe deposition threshold is typically set to 10% to 20% of the theoretically expected deposition density of the branch, based on the statistical lower limit of microsphere escape rate caused by arteriovenous fistula diameter in clinical practice. As a countermeasure, the real-time calculation module forcibly sets the topological weight corresponding to the leakage pathway in the initial three-dimensional spatial radiation dose distribution matrix generated in the preceding process to zero, blocking the proportion of dose subsequently predicted and allocated to the risk area from the root cause of the system.

[0097] S35. After the above multi-dimensional feature purification and abnormal working condition elimination, the real-time calculation module finally enters the dynamic resistance parameter calculation stage.

[0098] The real-time calculation module extracts the mean transient fluid resistance within the aforementioned dual physiological gating time window. At this point, the extracted resistance variable has been completely freed from active squeezing artifacts from cardiac contraction, spasm artifacts from catheter stimulation, and erroneous estimations of shunt flow. The real-time calculation module performs a differential calculation between this mean and the initial flow resistance coefficient of the corresponding node pre-allocated in step S1, thereby outputting a dynamic flow resistance increment that truly reflects the degree of physical and mechanical blockage in the microcirculation lumen. This increment variable will serve as a key input benchmark for subsequent spatial redistribution mapping of radioactive microspheres, thus completing the physical quantity closed loop from in-situ sensor measurement to fluid network topology deduction at the macroscopic level.

[0099] See attached document Figure 2 After obtaining the actual flow resistance evolution state at the microcirculation end, the real-time calculation module built into the computational center reconstructs the blood flow distribution ratio by inputting the dynamic flow resistance increment and the particle distribution transfer function, and performs time-domain accumulation to generate a real-time cumulative radiation dose distribution matrix. In this embodiment, the process of resistance mapping analysis and real-time dynamic updating of the radiation dose distribution matrix specifically includes the following sub-steps:

[0100] S41. Based on the general technical principle of parallel flow splitting in fluid networks, the blood perfusion volume of each branch downstream of a vascular bifurcation node is negatively correlated with the terminal flow resistance of that branch. Considering that the introduction of radioactive microspheres as solid particles in interventional therapy scenarios will affect the distribution law of two-phase flow due to the axial aggregation effect of particles and inertial centrifugal force, it is no longer completely equivalent to the linear resistance distribution law of pure blood flow. Based on the above-mentioned multiphase flow dynamics physical mechanism, the real-time calculation module reconstructs the real-time flow splitting topology of the pipeline network based on the initial flow resistance coefficient and dynamic flow resistance increment of the corresponding node obtained from the previous process.

[0101] Based on the technical objective of mapping macroscopic resistance evolution to spatial local perfusion ratio, the real-time calculation module uses the resistance and particle distribution transfer function formula to calculate the dynamic blood flow distribution ratio of specific vascular branches. The resistance and particle distribution transfer function formula is as follows: ;

[0102] in, Indicates time The The dynamic blood flow distribution ratio of each vascular branch; Indicates the first The initial resistance coefficient of each vascular branch; Indicates time The Dynamic flow resistance increment of a vascular branch; Indicates time The Real-time integrated flow resistance of individual vascular branches; This represents the deviation coefficient of the two-phase flow distribution; Indicates time The Nonlinear admittance term for each vascular branch; Indicates the total number of parallel sub-branches; Indicates a parallel sub-branch index; This represents the summation operation over all parallel sub-branches; Indicates the first The initial resistance coefficient of each vascular branch; Indicates time The Dynamic flow resistance increment of a vascular branch; Indicates time The Real-time integrated flow resistance of individual vascular branches; Indicates time The Nonlinear admittance term for each vascular branch; Indicates time The sum of the nonlinear admittance terms of all parallel sub-branches; Represents the topological anti-singularity regularization constant; Indicates time The sum of the nonlinear admittance terms of all parallel sub-branches and the sum of the topological anti-singularity regularization constant; Indicates the target blood vessel branch index; This represents the current time variable.

