Multi-modal pathological feature fusion uro-lithotripsy parameter optimization method and system

Abstract

The application discloses a urinary calculus crushing parameter optimization method and system based on multi-modal pathological feature fusion, relates to the technical field of health information science, and extracts a preoperative feature vector F of a patient before operation, and calculates a theoretical fragmentation energy efficiency I according to the preoperative feature vector F, so as to generate an initial laser lithotripsy parameter P, thereby improving the initial accuracy and scientificity of treatment. In the operation, the original signal data set DSA is generated by collecting acoustic signals through a miniature hydrophone, and the pulse fragmentation efficiency η at the time t is calculated t , and the pulse fragmentation efficiency sequence SSEQ is constructed, the objective and real-time quantification of the lithotripsy effect is realized, and the limitation of visual observation is overcome. Finally, the state evaluation result is generated by comparison, and the laser lithotripsy parameter optimization strategy is executed, adaptive regulation is constituted, and the efficiency and safety of the whole operation are ensured.

Description

Technical Field

[0001] This invention relates to the field of in vivo laser lithotripsy technology, specifically to a method and system for optimizing lithotripsy parameters for urinary stones by fusing multimodal pathological features. Background Technology

[0002] The convergence of information technology and the medical field has driven the development of computer-aided diagnosis and treatment technologies, particularly in medical image analysis and treatment decision support systems. Quantitative analysis of multi-dimensional patient data provides objective evidence for clinical treatment. In urology, urinary tract stones are a common condition, and one of the main treatment methods is endoscopic lithotripsy. The effectiveness of this procedure is highly dependent on the lithotripsy equipment and the settings of its operating parameters, including pulse energy and pulse frequency. To find an optimal combination of lithotripsy parameters tailored to the physical and chemical characteristics of stones in different patients, a comprehensive analysis and integration of the patient's preoperative multi-dimensional pathophysiological characteristics is necessary. This technology can be applied to various clinical scenarios, such as developing initial high-force fragmentation strategies for high-density, high-hardness calcium oxalate stones, or planning efficient pulverization schemes for loosely textured infected stones.

[0003] In current laser lithotripsy procedures, the selection of equipment operating parameters primarily relies on the operator's personal clinical experience. Doctors typically make an initial assessment based on the size and approximate density of the stones in preoperative CT images, and then manually adjust the parameters intraoperatively using endoscopic visual feedback. This approach is subjective and fails to fully utilize multimodal pathological information, such as the chemical environment reflected in the patient's urinalysis data and the heterogeneity of the stone's internal microstructure, for comprehensive evaluation. Furthermore, existing parameter setting methods are mostly static open-loop controls. That is, once the initial parameters are set, adjustments during the procedure can only be made based on qualitative and delayed feedback from visual observation. When encountering uneven stone hardness in different areas, or when dust clouds from lithotripsy obstruct the view, doctors struggle to obtain accurate real-time feedback on lithotripsy efficiency, making it impossible to optimize equipment parameters instantly and quantitatively.

[0004] The above situation is caused by the lack of an objective methodology that can effectively fuse and quantitatively analyze the discrete multi-modal pathological data of the patient and the continuous intraoperative physical feedback signal. This deficiency may bring a series of abnormal effects in the clinic: if the parameter setting is too conservative, it may lead to low stone crushing efficiency, prolonged operation time, and even incomplete removal of stones in a single operation, increasing the risk and economic burden of the patient needing a second operation. On the contrary, if the parameter setting is too aggressive, it may exceed the necessary energy threshold, causing unnecessary thermal damage or mechanical impact on the surrounding normal tissues such as the ureteral wall or the renal collecting system, leading to postoperative bleeding, ureteral stenosis and other complications. At the same time, due to the lack of real-time feedback, when dealing with complex stones with uneven texture, the energy utilization efficiency and treatment consistency of the entire process are difficult to guarantee. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization method and system, which solves the problems mentioned in the background art.

[0006] To achieve the above object, the present application is realized by the following technical scheme: a multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization method, comprising the following steps:

[0007] S1, retrieve patient medical data from a hospital database through an API interface, process it through a deep learning model, and extract a preoperative feature vector;

[0008] S2, calculate the theoretical fragmentation energy efficiency and generate the laser lithotripsy parameters according to the preoperative feature vector, combined with the fracture toughness in material mechanics and the influence of the fluid chemical environment;

[0009] S3, start lithotripsy according to the initial laser lithotripsy parameters, collect the sound waves generated by laser lithotripsy through a miniature hydrophone integrated at the end of an endoscope, and convert them into digital signals to generate an original signal dataset;

[0010] S4, calculate the real-time pulse fragmentation efficiency according to the original signal dataset, combined with the analysis idea of impact dynamics and acoustic emission technology, and construct a real-time pulse fragmentation efficiency sequence;

[0011] S5, calculate the average fragmentation energy efficiency according to the real-time pulse fragmentation efficiency sequence, and compare it with the theoretical fragmentation energy efficiency to generate a state evaluation result;

[0012] S6, execute the laser lithotripsy parameter optimization strategy according to the state evaluation result, combined with the current laser lithotripsy parameters and the nonlinear correction mechanism combined with the feedback deviation.

[0013] Preferably, S1 includes S11;

[0014] S11, retrieve patient medical data including urine routine test data DUR and CT image data DCT from a hospital database through an API interface;

[0015] According to the urine routine test data DUR, the urine pH value is extracted, according to the CT image data DCT, the 3D U-Net segmentation model is used to obtain the three-dimensional segmentation mask Ms of the CT image data DCT, and the stone volume Vs is obtained according to the three-dimensional segmentation mask Ms. The three-dimensional segmentation mask Ms is used as a filter to act on the CT image data DCT to obtain the original Hounsfield unit of the stone region and calculate the arithmetic mean of the original Hounsfield unit to obtain the average Hounsfield unit Hu. According to the three-dimensional segmentation mask Ms, the three-dimensional image data block VOL of the stone is extracted from the CT image data DCT and input into the 3D-CNN feature extraction model to obtain the stone texture heterogeneity parameter ;

[0016] According to the urine pH value, the stone volume Vs, the average Hounsfield unit Hu and the stone texture heterogeneity parameter , a preoperative feature vector F = [pH, Vs, Hu, ] is generated.

[0017] Preferably, S2 includes S21 and S22;

[0018] S21, according to the preoperative feature vector F, combining the fracture toughness in material mechanics and the influence of chemical environment, using the ln function to simulate the nonlinear characteristics of the internal brittleness of the stone for the stone texture heterogeneity parameter τ and the stone volume Vs, using the ln function to reflect the nonlinear relationship of the influence of stone hardness for the average Hounsfield unit Hu, and introducing the chemical environment factor brought by the urine pH value. The stone is most difficult to break under the standard of urine pH value 7, and the stone is easy to break when the urine pH value deviates from 7. Through the above calculation process, the original theoretical fragmentation energy efficiency I r is obtained. r The tanh function is used for normalization to obtain the theoretical fragmentation energy efficiency I.

