An ai planning and microsphere size optimization method for prostate arterial embolization surgery

Abstract

The application discloses an AI planning and microsphere particle size optimization method for a prostate arterial embolization operation, relates to the cross field of interventional radiology medical instruments and artificial intelligence, and specifically comprises the following steps: performing multimodal image preprocessing, acquiring and processing pelvic MRI images and ultrasonic Doppler blood flow parameters of a patient; reconstructing a three-dimensional topology of a prostate artery by using an improved 3DU-Net model and outputting geometric parameters; establishing a vascular flow resistance model based on the parameters to perform fluid dynamics simulation; constructing a multi-objective loss function to optimize the particle size of microspheres; monitoring blood flow in real time by using an optical flow method during the operation, constructing a judgment logic to output a stop injection warning, and integrating image processing, fluid simulation and particle size optimization to build an individualized operation planning framework, improve operation accuracy and reduce hidden dangers; with the aid of real-time blood flow monitoring and automatic judgment of an embolization end point during the operation, quantitative and standardized judgment of the embolization end point is realized, an integrated intelligent system is formed, and support is provided for safe and standardized operation.

Description

Technical Field

[0001] This invention relates to the intersection of interventional radiology medical devices and artificial intelligence, specifically to an AI planning and microsphere size optimization method for prostate artery embolization surgery. Background Technology

[0002] Benign prostatic hyperplasia (BPH) is a common urinary tract disease in elderly men. Prostatic artery embolization (PAE) is a mainstream minimally invasive treatment that embolizes the prostatic arteries to induce ischemic atrophy of the gland and relieve lower urinary tract symptoms. The core challenges of this procedure lie in the precise identification of the small branches of the prostatic arteries, the scientific selection of individualized microsphere sizes, and the objective determination of the embolization endpoint during the procedure. Currently, clinical PAE implementation heavily relies on physician experience and lacks integrated technical support including multimodal image fusion processing, individualized 3D vascular modeling, fluid dynamics simulation, and real-time intraoperative quantitative monitoring. This makes it difficult to consistently guarantee surgical precision and safety, necessitating the development of intelligent, individualized surgical planning and optimization techniques.

[0003] Current prostate artery embolization techniques have several shortcomings: In terms of image processing, traditional two-dimensional imaging cannot reconstruct the three-dimensional topology of the prostate artery, and conventional segmentation models lack sensitivity in identifying small branches, making it difficult to obtain accurate vascular geometric parameters. Regarding microsphere selection, particle size selection relies solely on clinical experience, without considering individualized blood flow and vascular characteristics for multi-objective optimization, easily leading to incomplete embolization, ectopic embolization, or reflux risks. In terms of simulation and monitoring, liquid-solid multiphase fluid dynamics are not used to simulate microsphere movement; intraoperatively, the physician relies solely on visual observation of DSA images to determine the embolization endpoint, lacking real-time quantitative blood flow monitoring and objective injection stoppage warning mechanisms. Current technologies do not integrate AI segmentation, fluid simulation, particle size optimization, and real-time intraoperative monitoring, resulting in low levels of individualization and intelligence, severely limiting surgical efficacy and safety.

[0004] Addressing the technical challenges of existing prostate artery embolization (PAE) procedures in areas such as image processing, microsphere selection, and intraoperative monitoring, this invention provides an AI-driven planning and microsphere size optimization method for PAE surgery. This invention achieves personalized and intelligent surgical planning throughout the entire process through multimodal image preprocessing, improved 3DU-Net vascular 3D reconstruction, liquid-solid multiphase fluid dynamics simulation, multi-target microsphere size optimization, and real-time intraoperative optical flow monitoring. It accurately determines the optimal microsphere size and objectively identifies the embolization endpoint, effectively reducing surgical risks and improving treatment outcomes. This addresses the core shortcomings of existing technologies and provides crucial technical support for the precise clinical implementation of PAE. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an AI planning and microsphere size optimization method for prostate artery embolization surgery. This method acquires data through multimodal image preprocessing, outputs geometric parameters through three-dimensional reconstruction of blood vessels, performs fluid dynamics simulation based on the parameters, constructs a multi-objective loss function to optimize microsphere size, and monitors blood flow velocity in real time during surgery using optical flow based on DSA video stream, constructs embolization endpoint determination logic according to expected thresholds, and outputs injection cessation warning, forming an integrated intelligent system from preoperative planning to intraoperative monitoring.

[0006] To solve the above-mentioned technical problems, this invention provides the following technical solution: an AI planning and microsphere size optimization method for prostate artery embolization surgery, the specific steps of which are as follows:

[0007] S100, Multimodal Image Preprocessing: Acquire pelvic MRI images and transrectal Doppler ultrasound blood flow parameters of the patient and perform preprocessing;

[0008] S200, 3D reconstruction of blood vessels: The improved 3DU-Net model is used to perform 3D topological reconstruction of the prostate artery in the preprocessed pelvic MRI images and output the vascular geometric parameters.