[0103] The physical significance of the aforementioned drag and particle distribution transfer function formulas lies in using a nonlinear power function to correct the distribution error of particles deviating from the blood flow streamline due to inertia in a pure fluid model. In this embodiment, the specific value range of the two-phase flow distribution deviation coefficient is set to 1.1 to 1.3, and its determination is based on the statistical calibration value of the radial force deviation of solid microspheres of different sizes in the laminar flow profile. As a preferred algorithm protection logic, the topological anti-singularity regularization constant introduced in the denominator has a value range of... to Between these, it is used to prevent matrix calculation collapse caused by the denominator approaching zero in division operations under extreme complete embolism conditions (i.e., when all branch resistances tend to infinity, making the local admittance approximately zero).

[0104] S42. To address the issue of converting the spatial flow distribution ratio into the actual radiometric absorbed dose dimension, the real-time calculation module performs cross-domain conversion of physical dimensions. The number of microspheres carried in each injection during the interventional procedure is directly related to the volumetric displacement of the injector. Based on this correspondence, the real-time calculation module extracts the time integral of the injection flow signal, which has already undergone global timestamp alignment in the preceding process, within the current single calculation cycle, thereby obtaining the injection physical volume of the microspheres.

[0105] After obtaining the physical volume, the real-time calculation module multiplies the injected physical volume by the pre-entered specific activity constant of the radioactive microspheres (i.e., the radioactivity per unit volume) to calculate the total radiation source intensity entering the main blood vessel of the target area within the current cycle. Subsequently, the real-time calculation module utilizes the time... The Using the dynamic blood flow distribution ratio of each vascular branch as a weighting coefficient, the total radiation source intensity is spatially and topologically discretely distributed to the blood supply areas of each terminal microvessel. Furthermore, by combining the equivalent physical density of the local lesion tissue with the decay characteristic energy constant of the carried isotopes, a physical conversion from activity to absorbed radiation dose is performed, thereby generating an incremental dose distribution matrix that maps the current transient injection characteristics. This conversion step achieves, from a physical causal perspective, the dimensionality reduction and reconstruction of the dose deposited in the internal microscopic lesion tissue from the external macroscopic injection mechanical motion.

[0106] S43. Considering that the conversion result of a single injection can only reflect transient evolution, in order to fully characterize the global radiation effect of the microspheres accumulating in the three-dimensional lesion over time, the real-time solution module performs time-domain iterative superposition calculation of the spatial dimension.

[0107] During the injection cycle of the interventional procedure, the real-time calculation module updates the incremental dose distribution matrix generated in each calculation cycle with the spatial coordinate voxels corresponding to the initial three-dimensional spatial radiation dose distribution matrix containing anatomical prior positions, using the time step associated with the hardware sampling grid as the update frequency. Considering the possible slight physiological positional shifts of the patient during the operation, as a preferred three-dimensional alignment protection strategy, the real-time calculation module performs condition alignment and affine transformation registration on the spatial coordinate axes of the two matrices before matrix superposition, in conjunction with the respiratory phase reference in the previously extracted macroscopic physiological synchronization signal. After confirming that the topological relationship of the voxel space is correct, the real-time calculation module uses a weighted linear accumulation algorithm to superimpose the incremental values ​​frame by frame into the historical cumulative basis matrix. For the specific voxel mapping addressing and memory pool overwriting mechanisms involved in this weighted linear accumulation algorithm, those skilled in the art can refer to the conventional standards in the field of medical image three-dimensional reconstruction and volume rendering, which are well-known technologies in this field and will not be elaborated here. Through continuous temporal iterative operations, the real-time calculation module generates and continuously refreshes the real-time cumulative radiation dose distribution matrix. This matrix intuitively and quantitatively reflects the cumulative internal radiation exposure dose borne by each tissue spatial node within the target area up to the current time, thus providing direct data support for the identification of radiation saturation under subsequent operating conditions and the determination of the backflow physical threshold.