[0019] The calculation expression of the theoretical fragmentation energy efficiency I is as follows:

[0020] ; ;

[0021] In the formula, V0 represents a preset standard stone volume, Hu0 represents a preset standard stone Hounsfield unit, which is obtained by taking the average value of the stone volume in the hospital database, the ln function represents the logarithmic function operation, and the exp function represents the natural exponential function operation.

[0022] Preferably, S22 matches the data according to the theoretical fragmentation efficiency I with a preset strategy database to obtain initial laser lithotripsy parameters P, the initial laser lithotripsy parameters P including pulse energy pe and pulse frequency pf, wherein the preset strategy database sets the logic as follows:

[0023] If the theoretical fragmentation efficiency I is less than a preset first-level theoretical fragmentation efficiency threshold I1, it indicates that the stone is hard and stable, a powerful fragmentation strategy is adopted, and a set of high-energy and low-frequency laser lithotripsy parameters are retrieved;

[0024] If the theoretical fragmentation efficiency I is greater than a preset second-level theoretical fragmentation efficiency threshold I2, it indicates that the stone is fragile and unstable, a pulverization strategy is adopted, and a set of low-energy and high-frequency laser lithotripsy parameters are retrieved;

[0025] If the preset first-level theoretical fragmentation efficiency threshold I1 is less than or equal to the theoretical fragmentation efficiency I and is less than or equal to the preset second-level theoretical fragmentation efficiency threshold I2, it indicates that the stone has both hard and stable characteristics, a balanced strategy is adopted, and a set of balanced energy and frequency laser lithotripsy parameters are retrieved.

[0026] Preferably, S3 includes S31;

[0027] S31, according to the initial laser lithotripsy parameters P, starts to perform the lithotripsy operation, when the laser pulse impacts the stone, the generated sound wave propagates in the surrounding physiological saline medium, reaches the micro hydrophone integrated in the distal end of the endoscope, and generates an analog electrical signal S at time t which is consistent with the change rule of the sound wave pressure. t and performs digital signal conversion, and the conversion operation is as follows:

[0028] First, the analog electrical signal S at time t is subjected to signal conditioning and is amplified through a low-noise amplifier. t Further, the amplified analog electrical signal S at time t is subjected to band-pass filtering. t The low-frequency noise caused by the perfusion flow and the high-frequency electronic noise generated by the device circuit which are irrelevant to the lithotripsy event are filtered out, and finally, the signal is digitized. t The denoised analog electrical signal S at time t is sent into a high-speed analog-to-digital converter, the continuous analog signal is snapped, the voltage value at each time point is converted into a discrete binary number, a structured digital matrix is obtained, and an original signal data set DSA is generated.

[0029] Preferably, S4 includes S41 and S42;

[0030] S41, according to the original signal data set DSA, performs mathematical transformation and analysis;

[0031] Perform time-domain analysis on the original signal dataset DSA to search for the maximum absolute value of the signal amplitude, labeled as the peak sound pressure level sp at time t. t The peak sound pressure sp at time t t Starting from the given time point and tracing back from that point, find the times when the signal first reaches 10% and 90% of its peak amplitude, respectively, and calculate the difference to obtain the pulse rise time tr. sp ;

[0032] Frequency domain analysis is performed on the original signal dataset DSA by executing a Fast Fourier Transform. Based on a preset spectral analysis threshold Z, the transformed spectrum is divided into low-frequency and high-frequency regions. The total energy E in the high-frequency region at time t is calculated by numerical integration of the spectrum. Ht The total energy E in the low-frequency region at time t Lt ;

[0033] Based on the peak sound pressure sp at time t t Pulse rise time tr sp The total energy E in the high-frequency region at time t Ht and the total energy E in the low-frequency region at time t Lt This generates an intermediate feature vector G.

[0034] Preferably, in step S42, based on the intermediate eigenvector G and combining the analytical concepts of impact dynamics and acoustic emission technology, the pulse fragmentation efficiency η at time t is calculated. t Based on the calculated pulse fragmentation efficiency η at time t t Add this to a time series to form the pulse fragmentation performance sequence SSEQ=[η1, η2, ..., η t ];

[0035] Among them, the pulse fragmentation efficiency η at time t t The calculation expression is as follows:

[0036] ;

[0037] In the formula, tanh represents the hyperbolic tangent function operation, and K W This represents the bulk modulus of water, specifically 2.2 GPa. T represents the preset laser pulse width, calibrated by the device itself. th This indicates the preset invalid sound pressure threshold.

[0038] Preferably, S5 includes S51;

[0039] S51. Based on the pulse fragmentation efficiency sequence SSEQ, the pulse fragmentation efficiency η at time t... tAs a starting point, the latest N pulse fragmentation efficiency η is traced back and a moving average is taken to obtain the latest average pulse fragmentation efficiency η μ Wherein, the specific setting method of N is as follows: the feedback time window △t is set by the relevant staff according to their own expectations, the feedback time window △t is multiplied by the pulse frequency pf to obtain the number of pulse fragmentation efficiency η that needs to be traced back;

[0040] The latest average pulse fragmentation efficiency η μ Is compared with the theoretical fragmentation efficiency I, and the state evaluation result is generated according to the comparison result;

[0041] If the latest average pulse fragmentation efficiency η μ <95%×theoretical fragmentation efficiency I, indicating that the actual efficiency is lower than expected, and the parameter needs to be increased to increase the efficiency, and the state label output is-1;

[0042] If 95%×theoretical fragmentation efficiency I≤latest average pulse fragmentation efficiency η μ ≤105%×theoretical fragmentation efficiency I, indicating that the current parameter is in an ideal state and does not need to be adjusted, and the state label output is 0;

[0043] If the latest average pulse fragmentation efficiency η μ > 105%×theoretical fragmentation efficiency I, indicating that the actual efficiency is higher than expected, and the parameter needs to be reduced to protect the tissue, and the state label output is 1.