[0009] S300, Fluid Dynamics Simulation: Based on the aforementioned vascular geometric parameters and blood flow parameters, a vascular flow resistance model is established, and a liquid-solid multiphase fluid dynamics simulation is performed to simulate the movement of microspheres within the blood vessel;

[0010] S400, Microsphere Particle Size Optimization: Construct a multi-objective loss function, and output the optimal microsphere particle size that minimizes the multi-objective loss function by iterating through the simulation results of microspheres with different diameters;

[0011] S500, Intraoperative Real-time Monitoring: Based on DSA video stream, the system uses optical flow method to monitor blood flow velocity in real time. Based on the expected flow velocity threshold and expected flow velocity decay rate threshold obtained in advance through fluid dynamics simulation of the optimal microsphere particle size, the system constructs the embolization endpoint determination logic and outputs a stop-injection warning.

[0012] Furthermore, the multimodal image preprocessing uses a 3.0T magnetic resonance scanner to acquire pelvic MRI images. The scanning sequence uses a T2-weighted sequence with a slice thickness of 2.0 mm and no gap scanning. The scanning field of view covers the entire prostate and internal iliac artery branches. The pelvic MRI images are corrected by N4 bias field and normalized to the [0,1] interval. Transrectal Doppler ultrasound examination uses a 5MHz to 9MHz linear array probe to measure the peak systolic velocity and end-diastolic velocity of the bilateral prostatic arteries and obtain the resistance index.

[0013] Furthermore, the 3D reconstruction of blood vessels employs an improved 3DU-Net model. This model introduces an attention mechanism on top of the standard 3DU-Net to enhance the segmentation sensitivity for small branches of the prostatic artery, and is paired with a multi-scale feature fusion module to improve the segmentation accuracy of vessels with different diameters. The improved 3DU-Net model uses a hybrid loss function combining Dice-Loss and Focal-Loss. The input layer size of the improved 3DU-Net model is 128×128×64 voxels, and the Adam optimizer is used during training with an initial learning rate of 1×10⁻. 4 The vascular segmentation mask is converted into a topology map by centerline extraction, and the branch length, diameter and vascular curvature parameters are output. The improved 3DU-Net model identifies the third-order branch structure of the prostatic artery and outputs the geometric parameters.

[0014] Furthermore, the liquid-solid multiphase fluid dynamics simulation uses the Navier-Stokes equations to simulate the blood flow field and couples a discrete phase model to simulate the trajectory of microspheres within the blood vessel; blood, as a non-Newtonian fluid, is described using the Carreau-Yasuda model to describe the viscosity-shear rate relationship, the expression of which is:

[0015]

[0016] in, Shear rate; Zero shear viscosity; For infinite shear viscosity; It is a time constant; For shape parameters; The power-law exponent; the stated The value ranges from 0.01 Pa·s to 0.10 Pa·s. The value ranges from 0.001 Pa·s to 0.01 Pa·s. The value ranges from 0.1s to 10s. The value range is from 1.0 to 2.0. The value range is from 0.2 to 0.8;

[0017] In the simulation boundary conditions, the inlet is set as a pulsating velocity inlet, and the velocity waveform is obtained by fitting the peak systolic velocity and resistance index measured by transrectal Doppler ultrasound; the outlet is set as a pressure outlet, and the pressure value is set according to the venous pressure; during the simulation, the microsphere diameter within the preset particle size range is iterated, and the pressure drop and backflow risk coefficient after embolism are output.

[0018] Furthermore, the microsphere particle size is optimized, and the optimized microspheres are self-developing biodegradable polylactic acid modified microspheres. The microspheres are prepared by grafting 2,3,5-triiodobenzoic acid onto lactone monomers via monomer copolymerization, followed by copolymerization with lactide, resulting in a microsphere density of 1.35 g / cm³. The iodine atoms in the 2,3,5-triiodobenzoic acid constitute developing iodine atoms, which are gradually released as the ester bonds of the polylactic acid backbone break and excreted from the body with metabolites. Three sizes of microspheres—100 μm to 300 μm, 300 μm to 500 μm, and 500 μm to 700 μm—are provided for simulation iteration.

[0019] Furthermore, in the self-developing biodegradable polylactic acid modified microspheres with optimized microsphere particle size, the 2,3,5-triiodobenzoic acid is grafted onto the lactone monomer through an esterification reaction to generate a triiodobenzoyl lactide intermediate, which is then subjected to ring-opening copolymerization with lactide; the developing iodine atoms are distributed on the side groups of the polylactic acid backbone and are gradually released as the ester bonds are hydrolyzed and broken; the released iodine atoms are metabolized by the kidneys in the form of iodides and excreted in the urine.

[0020] Furthermore, the multi-objective loss function constructed by the microsphere size optimization is:

[0021]

[0022] in, This represents the loss value corresponding to the microsphere particle size. The diameter of the microsphere; Weighted by efficacy; For safety weights; This is the backflow penalty coefficient; To simulate the pressure drop after embolization; The pressure drop under ideal complete closure is a preset normal value; This represents the proportion of microspheres undergoing reverse motion in the simulation. and It satisfies α+β=1.

[0023] Furthermore, the intraoperative real-time monitoring employs the Lucas-Kanade sparse optical flow method to track the contrast agent's motion characteristics; based on the assumption of constant brightness, the brightness correspondence of pixels in the spatiotemporal domain is established, and after Taylor expansion and neglecting higher-order terms, the optical flow constraint equation is obtained:

[0024]

[0025] in, The spatial gradient of brightness along the x-direction, in pixels. −1 ; The spatial gradient of brightness along the y-direction, in pixels. −1 ; The rate of change of brightness with time t, in seconds. −1 ; This represents the velocity component of a pixel in the x-direction, expressed in pixels per second. This represents the velocity component of a pixel in the y-direction, expressed in pixels per second.