[0108] See attached document Figure 2 After mapping the radiation distribution of microscopic lesions, the system needs to simultaneously establish a physical safety baseline to cope with extreme operating conditions. Based on the aforementioned technical objective of establishing a multi-dimensional defense system, the real-time calculation module built into the computational center monitors the real-time cumulative radiation dose distribution matrix and the catheter reflux pressure gradient. When the radiation saturation threshold or reflux threshold is reached, a low-level command is sent to the treatment execution module to stop the operation of the radiation-proof high-pressure infusion pump. In this embodiment, the process of multi-dimensional physical safety constraint verification and hardware-linked closed-loop blocking specifically includes the following sub-steps:

[0109] S51. In the actual anatomical setting of radioembolization therapy, collateral blood supply from normal liver parenchyma or gastrointestinal tissues inevitably exists around the tumor. These healthy physiological tissues have a biological upper limit to their radiation tolerance. To prevent radiation-induced organ damage complications, the real-time calculation module performs spatial voxel-level saturation boundary checks on the previously generated real-time cumulative radiation dose distribution matrix.

[0110] In the specific calculation process, the real-time calculation module extracts the set of voxels with non-target area anatomical labels from the matrix and calculates the cumulative mean radiation dose of local small neighborhoods (e.g., spatial clusters formed by three consecutive adjacent voxels). As a preferred multi-dimensional fault-tolerant determination method, the system uses the mean of the spatial neighborhood instead of a single extreme value for comparison. This design aims to avoid the risk of system false triggering caused by single-point artifacts due to intraoperative image registration deviation. Subsequently, the real-time calculation module compares the above mean with the system's preset radiation saturation threshold in real time. In this embodiment, the radiation saturation threshold is typically set to a range of 30 Gy to 40 Gy, and its determination is based on the standardized tolerance limit parameter of healthy liver tissue in radiobiology. When the mean dose of the neighborhood touches or exceeds the above threshold, the real-time calculation module directly sets the dose saturation flag in the system's internal register.

[0111] S52. In addition to the biochemical constraints of radiation exposure, the mechanical instability of the hydrodynamic state at the tip of the interventional catheter is another direct physical cause of medical accidents. When the microcirculatory network of the target vessel is extensively blocked by radioactive microspheres, the compliant expansion capacity of the downstream vascular bed is exhausted. If fluid is forcibly injected into the lumen at this time, the hydraulic pressure cannot be effectively transmitted and released distally. This local fluid obstruction will generate a strong reverse pressure gradient in the annular gap between the catheter tip and the inner wall of the vessel, which can easily cause the high-activity microspheres to reflux and escape into non-target vessels.

[0112] To achieve highly sensitive prospective prediction before macroscopic reflux occurs, threshold determination relying solely on absolute static pressure values ​​is easily affected by fluctuations in the patient's baseline blood pressure or respiratory work. Based on the aforementioned technical objective of eliminating baseline drift interference to accurately monitor catheter reflux pressure gradients, the real-time calculation module extracts the in-situ blood pressure signal and the smoothed bolus flow rate signal. Considering the requirement for synchronization of multi-source data, before feature fusion, the real-time calculation module performs time-domain alignment between the above physical quantity sequence and the dynamic flow resistance increment obtained from previous calculations, based on a hardware clock stamp. After alignment, the real-time calculation module uses the reflux risk state assessment formula to calculate the risk characteristic value coupled with the multidimensional physical quantities. The reflux risk state assessment formula is: ;

[0113] in, Indicates time The backflow risk assessment index; This represents the dimensionless normalization coefficient used to match clinical measurement units with standardized thresholds; Indicates time The first time derivative of the in situ blood pressure signal; Indicates time In situ blood pressure signal; Represents the time derivative; Indicates time Smooth injection flow rate; This represents the small flow noise floor constant used to prevent division by zero; Indicates time The sum of the smoothed injection flow rate and the small flow rate noise constant to prevent division by zero; Indicates time The reciprocal of the sum of the smoothed push flow rate and the small flow noise floor constant to prevent division by zero; Indicates time The dynamic flow resistance increment; Indicates the current time variable; It represents the product of the dimensional normalization coefficient, the first time derivative of the in situ blood pressure signal, the reciprocal of the sum of the smoothed injection flow rate and the division-to-zero constant, and the dynamic flow resistance increment.