[0044] Preferably, S6 comprises S61;

[0045] S61, according to the state evaluation result, execute the laser lithotripsy parameter optimization strategy, if the state label is 0, no parameter optimization is needed, if the state label is 1 or-1, parameter optimization is needed, and the parameter optimization method is as follows:

[0046] According to the theoretical fragmentation efficiency I, the latest average pulse fragmentation efficiency η μ , pulse energy pe and pulse frequency pf, combined with the current laser lithotripsy parameters, a nonlinear correction mechanism based on the feedback deviation of the latest average pulse fragmentation efficiency η μ And the theoretical fragmentation efficiency I is created using the hyperbolic tangent function, and the new laser pulse parameter P opt After optimization is obtained;

[0047] Wherein, the new laser pulse parameter P opt The calculation expression is as follows:

[0048] ;

[0049] In the formula, pe opt Indicates the new pulse energy after optimization, and pf optrepresents the optimized new pulse frequency;

[0050] The optimized new laser pulse parameters P opt are sent to the laser emitter, and the cycle from step three is repeated until the end of the operation.

[0051] The multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization system comprises a data acquisition module, a multi-modal fusion module, an acoustic signal acquisition module, a pulse fragmentation efficiency evaluation module, a state evaluation module and a parameter optimization module.

[0052] The data acquisition module is linked with a hospital database through an API interface, retrieves patient medical data from the hospital database, processes the data through a deep learning model, extracts preoperative feature vectors and sends them to the multi-modal fusion module through an internal data bus.

[0053] The multi-modal fusion module receives the preoperative feature vectors, calculates the theoretical fragmentation efficiency by combining the fracture toughness in material mechanics and the influence of the fluid chemical environment, generates laser lithotripsy parameters, and sends the laser lithotripsy parameters to the laser lithotripsy device and the acoustic signal acquisition module.

[0054] The acoustic signal acquisition module starts lithotripsy by receiving the initial laser lithotripsy parameters, combines the microhydrophone integrated at the end of the endoscope, acquires the acoustic waves generated by laser lithotripsy and converts them into digital signals through a high-speed digital-to-analog converter, generates an original signal data set and transmits it in real time to the pulse fragmentation efficiency evaluation module through a data interface.

[0055] The pulse fragmentation efficiency evaluation module receives the original signal data set, calculates the real-time pulse fragmentation efficiency by combining the analysis ideas of impact dynamics and acoustic emission technology, and constructs a real-time pulse fragmentation efficiency sequence, which is continuously sent to the state evaluation module.

[0056] The state evaluation module continuously receives the real-time pulse fragmentation efficiency sequence, calculates the average fragmentation efficiency, compares it with the theoretical fragmentation efficiency received by the multi-modal fusion module, generates a state evaluation result and sends it to the parameter optimization module.

[0057] The parameter optimization module receives the state evaluation result, combines the current laser lithotripsy parameters with a nonlinear correction mechanism based on feedback deviation, executes a laser lithotripsy parameter optimization strategy, and sends the optimized laser lithotripsy parameters to the laser lithotripsy device through a data interface to update the device operating parameters.

[0058] The present application provides a multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization method and system, which has the following beneficial effects:

[0059] (1) Firstly, in the preoperative stage, the multi-source heterogeneous data of CT image data DCT and urine routine test data DUR are deeply fused by a deep learning model, a personalized digital portrait of a specific stone is constructed, which contains its physical hardness, internal heterogeneity and chemical environment stability, so that the traditional empirical setting can be abandoned, and a set of theoretically optimal, clearly quantified initial treatment strategies can be generated. Further, in the operation execution process, the invention innovatively introduces real-time acoustic signal feedback based on in-vivo micro hydrophone, through the instant analysis of the signal in the dimensions of impact dynamics and acoustic emission, the real efficiency of single pulse lithotripsy is objectively and instantaneously quantified. Finally, the invention combines preoperative accurate prediction with intraoperative real-time feedback to construct a dynamic and adaptive parameter regulation system, ensuring that the energy output throughout the operation is always matched with the real-time feedback state of the stone, thereby fundamentally improving the efficiency, safety and consistency of the treatment effect of the lithotripsy operation.

[0060] (2) Firstly, the invention realizes deep quantitative evaluation of stone characteristics in the operation planning stage. Through deep mining of the CT image data DCT and urine routine test data DUR of the patient by a deep learning model, not only the stone volume Vs and the average Hounsfield unit Hu are obtained, but also the stone texture heterogeneity parameter τ and the pH environment of the urine are innovatively quantified. Based on this comprehensive preoperative feature vector F, the system can generate a highly personalized theoretical fragmentation energy efficiency I according to the principles of material mechanics and chemistry, and match a set of optimal initial laser lithotripsy parameters P accordingly. This abandons the traditional one-size-fits-all initial setting that relies on experience, making the treatment more targeted and efficient from the first laser pulse. On this basis, the real-time and non-destructive acquisition of the acoustic waves of the stone by the micro hydrophone generates the original signal data set DSA, providing objective physical feedback data for subsequent dynamic control and constituting the basis for decision-making from static planning to dynamic control.

[0061] (3) Further, in the operation execution stage, the invention intelligently processes the collected original signal data set DSA. Through an algorithm combining impact dynamics and acoustic emission analysis, the complex acoustic signal is converted into a directly quantified pulse fragmentation efficiency η t at time t. This enables the system to accurately perceive the real effect of each laser impact and construct the pulse fragmentation efficiency sequence SSEQ, which is the core prerequisite for realizing the optimal control of laser parameters. The system can continuously compare the recently averaged pulse fragmentation efficiency η μ calculated according to the sequence with the theoretically set fragmentation energy efficiency I, and once a deviation is found according to the state evaluation result, the adaptive correction mechanism based on control theory is started, automatically and smoothly optimizing and generating new laser pulse parameters P optThis real-time regulation capability ensures that the operation is always in the optimal efficiency range throughout the entire operation, effectively shortening the operation time, improving the stone clearance rate, and avoiding excessive damage to the surrounding tissues due to precise energy delivery, thereby significantly improving the overall safety and consistency of the operation. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization method step schematic diagram;

[0063] Figure 2 A multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization system block diagram schematic diagram;

[0064] Figure 3 A data trend chart of the optimized new laser pulse parameter P opt . DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0066] Embodiment 1

[0067] The present application provides a multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization method, please refer to Figure 1 , comprising the following steps:

[0068] S1, call patient medical data from the hospital database through the api interface, process through the deep learning model, and extract the preoperative feature vector;

[0069] S2, according to the preoperative feature vector, combining the fracture toughness in material mechanics and the influence of fluid chemical environment, the theoretical fragmentation energy efficiency is calculated and the laser lithotripsy parameter is generated;

[0070] S3, according to the initial laser lithotripsy parameter to start lithotripsy, through the micro hydrophone integrated in the endoscope, the sound wave generated by laser lithotripsy is collected and converted into digital signal, and the original signal data set is generated;

[0071] S4, according to the original signal data set, combining the analysis idea of impact dynamics and acoustic emission technology, the real-time pulse fragmentation efficiency is calculated, and the real-time pulse fragmentation efficiency sequence is constructed;

[0072] S5, according to the real-time pulse fragmentation efficiency sequence, the average fragmentation energy efficiency is calculated, and compared with the theoretical fragmentation energy efficiency, and the state evaluation result is generated;

[0073] S6, according to the state evaluation result, combining the current laser lithotripsy parameters and the nonlinear correction mechanism combined with the feedback deviation, the laser lithotripsy parameter optimization strategy is executed.