[0026] The position of the duct tip is automatically identified in the DSA video stream, and a region of interest (ROI) of 2cm × 2cm in front of the duct is locked. The velocity components of the pixels within the ROI are obtained by solving the optical flow constraint equation. and , where the subscript i represents the i-th feature point, i=1,2,…,m, and m is the total number of feature points tracked in the region of interest, which is dimensionless;

[0027] Based on the velocity components, obtain the average flow velocity scalar within the region of interest:

[0028]

[0029] in, The average flow velocity scalar within the region of interest, in pixels per second; For the first The velocity amplitude of each feature point, in pixels per second;

[0030] And continuously monitor the flow rate decay rate.

[0031] Furthermore, in the embolism endpoint determination logic, the flow velocity threshold is the expected flow velocity threshold obtained in advance through hydrodynamic simulation of the optimal microsphere particle size, and the flow velocity decay rate threshold is the corresponding expected flow velocity decay rate threshold; the expected flow velocity threshold is set to 5% to 20% of the initial flow velocity in the simulation, and the expected flow velocity decay rate threshold is set to 0.01 to 0.1; when the average flow velocity in the region of interest is lower than the flow velocity threshold, and the flow velocity decay rate is less than the flow velocity decay rate threshold, and this continues for a preset number of frames, the system triggers a stop-injection warning signal; the stop-injection warning signal is superimposed on the DSA screen with a color flow velocity vector image and output in conjunction with audio-visual prompts.

[0032] Compared with existing technologies, this AI planning and microsphere size optimization method for prostate artery embolization surgery has the following advantages:

[0033] I. This invention utilizes multimodal image preprocessing and 3D vascular reconstruction technology to accurately extract the 3D topological structure and geometric features of the prostate artery, enhancing the accuracy of identifying and segmenting small vascular branches, and providing stable and reliable vascular data support for surgical planning. Combined with liquid-solid multiphase fluid dynamics simulation to recreate the real motion of microspheres within the blood vessel, and relying on a multi-objective loss function to achieve individualized matching and optimization of microsphere particle size, the microsphere selection is highly compatible with the patient's vascular morphology and blood flow characteristics, effectively avoiding problems such as insufficient embolization and ectopic embolization. This technology deeply integrates image processing, fluid simulation, and particle size optimization to build a fully individualized surgical planning framework, significantly improving the accuracy of prostate artery embolization surgery, ensuring embolization effectiveness while reducing intraoperative safety risks, breaking away from the reliance on traditional experience-based diagnosis and treatment, and making surgical implementation more aligned with the patient's individual physiological state.

[0034] II. This invention utilizes real-time intraoperative blood flow monitoring and automatic determination of embolization endpoints. It employs optical flow to continuously track changes in blood flow velocity and combines this with preset thresholds to construct an objective embolization stoppage warning mechanism. This replaces the traditional subjective mode of manual visual judgment, achieving quantitative and standardized determination of the embolization endpoint. Real-time intraoperative warning signals are output and the blood flow status is presented intuitively, helping physicians accurately control embolization operation points, reducing human judgment bias, and effectively lowering the possibility of adverse events such as reflux and over-embolization. The entire technology forms an integrated intelligent system from preoperative planning and particle size optimization to intraoperative monitoring, standardizing surgical procedures, improving the stability and controllability of surgical implementation, providing solid support for the safe and standardized conduct of prostate artery embolization surgery, and optimizing patient treatment outcomes and safety.

[0035] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0037] Figure 1 A step-by-step framework diagram of an AI planning and microsphere size optimization method for prostate artery embolization surgery;

[0038] Figure 2 A flowchart illustrating the overall process of AI planning and microsphere size optimization for prostate artery embolization surgery;

[0039] Figure 3 This is a flowchart of a three-dimensional vascular reconstruction method for an AI-based planning and microsphere size optimization approach in prostate artery embolization surgery. Detailed Implementation

[0040] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0041] Example 1:

[0042] This embodiment addresses the prostate artery embolization surgery scenario for patients with benign prostatic hyperplasia (BPH) and moderate to severe lower urinary tract symptoms. It leverages AI technology to build a comprehensive surgical planning and microsphere size optimization system, forming a closed-loop management system from standardized image data processing to dynamic intraoperative monitoring. This system deeply matches the patient's individual physiological structure and clinical treatment needs, comprehensively improving the accuracy of surgical planning and the safety of intraoperative procedures. It provides an intelligent and standardized technical solution for minimally invasive embolization treatment of patients with moderate to severe BPH, solving the clinical challenges of traditional surgery that rely on physician experience, blind particle size selection, and ambiguous determination of the embolization endpoint.