[0114] The physical significance of the aforementioned backflow risk assessment formula lies in nonlinearly amplifying and cross-validating three early warning characteristics: the first derivative of pressure reflects the steep rate of lumen compliance hardening, the reciprocal of flow rate characterizes the stagnant state of fluid forward kinetic energy loss, and the resistance increment confirms the actual embolism situation in the distal pipeline network from a physical causal source. In this embodiment, the range of the small flow rate noise floor constant to prevent division by zero is set to... to The value is milliliters per second, determined based on the inherent creep leakage rate of the injection pump motor during standby. This parameter ensures that the algorithm will not crash due to the denominator approaching zero when the fluid is completely stationary.

[0115] The real-time calculation module monitors time at a fixed clock cycle. The system employs a reflux risk assessment index. When this index exceeds a preset reflux threshold for consecutive judgment time windows, the system sets a reflux danger flag. The specific length of this judgment time window is set to 0.5 to 1.5 seconds, based on the premise of covering at least one complete cardiac cycle, thereby effectively filtering out transient blood pressure pulse interference caused by occasional coughing or premature ventricular contractions. As a preferred parameter configuration, the reflux threshold is typically set between 0.8 and 1.2, based on the Navier-Stokes reflux resistance model derived from the fluid dynamics of the coaxial annular gap between the catheter and the blood vessel.

[0116] S53, after the above multi-dimensional state verification of dose boundary and fluid stability, the real-time calculation module enters the final hardware linkage execution stage.

[0117] As a closed-loop safety cutoff mechanism, the real-time calculation module performs concurrent logical OR operations on the dose saturation flag and the backflow danger flag in the internal memory. Once any abnormal flag is detected to be activated, the real-time calculation module immediately triggers a high-priority interrupt sequence, outputting a definite low-level instruction to the treatment execution module. This instruction is not encapsulated by complex application-layer network protocols, but is directly sent to the external hardware interrupt interface integrated into the main control circuit board of the radiation-proof high-pressure injection pump via the industrial-grade control bus. Upon receiving the low-level instruction, the external hardware interrupt interface bypasses the polling queuing delay of the lower-level software system, directly cuts off the power supply to the motor driver's power stage using pure low-level hardware logic gates, and simultaneously activates the mechanical brake mechanism of the drive shaft, thereby forcing the radiation-proof high-pressure injection pump to stop the injection operation. This low-level hardware cutoff mechanism, which bypasses the application software layer, can achieve a millisecond-level fast response, curbing irreversible risks such as excessive radiation or microsphere backflow leakage from the source of mechanical execution. For the specific trigger edge level configuration and optocoupler electrical isolation protection mechanism involved in the hardware interrupt control link, those skilled in the art can refer to the conventional reliability design standards for medical device electrical safety and industrial control systems, which are well-known technologies in the field and will not be elaborated here.

[0118] This embodiment provides a specific application scenario for yttrium-90 radioactive microsphere embolization therapy in patients with primary hepatocellular carcinoma.

[0119] During the preparatory stage of the interventional procedure, medical personnel insert a radiographic microcatheter at the front end of the interventional sensor into a branch of the blood supply artery of the target tumor in the liver via the femoral artery. At the same time, they attach a grounded patch electrode to the corresponding position on the patient's chest and abdomen to establish a dielectric monitoring circuit.

[0120] The computational center extracted high-resolution three-dimensional angiography images of the patient before surgery, identified 12 terminal branch nodes of the target area arterial tree, and, based on the measured hematocrit level of 0.42, invoked a modified Casson model to pre-allocate and generate the initial flow resistance coefficients for each node and an initial three-dimensional spatial radiation dose distribution matrix covering the tumor and surrounding normal liver parenchyma. After completing the basic network topology reconstruction, medical personnel activated the treatment execution module, and a radiation-proof high-pressure infusion pump injected a suspension carrying radioactive microspheres into a radiation-dedicated microcatheter at a rated rate of an average of 0.8 ml / s.