[0074] In this embodiment, in the preoperative planning stage, by extracting the preoperative feature vector F containing the urine pH value, the stone volume Vs, the average Hounsfield unit Hu and the stone texture heterogeneity parameter τ, a comprehensive digital image of the stone is constructed, and through the deep fusion of multi-modal data, the problem that the parameter setting relies on subjective experience and cannot fully utilize multi-dimensional pathological information is solved. Subsequently, the theoretical fragmentation energy efficiency I is calculated based on the preoperative feature vector F, and a set of scientific and quantitative initial laser lithotripsy parameters P is generated, so as to ensure that the treatment has high precision from the initial stage. In the intraoperative execution stage, the laser lithotripsy is started according to the initial laser lithotripsy parameters P, and the acoustic signal is collected through the microhydrophone to generate the original signal data set DSA, and then the pulse fragmentation efficiency η t at time t is calculated, which introduces objective physical feedback and overcomes the limitation that the traditional visual observation is easily blocked by dust clouds. Finally, by continuously comparing the recent average pulse fragmentation efficiency η μ calculated according to the pulse fragmentation efficiency sequence SSEQ with the theoretical fragmentation energy efficiency I, the state evaluation result is generated, and the optimization strategy is executed according to the state evaluation result to output the new laser pulse parameters P opt This complete adaptive regulation process solves the defects of static open-loop control in the background technology, and can ensure that the energy output parameters can be optimized and adjusted according to the stone state throughout the operation, thereby effectively shortening the operation time, avoiding excessive damage to the surrounding tissues, and improving the safety and consistency of the treatment.

[0075] Embodiment 2

[0076] This embodiment is an explanation and description in embodiment 1, please refer to Figure 1 , in particular: S1 includes S11;

[0077] S11, the patient's medical data including the urine routine test data DUR and the CT image data DCT are called from the hospital database through the api interface;

[0078] According to the urine routine test data DUR, the urine pH value is extracted;

[0079] According to the CT image data DCT, using a 3D U-Net segmentation model, a three-dimensional segmentation mask Ms of the CT image data DCT is obtained, wherein the value of the corresponding position of the voxel judged as a calculus is set to 1, and the value of the corresponding position of the voxel judged as not a calculus is set to 0, according to the total number of voxels with a value of 1 in the three-dimensional segmentation mask Ms and the physical size of a single voxel, the calculus volume Vs is obtained, the three-dimensional segmentation mask Ms is used as a filter to act on the CT image data DCT, the original Hounsfield unit of the calculus region is obtained, and the arithmetic mean of the original Hounsfield unit is calculated to obtain the average Hounsfield unit Hu;

[0080] According to the three-dimensional segmentation mask Ms, the three-dimensional image data block VOL of the calculus is extracted from the CT image data DCT and input into the value 3D-CNN feature extraction model, the activation value of the last full connection layer is extracted, the variance of all elements in the activation value is calculated, the activation value is reduced to a single heterogeneity index, and the calculus texture heterogeneity parameter is obtained.

[0081] According to the urine pH, calculus volume Vs, average Hounsfield unit Hu and calculus texture heterogeneity parameter , a preoperative feature vector F = [pH, Vs, Hu, ] is generated.

[0082] Among them, the 3D U-Net segmentation model adopts a classic encoder-decoder symmetric structure and contains a skip connection to fuse feature information at different levels, the encoder part is composed of 4 down-sampling modules in series, each module contains two 3x3x3 convolution layers and a 2x2x2 maximum pooling layer, the decoder part is composed of corresponding 4 up-sampling modules, up-sampling is performed through deconvolution, and it is realized based on the PyTorch framework, using a CT image data DCT of 500 anonymous patients and its corresponding three-dimensional segmentation mask Ms manually drawn by professional physicians as a training set, an Adam optimizer is used in the training process, the initial learning rate is set to 10 -4 , the batch size is 4, and the Dice loss function is used as the target for optimization, and a total of 100 cycles are iteratively trained.

[0083] The 3D-CNN feature extraction model adopts a structure including a convolution base and a full connection layer, and is used to learn deep texture features inside the calculus. The convolution base can be stacked by three 3D convolution blocks, each of which includes a 3x3x3 convolution layer, a ReLU activation function and a 2x2x2 maximum pooling layer. Two full connection layers are connected after the convolution base, and the number of neurons is 256 and 128 respectively. For training the model, we use the idea of transfer learning. First, we pre-train the model on an auxiliary task: using a data set composed of 500 examples of three-dimensional image data blocks VOL and their corresponding component labels confirmed by postoperative component analysis to train the 3D-CNN model for calculus component classification with a cross-entropy loss function as the target. After pre-training, the last classification layer is removed, and the activation values of the last full connection layer containing 128 neurons are used as a feature vector that can represent the advanced physical properties inside the calculus. In practical applications, the variance of this feature vector is calculated as the calculus texture heterogeneity parameter ;

[0084] S2 includes S21 and S22;

[0085] S21, according to the preoperative feature vector F, combines the fracture toughness in material mechanics and the influence of chemical environment to simulate the nonlinear characteristics of the brittleness inside the calculus using the ln function for the calculus texture heterogeneity parameter τ and the calculus volume Vs, uses the ln function to reflect the nonlinear relationship of the influence of calculus hardness for the average Hounsfield unit Hu, and introduces the chemical environment factor brought by the urine pH value. When the urine pH value is 7, the calculus is the most difficult to break, and when the urine pH value deviates from 7, the calculus is easy to break. Through the above calculation process, the original theoretical fragmentation energy efficiency I r is obtained. r The tanh function is used for normalization to obtain the theoretical fragmentation energy efficiency I.

[0086] The calculation expression of the theoretical fragmentation energy efficiency I is as follows:

[0087] ; ;

[0088] In the formula, V0 represents a preset standard calculus volume, which is obtained by calculating the average value of the calculus volumes of all patients in the hospital database, Hu0 represents a preset standard calculus Hounsfield unit, which is obtained by calculating the average value of the average Hounsfield unit Hu of all patients in the hospital database, the ln function represents a logarithmic function operation, and the exp function represents a natural exponential function operation.