[0043] First, multimodal image preprocessing was performed. A 3.0T MRI scanner was used to acquire pelvic MRI images of the patient. The scanning sequence was fixed at T2-weighted, with a slice thickness of 2.0mm and no gaps. The scanning field completely covered the entire prostate and internal iliac artery branches, comprehensively acquiring the morphology and distribution information of all vascular branches in the prostate region, eliminating any omissions of small branches or incomplete image coverage. The acquired pelvic MRI images were then subjected to N4 offset field correction and data normalization, uniformly standardizing the image data to the 0-1 range, completely eliminating magnetic field offsets and brightness unevenness generated during image acquisition. Interference factors such as grayscale deviation allow subsequent vascular segmentation and 3D reconstruction to be carried out based on clear, regular, and standardized image data, significantly improving the accuracy of subsequent model processing. At the same time, a 5MHz to 9MHz linear array probe is used to perform transrectal Doppler ultrasound examination, accurately measuring the peak systolic velocity and end-diastolic velocity of the bilateral prostatic arteries of the patient, and simultaneously obtaining the corresponding resistance index, so as to realistically restore the physiological blood flow state of the patient's prostatic arteries. This provides realistic blood flow input data for subsequent fluid dynamics simulation, avoiding the inconsistency between simulation results and the patient's actual situation due to deviations in blood flow parameters from the source.

[0044] Next, 3D reconstruction of blood vessels was performed using an improved 3DU-Net model that incorporates an attention mechanism and a multi-scale feature fusion module on top of the standard 3DU-Net. The attention mechanism significantly enhances the model's sensitivity to segmenting small branches of the prostatic artery, accurately capturing subtle vascular structures that are easily missed by conventional models. The multi-scale feature fusion module effectively improves the model's segmentation accuracy for vessels of different diameters, perfectly adapting to the uneven branch structure features of the prostatic artery. The model training was combined with a hybrid loss function of Dice-Loss and Focal-Loss, efficiently solving the problems of sample imbalance, blurred edge segmentation, and disconnection of small branches in the blood vessel segmentation task, ensuring that the blood vessel segmentation results closely match the real blood vessel contours. The input layer size of the improved 3DU-Net model was strictly set to 128×128×64 voxels, and the Adam optimizer was used during training with an initial learning rate of 1×10⁻. 4 This ensures stable model training, fast convergence, and unbiased segmentation results. By extracting the centerline, the vascular segmentation mask is transformed into a topological map, accurately identifying the three-level branch structure of the prostatic artery and outputting the vascular branch length, diameter, and vascular curvature geometric parameters. This fully presents the three-dimensional topological structure and core geometric parameters of the blood vessel, building a high-precision individualized vascular geometric model for subsequent fluid dynamics simulation and laying a solid core data foundation for simulation calculation.

[0045] Subsequently, fluid dynamics simulations were performed. Based on the previously acquired vascular geometry and blood flow parameters, a customized vascular flow resistance model was established. Liquid-solid multiphase fluid dynamics simulations were conducted, tailored to the patient's individualized vascular structure. The Navier-Stokes equations were used to simulate the blood flow field, fully reproducing the flow patterns, dynamic characteristics, and distribution of blood within the prostate artery. A coupled discrete-phase model was used to simulate the trajectory, deposition location, and embolism state of microspheres within the blood vessel, accurately presenting the entire process of changes in microspheres of different sizes as they flow through the blood. The Carreau-Yasuda model was used to describe the relationship between blood viscosity and shear rate; the expression for this model is: ,in, Shear rate; Zero shear viscosity; For infinite shear viscosity; It is a time constant; For shape parameters; The power-law exponent; the stated The value ranges from 0.01 Pa·s to 0.10 Pa·s. The value ranges from 0.001 Pa·s to 0.01 Pa·s. The value ranges from 0.1s to 10s. The value range is from 1.0 to 2.0. The value range is 0.2 to 0.8; it strictly conforms to the physical characteristics of non-Newtonian fluids in human blood, making the simulation results of blood flow infinitely close to the physiological reality; in the simulation boundary conditions, the inlet is set as a pulsating velocity inlet, and the velocity waveform is obtained by fitting the peak systolic velocity and resistance index measured by transrectal Doppler ultrasound; the outlet is set as a pressure outlet, and the pressure value is set according to the venous pressure, so that the simulation boundary conditions completely match the physiological pressure and flow velocity environment of human pelvic blood vessels. It iterates the microsphere diameter within the preset particle size range, outputs the pressure drop and backflow risk coefficient after embolism, and presents the embolism effect and safety hazards of microspheres of different particle sizes intuitively through quantitative data, providing a real, reliable and quantifiable reference for microsphere particle size optimization.