[0121] During the injection process, the system enters the multi-field coupling state observation and dynamic resistance increment extraction stage. The multi-channel synchronous data acquisition module concurrently records the in-situ blood pressure signal transmitted by the miniature pressure sensor, the original cross-tissue bioelectrical impedance signal generated by the monopolar microelectrode and the surface grounding patch electrode circuit, and the injection flow signal at a reference sampling rate of 1000 Hz. At the 45th second of the injection, due to the initial embolism of the local microcirculation network and the superposition of the patient's slight spontaneous breathing work, the flow field in the lumen exhibits non-steady-state fluctuations. The real-time calculation module constructs a physiological gating time window using the extracted baseline respiratory impedance drift waveform, effectively isolating dielectric artifacts caused by changes in thoracic cavity volume. The system then performs differential analysis on the transient resistance mean and the initial flow resistance coefficient within the gating window, and finds that the dynamic flow resistance increment of the main branch has climbed from the baseline of 0 to 4.2 mmHg / mL at the current moment. Based on this dynamic flow resistance increment, the real-time calculation module synchronously reconstructs the dynamic blood flow distribution ratio of each downstream branch and accumulates the incremental dose distribution matrix into the real-time cumulative radiation dose distribution matrix according to the time domain step.

[0122] To verify the defensive effectiveness and parameter dynamic tracking accuracy of this scheme under extreme conditions, this embodiment introduces a comparative verification mechanism. During the experimental verification phase, the control logic equipped with the multi-dimensional dynamic joint judgment algorithm of this invention is set as the embodiment of this invention, while the traditional control logic that relies solely on a fixed initial resistance prediction distribution and lacks a physiological artifact removal mechanism is set as the traditional comparative example. Both sets of logic receive the same measured sensor data from the same source on the same fluid injection physical platform and perform parallel computation.

[0123] Between 67.8 and 68.3 seconds into the injection, the simulated patient experienced a sudden cough, causing a rapid increase in intrathoracic pressure and a transient spike of up to 35 mmHg in the in situ blood pressure signal. Traditional comparative methods, unable to distinguish between mechanical obstruction and physiological artifacts, saw their reflux assessment index spike instantly and break through the threshold due to this volumetric artifact, leading to frequent erroneous shutdowns of the radiation-protected high-pressure infusion pump. However, the real-time calculation module in this embodiment of the invention, benefiting from the feature stripping effect of dual physiological gating, effectively suppressed spurious high-frequency spikes. During this interference phase, the calculated reflux risk assessment index fluctuated steadily at only around 0.6, successfully avoiding false triggers.

[0124] Subsequently, at 69.5 seconds after injection, the microcirculation at the distal end of the target vessel experienced actual deterioration of fluid resistance and mechanical obstruction due to particulate matter accumulation. In this embodiment, the real-time calculation module strictly aligned the first-order time derivative of the pressure (measured at 120 mmHg / s), the smoothed injection flow rate (reduced to 0.15 mL / s), and the dynamic flow resistance increment (measured at 18.5 mmHg / mL) at this moment based on a hardware clock stamp. Substituting these values ​​into the reflux risk assessment formula, combined with the set small flow rate noise floor constant to prevent division by zero, and the set value... The dimensional normalized coefficients were used for transient calculation, and the calculated reflux risk assessment index caused by the actual blockage climbed to 1.48, which directly exceeded the preset reflux threshold of 1.2. After confirming that the index exceeded the limit for more than 0.8 seconds (covering the patient's current cardiac cycle), the real-time calculation module immediately sent a low-level command to the external hardware interrupt interface through the industrial-grade data bus.

[0125] The underlying hardware gate circuit cuts off the motor drive power within 3 milliseconds, and the mechanical brake mechanism of the radiation-proof high-pressure injection pump forcibly locks the drive shaft, successfully cutting off the physical kinetic energy before the visible backflow of macroscopic microspheres occurs, ensuring the absolute safety of non-target tissues.