[0089] S22, according to the theoretical fragmentation efficiency I, data matching is performed with a preset strategy database to obtain initial laser lithotripsy parameters P, the initial laser lithotripsy parameters P including pulse energy pe and pulse frequency pf, wherein the preset strategy database sets the logic as follows:

[0090] If the theoretical fragmentation efficiency I is less than a preset first-level theoretical fragmentation efficiency threshold I1, it indicates that the stone is hard and stable, a high-energy and low-frequency laser lithotripsy parameter set is adopted;

[0091] If the theoretical fragmentation efficiency I is greater than a preset second-level theoretical fragmentation efficiency threshold I2, it indicates that the stone is fragile and unstable, a low-energy and high-frequency laser lithotripsy parameter set is adopted;

[0092] If the preset first-level theoretical fragmentation efficiency threshold I1 is less than or equal to the theoretical fragmentation efficiency I and the theoretical fragmentation efficiency I is less than or equal to the preset second-level theoretical fragmentation efficiency threshold I2, it indicates that the stone has both hard and stable characteristics, a balanced strategy is adopted, and a laser lithotripsy parameter set with balanced energy and frequency is adopted;

[0093] The first-level theoretical fragmentation efficiency threshold I1 and the second-level theoretical fragmentation efficiency threshold I2 are determined by analyzing the receiver operating characteristic curve according to the stone characteristics labeled by doctors in the hospital database as a training data set to obtain the first-level theoretical fragmentation efficiency threshold I1 and the second-level theoretical fragmentation efficiency threshold I2.

[0094] In this embodiment, the patient's urine routine test data DUR and CT image data DCT are obtained through the api interface. In order to solve the problem of insufficient information, the method uses a 3D U-Net segmentation model to process the CT image data DCT to accurately calculate the stone volume Vs and the average Hounsfield unit Hu, and uses a 3D-CNN feature extraction model to quantify the stone texture heterogeneity parameter τ, and then combines the urine pH value extracted from the urine routine test data DUR to form a comprehensive preoperative feature vector F. This process converts discrete, multi-modal raw data into standardized digital images, which is the basis for personalized analysis. Subsequently, the preoperative feature vector F is substituted into the preset formula to calculate the theoretical fragmentation efficiency I, and further according to the comparison result of the efficiency value and the first-level theoretical fragmentation efficiency threshold I1 and the second-level theoretical fragmentation efficiency threshold I2, a set of optimal initial laser lithotripsy parameters P is matched and output from the strategy database. The special advantage of this embodiment is that it completely shows a complete process of converting the subjective experience of doctors into objective and repeatable scientific decisions. By deeply mining and quantifying the key pathological features that have been ignored in the past, it has laid a unprecedented and highly personalized solid starting point for treatment before the operation starts, thereby fundamentally solving the pain points of strong blindness and poor consistency in the setting of initial parameters in the prior art.

[0095] Embodiment 3

[0096] This embodiment is an explanation and illustration in Embodiment 2, please refer to Figure 1 , specifically: S3 includes S31;

[0097] S31, according to the initial lithotripsy parameters P to start the operation, the laser pulse impact on the stone, the sound wave generated in the surrounding physiological saline medium, to the integrated in the end of the endoscope micro hydrophone, produce a with the sound wave pressure change rule consistent, continuous change in time t of the analog electrical signal S t And carry on the digital signal conversion, conversion operation as follows:

[0098] First, the analog signal S t at time t is signal conditioned, through the low noise amplifier to amplify it, further, the amplified analog signal S t at time t is passed through a band-pass filter, filter out the low frequency noise caused by the perfusion flow and the high frequency electronic noise generated by the device circuit, and finally, signal digitization, the denoised analog signal S t at time t is sent to a high-speed analog-to-digital converter, which takes a snapshot of the continuous analog signal and converts the voltage value at each time point to a discrete binary number to obtain a structured digital matrix and generate an original signal data set DSA;

[0099] S4 includes S41 and S42;

[0100] S41, according to the original signal data set DSA, mathematical transformation and analysis;

[0101] The original signal data set DSA is analyzed in the time domain, the maximum absolute value of the signal amplitude is searched, which is marked as the peak sound pressure sp t at time t, the time point where the peak sound pressure sp t at time t is located and is traced back from this time point, the time points where the signal first reaches 10% and 90% of the peak amplitude are found respectively and the difference is calculated to obtain the pulse rise time tr sp ;

[0102] The original signal data set DSA is analyzed in the frequency domain, a fast Fourier transform is performed, and in the transformed frequency spectrum, according to the preset frequency spectrum analysis threshold Z, the frequency spectrum is divided into low frequency and high frequency regions, and by integrating the frequency spectrum values, the high frequency region total energy E Ht at time t and the low frequency region total energy E LtWherein, the preset spectrum analysis threshold Z is found by spectrum analysis of known acoustic signal samples confirming the occurrence of effective brittle fracture, which can maximize the frequency point to distinguish high-frequency fracture signals from background noise;

[0103] According to the peak sound pressure sp at time t t , the pulse rise time tr sp , the total energy E of the high-frequency region at time t Ht , and the total energy E of the low-frequency region at time t Lt , the intermediate feature vector G is generated.

[0104] S42, according to the intermediate feature vector G, combined with the analysis idea of impact dynamics and acoustic emission technology, the pulse fragmentation efficiency η at time t is calculated t According to the calculated pulse fragmentation efficiency η at time t t , it is added to a time sequence to form a pulse fragmentation efficiency sequence SSEQ=[η1, η2, …, η t ];

[0105] Wherein, the pulse fragmentation efficiency η at time t t The calculation expression is as follows:

[0106] ;

[0107] In the formula, tanh represents the hyperbolic tangent function operation, K W represents the bulk modulus of water, and the specific value is 2.2 GPa, T represents the preset laser pulse width, which is calibrated by the device itself, sp th represents the preset invalid sound pressure threshold, the preset invalid sound pressure threshold sp th is determined by statistical analysis of the signal amplitude collected by the hydrophone in the perfusion state without stone operation.

[0108] In this embodiment, the method of converting intraoperative physical signals into quantifiable feedback indicators completely presents a complete methodology of converting invisible complex physical processes into quantifiable and analyzable digital indicators. According to the initial laser fragmentation parameters P, the operation is started, and the impact sound wave is captured by the micro hydrophone. After signal conditioning and digitization, a high-quality original signal data set DSA is generated. This process establishes an objective physical signal acquisition channel and solves the pain point of visual observation being easily disturbed. Subsequently, in order to realize the precise quantification of single impact, the original signal data set DSA is deeply processed to calculate the pulse fragmentation efficiency η at time t t, which is calculated according to the principle of combining impact dynamics and acoustic emission: first, the intermediate feature vector G is extracted from the original signal dataset DSA through time domain and frequency domain analysis, and then the performance is evaluated based on the vector, and when evaluating the quality of the impact cause, the peak sound pressure sp t The volume modulus K of water W The relative intensity of the impact is quantified by comparing the ratio of the peak sound pressure sp sp The relative sharpness of the impact is quantified by comparing the ratio of the peak sound pressure sp Ht The ratio of the total energy E t The impact force performance of the above laser impact and the energy distribution of the laser impact are evaluated, and are added to the pulse fragmentation performance sequence SSEQ. This method creates a new, objective quantitative standard for the surgical process, and the pulse fragmentation performance sequence SSEQ generated is not only a real-time, accurate quantification of the treatment effect, but also an indispensable data cornerstone for the implementation of all subsequent control strategies, solving the problem of lack of reliable real-time feedback signal source in the prior art.