[0046] Next, microsphere particle size optimization was performed, using self-reproducible polylactic acid (PLA) modified microspheres as the embolization material. These microspheres were prepared by grafting 2,3,5-triiodobenzoic acid onto lactone monomers via monomer copolymerization, followed by copolymerization with lactide. The microsphere density was precisely controlled to 1.35 g / cm³. The iodine atoms in the microspheres constitute visible iodine atoms, which can be clearly visualized under imaging equipment during the procedure, facilitating real-time observation of the microsphere distribution and embolization range by physicians. The visible iodine atoms are gradually released as the ester bonds of the PLA backbone break and excreted with metabolites, avoiding long-term accumulation of iodine in the body and causing adverse reactions, thus adapting to the human metabolic mechanism. Three microsphere sizes were selected for simulation iteration: 100μm to 300μm, 300μm to 500μm, and 500μm to 700μm, comprehensively covering the commonly used particle size range for clinical prostate artery embolization. A multi-objective loss function was constructed, with the following formula: ,in, This represents the loss value corresponding to the microsphere particle size. The diameter of the microsphere; Weighted by efficacy; For safety weights; This is the backflow penalty coefficient; To simulate the pressure drop after embolization; The pressure drop under ideal complete closure is a preset normal value; This represents the proportion of microspheres undergoing reverse motion in the simulation. and The simulation results of three microsphere sizes are iteratively processed to output the optimal microsphere particle size that minimizes the multi-objective loss function. This maximizes the efficacy of embolization while minimizing the risk of reflux, precisely balancing surgical efficacy and operational safety, and ensuring that the microsphere particle size is perfectly adapted to the patient's individual vascular structure and blood flow characteristics.

[0047] Finally, intraoperative real-time monitoring was performed. Based on the DSA video stream, the Lucas-Kanade sparse optical flow method was used to monitor blood flow velocity in real time, accurately tracking the movement characteristics of the contrast agent within the blood vessel and acquiring dynamic changes in blood flow velocity data in real time. Based on the assumption of constant brightness, the brightness correspondence of pixels in the spatiotemporal domain was established. After Taylor expansion and neglecting higher-order terms, the optical flow constraint equation was obtained, ensuring the accuracy and stability of blood flow velocity calculation. The catheter tip position was automatically identified in the DSA video stream, and a 2cm × 2cm region of interest was locked in front of the catheter. This narrowed the monitoring range and eliminated interference from irrelevant areas, significantly improving the accuracy and specificity of blood flow velocity monitoring. The velocity components of pixels within the region of interest were obtained by solving the optical flow constraint equation. Based on these velocity components, the average flow velocity scalar within the region of interest was obtained. The system continuously monitors the flow rate decay rate. Based on the expected flow rate threshold and expected flow rate decay rate threshold obtained through hydrodynamic simulation with the optimal microsphere size, the embolization endpoint determination logic is constructed. The expected flow rate threshold is set to 5% to 20% of the initial flow rate in the simulation, and the expected flow rate decay rate threshold is set to 0.01 to 0.1. When the average flow rate in the region of interest is lower than the flow rate threshold and the flow rate decay rate is less than the expected flow rate decay rate threshold, and this continues for a preset number of frames (10 to 30), the system triggers a stop-injection warning signal. The stop-injection warning signal is superimposed on the DSA screen with a color flow rate vector image and accompanied by audio-visual prompts, allowing physicians to intuitively and quickly obtain embolization status prompts, stop microsphere injection in time, and avoid complications such as vascular damage and tissue necrosis caused by excessive embolization, thus ensuring the safety, controllability, and effectiveness of the procedure throughout.

[0048] The AI-based planning and microsphere size optimization method for prostate artery embolization surgery described in this invention forms a fully intelligent solution for the treatment of benign prostatic hyperplasia with moderate to severe lower urinary tract symptoms. This method strengthens the data foundation through multimodal image preprocessing, making vascular and blood flow information more closely match the patient's actual physiological state. It utilizes an improved 3D segmentation model to complete precise vascular topology reconstruction, providing a reliable structural basis for subsequent simulations. Based on fluid dynamics simulation, it recreates the movement patterns of blood and microspheres, and combines multi-objective optimization logic to determine the appropriate microsphere size, effectively balancing embolization efficacy and surgical safety. During the procedure, optical flow is used to monitor blood flow changes in real time and a mechanism for determining the embolization endpoint is constructed, achieving precise and controllable embolization, significantly reducing the risk of over-embolization and reflux. Figure 1 As shown in the diagram, this approach eliminates the reliance on physician experience in traditional surgery, improves the standardization and precision of surgical planning, and provides safe, efficient, and individualized technical support for minimally invasive embolization treatment of patients with moderate to severe benign prostatic hyperplasia. It effectively improves patient prognosis and promotes the development of prostate artery embolization surgery towards intelligence and precision.

[0049] Example 2:

[0050] This embodiment is applied to the prostate artery embolization surgery scenario for patients with benign prostatic hyperplasia and recurrent refractory hematuria. Addressing the clinical characteristics of refractory hematuria patients—fragile blood vessels, abnormal blood flow, and a high risk of vascular rupture and bleeding—AI technology enables precise individualized surgical planning and intelligent optimization of microsphere size. This allows for comprehensive control of embolization intensity and surgical risks, rapidly blocking abnormal bleeding vessels and controlling hematuria symptoms while avoiding additional damage to normal blood vessels and tissues. It provides a safe, efficient, and precise minimally invasive embolization treatment for patients with recurrent refractory hematuria due to benign prostatic hyperplasia, fundamentally addressing the clinical pain points of traditional surgical planning, such as subjective biases, inappropriate particle size selection, and difficulty in controlling the degree of embolization.