[0126] See attached document Figure 3 , Figure 3 The solid black lines in the diagram represent embodiments of the present invention, while the dashed dark gray lines represent conventional comparative examples. In the initial injection phase (0-20 seconds), since the microspheres have not yet formed a large-area deposition, the local flow field deviates little from the baseline state, and both the solid black lines and the dashed dark gray lines maintain a low error level. As the injection process progresses and the target area microcirculation is gradually blocked by particles (after 30 seconds), because the conventional comparative example cannot perceive the actual evolution of fluid resistance, its estimated dose and actual deposition location exhibit a severe spatial misalignment. The dashed dark gray lines show a clear exponential divergence trend, with the maximum mapping error approaching 40%.

[0127] This invention, through the introduction of dynamic flow resistance increments and the drag-particle distribution transfer function, reconstructs the blood flow distribution ratio in real time. This allows for dynamic tracking of distribution deviations caused by microsphere axial aggregation and inertial centrifugal force. The solid black line remains within a 5% tolerance band throughout the entire injection cycle. This data directly demonstrates that the dynamic time-domain accumulation mechanism based on in-situ sensing and cross-domain dimensional conversion can eliminate the inherent prediction distortion of fixed flow resistance models under complex two-phase flow conditions, thus improving the accuracy of radiation deposition at the microscopic tissue level in interventional radiotherapy planning for tumors.

[0128] See attached document Figure 4 , Figure 4 The solid black line represents an embodiment of the present invention, the dashed dark gray line represents a conventional comparative example, and the dotted black line represents the reflux threshold. A sudden intrathoracic pressure pulse interference was introduced between 67.8 and 68.3 seconds of the test timeline. The conventional comparative example relies solely on static absolute pressure monitoring; during this phase, the dashed dark gray line, accompanied by sensor background noise, experienced violent oscillations due to the sudden volume artifact, momentarily breaking through the dotted black line, causing the system to trigger false alarms and resulting in frequent erroneous shutdowns of the radiation-proof high-pressure injection pump.

[0129] In this embodiment of the invention, the real-time calculation module utilizes the aligned multimodal synchronous timing characteristics output by the multi-channel synchronous data acquisition module to nonlinearly cross-couple the first-order time derivative of pressure, the reciprocal of flow rate, and the flow resistance increment. During the brief physiological artifact period, the black solid line effectively suppresses false high-frequency spikes thanks to the filtering effect of the judgment time window and the feature stripping of dual physiological gating. Furthermore, when the microcirculation fluid resistance truly deteriorates at 69.5 seconds after injection, triggering a physical precursor to backflow, the black solid line rapidly and stably penetrates the preset blocking boundary indicated by the black dotted line. This characteristic demonstrates that the backflow risk assessment formula established based on the Navier-Stokes backflow resistance model can achieve highly sensitive and forward-looking prediction of the lumen's mechanical instability state, while eliminating baseline drift interference and Gaussian background noise. Combined with the underlying cross-stack truncation mechanism of the external hardware interrupt interface, a highly reliable cross-domain fault-tolerant safety baseline is constructed.

Claims

1. A method for real-time optimization of interventional radiotherapy planning for tumors, characterized in that, include: The computational center performs dimensionality reduction processing on the three-dimensional angiography images before tumor interventional surgery, calculates and pre-allocates the initial flow resistance coefficient, and generates the initial three-dimensional spatial radiation dose distribution matrix. The multi-channel synchronous data acquisition module acquires in-situ physical and electrophysiological physical quantities corresponding to the tumor intervention injection stage, and outputs aligned multimodal synchronous temporal characteristics by combining the initial three-dimensional spatial radiation dose distribution matrix. The real-time solution module analyzes the multimodal synchronous timing features based on feature decoupling and cross-tissue dielectric cross-verification logic, excludes vasospasm and arteriovenous shunt states, and calculates the dynamic flow resistance increment corresponding to the diastolic phase of the target vessel terminal. The blood flow distribution ratio is reconstructed by inputting the dynamic flow resistance increment and the particle distribution transfer function formula, and then accumulated in the time domain to generate a real-time cumulative radiation dose distribution matrix. The system monitors the real-time cumulative radiation dose distribution matrix and the catheter reflux pressure gradient. When the radiation saturation threshold or reflux threshold is reached, a low-level command is sent to the treatment execution module to stop the operation of the radiation protection high-pressure infusion pump.

2. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 1, characterized in that, The specific steps for generating the initial three-dimensional spatial radiation dose distribution matrix include: The preoperative three-dimensional angiography image of the tumor interventional procedure was processed into a voxel skeleton as a dimension reduction process by using a vascular centerline extraction algorithm to identify bifurcation points and terminal points to construct a one-dimensional lumped parameter fluid network topology. Extract the vascular lumen geometric parameters of the one-dimensional lumen-parameter fluid network topology, call the modified Casson fluid rheology empirical model to obtain the blood dynamic viscosity, perform calculation and pre-allocate the initial flow resistance coefficient, and combine the node constraint condition of inlet total flow conservation to normalize the allocation weight to obtain the initial blood flow allocation ratio. The expected local radioactivity is obtained by multiplying the expected total activity of injected radioactive microspheres by the initial blood flow distribution ratio. The expected local radioactivity is then mapped to the corresponding three-dimensional voxel space coordinate system using a macroscopic medical internal radiation dose model, and the initial three-dimensional spatial radiation dose distribution matrix is ​​generated.

3. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 1, characterized in that, The steps for defining the output-aligned multimodal synchronization timing features specifically include: The in-situ blood pressure signal, injection flow signal and cross-tissue bioelectrical impedance raw signal are obtained through the interventional sensing front end and the treatment execution module; the low-frequency periodic baseline drift waveform is extracted and merged with the electrocardiogram waveform using the surface grounding patch electrode to generate a macroscopic physiological synchronization signal; The highest channel sampling rate is used as a benchmark to construct a resampling time grid. Hardware timestamps are used to map the in situ blood pressure signal, the injection flow rate signal, the original cross-tissue bioelectrical impedance signal and the macroscopic physiological synchronization signal to the resampling time grid. After time-domain interpolation calculation using a cubic spline interpolation algorithm, the data is spliced ​​to generate aligned data. The aligned data is then subjected to dimensional expansion and correlation mapping with the initial three-dimensional spatial radiation dose distribution matrix to output the aligned multimodal synchronous temporal features.

4. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 3, characterized in that, The step of calculating the dynamic flow resistance increment corresponding to the diastolic phase of the target vessel terminal specifically includes: Fast Fourier transform is performed on the in-situ blood pressure signal and the injection flow rate signal included in the multimodal synchronous timing features to extract the pressure complex spectrum and flow rate complex spectrum, and the complex fluid impedance at the harmonic frequency point is calculated using the complex fluid impedance extraction formula. The phase angle of the complex fluid impedance corresponding to the fundamental frequency is extracted. When the flow rate advance corresponding to the phase angle exceeds the normal compliance phase threshold and the low-pass mean baseline of the in situ blood pressure signal shows a step upward, it is confirmed as a vasospasm state. The effective fluid resistance is calculated using the vasospasm transient compensation formula in feature decoupling. A dual physiological gated time window is constructed using the macroscopic physiological synchronization signal contained in the multimodal synchronization time series feature, and combined with the original cross-tissue bioelectrical impedance signal to generate a net cross-tissue bioelectrical impedance sequence that eliminates the interference of macroscopic tissue volume changes. The effective fluid resistance and the net transtissue bioelectrical impedance sequence are input into an arteriovenous shunt cross-validation observation model based on a Kalman filter framework for state determination. After excluding the arteriovenous shunt state, the mean transient fluid resistance within the dual physiological gating time window is extracted. The difference between the average transient fluid resistance and the initial flow resistance coefficient of the pre-assigned corresponding node is calculated, and the dynamic flow resistance increment is output.

5. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 4, characterized in that, In the state determination step: In the filter recursive update logic, when the flow resistance feature maintains a low baseline and does not increase with the increase of injection volume, and the slope of the net trans-tissue bioelectrical impedance sequence is lower than the lower dielectric limit threshold, the estimated value of the true microsphere deposition density state approaches zero. When the estimated value of the actual microsphere deposition density is lower than the preset safe deposition threshold, it is determined that there is arteriovenous shunt in the target blood vessel branch, and the topological weight of the corresponding pathway is forcibly reset to zero.

6. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 1, characterized in that, The step of generating the real-time cumulative radiation dose distribution matrix includes: Based on the initial flow resistance coefficient and the dynamic flow resistance increment of the corresponding node, the dynamic blood flow distribution ratio is calculated using the drag and particle distribution transfer function formula; The physical volume of the injection is obtained by extracting the time integral of the injection flow rate signal within the calculation period, and multiplying it with the specific activity constant of the radioactive microspheres to obtain the total radiation source intensity; the dynamic blood flow distribution ratio is used as a weighting coefficient to perform physical conversion with the equivalent physical density and the isotope decay characteristic energy constant to generate an incremental dose distribution matrix. Using the hardware sampling grid associated with the time step as the update frequency, the incremental dose distribution matrix is ​​three-dimensionally aligned with the spatial coordinate voxels of the initial three-dimensional spatial radiation dose distribution matrix, and then superimposed frame by frame onto the historical cumulative basis matrix using a weighted linear accumulation algorithm, thereby continuously refreshing the real-time cumulative radiation dose distribution matrix.

7. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 6, characterized in that, In the step of 3D alignment of spatial coordinate voxels: By combining the respiratory phase reference in the extracted macroscopic physiological synchronization signal, the incremental dose distribution matrix and the spatial coordinate axes of the initial three-dimensional spatial radiation dose distribution matrix are aligned and affinely transformed for registration.

8. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 1, characterized in that, The specific steps for stopping the operation of the radiation-proof high-pressure injection pump include: Extract the set of non-target voxels from the real-time cumulative radiation dose distribution matrix to calculate the average cumulative radiation dose in a local small neighborhood. When the radiation saturation threshold is exceeded, set the dose saturation flag in the internal register. After performing time-domain alignment of the in-situ blood pressure signal, the injection flow rate signal and the dynamic flow resistance increment based on the hardware timestamp, the reflux risk assessment index is calculated using the reflux risk status assessment formula; when the reflux risk assessment index exceeds the reflux threshold within a continuous judgment time window, the reflux danger flag is set. A logical OR operation is performed on the dose saturation flag and the backflow danger flag. When the abnormal flag is detected to be activated, a high-priority interrupt sequence is triggered to send the level pull-down instruction to stop the operation of the radiation protection high-pressure injection pump. The radiation saturation threshold is set to a value range of 30 Gy to 40 Gy; the backflow threshold is set to a value range of 0.8 to 1.

2.

9. The method for real-time optimization of interventional radiotherapy planning for tumors according to claim 8, characterized in that, In the step of sending the low-level command to stop the operation of the radiation-proof high-pressure injection pump: The low-level instruction is sent via the control bus to the external hardware interrupt interface integrated on the main control circuit board. After receiving the instruction, the external hardware interrupt interface directly cuts off the power supply to the motor driver and simultaneously activates the mechanical brake mechanism of the transmission shaft to forcibly stop the injection.

10. A method for real-time optimization of interventional radiotherapy planning for tumors according to claim 2, characterized in that, In the step of extracting the vascular lumen geometric parameters of the one-dimensional lumped parameter fluid network topology: When the extracted blood vessel branch radius is less than the lower limit threshold of imaging resolution, the blood vessel branch radius is forcibly corrected to the lower limit threshold of imaging resolution. The lower limit threshold for imaging resolution is typically set between 0.1 mm and 0.3 mm.

Images