[0109] Embodiment 4

[0110] This embodiment is an explanation and description in embodiment 3, please refer to Figure 1 and Figure 3 Specifically, S5 includes S51;

[0111] S51, according to the pulse fragmentation performance sequence SSEQ, taking the pulse fragmentation performance η t at time t as the starting point, backtracking the last N pulse fragmentation performances η and performing sliding average to obtain the last average pulse fragmentation performance η μ , wherein the specific setting method of N is as follows: the relevant staff sets the feedback time window △t according to the expected feedback time window △t, multiplies the feedback time window △t by the pulse frequency pf to obtain the number of pulse fragmentation performances η that need to be backtracked;

[0112] Compare the last average pulse fragmentation performance η μ with the theoretical fragmentation energy efficiency I, and generate a state evaluation result according to the comparison result;

[0113] If the last average pulse fragmentation performance η μ is less than 95% of the theoretical fragmentation energy efficiency I, it means that the actual efficiency is lower than expected, and the parameters need to be enhanced to increase the efficiency, and the output state label is -1;

[0114] If 95% x theoretical fragmentation energy efficiency I ≤ recent average pulse fragmentation efficiency η μ ≤ 105% x theoretical fragmentation energy efficiency I, indicating that the current parameters are in an ideal state, no adjustment is needed, and the output state label is 0;

[0115] If recent average pulse fragmentation efficiency η μ > 105% x theoretical fragmentation energy efficiency I, indicating that the actual efficiency is higher than expected, low parameters are needed to protect the tissue, and the output state label is 1;

[0116] S6 includes S61;

[0117] S61, according to the state evaluation result, execute the laser lithotripsy parameter optimization strategy, if the state label is 0, no parameter optimization is needed, if the state label is 1 or -1, parameter optimization is needed, the parameter optimization method is as follows:

[0118] According to the theoretical fragmentation energy efficiency I, the recent average pulse fragmentation efficiency η μ , the pulse energy pe and the pulse frequency pf, combined with the current laser lithotripsy parameters, a nonlinear correction mechanism based on the recent average pulse fragmentation efficiency η μ and the feedback deviation of the theoretical fragmentation energy efficiency I is created using the hyperbolic tangent function, to obtain the new laser pulse parameter P opt ;

[0119] Wherein, the new laser pulse parameter P opt The calculation expression is as follows:

[0120] ;

[0121] In the formula, pe opt represents the new pulse energy, pf opt represents the new pulse frequency;

[0122] The new laser pulse parameter P opt is sent to the laser emitter, and the cycle from step three is started until the operation is completed;

[0123] The new laser pulse parameter P opt The calculation example is as follows:

[0124] Urine pH: 5.8;

[0125] Stone volume Vs: 210mm 3 , average Hounsfield unit Hu: 1400, stone texture heterogeneity parameter τ: 0.8;

[0126] Preoperative feature vector F = [5.8, 210, 1400, 0.8];

[0127] Primary theoretical fragmentation energy efficiency threshold I1: 0.4, secondary theoretical fragmentation energy efficiency threshold I2: 0.8;

[0128] Theoretical fragmentation energy efficiency I calculation example is as follows:

[0129] ;

[0130] ;

[0131] Because the primary theoretical fragmentation energy efficiency threshold I1 < the theoretical fragmentation energy efficiency I < the secondary theoretical fragmentation energy efficiency threshold I2, the balance strategy is adopted, and the initial laser fragmentation parameter P retrieved is: pulse energy pe: 1.2 J, pulse frequency pf: 15 Hz;

[0132] The pulse signal in the original signal data set DSA is analyzed, and the following parameters are extracted:

[0133] Peak sound pressure sp at time t t : 380 MPa, pulse rise time tr sp : 0.09 μs;

[0134] Total energy EH in high frequency region at time t t : 5.5 J, total energy EL in low frequency region at time t t : 4.5 J;

[0135] Device constant: preset invalid sound pressure threshold sp th : 20 MPa;

[0136] Bulk modulus K of water W : 2200 MPa; preset laser pulse width T: 300 μs;

[0137] Feedback time window △t set by the operator: 1.5 s;

[0138] Pulse fragmentation efficiency η at time t t The calculation example is as follows:

[0139] ;

[0140] The number of backtracking N is calculated: 22;

[0141] The latest 22 pulse fragmentation efficiencies η in the pulse fragmentation efficiency sequence SSEQ are slidingly averaged to obtain the latest average pulse fragmentation efficiency η μ : 0.452;

[0142] The latest average pulse fragmentation efficiency η μ < 95% × the theoretical fragmentation energy efficiency I, output the label output state label as -1, and execute the parameter optimization strategy;

[0143] optimized new laser pulse parameters P opt The calculation example is as follows:

[0144] ;

[0145] The optimized new laser pulse parameters P opt are sent to the laser emitter, and the next cycle starts from step S3, and the parameters are continuously fine-tuned.

[0146] In this embodiment, first, the pulse fragmentation efficiency sequence SSEQ is processed to generate a state evaluation result: taking the latest pulse fragmentation efficiency η t at time t in the sequence as the starting point, the latest N efficiency values are traced back to calculate the latest average pulse fragmentation efficiency η μ , which can reflect the recent stone stability trend; the latest average pulse fragmentation efficiency η μ is then compared with the theoretical fragmentation efficiency I, and the state evaluation result is generated according to the preset threshold. After obtaining the state evaluation result, the regulation and control is immediately executed: if the state label is -1 or 1, the optimization strategy will be started, and the hyperbolic tangent function tanh is used to convert the deviation between the theoretical fragmentation efficiency I and the latest average pulse fragmentation efficiency η μ into a bounded and smooth correction signal, and finally the optimized new laser pulse parameters P opt are calculated and output to the laser emitter. This method completely shows a set of decision and execution logic for converting continuous perception data into accurate control instructions. By introducing a clinically set feedback time window △t to dynamically calculate the moving average window, this method processes the complex pulse fragmentation efficiency sequence SSEQ into a stable and clinically meaningful state evaluation result. Then, through a nonlinear correction mechanism based on control theory, this method converts this state evaluation result into a set of smooth and safe optimized new laser pulse parameters P opt . This complete autonomous regulation and control process from evaluation to decision to execution solves the pain points of manual adjustment lag, inconsistency and lack of scientific basis.