[0051] First, multimodal image preprocessing was performed. A 3.0T MRI scanner was used to acquire pelvic MRI images of the patient. T2-weighted sequences were used for scanning, with a slice thickness of 2.0mm to perform seamless scanning, ensuring the entire prostate and internal iliac artery branches were fully covered. This allowed for the complete capture of abnormally dilated, thin-walled, and easily bleeding vascular branches in the prostate region, comprehensively acquiring the morphology, location, and distribution information of the diseased vessels. This provided complete and clear imaging evidence for subsequent 3D vascular reconstruction. N4 bias field correction was performed on the pelvic MRI images, and normalization was performed to the 0-1 range to correct the MRI image distortion. Addressing issues such as grayscale deviation, magnetic field inhomogeneity, and artifact interference, this method makes the vascular outline and tissue boundaries clearer and sharper, significantly improving the accuracy and completeness of subsequent vascular segmentation. Transrectal Doppler ultrasound examination using a 5MHz to 9MHz linear array probe accurately measures the peak systolic velocity and end-diastolic velocity of both prostatic arteries and calculates the resistance index. This precisely captures abnormal blood flow changes caused by recurrent hematuria, realistically reflecting the hemodynamic state of the patient's diseased vessels. It provides blood flow data that closely matches the lesion characteristics for subsequent fluid dynamics simulation, preventing the simulation results from deviating from the patient's actual physiological state.

[0052] Next, 3D reconstruction of blood vessels was performed. An improved 3DU-Net model incorporating an attention mechanism and a multi-scale feature fusion module was used for prostate artery segmentation. The attention mechanism significantly enhances the ability to identify bleeding-related small vascular branches, accurately locating vulnerable segments prone to bleeding and preventing missed segmentation of bleeding branches. The multi-scale feature fusion module adapts to the irregular diameter and significant variations in thickness of the patient's diseased blood vessels, comprehensively improving the completeness and accuracy of segmentation for vascular branches of different thicknesses. The model was trained using a hybrid loss function of Dice-Loss and Focal-Loss, effectively reducing segmentation errors caused by blurred vessel edges, small branches, and structural distortion, ensuring that the segmentation results closely match the patient's actual vascular topology. The improved 3DU-Net model's input layer was strictly set to 128×128×64 voxels, and training used the Adam optimizer with an initial learning rate of 1×10⁻. 4This ensures rapid model convergence and stable, unbiased segmentation results. Centerline extraction converts the segmented blood vessel data into a topological map, accurately identifying the three-level branches of the prostatic artery and outputting geometric parameters such as branch length, diameter, and vessel curvature. This comprehensively presents the spatial structure and geometric features of the diseased blood vessels, building a high-precision, individualized vascular model for subsequent fluid dynamics simulations. This ensures the accuracy and reliability of the foundational data for simulation calculations. Figure 3 As shown.

[0053] Subsequently, fluid dynamics simulations were performed. A vascular flow resistance model was built based on the patient's specific vascular geometry and blood flow parameters. Liquid-solid multiphase fluid dynamics simulations were conducted targeting the vascular lesion characteristics of patients with refractory hematuria. The Navier-Stokes equations were used to simulate the blood flow field distribution, fully reproducing the blood flow dynamics and turbulent state within the abnormal bleeding vessels. A discrete phase model was combined to simulate the movement, deposition, aggregation, and embolization processes of microspheres within the blood vessels, accurately presenting the occlusion efficiency and coverage of the bleeding vessels by the microspheres. The Carreau-Yasuda model was used to characterize the changes in blood viscosity, strictly adhering to the non-Newtonian law of human blood. Fluid properties ensure that the blood flow simulation results fully match the actual physiological situation. The simulation sets pulsating velocity inlet and pressure outlet boundary conditions. The inlet velocity is generated by fitting the patient's ultrasound blood flow data, and the outlet pressure matches the human venous pressure standard. This allows the simulation environment to 100% replicate the physiological pressure and flow velocity state of the patient's pelvic blood vessels. Iteratively, the microsphere diameter within the preset particle size range is selected, and the pressure drop and reflux risk coefficient after embolization simulation are output. The quantitative data clarifies the sealing effect and reflux risk of different particle sizes of microspheres on bleeding vessels, providing a direct and accurate quantitative reference for microsphere particle size optimization, ensuring that the particle size selection is fully adapted to the sealing requirements of bleeding vessels.

[0054] Following microsphere size optimization, self-reproducible polylactic acid (PLA) modified microspheres were selected as the embolization material. These microspheres were prepared by grafting 2,3,5-triiodobenzoic acid onto a lactone monomer via esterification to generate a triiodobenzoyl lactide intermediate, which was then ring-opening copolymerized with lactide. The microsphere density was stable at 1.35 g / cm³. Iodine atoms were clearly visualized and distributed on the side groups of the PLA backbone, allowing for precise intraoperative localization of the embolization site and occlusion range. The visualized iodine atoms were gradually released as the ester bonds broke through hydrolysis. The released iodine atoms were metabolized by the kidneys as iodides and excreted in the urine. The residual risk in the body is perfectly matched to the metabolic state and safety needs of hematuria patients. A multi-objective loss function was constructed, and simulation data of microspheres with three sizes of 100μm to 300μm, 300μm to 500μm, and 500μm to 700μm were iteratively applied. Taking into account the two core indicators of embolization pressure drop and reflux risk, the optimal microsphere particle size that minimizes the multi-objective loss function was determined. This allows the microspheres to quickly block bleeding vessels and control hematuria symptoms, while avoiding problems such as insufficient embolization, excessive embolization, and vascular reflux caused by particle size that is too large or too small. This precisely balances the therapeutic effect and surgical safety.