[0147] Embodiment 5

[0148] The multi-modal pathological feature fusion urinary stone fragmentation parameter optimization system, please refer to Figure 2 , specifically: including a data acquisition module, a multi-modal fusion module, an acoustic signal acquisition module, a pulse fragmentation efficiency evaluation module, a state evaluation module, and a parameter optimization module.

[0149] The data acquisition module is linked with the hospital database through an API interface, retrieves patient medical data from the hospital database, processes the data through a deep learning model, extracts preoperative feature vectors, and sends the feature vectors to the multi-modal fusion module through an internal data bus;

[0150] The multi-modal fusion module receives the preoperative feature vectors, combines the fracture toughness in material mechanics and the influence of fluid chemical environment, calculates the theoretical fragmentation energy efficiency, generates laser lithotripsy parameters, and sends the laser lithotripsy parameters to the laser lithotripsy device and the acoustic signal acquisition module;

[0151] The acoustic signal acquisition module starts the lithotripsy by receiving the initial laser lithotripsy parameters, combines the micro hydrophone integrated at the end of the endoscope, acquires the acoustic waves generated by the laser lithotripsy, converts the acoustic waves into digital signals through a high-speed digital-to-analog converter, generates an original signal data set, and transmits the data set to the pulse fragmentation efficiency evaluation module in real time through a data interface;

[0152] The pulse fragmentation efficiency evaluation module receives the original signal data set, combines the analysis ideas of impact dynamics and acoustic emission technology, calculates the real-time pulse fragmentation efficiency, constructs a real-time pulse fragmentation efficiency sequence, and continuously sends the sequence to the state evaluation module;

[0153] The state evaluation module continuously receives the real-time pulse fragmentation efficiency sequence, calculates the average fragmentation energy efficiency, compares the average fragmentation energy efficiency with the theoretical fragmentation energy efficiency received by the multi-modal fusion module, generates a state evaluation result, and sends the result to the parameter optimization module;

[0154] The parameter optimization module receives the state evaluation result, combines the current laser lithotripsy parameters with a nonlinear correction mechanism based on feedback deviation, executes a laser lithotripsy parameter optimization strategy, and sends the optimized laser lithotripsy parameters to the laser lithotripsy device through a data interface to update the device operating parameters.

[0155] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multimodal pathological feature fused urolithotripsy parameter optimization system, characterized by: The data acquisition module, the multi-modal fusion module, the acoustic signal acquisition module, the pulse fragmentation efficiency evaluation module, the state evaluation module, and the parameter optimization module are included. The data acquisition module is linked with a hospital database through an API interface, and patient medical data is called from the hospital database, wherein the patient medical data includes urine routine test data DUR and CT image data DCT; the patient medical data is processed by a deep learning model to extract a preoperative feature vector and sent to the multi-modal fusion module through an internal data bus; The multimodal fusion module receives the preoperative feature vector, combines the fracture toughness in material mechanics and the influence of fluid chemical environment, calculates the theoretical fragmentation energy efficiency, generates the laser lithotripsy parameters, and sends the laser lithotripsy parameters to the laser lithotripsy device; specifically including S21 and S22: S21, according to the preoperative feature vector F, combining the fracture toughness in material mechanics and the influence of chemical environment, using ln function to simulate the nonlinear characteristics of the internal fragility of the stone for the stone texture heterogeneity parameter τ and the stone volume Vs, and introducing the chemical environment factor brought by the urine pH value, under the standard of urine pH value 7, the stone is the hardest to break, and when the urine pH value deviates from 7, it is easy to break, through calculation to obtain the original theoretical fragmentation energy efficiency I r , the original theoretical fragmentation energy efficiency I r is normalized by using the tanh function to obtain the theoretical fragmentation energy efficiency I; The theoretical fragmentation efficiency I is calculated according to the following formula: ； ； In the formula, V0 represents a preset standard stone volume, which is obtained by taking the average stone volume in the hospital database, ln represents a logarithmic function operation, exp represents a natural exponential function operation, Hu0 represents a preset standard stone Hounsfield unit, and Hu represents an average Hounsfield unit; a three-dimensional image data block VOL of the stone is extracted from the CT image data DCT according to the three-dimensional segmentation mask Ms and input into a 3D-CNN feature extraction model, the activation value of the last full connection layer is extracted, the variance of all elements in the activation value is calculated, and a stone texture heterogeneity parameter τ is obtained; The laser lithotripsy device starts the lithotripsy by receiving initial laser lithotripsy parameters, the acoustic signal acquisition module collects acoustic waves generated by laser lithotripsy through a micro hydrophone integrated at the end of an endoscope and converts the acoustic waves into digital signals through a high-speed digital-to-analog converter, generates an original signal data set, and transmits the original signal data set to the pulse fragmentation efficiency evaluation module in real time through a data interface; The pulse fragmentation efficiency evaluation module receives the original signal data set, combines the analysis ideas of impact dynamics and acoustic emission technology, calculates the real-time pulse fragmentation efficiency, constructs a real-time pulse fragmentation efficiency sequence, and continuously sends the real-time pulse fragmentation efficiency sequence to the state evaluation module; specifically including S41 and S42: S41, performing mathematical transformation and analysis according to the original signal data set DSA; Perform time-domain analysis on the original signal dataset DSA to search for the maximum absolute value of the signal amplitude, labeled as the peak sound pressure level sp at time t. t The peak sound pressure sp at time t t Starting from the given time point and tracing back from that point, find the times when the signal first reaches 10% and 90% of its peak amplitude, respectively, and calculate the difference to obtain the pulse rise time tr. sp ; The original signal data set DSA is subjected to frequency domain analysis, a fast Fourier transform is performed, and in the transformed frequency spectrum, the frequency spectrum is divided into a low frequency region and a high frequency region according to a preset frequency spectrum analysis threshold Z, and by integrating the frequency spectrum values, the total energy E Ht of the high frequency region at time t is calculated Lt ; • peak sound pressure sp at time t t • pulse rise time tr sp • total energy E in high frequency region at time t Ht • total energy E in low frequency region at time t Lt • generate intermediate feature vector G S42、According to the intermediate eigenvector G, combining the analysis idea of impact dynamics and acoustic emission technology, the pulse fragmentation efficiency η at time t is calculated t According to the calculated pulse fragmentation efficiency η at time t t , added to a time sequence to form a pulse fragmentation efficiency sequence SSEQ=[η1,η2,……,η t ] wherein the pulse fragmentation efficiency η at time t t The calculation expression is as follows: ； In the formula, tanh represents hyperbolic tangent function operation, K W represents the volume modulus of water, and the specific value is 2.2 GPa, T represents a preset laser pulse width, which is calibrated by the device itself, sp th represents a preset invalid sound pressure threshold value; The state evaluation module continuously receives the real-time pulse fragmentation efficiency sequence, calculates the average fragmentation efficiency, compares it with the theoretical fragmentation efficiency calculated by the multi-modal fusion module, generates a state evaluation result, and sends the state evaluation result to the parameter optimization module; The parameter optimization module receives the state evaluation result, combines the current laser lithotripsy parameters with a nonlinear correction mechanism based on feedback deviation, executes a laser lithotripsy parameter optimization strategy, and sends the optimized laser lithotripsy parameters to the laser lithotripsy device to update the device working parameters through a data interface.

2. The multimodal pathological feature fused uro-lithotripsy parameter optimization system of claim 1, wherein: The data acquisition module includes S11; S11, according to the urine routine test data DUR, extracts the urine pH value, uses a 3D U-Net segmentation model to obtain a three-dimensional segmentation mask Ms of the CT image data DCT according to the CT image data DCT, obtains the stone volume Vs according to the three-dimensional segmentation mask Ms, uses the three-dimensional segmentation mask Ms as a filter to process the CT image data DCT, obtains the original Hounsfield unit of the stone region, calculates the arithmetic mean of the original Hounsfield unit, and obtains the average Hounsfield unit Hu. According to the urine pH, stone volume Vs, average Hounsfield unit Hu and stone texture heterogeneity parameter τ, a preoperative feature vector F=[pH, Vs, Hu, τ] is generated.

3. The multimodal pathological feature fused uro-lithotripsy parameter optimization system of claim 2, wherein: S22, according to the theoretical fragmentation efficiency I, data matching is performed with the preset strategy database to obtain initial laser lithotripsy parameters P, the initial laser lithotripsy parameters P including pulse energy pe and pulse frequency pf, wherein the preset strategy database sets the logic as follows: If the theoretical fragmentation efficiency I is less than the preset first-level theoretical fragmentation efficiency threshold I1, it indicates that the stone is hard and stable, and a strong fragmentation strategy is adopted, and a set of high-energy and low-frequency laser lithotripsy parameters are retrieved; If the theoretical fragmentation efficiency I is greater than the preset second-level theoretical fragmentation efficiency threshold I2, it indicates that the stone is fragile and unstable, and a pulverization strategy is adopted, and a set of low-energy and high-frequency laser lithotripsy parameters are retrieved; If the preset first-level theoretical fragmentation efficiency threshold I1 is less than or equal to the theoretical fragmentation efficiency I and is less than or equal to the preset second-level theoretical fragmentation efficiency threshold I2, it indicates that the stone has both hard and stable characteristics, and a balanced strategy is adopted, and a set of balanced energy and frequency laser lithotripsy parameters are retrieved.

4. The multimodal pathological feature fused uro-lithotripsy parameter optimization system of claim 3, wherein: The laser lithotripsy device and the acoustic signal acquisition module perform S31; S31, according to the initial laser lithotripsy parameters P to start performing lithotripsy operation, when the laser pulse impact calculus, the sound wave produced in the surrounding physiological saline medium propagation, to the integrated in the end of the endoscope micro hydrophone, produce a consistent with the sound wave pressure change law, a continuous change in time t analog electrical signal S t And digital signal conversion, conversion operation as follows: First, the analog signal S t is conditioned, amplified by a low-noise amplifier, and further amplified S t is passed through a band-pass filter, which filters out low-frequency noise caused by the perfusion flow and high-frequency electronic noise generated by the device circuitry, which is not related to the stone event, and finally, the analog signal S t is digitized, and the denoised analog signal S is fed into a high-speed analog-to-digital converter, which takes a snapshot of the continuous analog signal, converts the voltage value at each time point into a discrete binary number, and obtains a structured digital matrix to generate the original signal dataset DSA.

5. The multimodal pathological feature fused uro-lithotripsy parameter optimization system of claim 4, wherein: The state evaluation module includes S51; S51, according to the pulse fragmentation efficiency sequence SSEQ, the pulse fragmentation efficiency η at time t t As a starting point, the last N pulse fragmentation efficiencies η are traced back and a moving average is taken to obtain the last average pulse fragmentation efficiency η μ Wherein, the specific setting method of N is as follows: the feedback time window Δt is set by the relevant staff according to their own expectations, the feedback time window Δt is multiplied by the pulse frequency pf to obtain the number of pulse fragmentation efficiencies η that need to be traced back. calculating a recent average pulse fragmentation efficiency η μ comparing the recent average pulse fragmentation efficiency η to a theoretical fragmentation efficiency I, and generating a status assessment based on the comparison. If the recent average pulse fragmentation efficiency η μ <95% x theoretical fragmentation efficiency I, indicating that the actual efficiency is lower than expected, the enhancement parameter needs to be increased to increase the efficiency, and the output state label is -1; If 95% x theoretical fragmentation energy efficiency I ≤ recent average pulse fragmentation efficiency η μ ≤ 105% x theoretical fragmentation energy efficiency I, it indicates that the current parameters are in an ideal state and no adjustment is needed, and the output state label is 0; If the recent average pulse fragmentation efficiency η μ > 105% x theoretical fragmentation efficiency I, indicates that the actual efficiency is higher than expected, and low parameters are needed to protect the tissue, the output status flag is 1.

6. The multi-modal pathological feature fusion urinary stone lithotripsy parameter optimization system of claim 5, wherein: The parameter optimization module includes S61; S61, according to the state evaluation result, a laser lithotripsy parameter optimization strategy is executed, if the state label is 0, no parameter optimization is needed, if the state label is 1 or -1, parameter optimization is needed, and the parameter optimization method is as follows: According to the theoretical fragmentation energy efficiency I, the recent average pulse fragmentation efficiency η μ , the pulse energy peand the pulse frequency pf, combined with the current laser fragmentation parameters, a nonlinear correction mechanism based on the recent average pulse fragmentation efficiency η μ deviation from the feedback of the theoretical fragmentation energy efficiency I, the optimized new laser pulse parameters P opt ; wherein the optimized new laser pulse parameters P opt The calculation expression is as follows: ； where pe opt denotes the optimized new pulse energy, pf opt denotes the optimized new pulse frequency; The optimized new laser pulse parameters P opt are sent to the laser emitter and the cycle starts again from step three until the end of the surgery.

Images