[0055] Finally, real-time intraoperative monitoring was performed. Utilizing the DSA video stream, the Lucas-Kanade sparse optical flow method was used to monitor blood flow velocity in real time, continuously tracking the movement of the contrast agent within the bleeding vessel and capturing subtle changes and attenuation trends in blood flow velocity. Based on the assumption of constant brightness, a spatiotemporal brightness correspondence between pixels was established. After Taylor expansion and neglecting higher-order terms, the optical flow constraint equation was obtained, ensuring the accuracy and stability of blood flow velocity calculation. The catheter tip was automatically located in the DSA video stream, and a 2cm × 2cm region of interest was locked in front of it, focusing on the core monitoring area of ​​the bleeding vessel and eliminating interference from surrounding normal tissue and blood vessels, significantly improving the specificity and effectiveness of the monitoring data. The blood flow velocity components within the region were obtained by solving the optical flow constraint equation, and the calculation was performed... The system averages the flow rate and continuously monitors the flow rate decay rate. It establishes an embolization endpoint determination logic by combining the expected flow rate threshold corresponding to the optimal particle size and the expected flow rate decay rate threshold. The expected flow rate threshold is set to 5% to 20% of the simulated initial flow rate, and the expected flow rate decay rate threshold is set to 0.01 to 0.1. When the monitored data meets the threshold conditions and continues for a preset number of frames (10 to 30), the system immediately outputs a stop-injection warning. A color flow rate vector image is overlaid with a DSA display, accompanied by audio-visual prompts, allowing physicians to quickly and intuitively grasp the embolization completion status, promptly terminate microsphere injection, avoid damage to normal vascular tissue due to excessive embolization, and ensure that bleeding vessels are effectively blocked. This rapidly and thoroughly relieves the patient's intractable hematuria symptoms, ensuring the precision, safety, and therapeutic effect of the procedure throughout.

[0056] The AI ​​planning and microsphere size optimization method proposed in this invention has significant adaptability and practical value for the clinical scenario of benign prostatic hyperplasia complicated with recurrent refractory hematuria. The method focuses entirely on the fragility and abnormal blood flow characteristics of the diseased blood vessels. Multimodal image processing clearly captures easily bleeding vascular branches, and high-precision 3D vascular reconstruction fully presents the topological structure of the diseased blood vessels, providing precise data support for individualized treatment. Fluid dynamics simulation simulates the interaction mechanism between microspheres and bleeding vessels, and multi-objective optimization determines the optimal embolization particle size, rapidly sealing bleeding vessels while avoiding the risks of insufficient embolization or excessive damage. Intraoperative real-time monitoring accurately determines the embolization endpoint, providing visual early warnings to guide surgical procedures, effectively controlling hematuria symptoms and reducing the incidence of complications, such as... Figure 2 As shown in the figure. This approach solves the problems of vague planning, blind particle size selection, and difficulty in controlling the degree of embolism in traditional hematuria embolism treatment, providing a safe and reliable intelligent treatment plan for complex benign prostatic hyperplasia cases, significantly improving the effectiveness and safety of interventional surgery, and contributing to the improvement of the treatment level of clinical difficult cases.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for AI planning and microsphere size optimization in prostate artery embolization surgery, characterized in that, The specific steps of this method are as follows: S100, Multimodal Image Preprocessing: Acquire pelvic MRI images and transrectal Doppler ultrasound blood flow parameters of the patient and perform preprocessing; S200, 3D reconstruction of blood vessels: The improved 3DU-Net model is used to perform 3D topological reconstruction of the prostate artery in the preprocessed pelvic MRI images and output the vascular geometric parameters. S300, Fluid Dynamics Simulation: Based on the aforementioned vascular geometric parameters and blood flow parameters, a vascular flow resistance model is established, and a liquid-solid multiphase fluid dynamics simulation is performed to simulate the movement of microspheres within the blood vessel; S400, Microsphere Particle Size Optimization: Construct a multi-objective loss function, and output the optimal microsphere particle size that minimizes the multi-objective loss function by iterating through the simulation results of microspheres with different diameters; S500, Intraoperative Real-time Monitoring: Based on DSA video stream, the system uses optical flow method to monitor blood flow velocity in real time. Based on the expected flow velocity threshold and expected flow velocity decay rate threshold obtained in advance through fluid dynamics simulation of the optimal microsphere particle size, the system constructs the embolization endpoint determination logic and outputs a stop-injection warning.

2. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 1, characterized in that, In step S100, the multimodal image preprocessing uses a 3.0T magnetic resonance scanner to acquire pelvic MRI images. The scanning sequence uses a T2-weighted sequence with a slice thickness of 2.0 mm and no gap scanning. The scanning field covers the entire prostate and internal iliac artery branches. The pelvic MRI images are corrected by N4 bias field and normalized to the [0,1] interval. Transrectal Doppler ultrasound examination uses a 5MHz to 9MHz linear array probe to measure the peak systolic velocity and end-diastolic velocity of the bilateral prostatic arteries and obtain the resistance index.

3. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 1, characterized in that, In step S200, the 3D reconstruction of blood vessels employs an improved 3DU-Net model. This model introduces an attention mechanism to enhance the segmentation sensitivity of small branches of the prostatic artery, building upon the standard 3DU-Net model. It also incorporates a multi-scale feature fusion module to improve the segmentation accuracy of vessels with different diameters. The improved 3DU-Net model uses a hybrid loss function combining Dice-Loss and Focal-Loss. The input layer size of the improved 3DU-Net model is 128×128×64 voxels, and the Adam optimizer is used during training with an initial learning rate of 1×10⁻⁻⁶. 4 The vascular segmentation mask is converted into a topology map by centerline extraction, and the branch length, diameter and vascular curvature parameters are output. The improved 3DU-Net model identifies the third-order branch structure of the prostatic artery and outputs the geometric parameters.

4. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 1, characterized in that, In step S300, the liquid-solid multiphase fluid dynamics simulation uses the Navier-Stokes equations to simulate the blood flow field and couples a discrete phase model to simulate the trajectory of the microspheres within the blood vessel; blood, as a non-Newtonian fluid, is described using the Carreau-Yasuda model to describe the viscosity-shear rate relationship, the expression of which is: in, Shear rate; Zero shear viscosity; For infinite shear viscosity; It is a time constant; For shape parameters; The power-law exponent; the stated The value ranges from 0.01 Pa·s to 0.10 Pa·s. The value ranges from 0.001 Pa·s to 0.01 Pa·s. The value ranges from 0.1s to 10s. The value range is from 1.0 to 2.

0. The value range is from 0.2 to 0.8; In the simulation boundary conditions, the inlet is set as a pulsating velocity inlet, and the velocity waveform is obtained by fitting the peak systolic velocity and resistance index measured by transrectal Doppler ultrasound; the outlet is set as a pressure outlet, and the pressure value is set according to the venous pressure; during the simulation, the microsphere diameter within the preset particle size range is iterated, and the pressure drop and backflow risk coefficient after embolism are output.

5. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 1, characterized in that, In step S400, the microsphere particle size is optimized. The optimized microspheres are self-developing biodegradable polylactic acid modified microspheres. The microspheres are prepared by grafting 2,3,5-triiodobenzoic acid onto a lactone monomer via monomer copolymerization, followed by copolymerization with lactide. The microsphere density is 1.35 g / cm³. The iodine atoms in the 2,3,5-triiodobenzoic acid constitute developing iodine atoms, which are gradually released as the ester bonds of the polylactic acid backbone break and excreted with metabolites. Three sizes of microspheres, 100 μm to 300 μm, 300 μm to 500 μm, and 500 μm to 700 μm, are set for simulation iteration.

6. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 5, characterized in that, In step S400, in the self-developing biodegradable polylactic acid modified microspheres with optimized microsphere particle size, 2,3,5-triiodobenzoic acid is grafted onto the lactone monomer through an esterification reaction to generate a triiodobenzoyl lactide intermediate, which is then subjected to ring-opening copolymerization with lactide; the developing iodine atoms are distributed on the side groups of the polylactic acid backbone and are gradually released as the ester bonds are hydrolyzed and broken; the released iodine atoms are metabolized by the kidneys in the form of iodides and excreted in the urine.

7. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 1, characterized in that, In step S400, the multi-objective loss function constructed by the microsphere size optimization is: in, This represents the loss value corresponding to the microsphere particle size. The diameter of the microsphere; Weighted by efficacy; For safety weights; This is the backflow penalty coefficient; To simulate the pressure drop after embolization; The pressure drop under ideal complete closure is a preset normal value; This represents the proportion of microspheres undergoing reverse motion in the simulation. and It satisfies α+β=1.

8. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 1, characterized in that, In step S500, the intraoperative real-time monitoring uses the Lucas-Kanade sparse optical flow method to track the movement characteristics of the contrast agent; Based on the assumption of constant brightness, the brightness correspondence of pixels in the spatiotemporal domain is established. After Taylor expansion and neglecting higher-order terms, the optical flow constraint equation is obtained. The position of the duct tip is automatically identified in the DSA video stream, and a region of interest of 2cm×2cm in front of the duct is locked. The velocity components of the pixels in the region of interest are obtained by solving the optical flow constraint equation. The average flow velocity scalar in the region of interest is obtained based on the velocity components, and the flow velocity decay rate is continuously monitored.

9. The AI ​​planning and microsphere size optimization method for prostate artery embolization surgery according to claim 1, characterized in that, In step S500, in the embolism endpoint determination logic, the flow velocity threshold is the expected flow velocity threshold obtained in advance through fluid dynamics simulation for the optimal microsphere particle size, and the flow velocity decay rate threshold is the corresponding expected flow velocity decay rate threshold; the expected flow velocity threshold is set to 5% to 20% of the initial flow velocity in the simulation, and the expected flow velocity decay rate threshold is set to 0.01 to 0.1; when the average flow velocity in the region of interest is lower than the flow velocity threshold, and the flow velocity decay rate is less than the flow velocity decay rate threshold, and this continues for a preset number of frames, the system triggers a stop-injection warning signal; the stop-injection warning signal is superimposed on the DSA screen with a color flow velocity vector image and output in conjunction with audio-visual prompts.

Images